CN106682736A - Image identification method and apparatus - Google Patents

Abstract

The present disclosure relates to an image recognition method and device. The method includes: inputting the image to be recognized into a trained first convolutional neural network, the first convolutional neural network includes P groups of convolutional layers, each group of convolutional layers includes two convolutional layers, and the two The convolution kernels of the convolution layer are respectively the first convolution kernel with a size of 1*N and the second convolution kernel with a size of N*1; through each group of convolution layers of the first convolution neural network pair The image to be recognized is subjected to feature extraction to obtain the features to be extracted of each set of convolutional layers; the recognition result of the image to be recognized is determined according to the features to be extracted of each set of convolutional layers. The technical solution provided by the present disclosure solves the problems of large calculation amount and high calculation cost caused by adopting a square convolution kernel in the related art.

Description

Image recognition method and device

technical field

The present disclosure relates to the field of convolution technology, in particular to an image recognition method and device.

Background technique

Convolutional Neural Network (CNN) is an efficient image recognition method. In CNN, the image undergoes a series of feature extraction and processing of convolutional layers, activation layers, pooling layers, and fully connected layers to obtain the recognition result of the image to be recognized. In related technologies, the calculation principle of the convolution layer is to use a square convolution kernel to slide on the input rectangular area in the form of a sliding window, and calculate the product of the convolution kernel and the value of the position every time it slides to a new position , the calculation amount of the square convolution kernel is O(N ² ), the calculation amount is relatively large, and the calculation cost is relatively high.

Contents of the invention

In order to overcome the problems existing in the related technologies, the embodiments of the present disclosure provide an image recognition method and device, which are used to reduce the calculation amount of convolution and reduce the calculation cost.

According to a first aspect of an embodiment of the present disclosure, an image recognition method is provided, which may include:

Input the image to be identified into the trained first convolutional neural network, the first convolutional neural network includes P groups of convolutional layers, each group of convolutional layers includes two convolutional layers, and the two convolutional layers The convolution kernels are the first convolution kernel with a size of 1*N and the second convolution kernel with a size of N*1;

Carrying out feature extraction on the image to be identified through each group of convolutional layers of the first convolutional neural network to obtain the features to be extracted of each group of convolutional layers;

The recognition result of the image to be recognized is determined according to the features to be extracted of each group of convolutional layers.

In one embodiment, the method also includes:

Each convolutional layer in the second convolutional neural network is decomposed into the two convolutional layers to obtain the first convolutional neural network, and each convolutional layer in the second convolutional neural network includes A square convolution kernel.

In one embodiment, the convolution kernel of each convolution layer in the second convolutional neural network is a square convolution kernel with a size of N*N, and the second convolutional neural network includes P convolutional layers;

The decomposing each convolutional layer in the second convolutional neural network into the two convolutional layers includes:

Decomposing the square convolution kernel whose size is N*N in each convolutional layer in the second convolutional neural network into a first convolution kernel whose size is 1*N and a second convolution whose size is N*1 Kernel, the product value of the first convolution kernel and the second convolution kernel is the square convolution kernel;

Use the first convolution kernel and the second convolution kernel to replace the square convolution kernels of the corresponding convolution layers in the second convolutional neural network to obtain a set of the first convolutional neural network convolutional layer.

In one embodiment, feature extraction is performed on the image to be recognized through each group of convolutional layers of the first convolutional neural network, including:

Using a convolutional layer in each group of convolutional layers of the first convolutional neural network to perform feature extraction on the input information, to obtain an intermediate value of the features to be extracted in each group of convolutional layers;

Inputting the intermediate value of the features to be extracted into another convolution layer in each set of convolution layers for feature extraction, to obtain the features to be extracted of each set of convolution layers.

According to a second aspect of an embodiment of the present disclosure, an image recognition device is provided, which may include:

The input module is configured to input the image to be recognized into the trained first convolutional neural network, the first convolutional neural network includes P groups of convolutional layers, and each group of convolutional layers includes two convolutional layers, so The convolution kernels of the two convolution layers are respectively a first convolution kernel with a size of 1*N and a second convolution kernel with a size of N*1;

The feature extraction module is configured to perform feature extraction on the image to be recognized through each set of convolutional layers of the first convolutional neural network, and obtain the features to be extracted of each set of convolutional layers;

The determining module is configured to determine the recognition result of the image to be recognized according to the features to be extracted of each set of convolutional layers.

