CN105184778A - Detection method and device - Google Patents

Abstract

The embodiment of the present invention discloses a detection method and device. The method includes: inputting a template image and a test image into a trained Siamese network, and performing forward calculations respectively to obtain the results of the template image and the test image. feature vector; calculating the similarity between the template image and the feature vector of the test image; determining whether the test image matches the template image according to the similarity. Since the Siamese network can extract the advanced features of the image, the embodiment of the present invention can accurately judge whether the template image matches the test image by inputting the image into the trained Siamese network, and the judgment accuracy is high.

Description

A detection method and device

technical field

The invention relates to the field of AOI, in particular to a detection method and device.

Background technique

Image matching refers to the technology of judging whether two images are similar by using image features. It has been used in various fields, such as component detection and face recognition, etc. by using image matching technology. The robustness is low, and it is easy to be affected by the environment such as lighting, which may cause matching errors, and the matching accuracy is low.

Take image matching in the field of component detection as an example: Printed circuit board (PCB board for short) refers to a circuit board that provides connections for various electronic components. As electronic equipment becomes more and more complex, the required parts are naturally more and more Many, the lines and connections on the PCB are becoming more and more dense, so it is inevitable that there will be missing parts when soldering/hand-inserting electronic components, so it is necessary to check whether there are missing parts on the PCB board after the PCB board is soldered.

At present, the detection of missing parts of electronic components on PCB boards is carried out by manual detection. This kind of scheme is time-consuming, high in cost and low in efficiency. Template matching missing parts detection method, but this scheme is susceptible to detection errors due to the influence of lighting and other environments, and some are implemented based on color histograms or some low-level features, but these schemes are due to unreliable color information of electronic components or The low-level features of electronic components are not obvious and lead to detection errors.

Contents of the invention

The embodiment of the present invention provides a detection method and device, in order to improve the accuracy of image similarity detection and to have high reliability.

The first aspect of the embodiments of the present invention provides a detection method, including:

The template image and the test image are input into the Siamese network after training, and the feature vectors of the template image and the test image are obtained by forward calculation respectively;

Calculating the similarity between the template image and the feature vector of the test image;

Determine whether the test image matches the template image according to the similarity.

The second aspect of the embodiment of the present invention provides a detection device, including:

Image input module, for template image and test image input Siamese network after training, and do forward calculation respectively to obtain the feature vector of described template image and described test image;

Calculation module, for calculating the similarity between the feature vector of the module image and the test image;

A determining module, configured to determine whether the test image matches the template image according to the similarity.

It can be seen that in the technical solution provided by the embodiments of the present invention, the template image and the test image are input into the trained Siamese network, and forward calculations are performed respectively to obtain the feature vectors of the template image and the test image; The similarity between the feature vector of the template image and the test image; determining whether the test image matches the template image according to the similarity. Since the Siamese network can extract the advanced features of the image, inputting the image into the trained Siamese network can accurately judge whether the template image matches the test image, and the judgment accuracy is high.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1-a is the network structural diagram of the Siamese network that the embodiment of the present invention provides;

Fig. 1-b is a schematic flow chart of a detection method provided by the first embodiment of the present invention;

Fig. 2 is a schematic flow chart of a detection method provided by the second embodiment of the present invention;

Fig. 3 is a schematic structural diagram of a detection device provided by a third embodiment of the present invention;

Fig. 4 is a schematic structural diagram of a detection device provided by a fourth embodiment of the present invention;

Fig. 5 is a schematic structural diagram of a detection device provided by a fifth embodiment of the present invention.

Detailed ways

The embodiment of the present invention provides a detection method and device in the embodiment of the present invention, in order to improve the accuracy of image similarity detection and to have high reliability.

In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only It is an embodiment of a part of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

The terms "first", "second" and "third" in the specification and claims of the present invention and the above drawings are used to distinguish different objects, rather than to describe a specific order. Furthermore, the term "comprise", as well as any variations thereof, is intended to cover a non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes unlisted steps or units, or optionally further includes For other steps or units inherent in these processes, methods, products or apparatuses.

A kind of detection method of the embodiment of the present invention, a kind of detection method comprises:

Input the template image and the test image into the Siamese network after training, and perform forward calculations respectively to obtain the feature vector of the template image and the test image; calculate the similarity between the template image and the feature vector of the test image ; Determine whether the test image matches the template image according to the similarity.

Referring first to Fig. 1, Fig. 1-a is a network structure diagram of a Siamese network provided by an embodiment of the present invention; Fig. 1-b is a schematic flowchart of a detection method provided by the first embodiment of the present invention. Among them, as shown in Figure 1-b, a detection method provided by the first embodiment of the present invention may include:

S101. Input the template image and the test image into the trained Siamese network, and perform forward calculation respectively to obtain feature vectors of the template image and the test image. Among them, see Figure 1-a, the Siamese network is a deep neural network. According to the characteristics of the Siamese network, the features of each level of the image can be calculated, including low-level features and high-level features, so that the image can be analyzed according to the high-level features of the image. will get more accurate results. In order to perform image matching, the Siamese network is designed to include two convolutional neural networks with the same structure and shared parameters. When using the Siamese network for image matching, the template image and the test image of the image are input respectively, so that two input images can be calculated separately. The eigenvectors, and then calculate the similarity of the two eigenvectors, so as to judge the similarity of the two images according to the value of the similarity, that is, to judge whether the two images match.

Among them, the trained Siamese network has the ability to accurately match the template image and the test image due to the training and learning of a large number of sample images.

Wherein, the template image is a standard picture for comparison, and the test image is compared with the template image to determine whether it matches the template image.

S102. Calculate the similarity between the feature vectors of the template image and the test image.

Wherein, the similarity is a parameter for judging the similarity between the template image and the test image.

Optionally, in some possible implementations of the present invention, after calculating the feature vectors of the module image and the test image, the Euclidean distance of the feature vectors can be calculated to obtain the similarity between the feature vectors of the template image and the test image.

S103. Determine whether the test image matches the template image according to the similarity.

Among them, since the similarity is a measure of the similarity of the features of the two images input by the Siamese network, it can be used to judge whether the template image is similar to the test image, that is, to judge whether the template image matches the test image.

