CN119206318A

CN119206318A - Method for identifying camera modules based on artificial intelligence

Info

Publication number: CN119206318A
Application number: CN202411236105.7A
Authority: CN
Inventors: 姚乐; 张果
Original assignee: Shenzhen Yungli Optronics Co ltd
Current assignee: Shenzhen Yungli Optronics Co ltd
Priority date: 2024-09-04
Filing date: 2024-09-04
Publication date: 2024-12-27

Abstract

The invention provides a method for identifying a camera module based on artificial intelligence, which comprises the steps of collecting an appearance image of the camera module, obtaining an image collection mode of the camera module, obtaining a target image collected by the camera module, extracting characteristic attributes of the target image, analyzing first classification of the camera module based on the appearance image, analyzing second classification of the camera module based on the image collection mode, analyzing third classification of the camera module based on the characteristic attributes, and carrying out fusion analysis on the first classification, the second classification and the third classification based on an artificial intelligence model to identify the target classification of the camera module. In the invention, the appearance of the camera module, the image acquisition mode and the characteristic attribute of the target image are comprehensively considered, and the precise identification and classification of the camera module are realized through a fusion analysis technology.

Description

Method for identifying camera module based on artificial intelligence

Technical Field

The invention relates to the technical field of data processing, in particular to a method for identifying a camera module based on artificial intelligence.

Background

In the conventional camera module recognition technology, the manual experience or a single feature analysis method is often relied on. For example, the type of the camera module is judged only by rough observation of the appearance, and the accuracy is low, and the performance characteristics and the image acquisition capability inside the camera module cannot be deeply known. The analysis of the image acquisition mode is also simpler, and only some basic parameter settings are usually focused, so that the adaptability and stability of the image acquisition mode under different environments and application scenes are ignored.

In the aspect of processing a target image, the traditional method simply analyzes the basic attributes of the image such as definition, color and the like, lacks comprehensive deep mining of the image characteristic attributes, and cannot accurately infer the performance and characteristics of the camera module according to the image characteristics.

In addition, existing identification methods present significant limitations in facing increasingly diverse and complex camera module markets. The camera modules produced by different manufacturers have different characteristics and parameters, and the traditional method is difficult to realize the accurate classification and identification of various types of camera modules.

Disclosure of Invention

The invention mainly aims to provide a method for identifying a camera module based on artificial intelligence, which aims to overcome the defect that the camera module cannot be accurately identified at present.

In order to achieve the above purpose, the invention provides a method for identifying a camera module based on artificial intelligence, which comprises the following steps:

Acquiring an appearance image of the camera module, acquiring an image acquisition mode of the camera module, acquiring a target image acquired by the camera module, and extracting characteristic attributes of the target image;

Analyzing the first classification of the camera module based on the appearance image, analyzing the second classification of the camera module based on the image acquisition mode, and analyzing the third classification of the camera module based on the characteristic attribute;

And carrying out fusion analysis on the first classification, the second classification and the third classification based on the artificial intelligence model, and identifying to obtain the target classification of the camera module.

Further, the image acquisition mode of the camera module comprises:

detecting hardware interface information of the camera module, comparing the hardware interface information with preset standard interface parameters, and determining an acquisition mode category corresponding to the type of a connection interface of the hardware interface information to obtain interface association acquisition mode information;

Analyzing a signal transmission mode of the camera module in a working state, and determining related acquisition mode characteristics according to the signal transmission mode to obtain signal mode acquisition mode information;

Analyzing a driving program of the camera module, extracting driving characteristics including configuration parameters of image acquisition and an instruction set, and determining driving characteristic association acquisition mode information according to a preset corresponding relation library of the driving characteristics and the acquisition modes;

And combining the interface association acquisition mode information, the signal mode acquisition mode information and the driving characteristic association acquisition mode information, and obtaining an image acquisition mode of the camera module by applying a preset fusion algorithm.

Further, the extracting the characteristic attribute of the target image includes:

Performing color space conversion processing on the target image, converting the target image from an RGB color space to an HSV color space, and counting the mean value, variance and peak value of each channel of the converted color space to obtain color characteristic attributes;

Performing edge extraction on the target image by using an edge detection algorithm, and calculating the length, density and curvature of the edge to obtain edge characteristic attributes;

Analyzing the texture of the target image by adopting a texture analysis algorithm, and extracting roughness, contrast and direction degree of the texture to obtain texture characteristic attributes;

Dividing a target image into a plurality of sub-blocks, calculating a gray level co-occurrence matrix of each sub-block, extracting energy, entropy and correlation characteristics from the gray level co-occurrence matrix, and synthesizing the characteristics of each sub-block to obtain the spatial distribution characteristic attribute of the target image;

And carrying out fusion processing on the color characteristic attribute, the edge characteristic attribute, the texture characteristic attribute and the spatial distribution characteristic attribute to obtain the characteristic attribute of the target image.