In one embodiment, the device further includes:

The decomposition module is configured to decompose each convolutional layer in the second convolutional neural network into the two convolutional layers to obtain the first convolutional neural network and the second convolutional neural network. Each convolution layer includes a square convolution kernel.

In one embodiment, the convolution kernel of each convolution layer in the second convolutional neural network is a square convolution kernel with a size of N*N, and the second convolutional neural network includes P convolutional layers;

The decomposition modules include:

The decomposition submodule is configured to decompose the square convolution kernel with a size of N*N of each convolution layer in the second convolutional neural network into a first convolution kernel with a size of 1*N and a size of N *1 second convolution kernel, the product value of the first convolution kernel and the second convolution kernel is the square convolution kernel;

The determination submodule is configured to use the first convolution kernel and the second convolution kernel to respectively replace the square convolution kernels of the corresponding convolution layers in the second convolution neural network to obtain the first A set of convolutional layers for a convolutional neural network.

In one embodiment, the feature extraction module includes:

The first extraction submodule is configured to use one convolutional layer in each group of convolutional layers of the first convolutional neural network to perform feature extraction on input information, and obtain the features to be extracted of each group of convolutional layers. Median;

The second extraction sub-module is configured to input the intermediate value of the features to be extracted into another convolution layer in each set of convolution layers for feature extraction, and obtain the features to be extracted of each set of convolution layers.

According to a third aspect of an embodiment of the present disclosure, an image recognition device is provided, including:

processor;

memory for storing processor-executable instructions;

Wherein, the processor is configured as:

Input the image to be identified into the trained first convolutional neural network, the first convolutional neural network includes P groups of convolutional layers, each group of convolutional layers includes two convolutional layers, and the two convolutional layers The convolution kernels are the first convolution kernel with a size of 1*N and the second convolution kernel with a size of N*1;

Carrying out feature extraction on the image to be identified through each group of convolutional layers of the first convolutional neural network to obtain the features to be extracted of each group of convolutional layers;

The recognition result of the image to be recognized is determined according to the features to be extracted of each group of convolutional layers.

The technical solutions provided by the embodiments of the present disclosure may include the following beneficial effects: the convolution kernels of the two convolution layers included in each group of convolution layers in the first convolution neural network are respectively 1*Nth A convolution kernel and a second convolution kernel with a size of N*1, the calculation amount of the convolution kernel of each group of convolution layers is 2*O(N), in the case of a relatively large convolution kernel, in this disclosure The calculation amount 2*O(N) of each group of convolutional layers is much lower than the calculation amount O(N ² ) of the square convolution kernel in the related technology, which solves the calculation caused by the use of the square convolution kernel in the related technology The problem of large volume and high computational cost.

The first convolutional neural network can be obtained by splitting the square convolution kernel of each convolutional layer in the second convolutional neural network into two orthogonal convolutional kernels. Since the two orthogonal convolutional kernels The product is the corresponding square convolution kernel. Therefore, the feature extraction effect of the image to be recognized using each set of convolutional layers in the first convolutional neural network is the same as that of each corresponding convolutional layer in the second convolutional neural network. The feature extraction effects of the square convolution kernels are the same, so the present disclosure reduces the calculation amount of convolution and reduces the calculation cost on the basis of the same convolution effect.

By directly using the first feature map extracted by one convolutional layer in each set of convolutional layers as the input of another convolutional layer, the feature extraction effect of each set of convolutional layers can be compared with the second convolutional neural network The feature extraction effect of the corresponding one of the convolutional layers is the same.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description serve to explain the principles of the invention.

Fig. 1 is a flowchart of an image recognition method according to an exemplary embodiment.

Fig. 2 is a flow chart of obtaining a first convolutional neural network through a second convolutional neural network according to a first exemplary embodiment.

Fig. 3A is a flow chart of feature extraction of an image to be recognized through each group of convolutional layers of the first convolutional neural network according to a second exemplary embodiment.

Fig. 3B is a schematic diagram showing convolution of input information by using a square convolution kernel in the second convolutional neural network according to an exemplary embodiment.