It can be seen that in the solution of this embodiment, the template image and the test image are input into the Siamese network after training, and the feature vectors of the template image and the test image are obtained by forward calculation respectively; the template image is calculated A similarity with the feature vector of the test image; determining whether the test image matches the template image according to the similarity. Since the Siamese network can extract the advanced features of the image, inputting the image into the trained Siamese network can accurately judge whether the template image matches the test image, and the judgment accuracy is high.

Optionally, in some possible implementations of the present invention, the template image is the corresponding component image when the component exists and is in a standard position, the test image is the component image that needs to be tested, and the Determine whether the test image matches the template image, including:

Determine whether the component corresponding to the test image exists according to the similarity.

Wherein, the template image can be a PCB template image, so that the input component image or test image corresponding to it is also a PCB image, and the template image can also be an image containing each independent component intercepted from the PCB image, so as to correspond to it The input test image is also an image containing individual components.

Optionally, in some possible implementations of the present invention, since the missing part detection is performed on each component, generally the template image input in the Siamese network is an image containing an independent component, then the input in the training phase The component image is also the component image of the position intercepted from the PCB image, and the image input in the test stage is also the component image of the position intercepted from the PCB image in different situations.

Optionally, in some possible implementations of the present invention, when the template image is a template image of a component, it can be understood that the template image is a component image in which the component exists and the position of the component is in a correct position, and the test image is based on the template The image judges whether there is a missing part.

It can be understood that since the Siamese network can judge whether the test image matches the template image, when the template image and the test image are component images, it can be determined according to the similarity whether the test image of the component image matches the template image, that is, determine the test image Whether the corresponding element exists.

Optionally, in some possible implementation manners of the present invention, before inputting the template image and the test image into the trained Siamese network, the method further includes:

Obtaining the position of the template image and the element in the test image by template matching and aligning the template image and the test image;

The template image and the test image are normalized to trigger the execution of the step of inputting the template image and the test image into the trained Siamese network.

It can be understood that since the component image is an image intercepted from the template image, in order to make the detection result more accurate, it is necessary to locate the component in the intercepted component image at the center of the image, and at the same time normalize the size of the image to ensure the accuracy of subsequent processing. This process is called preprocessing.

Optionally, in some possible implementation manners of the present invention, the image may not be preprocessed.

Optionally, in some possible implementation manners of the present invention, before inputting the template image and the test image into the trained Siamese network, the method further includes:

Create a positive sample set and a negative sample set for the component image pair, wherein the positive sample set includes a component image pair formed by the template image and the component image when the component exists, and the negative sample set includes the The component image pair formed by the template image and the component image when the component does not exist, the negative sample set also includes the component image pair formed by the template image and component images when other components exist;

Using the positive sample set and the negative sample set to train the Siamese network, so as to trigger the execution of the step of inputting the template image and the test image into the trained Siamese network.

It should be noted that, optionally, in some possible implementations of the present invention, when the template image and the test image are PCB images, the positive sample set in the sample set includes the template image and the position of each component on the template image. The component image pair composed of the component image when the component is correct, the negative sample set includes the component image pair composed of the template image and the component image when there may be no component in each component position, the negative sample set also includes the template image and each component A component image pair composed of component images when all components exist at the position, but a wrong component may be inserted.

Optionally, in other possible implementations of the present invention, when both the template image and the test image are images containing individual components that are intercepted from the PCB image, the positive sample set in the sample set includes the template image The component image pair composed of the image containing the component in the template image, the negative sample set includes the component image pair composed of the template image and the image that does not contain the component, and the negative sample set also includes the template image and the component that is not in the template image , but contains component images of other wrong components.

Optionally, in some possible implementation manners of the present invention, the set of positive samples includes the template image, and samples in which components exist and are taken in various scenes, such as positive samples taken in bad light images, and positive sample images taken from different positions or angles, or positive sample images taken in other complex scenes, the Siamese network is trained using the positive sample images taken in each scene, so that the Siamese network has a stronger recognition Ability to correctly detect non-missing component images taken in different scenarios in the subsequent test stage, and improve the recognition rate.

It can be understood that when using the Siamese network to detect missing parts in images, it is first necessary to train and learn the Siamese network using samples to obtain the parameters of the Siamese network, and then use the trained Siamese network to accurately detect missing parts in component images. When training the Siamese network, it is necessary to use positive sample sets and negative sample sets in various scenarios for training, so that the Siamese network can fully learn the picture when the component is not a missing part, and the Siamese network can fully learn Only by seeing the pictures of missing components, can the missing parts detection be carried out correctly in the future.

Optionally, in some possible implementations of the present invention, the negative sample set includes the template image, and an image when the component does not exist, or when the component exists but the welding position is obviously wrong, or not Images etc. when the components are correct.

It can be understood that taking as many positive sample sets and negative sample sets as possible in different scenarios to train the Siamese network can make the learning effect of the Siamese network better, so that the accuracy of subsequent missing parts detection and recognition is high.

Optionally, in some possible implementation manners of the present invention, the positive sample set and the negative sample set of the component image include training samples and testing samples.

Wherein, the training sample refers to training the Siamese network, and the test sample refers to testing the effect of the trained Siamese network, and the two together constitute a sample in the training phase of the Siamese network.

Optionally, in some possible implementation manners of the present invention, the creating the component image alignment positive sample set and negative sample set includes:

Capture printed circuit board images;

Taking the printed circuit board template image as a reference, intercepting the component image on the printed circuit board image;

The image of the component is collected to align a positive sample set and a negative sample set.

Optionally, in some possible implementation manners of the present invention, the component image may be marked after the component image is captured on the printed circuit board icon.

It can be understood that since it is necessary to detect missing parts for each component, when training the Siamese network, it is necessary to intercept a sample set of images of each component on the printed circuit board image for training. Moreover, in order to distinguish each component image during training, it is necessary to mark each component image to distinguish different components before training.

Optionally, in some possible implementations of the present invention, a camera can be set up on the production line to collect images of different types of PCB boards in batches, and use board tracking technology to avoid repeatedly photographing a certain PCB board. In this way, each type of PCB board contains multiple image samples, and each image sample corresponds to a certain type of PCB board, so that the obtained component images on the PCB board also come from different boards. Ensure sample diversity.