Further, the analyzing the first classification of the camera module based on the appearance image, the analyzing the second classification of the camera module based on the image acquisition mode, and the analyzing the third classification of the camera module based on the characteristic attribute comprises:

matching the appearance image with the appearance images of all types of camera modules stored in a database to obtain the probability of the camera modules being of all types as a first classification;

Matching the image acquisition mode with the image acquisition modes of all types of camera modules stored in a database to obtain the probability of the camera modules being of all types as a second classification;

and matching the characteristic attribute with the characteristic attribute of each type of camera module stored in the database to obtain the probability of each type of camera module as a third classification.

Further, the performing fusion analysis on the first classification, the second classification and the third classification based on the artificial intelligence model, and identifying the target classification of the camera module, includes:

Constructing a multidimensional feature space, and mapping the first classification, the second classification and the third classification to different dimensional axes of the multidimensional feature space to form a feature vector point;

The method comprises the steps of obtaining an artificial intelligent model, wherein the artificial intelligent model is a self-encoder based on deep learning, and learns the feature vector point distribution of a camera module of a known type in a multidimensional feature space in advance to obtain a potential distribution mode and rule of the feature space;

Inputting the feature vector points of the camera module into the artificial intelligent model, and calculating the reconstruction errors of the feature vector points in a feature space based on a self-encoder;

judging whether the reconstruction error is in a threshold range or not;

If the reconstruction error is within the threshold range, finding the nearest known type as the target classification according to the characteristic mapping relation inside the self-encoder and the similarity measure with the known type;

If the reconstruction error exceeds the threshold range, the reconstruction error is judged to be of a new type or an abnormal type, and a further manual intervention mechanism is triggered.

Further, the fusion analysis is performed on the first classification, the second classification and the third classification based on the artificial intelligence model, and after the target classification of the camera module is obtained by recognition, the method comprises the following steps:

Encoding the target classification to obtain a first code;

Respectively hashing the appearance image, the image acquisition mode and the characteristic attribute to obtain a corresponding first hash value, a corresponding second hash value and a corresponding third hash value;

reconstructing the first hash value, the second hash value and the third hash value to generate a hash value matrix;

And screening the hash value matrix based on the first code to generate an identification code for identifying the image which is acquired by the camera module subsequently.

The invention also provides a device for identifying the camera module based on artificial intelligence, which comprises:

the device comprises an acquisition unit, a characteristic attribute extraction unit and a control unit, wherein the acquisition unit is used for acquiring an appearance image of a camera module;

The system comprises an appearance image, an analysis unit, a third classification unit, a fourth classification unit, a fifth classification unit, a sixth classification unit and a fourth classification unit, wherein the appearance image is used for displaying an appearance image;

and the identification unit is used for carrying out fusion analysis on the first classification, the second classification and the third classification based on the artificial intelligent model, and identifying and obtaining the target classification of the camera module.

The invention also provides a computer device comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of any of the methods described above when the computer program is executed.

The invention also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the method of any of the preceding claims.

The method for identifying the camera module based on the artificial intelligence comprises the steps of collecting an appearance image of the camera module, obtaining an image collection mode of the camera module, obtaining a target image collected by the camera module, extracting characteristic attributes of the target image, analyzing first classification of the camera module based on the appearance image, analyzing second classification of the camera module based on the image collection mode, analyzing third classification of the camera module based on the characteristic attributes, and carrying out fusion analysis on the first classification, the second classification and the third classification based on an artificial intelligence model to identify the target classification of the camera module. In the invention, the appearance of the camera module, the image acquisition mode and the characteristic attribute of the target image are comprehensively considered, and the precise identification and classification of the camera module are realized through a fusion analysis technology.

Drawings

FIG. 1 is a schematic diagram of steps of a method for identifying a camera module based on artificial intelligence in an embodiment of the invention;

FIG. 2 is a block diagram of an apparatus for identifying camera modules based on artificial intelligence in accordance with an embodiment of the present invention;

Fig. 3 is a block diagram schematically illustrating a structure of a computer device according to an embodiment of the present invention.