Fig. 3C is a schematic diagram showing convolution of input information by using a convolution kernel corresponding to the square convolution kernel in Fig. 3B in the first convolutional neural network according to an exemplary embodiment.

Fig. 3D is a schematic diagram showing the convolution of the output information in Fig. 3C by using another convolution kernel corresponding to the square convolution kernel in Fig. 3B in the first convolutional neural network according to an exemplary embodiment.

Fig. 4 is a block diagram of an image recognition device according to an exemplary embodiment.

Fig. 5 is a block diagram of another image recognition device according to an exemplary embodiment.

Fig. 6 is a block diagram showing an image recognition device according to an exemplary embodiment.

detailed description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with aspects of the invention as recited in the appended claims.

Fig. 1 is the flow chart of the image recognition method shown according to an exemplary embodiment; The method includes the following steps:

In step 101, input the image to be recognized into the trained first convolutional neural network, the first convolutional neural network includes P groups of convolutional layers, each group of convolutional layers includes two convolutional layers, and the two convolutional layers The convolution kernels of the layers are the first convolution kernel with a size of 1*N and the second convolution kernel with a size of N*1.

In one embodiment, the first convolutional neural network can be obtained by processing the second convolutional neural network, and the first convolutional neural network can be obtained by decomposing each convolutional layer in the second convolutional neural network , can refer to the embodiment shown in FIG. 2 , and will not be described in detail here.

In one embodiment, the convolutional layers in the first convolutional neural network appear in groups, and the convolutional kernels of the two convolutional layers of each group of convolutional layers are a first convolutional kernel with a size of 1*N And a second convolution kernel of size N*1.

In one embodiment, the second convolutional neural network can be a convolutional neural network trained according to relevant training methods in the related art, and the convolution kernel in the convolution layer is a square convolution kernel, which is N*N convolution nuclear.

In step 102, feature extraction is performed on the image to be recognized through each set of convolutional layers of the first convolutional neural network to obtain features to be extracted for each set of convolutional layers.

In one embodiment, the step of extracting features of the image to be recognized through each group of convolutional layers of the first convolutional neural network may refer to the embodiment shown in FIG. 3A , which will not be described in detail here.

In step 103, the recognition result of the image to be recognized is determined according to the features to be extracted of each group of convolutional layers.

In one embodiment, the features to be extracted by each group of convolutional layers can be processed according to the method of using a convolutional neural network in the related art to obtain a recognition result, for example, using a pooling layer, a fully connected layer, etc. for each The feature to be extracted of the group convolutional layer is processed to obtain the recognition result, which will not be described in detail here.

In this embodiment, the convolution kernels of the two convolution layers included in each group of convolution layers in the first convolutional neural network are the first convolution kernel with a size of 1*N and the first convolution kernel with a size of N*1. The second convolution kernel, the calculation amount of the convolution kernel of each group of convolution layers is 2*O(N), in the case of relatively large convolution kernels, the calculation amount of each group of convolution layers in this disclosure is 2* O(N) is much lower than the calculation amount O(N ² ) of the square convolution kernel in the related art, which solves the problems of large calculation amount and high calculation cost caused by the use of the square convolution kernel in the related art.

In one embodiment, the method also includes:

Each convolutional layer in the second convolutional neural network is decomposed into two convolutional layers to obtain the first convolutional neural network, and each convolutional layer in the second convolutional neural network includes a square convolutional kernel.

In one embodiment, the convolution kernel of each convolution layer in the second convolutional neural network is a square convolution kernel with a size of N*N, and the second convolutional neural network includes P convolutional layers;

Decompose each convolutional layer in the second convolutional neural network into two convolutional layers, including:

Decomposing the square convolution kernel whose size is N*N in each convolutional neural network in the second convolutional neural network into a first convolution kernel whose size is 1*N and a second convolution kernel whose size is N*1, The product value of the first convolution kernel and the second convolution kernel is a square convolution kernel;

Using the first convolution kernel and the second convolution kernel to respectively replace the square convolution kernels of the corresponding convolution layers in the second convolution neural network to obtain a set of convolution layers of the first convolution neural network.