Optionally, in some possible implementation manners of the present invention, the intercepting component images on the printed circuit board includes:

The component image is automatically captured using the position information of the component on the printed board image.

It can be understood that when the position information of the component is known, the component image can be automatically intercepted according to the position information.

Optionally, in some possible implementations of the present invention, the position information of the obtained component image may be obtained through the component position information recorded in the template file, or through the manually marked position information.

Optionally, in some possible implementation manners of the present invention, the marking the component image includes:

Label according to component category information.

It can be understood that it is necessary to distinguish the types of components, so that the types of components can be recorded during training to accurately detect missing parts of components.

Optionally, in some other possible implementation manners of the present invention, the components may also be marked in other ways that can distinguish the components.

Optionally, in some possible implementation manners of the present invention, before collecting the component image alignment positive sample set and negative sample set, the method further includes:

Obtaining the position of the component in the component image by template matching and aligning the component image;

The component image is normalized to trigger the execution of the step of acquiring the positive sample set and the negative sample set of the component image pair.

It can be understood that, similar to the process of testing the Siamese network, when the sample image of the Siamese network is collected, the image is aligned so that the component is located in the center of the image, and the image is normalized. This process is called image pre-processing. During the processing process, preprocessing the image will make the Siamese network training effect better.

Optionally, in some possible implementations of the present invention, if the component image is preprocessed when the Siamese network is trained, then when the trained Siamese network is used to detect missing parts, the component image also needs to be preprocessed. Preprocessing, if the component image is not preprocessed when the Siamese network is trained, then the component image is not preprocessed when the trained Siamese network is used for missing parts detection.

Optionally, in some possible implementation manners of the present invention, the determining whether the element corresponding to the test image exists according to the similarity includes:

judging whether the similarity is within a preset range;

If the similarity is within a preset range, it is determined that the element exists;

If the similarity is not within the preset range, it is determined that the element does not exist.

It can be understood that since the similarity is used to characterize the degree of similarity between the template image and the test image, when the similarity is within a preset range, it means that the template image is similar to the test image, then the component image exists, that is, the component does not Missing parts, otherwise, components missing parts.

Optionally, in some possible implementations of the present invention, when the similarity is greater than or equal to a preset value, the component exists, that is, there are no missing parts in the component; when the similarity is smaller than the preset value, then the element does not exist.

In order to facilitate a better understanding and implementation of the above-mentioned solutions of the embodiments of the present invention, some specific application scenarios are used for illustration below.

Please refer to Fig. 2, Fig. 2 is a schematic flow chart of a detection method provided by the second embodiment of the present invention, wherein, as shown in Fig. 2, a detection method provided by the second embodiment of the present invention may include:

S201. Collect images of printed circuit boards.

It can be understood that since it is necessary to use the Siamese network for missing parts detection, it is first necessary to collect sample images.

Among them, the Siamese network refers to a deep neural network. According to the characteristics of the Siamese network, the features of each level of the component can be calculated, including low-level features and high-level features, so that it will be more accurate when analyzing the image according to the high-level features of the image. the result of. In order to detect missing parts for components, the Siamese network is designed to include two convolutional neural networks with the same structure and shared parameters. When using the Siamese network for missing parts detection, the template image and component image of the image are input respectively, so that they can be calculated separately. The eigenvectors of the two input images, and then calculate the similarity of the two eigenvectors, so as to judge whether it is a type of component according to the value of the similarity, so as to determine whether it is a missing part. Firstly, it is necessary to train the Siamese network to obtain the parameters of the Siamese network. During the training process, the samples of images collected in various situations are used to train the missing and existing images in different situations. Then, according to the trained Siamese network, the components are detected for missing parts. The images input by the two convolutional neural networks are different in the training phase and the testing phase. In the training phase, one of them inputs the template image, and the other inputs Component images, during the test phase, one of them is input as a template image, and the other is input as a test image. Moreover, in the training phase, the feature vectors of the two images are obtained by performing forward calculations respectively, and then the Euclidean distance of the two feature vectors is calculated to obtain the similarity.

Wherein, the template picture is a standard picture for comparison. The template image can be a PCB template image, so that the input component image or test image corresponding to it is also a PCB image, and the template image can also be an image intercepted from the PCB image that contains each independent component, so that the corresponding input image An image is also an image comprising individual elements.

Optionally, in some possible implementations of the present invention, since the missing part detection is performed on each component, generally the template image input in the Siamese network is an image containing an independent component, then the input in the training phase The component image is also the component image of the position intercepted from the PCB image, and the image input in the test stage is also the component image of the position intercepted from the PCB image in different situations.

It can be understood that the PCB image needs to be collected first before the component image sample can be intercepted.

Optionally, in some possible implementations of the present invention, a camera can be set up on the production line to collect images of different types of PCB boards in batches, and use board tracking technology to avoid repeatedly photographing a certain PCB board. In this way, each type of PCB board contains multiple image samples, and each image sample corresponds to a certain type of PCB board, so that the obtained component images on the PCB board also come from different boards. Ensure sample diversity.

S202. Using the printed circuit board template image as a reference, intercept the component image on the printed circuit board image and mark the component image.

It can be understood that since it is necessary to detect missing parts for each component, when training the Siamese network, it is necessary to intercept a sample set of images of each component on the printed circuit board image for training. Moreover, in order to distinguish each component image during training, it is necessary to mark each component image to distinguish different components before training.

Optionally, in some possible implementation manners of the present invention, the intercepting component images on the printed circuit board includes:

The component image is automatically captured using the position information of the component on the printed board image.

It can be understood that when the position information of the component is known, the component image can be automatically intercepted according to the position information.

Optionally, in some possible implementations of the present invention, the position information of the obtained component image may be obtained through the component position information recorded in the template file, or through the manually marked position information.

Optionally, in some possible implementation manners of the present invention, the marking the component image includes:

Label according to component category information.

It can be understood that it is necessary to distinguish the types of components, so that the types of components can be recorded during training to accurately detect missing parts of components.

Optionally, in some other possible implementation manners of the present invention, the components may also be marked in other ways that can distinguish the components.

S203. Align the positive sample set and the negative sample set with the image of the acquisition component.