The implementation, functional features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

Referring to fig. 1, in one embodiment of the present invention, a method for identifying a camera module based on artificial intelligence is provided, including the following steps:

step S1, acquiring an appearance image of a camera module, acquiring an image acquisition mode of the camera module, acquiring a target image acquired by the camera module, and extracting characteristic attributes of the target image;

step S2, analyzing the first classification of the camera module based on the appearance image, analyzing the second classification of the camera module based on the image acquisition mode, and analyzing the third classification of the camera module based on the characteristic attribute;

And step S3, carrying out fusion analysis on the first classification, the second classification and the third classification based on the artificial intelligence model, and identifying to obtain the target classification of the camera module.

The conventional method generally depends on appearance observation or simple image parameter analysis, and cannot comprehensively and accurately identify the type of the camera module. In the embodiment, through fusing an artificial intelligence technology and multidimensional analysis, misjudgment and missed judgment are reduced, and a large amount of camera module data can be rapidly processed.

As described in step S1 above, first, the camera module is photographed from a plurality of angles using the high resolution image capturing apparatus, ensuring that the appearance characteristics including shape, size, surface texture, etc. can be completely obtained. The collected images are subjected to preliminary preprocessing, such as noise removal, brightness adjustment, contrast adjustment and the like, so that the image quality is improved, and a clear data source is provided for subsequent analysis.

Meanwhile, the hardware configuration and the driving program of the camera module are analyzed, and the basic parameter settings of image acquisition, such as resolution, frame rate, exposure time and the like, are known. The data transmission mode and protocol of the camera module in actual work can be monitored through communication with the camera module, and the characteristics of the image acquisition mode of the camera module can be comprehensively mastered, wherein the data transmission mode and protocol comprise data format, transmission rate and the like.

Further, the actual image acquired by the camera module is subjected to deep analysis, and various characteristic attributes of the image are extracted by using an image processing algorithm. For example, analyzing color characteristics of an image, including color distribution, color contrast, color saturation, etc., analyzing texture characteristics of an image, such as roughness, direction, periodicity, etc., of textures, detecting edge characteristics of an image, including sharpness, number, shape, etc., of edges, and analyzing spatial distribution characteristics of an image, such as positional relationship of objects in an image, distribution density, etc.

As described in the above step S2, the collected camera module appearance image is compared with the pre-established appearance feature database by using the image recognition and pattern matching technology. According to the similarity of appearance characteristics, the camera modules are divided into different appearance categories, such as round camera modules, square camera modules, special-shape camera modules and the like. Meanwhile, the camera module can be further subdivided into a metal shell camera module, a plastic shell camera module and the like according to different surface materials. Meanwhile, the camera modules are classified according to the acquired image acquisition mode parameters and characteristics. For example, the camera module is divided into a high-frame-rate camera module, a middle-frame-rate camera module and a low-frame-rate camera module according to the frame rate, a high-definition camera module, a standard-definition camera module and the like according to the resolution, and the camera module can be divided into different functional categories according to the specificity of the acquisition mode, such as whether panoramic shooting, dynamic tracking and the like are supported.

Meanwhile, based on the extracted characteristic attribute of the target image, the camera module is classified by using data analysis and a machine learning algorithm. For example, the camera module is classified into a color-rich camera module, a color-single camera module and the like according to the color characteristic attribute, a texture-clear camera module, a texture-fuzzy camera module and the like according to the texture characteristic, and a corresponding classification such as an edge-sharp camera module, an object-distribution-uniform camera module and the like can be performed according to the edge characteristic and the spatial distribution characteristic.

As described in the above step S3, an artificial intelligence model capable of fusing multidimensional information is constructed in advance based on a deep learning algorithm such as a convolutional neural network, a cyclic neural network, or the like. The model is subjected to a large amount of training data learning and optimization, so that the results of the first classification, the second classification and the third classification can be accurately understood and processed, and comprehensive analysis and judgment can be performed according to the information.

In the present embodiment, the results of the first classification, the second classification, and the third classification are input as input data into the artificial intelligence model. The model performs fusion analysis on the classification results through calculation and logic processing. For example, each classification result is given a different weight by means of weighted fusion, and the importance of the final target classification is adjusted according to the weight. Meanwhile, the potential association and the characteristics between the classification results can be mined by applying the characteristic extraction and pattern recognition technology. After fusion analysis, the artificial intelligent model outputs a final target classification result. The result is a comprehensive assessment and classification of the camera module, including its type information. For example, the model can output that the camera module belongs to specific classification categories such as smart mobile phone special high definition digtal camera module, security protection is panoramic camera module, industrial automation is with high low temperature resistant camera module.