In one embodiment, each group of convolutional layers of the first convolutional neural network is used for feature extraction of the image to be recognized, including:

Utilizing a convolutional layer in each group of convolutional layers of the first convolutional neural network to perform feature extraction on the input information, and obtain an intermediate value of features to be extracted in each group of convolutional layers;

The intermediate value of the features to be extracted is input into another convolution layer in each set of convolution layers for feature extraction, and the features to be extracted of each set of convolution layers are obtained.

For details on how to perform image recognition, please refer to subsequent embodiments.

The technical solutions provided by the embodiments of the present disclosure are described below with a specific embodiment.

Fig. 2 is a flow chart of an image recognition method shown according to an exemplary embodiment 1; this embodiment utilizes the above-mentioned method provided by the embodiment of the present disclosure, and how the electronic device according to each convolutional layer of the second convolutional neural network It is decomposed into two convolutional layers as an example for illustration, as shown in Figure 2, including the following steps:

In step 201, the square convolution kernel whose size is N*N in each convolutional layer in the second convolutional neural network is decomposed into a first convolution kernel whose size is 1*N and a first convolution kernel whose size is N*1. Two convolution kernels, the product of the first convolution kernel and the second convolution kernel is a square convolution kernel.

In one embodiment, how to decompose a convolution layer is illustrated by taking a square convolution kernel as an example of a 3*3 convolution kernel. For example, a square convolution kernel is Can be split into a 1*3 first convolution kernel [110] and a 3*1 second convolution kernel In this embodiment, the first convolution kernel is a row vector, and the second convolution kernel is a column vector.

In an embodiment, by splitting each square convolution kernel into a first convolution kernel and a second convolution kernel, a convolution layer is split into a group of convolution layers.

In step 202, use the first convolution kernel and the second convolution kernel to respectively replace the square convolution kernels of the corresponding convolution layers in the second convolution neural network to obtain a set of convolution layers of the first convolution neural network .

In this embodiment, the first convolutional neural network can be obtained by splitting the square convolution kernel of each convolutional layer in the second convolutional neural network into two orthogonal convolutional kernels. The product of the intersecting convolution kernel is the corresponding square convolution kernel. Therefore, the feature extraction effect of the image to be recognized using each group of convolution layers in the first convolutional neural network is the same as that of the corresponding one in the second convolutional neural network. The feature extraction effects of the square convolution kernels of each convolution layer are exactly the same, so the present disclosure reduces the calculation amount of convolution and reduces the calculation cost on the basis of the same convolution effect.

Fig. 3A is a flow chart showing feature extraction of an image to be recognized through each group of convolutional layers of the first convolutional neural network according to a second exemplary embodiment; A schematic diagram of a square convolution kernel in the second convolutional neural network convoluting input information, FIG. 3C is a volume that uses the square convolution kernel corresponding to FIG. 3B in the first convolutional neural network shown according to an exemplary embodiment A schematic diagram of the convolution of the input information by the product kernel, and FIG. 3D is a schematic diagram of using another convolution kernel corresponding to the square convolution kernel of FIG. 3B in the first convolutional neural network to perform the output information of FIG. 3C according to an exemplary embodiment. Schematic diagram of convolution; this embodiment uses the above-mentioned method provided by the embodiment of the present disclosure to illustrate how an electronic device performs feature extraction of an image to be recognized through each set of convolution layers, as shown in FIG. 3A , including the following steps:

In step 301, feature extraction is performed on the input information by using one convolutional layer in each set of convolutional layers of the first convolutional neural network to obtain an intermediate value of features to be extracted in each set of convolutional layers.

In step 302, the intermediate value of the features to be extracted is input into another convolution layer in each set of convolution layers for feature extraction, and the features to be extracted of each set of convolution layers are obtained.