Wherein, the positive sample set includes the component image pair formed by the template image and the component image when the component exists, and the negative sample set includes the template image and the component when the component does not exist The component image pair formed by the image, the negative sample set also includes the component image pair formed by the template image and component images when other components exist.

It should be noted that, optionally, in some possible implementations of the present invention, when the template image and the test image are PCB images, the positive sample set in the sample set includes the template image and the position of each component on the template image. The component image pair composed of the component image when the component is correct, the negative sample set includes the component image pair composed of the template image and the component image when there may be no component in each component position, the negative sample set also includes the template image and each component A component image pair composed of component images when all components exist at the position, but a wrong component may be inserted.

Optionally, in other possible implementations of the present invention, when both the template image and the test image are images containing individual components that are intercepted from the PCB image, the positive sample set in the sample set includes the template image The component image pair composed of the image containing the component in the template image, the negative sample set includes the component image pair composed of the template image and the image that does not contain the component, and the negative sample set also includes the template image and the component that is not in the template image , but contains component images of other wrong components.

Optionally, in some possible implementation manners of the present invention, the set of positive samples includes the template image, and samples in which components exist and are taken in various scenes, such as positive samples taken in bad light images, and positive sample images taken from different positions or angles, or positive sample images taken in other complex scenes, the Siamese network is trained using the positive sample images taken in each scene, so that the Siamese network has a stronger recognition Ability to correctly detect non-missing component images taken in different scenarios in the subsequent test stage, and improve the recognition rate.

It can be understood that when using the Siamese network to detect missing parts in images, it is first necessary to train and learn the Siamese network using samples to obtain the parameters of the Siamese network, and then use the trained Siamese network to accurately detect missing parts in component images. When training the Siamese network, it is necessary to use positive sample sets and negative sample sets in various scenarios for training, so that the Siamese network can fully learn the picture when the component is not a missing part, and the Siamese network can fully learn Only by seeing the pictures of missing components can the subsequent detection of missing parts be carried out correctly.

Optionally, in some possible implementations of the present invention, the negative sample set includes the template image, and an image when the component does not exist, or when the component exists but the welding position is obviously wrong, or not Images etc. when the components are correct.

It can be understood that taking as many positive sample sets and negative sample sets as possible in different scenarios to train the Siamese network can make the learning effect of the Siamese network better, so that the accuracy of subsequent missing parts detection and recognition is high.

Optionally, in some possible implementation manners of the present invention, the positive sample set and the negative sample set of the component image include training samples and test samples.

Wherein, the training sample refers to training the Siamese network, and the test sample refers to testing the effect of the trained Siamese network, and the two together constitute a sample in the training phase of the Siamese network.

Optionally, in some possible implementation manners of the present invention, before collecting the component image alignment positive sample set and negative sample set, the method further includes:

Obtaining the position of the component in the component image by template matching and aligning the component image;

The component image is normalized.

It can be understood that since the component image is an image intercepted from the template image, in order to make the detection result more accurate, it is necessary to locate the component in the intercepted component image at the center of the image, and at the same time normalize the size of the image to ensure the accuracy of subsequent processing. This process is called preprocessing.

Optionally, in some possible implementation manners of the present invention, the image may not be preprocessed.

S204, using the positive sample set and the negative sample set to train the Siamese network.

It can be understood that the Siamese network needs to be trained with positive sample sets and negative sample sets, so that the Siamese network has the ability to detect missing parts.

S205 , input the template image and the test image into the trained Siamese network, and perform forward calculation respectively to obtain feature vectors of the template image and the test image.

S206. Calculate the similarity between the feature vectors of the module image and the test image.

Wherein, the template image is the corresponding component image when the component exists and is in the standard position, and the test image is the component image that needs to be tested

Wherein, the similarity is a measure of the similarity between the template image input by the Siamese network and the characteristics of the test image, and the similarity is to use the advanced features of the template image and the test image learned by the Siamese network to judge whether the template image and the test image are similar value of degree.

Optionally, in some possible implementation manners of the present invention, before inputting the template image and the test image into the trained Siamese network, the method further includes:

Obtaining the position of the template image and the element in the test image by template matching and aligning the template image and the test image;

The template image and the test image are normalized.

It can be understood that since the component image is an image intercepted from the template image, in order to make the detection result more accurate, it is necessary to locate the component in the intercepted component image at the center of the image, and at the same time normalize the size of the image to ensure the accuracy of subsequent processing. This process is called preprocessing.

Optionally, in some possible implementation manners of the present invention, the image may not be preprocessed.

Optionally, in some possible implementations of the present invention, if the component image is preprocessed when the Siamese network is trained, then when the trained Siamese network is used to detect missing parts, the component image also needs to be preprocessed. Preprocessing, if the component image is not preprocessed when the Siamese network is trained, then the component image is not preprocessed when the trained Siamese network is used for missing parts detection.

S207. Determine whether the component corresponding to the test image exists according to the similarity.

It can be understood that since the similarity is the value of the degree of similarity between the template image and the test image by using the advanced features of the template image and the test image learned by the Siamese network, the similarity can be used to determine whether the component is similar to the template image and the test image , that is, to determine whether the component corresponding to the test image exists, that is, to determine whether the component is missing.

Optionally, in some possible implementation manners of the present invention, the determining whether the element corresponding to the test image exists according to the similarity includes:

judging whether the similarity is within a preset range;

If the similarity is within a preset range, the element exists;

If the similarity is not within the preset range, the element does not exist.

It can be understood that since the similarity is used to characterize the degree of similarity between the template image and the test image, when the similarity is within a preset range, it means that the template image is similar to the test image, then the component image exists, that is, the component does not Missing parts, otherwise, components missing parts.

Optionally, in some possible implementations of the present invention, when the similarity is greater than or equal to a preset value, the component exists, that is, there are no missing parts in the component; when the similarity is smaller than the preset value, then the element does not exist.

It can be seen that in the solution of this embodiment, the template image and the test image are input into the trained Siamese network, and forward calculations are performed respectively to obtain the feature vectors of the template image and the test image, and finally Calculating the similarity of two feature vectors, wherein the template image is the corresponding component image when the component exists and is in a standard position, and the test image is the component image that needs to be tested; determine the test image according to the similarity Whether the corresponding element exists. Since the Siamese network can extract the advanced features of the image, inputting the image into the trained Siamese network can accurately determine whether the component image is missing, and the recognition accuracy is high.