In the embodiment, three key aspects of the appearance, the image acquisition mode and the target image characteristic attribute of the camera module are comprehensively considered, the characteristics of the camera module are comprehensively reflected, and the accuracy and the reliability of identification are improved. By applying the artificial intelligence technology of deep learning to perform fusion analysis on multidimensional data, potential relations and features between the data can be automatically learned and mined, and more intelligent identification and classification are realized. Compared with the traditional method, the method can rapidly process a large amount of camera module data, provide more accurate classification results, and provide powerful support for the production, application and management of the camera module.

In an embodiment, the acquiring the image acquisition mode of the camera module includes:

In this embodiment, in the conventional method, the judgment of the image acquisition mode of the camera module is often based on a single factor, for example, only by means of driver information or simple signal observation, which results in inaccurate and incomplete judgment. The scheme aims at realizing accurate acquisition mode judgment by integrating multiple factors.

Specifically, first, a hardware detection device or a software tool is used to perform comprehensive scanning and analysis on a physical interface of the camera module. These detection tools can read detailed information of the electrical characteristics, physical structures, etc. of the interface. For example, for a common USB interface camera module, the detection device may identify key parameters such as the version of the interface (e.g. USB 2.0 or USB 3.0), the power supply capability of the interface, and the data transmission rate.

A database containing standard parameters of various common camera module hardware interfaces is pre-established. This database covers detailed specifications, features and corresponding acquisition mode category information for different interface types. And comparing the detected hardware interface information of the camera module with standard parameters in a database one by one. For example, if the detected data transmission rate of the interface is within a certain specific range and the physical structure of the interface conforms to the characteristics of a standard interface type, the connection interface type thereof can be preliminarily determined. And determining the acquisition mode category of the connection interface type of the camera module according to the comparison result. For example, some specific types of interfaces are primarily used for continuous image acquisition, while others are more suitable for specific modes of image acquisition (e.g., high-speed continuous shooting mode). And the information is arranged into interface association acquisition mode information, so that basic data is provided for subsequent comprehensive analysis.

Further, signal monitoring equipment and software are used for monitoring signal transmission of the camera module in a working state in real time. These devices are capable of capturing various characteristics of the signal, including frequency, amplitude, phase, etc. of the signal. For example, for a camera module adopting a wireless transmission mode, the signal monitoring device can monitor the transmission frequency of a wireless signal, the change rule of the signal intensity and the like. And analyzing the characteristics of the signal according to the monitored signal transmission mode and determining the related acquisition mode characteristics. For example, if the frequency of the signal is stable and within a specific range, it means that the camera module adopts an image acquisition mode with a fixed frame rate, and if the amplitude of the signal is changed greatly and has a certain regularity, it means that it is performing certain dynamic image acquisition (such as signal change in auto-focusing process). And sorting and summarizing the collection mode characteristics obtained by analysis to form signal mode collection mode information. The information describes the characteristics of the image acquisition mode of the camera module in terms of signal transmission.

Further, a driver parsing tool is used to perform deep analysis on the driver of the camera module. The tool can read various configuration parameters, instruction sets and key information related to image acquisition in the driver. For example, key information about parameters of image resolution setting, instruction set of frame rate control, image encoding mode, and the like in the driver is extracted. A database containing the correspondence between various driving characteristics and corresponding acquisition modes is pre-constructed. The database is built by analyzing the drivers of a large number of different camera modules and researching the actual collection mode. For example, a particular combination of driver configuration parameters corresponds to a particular image acquisition mode (e.g., night shooting mode). And searching and matching in the corresponding relation library according to the extracted driving characteristics. Acquisition mode information associated with the drive feature is determined. For example, if the extracted driver parameters show that high resolution image acquisition and fast auto-focus functions are supported, it can be inferred that their corresponding acquisition modes are modes suitable for high quality image capture.

And finally, a fusion algorithm which comprehensively considers the interface association acquisition mode information, the signal mode acquisition mode information and the driving characteristic association acquisition mode information is preconfigured. The algorithm can adopt a mode of combining a plurality of technical means such as weighted average, fuzzy logic judgment, neural network and the like. For example, each information source is assigned a different weight depending on its reliability and importance. And inputting the information of the acquisition modes of the three sources into a fusion algorithm for comprehensive analysis. And the algorithm performs fusion processing on the information according to preset rules and logic. For example, if the interface association acquisition mode information indicates that the camera module is suitable for a specific continuous image acquisition mode, and the signal mode acquisition mode information and the driving feature association acquisition mode information both have the corresponding feature support conclusion, the fusion algorithm can more certainly judge that the image acquisition mode of the camera module belongs to the continuous image acquisition mode. And outputting the finally determined image acquisition mode of the camera module through the processing of the fusion algorithm. The result is that the accurate judgment of the information in a plurality of aspects is integrated, and the detailed information of the type, the characteristics, the parameters and the like of the acquisition mode is included.