In one embodiment, referring to FIG. 3B, the image 31 to be recognized with the input information 31 being 5*5 is used for feature extraction through the source convolution kernel in the second convolutional neural network, that is, the square convolution kernel labeled 36 Obtaining the feature labeled 35, the square convolution kernel 36 in the second convolutional neural network can be decomposed into the first convolution kernel 32 as [110], and the second convolution kernel 34 as

In one embodiment, in step 301, referring to FIG. 3C , the input information 31 is input to a convolutional layer of a group of convolutional layers in the first convolutional neural network, and convolution processing is performed through the first convolutional kernel 32 , to obtain the intermediate value of the feature to be extracted of the group of convolutional layers, that is, the feature labeled 33; in step 302, referring to Figure 3D, by inputting the feature labeled 33 into another one of the first convolutional neural networks Convolution layer, and use the second convolution kernel 34 to perform convolution processing to obtain the feature labeled 35. It can be seen from Figure 3B-3D that the effect of using a 1*N convolution kernel and an N*1 convolution kernel to convolve the image to be recognized in this disclosure is the same as that of using a square convolution kernel for convolution in related technologies The effect is the same.

In this embodiment, by directly using the first feature map extracted by one convolutional layer in each group of convolutional layers as the input of another convolutional layer, the feature extraction effect of each group of convolutional layers can be compared with the first feature map. The feature extraction effect of the corresponding convolutional layer in the two convolutional neural networks is exactly the same.

Corresponding to the foregoing embodiments of the image recognition method, the present disclosure also provides embodiments of an image recognition device and an electronic device to which it is applied.

Fig. 4 is a block diagram of an image recognition device according to an exemplary embodiment. As shown in Fig. 4, the image recognition device includes:

The input module 410 is configured to input the image to be recognized into the trained first convolutional neural network, the first convolutional neural network includes P groups of convolutional layers, and each group of convolutional layers includes two convolutional layers, two The convolution kernels of the convolution layer are the first convolution kernel with a size of 1*N and the second convolution kernel with a size of N*1;

The feature extraction module 420 is configured to perform feature extraction on the image to be recognized through each group of convolutional layers of the first convolutional neural network, and obtain the features to be extracted of each group of convolutional layers;

The determination module 430 is configured to determine the recognition result of the image to be recognized according to the features to be extracted of each group of convolutional layers.

Fig. 5 is a block diagram of another image recognition device according to an exemplary embodiment. As shown in Fig. 5, on the basis of the embodiment shown in Fig. 4 above, in one embodiment, the device further includes:

The decomposition module 440 is configured to decompose each convolutional layer in the second convolutional neural network into two convolutional layers to obtain the first convolutional neural network, and each convolutional layer in the second convolutional neural network Contains a square convolution kernel.

In one embodiment, the convolution kernel of each convolution layer in the second convolutional neural network is a square convolution kernel with a size of N*N, and the second convolutional neural network includes P convolutional layers;

Decomposition module 440 includes:

The decomposition sub-module 441 is configured to decompose the square convolution kernel whose size is N*N in each convolutional neural network in the second convolutional neural network into a first convolution kernel whose size is 1*N and whose size is N* The second convolution kernel of 1, the product value of the first convolution kernel and the second convolution kernel is a square convolution kernel;

The determination sub-module 442 is configured to use the first convolution kernel and the second convolution kernel to respectively replace the square convolution kernels of the corresponding convolution layers in the second convolutional neural network to obtain a set of convolutional neural networks of the first convolutional neural network convolutional layer.

In one embodiment, feature extraction module 420 includes:

The first extraction sub-module 421 is configured to use one convolutional layer in each group of convolutional layers of the first convolutional neural network to perform feature extraction on input information, and obtain the middle of the features to be extracted in each group of convolutional layers value;

The second extraction sub-module 422 is configured to input the intermediate value of the features to be extracted into another convolution layer in each set of convolution layers for feature extraction, and obtain the features to be extracted of each set of convolution layers.

For the implementation process of the functions and effects of each unit in the above device, please refer to the implementation process of the corresponding steps in the above method for details, and will not be repeated here.

As for the device embodiment, since it basically corresponds to the method embodiment, for related parts, please refer to the part description of the method embodiment. The device embodiments described above are only illustrative, and 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 can also be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. It can be understood and implemented by those skilled in the art without creative effort.

Fig. 6 is a block diagram showing an image recognition device according to an exemplary embodiment. For example, apparatus 600 may be an electronic device such as a tablet computer, a smartphone, or the like.

6, device 600 may include one or more of the following components: processing component 602, memory 604, power supply component 606, multimedia component 608, audio component 610, input/output (I/O) interface 612, sensor component 614, and communication component 616 .