The embodiment of the present invention also provides a component detection device, which includes:

The image input module is used to input the template image and the test image into the trained Siamese network, and perform forward calculations respectively to obtain the feature vectors of the template image and the test image, wherein the template image is that the element exists and The corresponding component image at the standard position, the test image is the component image that needs to be tested;

Calculation module, for calculating the similarity between the feature vector of the module image and the test image;

A determining module, configured to determine whether the element corresponding to the test image exists according to the similarity.

Specifically, please refer to FIG. 3 . FIG. 3 is a schematic structural diagram of a detection device provided by a third embodiment of the present invention, wherein, as shown in FIG. 3 , a detection device 300 provided by a third embodiment of the present invention may include :

An image input module 310 , a calculation module 320 and a determination module 330 .

Wherein, the image input module 310 is configured to input the template image and the test image into the trained Siamese network, and perform forward calculation to obtain feature vectors of the template image and the test image respectively.

Among them, see Figure 1-a, the Siamese network is a deep neural network. According to the characteristics of the Siamese network, the features of each level of the image can be calculated, including low-level features and high-level features, so that the image can be analyzed according to the high-level features of the image. will get more accurate results. In order to perform image matching, the Siamese network is designed to include two convolutional neural networks with the same structure and shared parameters. When using the Siamese network for image matching, the template image and the test image of the image are input respectively, so that two input images can be calculated separately. The eigenvectors, and then calculate the similarity of the two eigenvectors, so as to judge the similarity of the two images according to the value of the similarity, that is, to judge whether the two images match.

Among them, the trained Siamese network has the ability to accurately match the template image and the test image due to the training and learning of a large number of sample images.

Wherein, the template image is a standard picture for comparison, and the test image is compared with the template image to determine whether it matches the template image.

Calculation module 320, configured to calculate the similarity between the feature vectors of the template image and the test image.

Wherein, the similarity is a parameter for judging the similarity between the template image and the test image.

Optionally, in some possible implementations of the present invention, after calculating the feature vectors of the module image and the test image, the Euclidean distance of the feature vectors can be calculated to obtain the similarity between the feature vectors of the template image and the test image.

A determining module 330, configured to determine whether the test image matches the template image according to the similarity.

Among them, since the similarity is a measure of the similarity of the features of the two images input by the Siamese network, it can be used to judge whether the template image is similar to the test image, that is, to judge whether the template image matches the test image. It can be understood that the functions of each functional module of the detection device 300 in this embodiment can be specifically implemented according to the method in the above method embodiment, and the specific implementation process can refer to the relevant description of the above method embodiment, and will not be repeated here.

It can be seen that in the solution of this embodiment, the detection device 300 inputs the template image and the test image into the trained Siamese network, and performs forward calculations respectively to obtain the feature vectors of the template image and the test image; the detection device 300 calculates the similarity between the feature vectors of the template image and the test image; the detecting device 300 determines whether the test image matches the template image according to the similarity. Since the Siamese network can extract the advanced features of the image, inputting the image into the trained Siamese network can accurately judge whether the template image matches the test image, and the judgment accuracy is high.

Please refer to FIG. 4. FIG. 4 is a schematic structural diagram of a detection device provided by a fourth embodiment of the present invention, wherein, as shown in FIG. 4, a detection device 400 provided by the fourth embodiment of the present invention may include:

An image input module 410 , a calculation module 420 and a determination module 430 .

Wherein, the image input module 410 is configured to input the template image and the test image into the trained Siamese network, and perform forward calculation to obtain feature vectors of the template image and the test image respectively.

Among them, see Figure 1-a, the Siamese network is a deep neural network. According to the characteristics of the Siamese network, the features of each level of the image can be calculated, including low-level features and high-level features, so that the image can be analyzed according to the high-level features of the image. will get more accurate results. In order to perform image matching, the Siamese network is designed to include two convolutional neural networks with the same structure and shared parameters. When using the Siamese network for image matching, the template image and the test image of the image are input respectively, so that two input images can be calculated separately. The eigenvectors, and then calculate the similarity of the two eigenvectors, so as to judge the similarity of the two images according to the value of the similarity, that is, to judge whether the two images match.

Among them, the trained Siamese network has the ability to accurately match the template image and the test image due to the training and learning of a large number of sample images.

Wherein, the template image is a standard picture for comparison, and the test image is compared with the template image to determine whether it matches the template image.

Calculation module 420, configured to calculate the similarity between the feature vectors of the template image and the test image.

Wherein, the similarity is a parameter for judging the similarity between the template image and the test image.

Optionally, in some possible implementations of the present invention, after calculating the feature vectors of the module image and the test image, the Euclidean distance of the feature vectors can be calculated to obtain the similarity between the feature vectors of the template image and the test image.

A determining module 430, configured to determine whether the test image matches the template image according to the similarity.

Among them, since the similarity is a measure of the similarity of the features of the two images input by the Siamese network, it can be used to judge whether the template image is similar to the test image, that is, to judge whether the template image matches the test image.

Optionally, in some possible implementations of the present invention, the template image is the corresponding component image when the component exists and is in a standard position, the test image is the component image that needs to be tested, and the determination module 430 specifically Used for:

Determine whether the component corresponding to the test image exists according to the similarity.

Wherein, the template image can be a PCB template image, so that the input component image or test image corresponding to it is also a PCB image, and the template image can also be an image containing each independent component intercepted from the PCB image, so as to correspond to it The input test image is also an image containing individual components.

Optionally, in some possible implementations of the present invention, since the missing part detection is performed on each component, generally the template image input in the Siamese network is an image containing an independent component, then the input in the training phase The component image is also the component image of the position intercepted from the PCB image, and the image input in the test stage is also the component image of the position intercepted from the PCB image in different situations.

Optionally, in some possible implementations of the present invention, when the template image is a template image of a component, it can be understood that the template image is a component image in which the component exists and the position of the component is in a correct position, and the test image is based on the template The image judges whether there is a missing part.