The processing component 602 generally controls the overall operations of the device 600, such as those associated with display, telephone calls, data communications, camera operations, and recording operations. The processing element 602 may include one or more processors 620 to execute instructions to complete all or part of the steps of the above method. Additionally, processing component 602 may include one or more modules that facilitate interaction between processing component 602 and other components. For example, processing component 602 may include a multimedia module to facilitate interaction between multimedia component 608 and processing component 602 .

The memory 604 is configured to store various types of data to support operations at the device 600 . Examples of such data include instructions, messages, pictures, etc. for any application or method operating on device 600 . The memory 604 can be implemented by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Magnetic Memory, Flash Memory, Magnetic or Optical Disk.

The power supply component 606 provides power to various components of the device 600 . Power components 606 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for device 600 .

The multimedia component 608 includes a screen that provides an output interface between the device 600 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensor may not only sense a boundary of a touch or a swipe action, but also detect duration and pressure associated with the touch or swipe operation. In some embodiments, the multimedia component 608 includes a front camera and/or a rear camera. When the device 600 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capability.

The audio component 610 is configured to output and/or input audio signals. For example, the audio component 610 includes a microphone (MIC) configured to receive external audio signals when the device 600 is in operation modes, such as call mode, recording mode and voice recognition mode. Received audio signals may be further stored in memory 604 or sent via communication component 616 . In some embodiments, the audio component 610 also includes a speaker for outputting audio signals.

The I/O interface 612 provides an interface between the processing component 602 and a peripheral interface module. The peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include, but are not limited to: a home button, volume buttons, start button, and lock button.

Sensor assembly 614 includes one or more sensors for providing status assessments of various aspects of device 600 . For example, the sensor component 614 can detect the open/closed state of the device 600, the relative positioning of components, such as the display and keypad of the device 600, the sensor component 614 can also detect a change in the position of the device 600 or a component of the device 600, the user Presence or absence of contact with device 600 , device 600 orientation or acceleration/deceleration and temperature change of device 600 . The sensor assembly 614 may include a proximity sensor configured to detect the presence of nearby objects in the absence of any physical contact. Sensor assembly 614 may also include optical sensors, such as CMOS or CCD image sensors, for use in imaging applications. In some embodiments, the sensor component 614 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a distance sensor, a pressure sensor or a temperature sensor.

The communication component 616 is configured to facilitate wired or wireless communication between the apparatus 600 and other devices. The device 600 can access wireless networks based on communication standards, such as WIFI, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 616 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 616 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.

In an exemplary embodiment, apparatus 600 may be programmed by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable Gate array (FPGA), controller, microcontroller, microprocessor or other electronic component implementation for performing the method described above, including:

Input the image to be recognized into the trained first convolutional neural network, the first convolutional neural network includes P groups of convolutional layers, each group of convolutional layers includes two convolutional layers, and the convolution kernels of the two convolutional layers Respectively, the first convolution kernel with a size of 1*N and the second convolution kernel with a size of N*1;

Through each group of convolutional layers of the first convolutional neural network, feature extraction is performed on the image to be recognized, and the features to be extracted of each group of convolutional layers are obtained;

The recognition result of the image to be recognized is determined according to the features to be extracted of each group of convolutional layers.

In an exemplary embodiment, there is also provided a non-transitory computer-readable storage medium including instructions, such as the memory 604 including instructions, which can be executed by the processor 620 of the device 600 to implement the above method. For example, the non-transitory computer readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like.

Other embodiments of the disclosure will be readily apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, and these modifications, uses or adaptations follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not disclosed in the present disclosure . The specification and examples are to be considered exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It should be understood that the present disclosure is not limited to the precise constructions which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims (9)

1. a kind of image-recognizing method, it is characterised in that methods described includes：

The first convolutional neural networks that images to be recognized input has been trained, first convolutional neural networks include P group convolution Layer, each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively the first convolution that size is 1*N Core and the second convolution kernel that size is N*1；

Feature extraction is carried out to the images to be recognized by each group of convolutional layer of first convolutional neural networks, institute is obtained State the feature to be extracted of each group of convolutional layer；

The recognition result of the images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.