It can be understood that since the Siamese network can judge whether the test image matches the template image, when the template image and the test image are component images, it can be determined according to the similarity whether the test image of the component image matches the template image, that is, determine the test image Whether the corresponding element exists.

Optionally, in some possible implementation manners of the present invention, the component detection device 400 further includes:

A preprocessing module 440, configured to use template matching to obtain the positions of elements in the template image and the test image and align the template image and the test image;

The template image and the test image are normalized to trigger the image input module to execute the step of inputting the template image and the test image into the trained Siamese network.

It can be understood that since the component image is an image intercepted from the template image, in order to make the detection result more accurate, it is necessary to locate the component in the intercepted component image at the center of the image, and at the same time normalize the size of the image to ensure the accuracy of subsequent processing. This process is called preprocessing.

Optionally, in some possible implementation manners of the present invention, the image may not be preprocessed.

Optionally, in some possible implementation manners of the present invention, the component detection device 400 further includes:

A sample creation module 450, configured to create a positive sample set and a negative sample set for the component image alignment, wherein the positive sample set includes a component image pair composed of the template image and the component image when the component exists, The negative sample set includes a component image pair formed by the template image and the component image when the component does not exist, and the negative sample set also includes components formed by the template image and component images when other components exist image pair;

A training module 460, configured to use the positive sample set and the negative sample set to train the Siamese network, so as to trigger the image input module to execute the step of inputting the template image and the test image into the trained Siamese network. Network steps.

It should be noted that, optionally, in some possible implementations of the present invention, when the template image and the test image are PCB images, the positive sample set in the sample set includes the template image and the position of each component on the template image. The component image pair composed of the component image when the component is correct, the negative sample set includes the component image pair composed of the template image and the component image when there may be no component in each component position, the negative sample set also includes the template image and each component A component image pair composed of component images when all components exist at the position, but a wrong component may be inserted.

Optionally, in other possible implementations of the present invention, when both the template image and the test image are images containing individual components that are intercepted from the PCB image, the positive sample set in the sample set includes the template image The component image pair composed of the image containing the component in the template image, the negative sample set includes the component image pair composed of the template image and the image that does not contain the component, and the negative sample set also includes the template image and the component that is not in the template image , but contains component images of other wrong components.

Optionally, in some possible implementation manners of the present invention, the set of positive samples includes the template image, and samples in which components exist and are taken in various scenes, such as positive samples taken in bad light images, and positive sample images taken from different positions or angles, or positive sample images taken in other complex scenes, the Siamese network is trained using the positive sample images taken in each scene, so that the Siamese network has a stronger recognition Ability to correctly detect non-missing component images taken in different scenarios in the subsequent test stage, and improve the recognition rate.

It can be understood that when using the Siamese network to detect missing parts in images, it is first necessary to train and learn the Siamese network using samples to obtain the parameters of the Siamese network, and then use the trained Siamese network to accurately detect missing parts in component images. When training the Siamese network, it is necessary to use positive sample sets and negative sample sets in various scenarios for training, so that the Siamese network can fully learn the picture when the component is not a missing part, and the Siamese network can fully learn Only by seeing the pictures of missing components, can the missing parts detection be carried out correctly in the future.

Optionally, in some possible implementations of the present invention, the negative sample set includes the template image, and an image when the component does not exist, or when the component exists but the welding position is obviously wrong, or not Images etc. when the components are correct.

It can be understood that taking as many positive sample sets and negative sample sets as possible in different scenarios to train the Siamese network can make the learning effect of the Siamese network better, so that the accuracy of subsequent missing parts detection and recognition is high.

Optionally, in some possible implementation manners of the present invention, the positive sample set and the negative sample set of the component image include training samples and testing samples.

Wherein, the training sample refers to training the Siamese network, and the test sample refers to testing the effect of the trained Siamese network, and the two together constitute a sample in the training phase of the Siamese network.

Optionally, in some possible implementations of the present invention, the sample creation module 450 includes:

An image acquisition unit 451, which acquires an image of a printed circuit board;

The intercepting unit 452 is configured to use the printed circuit board template image as a reference to intercept the component image on the printed circuit board image;

The sample collection unit 453 is configured to collect the component image alignment positive sample set and negative sample set.

Optionally, in some possible implementation manners of the present invention, the component image may be marked after the component image is captured on the printed circuit board icon.

It can be understood that since it is necessary to detect missing parts for each component, when training the Siamese network, it is necessary to intercept a sample set of images of each component on the printed circuit board image for training. Moreover, in order to distinguish each component image during training, it is necessary to mark each component image to distinguish different components before training.

Optionally, in some possible implementations of the present invention, a camera can be set up on the production line to collect images of different types of PCB boards in batches, and use board tracking technology to avoid repeatedly photographing a certain PCB board. In this way, each type of PCB board contains multiple image samples, and each image sample corresponds to a certain type of PCB board, so that the obtained component images on the PCB board also come from different boards. Ensure sample diversity.

Optionally, in some possible implementation manners of the present invention, the intercepting unit 452 intercepts the component image on the printed circuit board specifically as follows:

The component image is automatically captured using the position information of the component on the printed board image.

It can be understood that when the position information of the component is known, the component image can be automatically intercepted according to the position information.

Optionally, in some possible implementations of the present invention, the position information of the obtained component image may be obtained through the component position information recorded in the template file, or through the manually marked position information.

Optionally, in some possible implementation manners of the present invention, the intercepting unit 452 marks the component image specifically as follows:

Label according to component category information.

It can be understood that it is necessary to distinguish the types of components, so that the types of components can be recorded during training to accurately detect missing parts of components.

Optionally, in some other possible implementation manners of the present invention, the components may also be marked in other ways that can distinguish the components.

Optionally, in some possible implementations of the present invention, the sample creation module 450 also includes:

A preprocessing unit 454, configured to use template matching to obtain the position of the component in the component image and align the component image;

The component image is normalized to trigger the sample acquisition unit to execute the step of acquiring a positive sample set and a negative sample set for the component image pair.

It can be understood that, similar to the process of testing the Siamese network, when the sample image of the Siamese network is collected, the image is aligned so that the component is located in the center of the image, and the image is normalized. This process is called image pre-processing. During the processing process, preprocessing the image will make the Siamese network training effect better.

Optionally, in some possible implementations of the present invention, if the component image is preprocessed when the Siamese network is trained, then when the trained Siamese network is used to detect missing parts, the component image also needs to be preprocessed. Preprocessing, if the component image is not preprocessed when the Siamese network is trained, then the component image is not preprocessed when the trained Siamese network is used for missing parts detection.

Optionally, in some possible implementation manners of the present invention, the determining module 430 is specifically configured to:

judging whether the similarity is within a preset range;

If the similarity is within a preset range, it is determined that the element exists;

If the similarity is not within the preset range, it is determined that the element does not exist.

It can be understood that since the similarity is used to characterize the degree of similarity between the template image and the test image, when the similarity is within a preset range, it means that the template image is similar to the test image, then the component image exists, that is, the component does not Missing parts, otherwise, components missing parts.

Optionally, in some possible implementations of the present invention, when the similarity is greater than or equal to a preset value, the component exists, that is, there are no missing parts in the component; when the similarity is smaller than the preset value, then the element does not exist.

It can be understood that the functions of each functional module of the detection device 400 in this embodiment can be specifically implemented according to the method in the above method embodiment, and the specific implementation process can refer to the relevant description of the above method embodiment, and will not be repeated here.

It can be seen that in the solution of this embodiment, the detection device 400 inputs the template image and the test image into the trained Siamese network, and performs forward calculations respectively to obtain the feature vectors of the template image and the test image; the detection device 400 calculates the similarity between the template image and the feature vector of the test image; the detecting device 400 determines whether the test image matches the template image according to the similarity. Since the Siamese network can extract the advanced features of the image, inputting the image into the trained Siamese network can accurately judge whether the template image matches the test image, and the judgment accuracy is high.

Referring to FIG. 5 , FIG. 5 is a schematic structural diagram of a detection device provided by a fifth embodiment of the present invention. As shown in FIG. 5 , a detection device 500 provided by a fifth embodiment of the present invention may include: at least one bus 501 , at least one processor 502 connected to the bus, and at least one memory 503 connected to the bus. Wherein, the processor 502 calls the code stored in the memory 503 through the bus 501 to input the template image and the test image into the trained Siamese network, and performs forward calculations respectively to obtain the values of the template image and the test image. A feature vector; calculating a similarity between the template image and the feature vector of the test image; determining whether the test image matches the template image according to the similarity.

Among them, the trained Siamese network has the ability to accurately match the template image and the test image due to the training and learning of a large number of sample images.

Wherein, the template image is a standard picture for comparison, and the test image is compared with the template image to determine whether it matches the template image.

Wherein, the similarity is a parameter for judging the similarity between the template image and the test image.

Optionally, in some possible implementations of the present invention, the template image is a corresponding component image when the component exists and is in a standard position, the test image is a component image that needs to be tested, and the processor 502 according to When the similarity determines whether the test image matches the template image, it is also used for:

Determine whether the component corresponding to the test image exists according to the similarity.

Wherein, the template image can be a PCB template image, so that the input component image or test image corresponding to it is also a PCB image, and the template image can also be an image containing each independent component intercepted from the PCB image, so as to correspond to it The input test image is also an image containing individual components.

Optionally, in some possible implementations of the present invention, since the missing part detection is performed on each component, generally the template image input in the Siamese network is an image containing an independent component, then the input in the training phase The component image is also the component image of the position intercepted from the PCB image, and the image input in the test stage is also the component image of the position intercepted from the PCB image in different situations.

Wherein, the similarity is a measure of the similarity of the features of the two images input by the Siamese network, and because the Siamese network after training can be prepared to identify whether the component images taken in different scenes are missing parts, so it can be based on The similarity value judges whether the component is missing or not.

Optionally, in some possible implementation manners of the present invention, before inputting the template image and the test image into the trained Siamese network, the processor 502 is further configured to:

Obtaining the position of the template image and the element in the test image by template matching and aligning the template image and the test image;

The template image and the test image are normalized to trigger the execution of the step of inputting the template image and the test image into the trained Siamese network.

Optionally, in some possible implementation manners of the present invention, before inputting the template image and the test image into the trained Siamese network, the processor 502 is further configured to:

Create a positive sample set and a negative sample set for the component image pair, wherein the positive sample set includes a component image pair formed by the template image and the component image when the component exists, and the negative sample set includes the A component image pair composed of the template image and the component image when the component does not exist, the negative sample set also includes a component image pair formed by the template image and component images when other components exist;

Using the positive sample set and the negative sample set to train the Siamese network, so as to trigger the execution of the step of inputting the template image and the test image into the trained Siamese network.

Optionally, in some possible implementation manners of the present invention, the positive sample set and the negative sample set of the component image include training samples and test samples.

Wherein, the training sample refers to training the Siamese network, and the test sample refers to testing the effect of the trained Siamese network, and the two together constitute a sample in the training phase of the Siamese network.

Optionally, in some possible implementation manners of the present invention, when creating the component image alignment positive sample set and negative sample set, the processor 502 is specifically configured to:

Capture printed circuit board images;

Taking the printed circuit board template image as a reference, intercepting the component image on the printed circuit board image;

The image of the component is collected to align a positive sample set and a negative sample set.

Optionally, in some possible implementations of the present invention, a camera can be set up on the production line to collect images of different types of PCB boards in batches, and use board tracking technology to avoid repeatedly photographing a certain PCB board. In this way, each type of PCB board contains multiple image samples, and each image sample corresponds to a certain type of PCB board, so that the obtained component images on the PCB board also come from different boards. Ensure that the samples are not repeated and the number is sufficient.

Optionally, in some possible implementation manners of the present invention, the intercepting the component image on the printed circuit board, the processor 502 is specifically configured to:

The component image is automatically captured using the position information of the component on the printed board image.

It can be understood that when the position information of the component is known, the component image can be automatically intercepted according to the position information.

Optionally, in some possible implementation manners of the present invention, in annotating the component image, the processor 502 is specifically configured to:

Label according to component category information.

Optionally, in some other possible implementation manners of the present invention, the components may also be marked in other ways that can distinguish the components.