2. method according to claim 1, it is characterised in that methods described also includes：

Each convolutional layer in second convolutional neural networks is decomposed into into described two convolutional layers, the first convolution god is obtained Jing networks, each convolutional layer in second convolutional neural networks includes a square convolution kernel.

3. method according to claim 2, it is characterised in that each convolutional layer in second convolutional neural networks Convolution kernel be size for N*N square convolution kernel, second convolutional neural networks include P convolutional layer；

Described each convolutional layer by the second convolutional neural networks is decomposed into described two convolutional layers, including：

The size of each convolutional layer in second convolutional neural networks is decomposed into into size for 1* for the square convolution kernel of N*N First convolution kernel of N and size for N*1 the second convolution kernel, first convolution kernel is with the product value of second convolution kernel The square convolution kernel；

Corresponding volume in second convolutional neural networks is replaced respectively using first convolution kernel and second convolution kernel The square convolution kernel of lamination, obtains one group of convolutional layer of first convolutional neural networks.

4. method according to claim 1, it is characterised in that each group by first convolutional neural networks Convolutional layer carries out feature extraction to the images to be recognized, including：

Feature is carried out using a convolutional layer in each group of convolutional layer of first convolutional neural networks to input information to carry Take, obtain the intermediate value of the feature to be extracted of each group of convolutional layer；

Another convolutional layer that the intermediate value of the feature to be extracted is input in each group of convolutional layer is carried out into feature extraction, Obtain the feature to be extracted of each group of convolutional layer.

5. a kind of pattern recognition device, it is characterised in that described device includes：

Input module, is configured to the first convolutional neural networks for having trained images to be recognized input, the first convolution god Jing networks include P group convolutional layers, and each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively greatly Little the first convolution kernel and size for 1*N is second convolution kernel of N*1；

Characteristic extracting module, is configured to each group of convolutional layer of first convolutional neural networks to the figure to be identified As carrying out feature extraction, the feature to be extracted of each group of convolutional layer is obtained；

Determining module, is configured to determine the identification of the images to be recognized according to the feature to be extracted of each group of convolutional layer As a result.

6. device according to claim 5, it is characterised in that described device also includes：

Decomposing module, is configured to for each convolutional layer in the second convolutional neural networks to be decomposed into described two convolutional layers, First convolutional neural networks are obtained, each convolutional layer in second convolutional neural networks includes a square convolution Core.

7. device according to claim 6, it is characterised in that each convolutional layer in second convolutional neural networks Convolution kernel be size for N*N square convolution kernel, second convolutional neural networks include P convolutional layer；

The decomposing module includes：

Decompose submodule, it is the square of N*N to be configured to the size of each convolutional layer in second convolutional neural networks Convolution kernel be decomposed into size be 1*N the first convolution kernel and size for N*1 the second convolution kernel, first convolution kernel with it is described The product value of the second convolution kernel is the square convolution kernel；

Determination sub-module, is configured with first convolution kernel and second convolution kernel replaces respectively second convolution The square convolution kernel of corresponding convolutional layer in neutral net, obtains one group of convolutional layer of first convolutional neural networks.

8. device according to claim 5, it is characterised in that the characteristic extracting module includes：

First extracting sub-module, a convolution being configured to, with each group of convolutional layer of first convolutional neural networks Layer carries out feature extraction to input information, obtains the intermediate value of the feature to be extracted of each group of convolutional layer；

Second extracting sub-module, is configured to the intermediate value of just described feature to be extracted and is input in each group of convolutional layer Another convolutional layer carries out feature extraction, obtains the feature to be extracted of each group of convolutional layer.

9. a kind of pattern recognition device, it is characterised in that described device includes：

Processor；

For storing the memorizer of processor executable；

Wherein, the processor is configured to：

The first convolutional neural networks that images to be recognized input has been trained, first convolutional neural networks include P group convolution Layer, each group of convolutional layer includes two convolutional layers, and the convolution kernel of described two convolutional layers is respectively the first convolution that size is 1*N Core and the second convolution kernel that size is N*1；

Feature extraction is carried out to the images to be recognized by each group of convolutional layer of first convolutional neural networks, institute is obtained State the feature to be extracted of each group of convolutional layer；

The recognition result of the images to be recognized is determined according to the feature to be extracted of each group of convolutional layer.