Optionally, in some possible implementation manners of the present invention, before collecting the component image alignment positive sample set and negative sample set, the processor 502 is further configured to:

Obtaining the position of the component in the component image by template matching and aligning the component image;

The component image is normalized to trigger the execution of the step of acquiring the positive sample set and the negative sample set of the component image pair.

Optionally, in some possible implementations of the present invention, the processor 502 determines whether the element corresponding to the test image exists according to the similarity, and the processor 502 is specifically configured to:

judging whether the similarity is within a preset range;

If the similarity is within a preset range, it is determined that the element exists;

If the similarity is not within the preset range, it is determined that the element does not exist.

It can be understood that the functions of each functional module of the component detection device 500 in this embodiment can be specifically realized according to the method in the above-mentioned method embodiment, and the specific implementation process can refer to the relevant description of the above-mentioned method embodiment, and will not be repeated here. .

It can be understood that the functions of each functional module of the detection device 500 in this embodiment can be specifically implemented according to the method in the above method embodiment, and the specific implementation process can refer to the relevant description of the above method embodiment, and will not be repeated here.

It can be seen that in the solution of this embodiment, the detection device 500 inputs the template image and the test image into the trained Siamese network, and performs forward calculations respectively to obtain the feature vectors of the template image and the test image; the detection device 500 calculates the similarity between the template image and the feature vector of the test image; the detecting device 500 determines whether the test image matches the template image according to the similarity. Since the Siamese network can extract the advanced features of the image, inputting the image into the trained Siamese network can accurately judge whether the template image matches the test image, and the judgment accuracy is high.

An embodiment of the present invention also provides a computer storage medium, wherein the computer storage medium can store a program, and the program includes some or all steps of any detection method described in the above method embodiments when executed.

It should be noted that for the foregoing method embodiments, for the sake of simple description, they are expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence. Because of the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

In the foregoing embodiments, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or can be Integrate into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be an embedded device, a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media capable of storing program codes such as U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk.

As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still understand the foregoing The technical solutions recorded in each embodiment are modified, or some of the technical features are replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (12)

1. a detection method, is characterized in that, described method comprises:

Template image and test pattern are inputted the Siamese network after training, and obtains the proper vector of described template image and described test pattern respectively as forward calculation;

Calculate the similarity of the proper vector of described template image and described test pattern;

Determine whether described test pattern mates with described template image according to described similarity.

2. method according to claim 1, it is characterized in that, the part drawing picture that described template image is corresponding when being element existence and being in normal place, described test pattern is the part drawing picture needing to carry out testing, describedly determine whether described test pattern mates with described template image, comprising according to described similarity:

Determine whether the element that described test pattern is corresponding exists according to described similarity.

3. method according to claim 2, is characterized in that, described described template image and described test pattern are inputted the Siamese network after training before, described method also comprises:

Template matches is utilized to obtain the position of described template image and element in described test pattern and align with described test pattern to described template image;

Described template image and described test pattern are normalized, perform the described step described template image and described test pattern being inputted the Siamese network after training to trigger.

4. according to the method in claim 2 or 3, it is characterized in that, described described template image and described test pattern are inputted the Siamese network after training before, described method also comprises:

Create described part drawing picture and align sample set and negative sample set, wherein, described positive sample set comprises the part drawing picture pair of described element image construction when described template image and described element exist, described negative sample set comprises the part drawing picture pair of described element image construction when described template image and described element do not exist, and described negative sample set also comprises the part drawing picture pair of element image construction when described template image and other element exist;

Utilize described positive sample set and described negative sample set to train described Siamese network, perform the described step described template image and described test pattern being inputted the Siamese network after training to trigger.

5. method according to claim 4, is characterized in that, the described part drawing picture of described establishment aligns sample set and negative sample set, comprising:

Gather printed circuit board image;

With printed circuit board (PCB) template image for reference, described printed circuit board image intercepts part drawing picture;

Gather described part drawing picture and align sample set and negative sample set.

6. method according to claim 5, before the described part drawing picture of described collection aligns sample set and negative sample set, described method also comprises:

Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;

Described part drawing picture is normalized, performs to trigger the step that the described part drawing picture of described collection aligns sample set and negative sample set.

7. method according to claim 6, is characterized in that, describedly determines whether the element that described test pattern is corresponding exists, and comprising according to described similarity:

Judge described similarity whether in preset range;

If described similarity is in preset range, be then judged to be that described element exists;

If described similarity is not in preset range, be then judged to be that described element does not exist.

8. a pick-up unit, is characterized in that, described device comprises:

Image input module, for template image and test pattern are inputted the Siamese network after training, and obtains the proper vector of described template image and described test pattern respectively as forward calculation;

Computing module, for calculating the similarity of the proper vector of described module map picture and described test pattern;

According to described similarity, determination module, for determining whether described test pattern mates with described template image.

9. device according to claim 8, is characterized in that, described template image is element corresponding part drawing picture when existing and be in normal place, and described test pattern is the part drawing picture needing to carry out testing, described determination module specifically for:

Determine whether the element that described test pattern is corresponding exists according to described similarity.

10. device according to claim 9, is characterized in that, described device also comprises:

Pretreatment module, obtains the position of described template image and element in described test pattern for utilizing template matches and aligns with described test pattern to described template image;

Described template image and described test pattern are normalized, perform the described step described template image and described test pattern being inputted the Siamese network after training to trigger described image input module.

11. devices according to claim 9 or 10, it is characterized in that, described device also comprises:

Sample creation module, sample set and negative sample set is aligned for creating described part drawing picture, wherein, described positive sample set comprises the part drawing picture pair of described element image construction when described template image and described element exist, described negative sample set comprises the part drawing picture pair of described element image construction when described template image and described element do not exist, and described negative sample set also comprises the part drawing picture pair of element image construction when described template image and other element exist;

Training module, for utilizing described positive sample set and described negative sample set to train described Siamese network, perform the described step described template image and described test pattern being inputted the Siamese network after training to trigger described image input module.

12. devices according to claim 11, is characterized in that, described sample creation module comprises:

Image acquisition units, for gathering printed circuit board image;

Interception unit, for printed circuit board (PCB) template image for reference, described printed circuit board image intercepts part drawing picture;

Sample collection unit, aligns sample set and negative sample set for gathering described part drawing picture.