In the embodiment, the information of three key aspects of a hardware interface, a signal transmission mode and a driver is integrated to determine the image acquisition mode, so that the limitation of single information source is overcome, and the accuracy and the reliability of judgment are improved.

In an embodiment, the extracting the characteristic attribute of the target image includes:

In this embodiment, first, the target image is converted from the common RGB color space to the HS color space, because the HSV space is more consistent with the human perception of color, and can better separate hue, saturation and brightness information of the color. The conversion process is realized by using an image processing algorithm, so that the accuracy and the high efficiency of conversion are ensured. In the converted HSV color space, three channels of H (hue), S (saturation) and V (brightness) are respectively subjected to statistical analysis. The average value of each channel is calculated to reflect the average level of the color value of the channel, the variance reflects the discrete degree of the color value, namely the diversity of the color, and the peak value represents the color value with the highest occurrence frequency in the channel and the intensity thereof. Through the statistics, the characteristic distribution condition of the target image in terms of color can be comprehensively described.

Further, edge detection algorithms, such as Canny algorithm, sobel algorithm and the like, are applied to detect edges of the target image. First, the length of the edge, i.e., the total length of all edge segments in the image, is calculated reflecting the complexity and size of the object boundary. The density represents the number of edge line segments in a unit area and can be used for measuring the detail richness and texture complexity of objects in an image. The curvature describes the degree of curvature of the edge curve and is of great importance for identifying objects with specific shape characteristics, such as circles, arcs and the like.

Further, texture analysis algorithms, such as a gray level co-occurrence matrix method, a wavelet transformation method and the like, are adopted to carry out deep analysis on textures of the target image. The algorithm can capture the gray scale relation and the spatial distribution rule among pixels in the image, so that the texture characteristics of the image are revealed.

Wherein roughness represents the roughness or smoothness of the texture, as determined by calculating the gradient of variation in pixel gray values or counting the differences between pixels. The contrast reflects the degree of difference in gray values in the texture, i.e., the contrast intensity between the light and dark areas. The direction degree describes the main direction of the texture or trend of the lines of the texture, and can be obtained by calculating the correlation of the gray values of the pixels or using a direction filter.

Further, the target image is divided into a plurality of sub-blocks, and a method of uniform blocking or adaptive blocking according to image content can be adopted. The self-adaptive partitioning can be flexibly divided according to the distribution and structure of objects in the images, and is better suitable for different types of images. For each sub-block, its gray level co-occurrence matrix is calculated. The gray level co-occurrence matrix reflects the joint probability distribution of gray values in the image over a certain spatial distance and direction. And extracting the characteristics of energy, entropy, relativity and the like from the gray level co-occurrence matrix. The energy represents the square sum of matrix element values, reflects the uniformity and regularity of the image texture, the entropy represents the complexity and randomness of the image texture, and the correlation describes the similarity of the matrix elements in the row or column direction and reflects the linear characteristics of the image texture. And further, the features of the sub-blocks are integrated to obtain the spatial distribution feature attribute of the target image. A simple averaging method, a weighted averaging method or a more complex fusion algorithm can be adopted to reasonably synthesize according to the factors of the position, importance and the like of the sub-blocks.

And finally, fusing the color characteristic attribute, the edge characteristic attribute, the texture characteristic attribute and the spatial distribution characteristic attribute through a fusion algorithm. The method can be used for combining the feature vectors into a comprehensive feature vector according to a certain sequence, or can be used for distributing different weights for the feature vectors according to the importance of the features by using a weighting fusion mode, and then carrying out weighting summation. And a fusion algorithm can be realized by using programming, and the extracted characteristic attribute data are processed according to a fusion strategy. In the fusion process, the dimension consistency of the features and the compatibility of data types are required to be considered, so that the accuracy and the effectiveness of the fusion result are ensured. And obtaining the comprehensive characteristic attribute description of the target image after fusion processing. The characteristic attribute contains key information of the image in various aspects such as color, edge, texture, spatial distribution and the like, and can comprehensively and accurately reflect the characteristics of the target image.

In this embodiment, the feature extraction method of four key dimensions of color, edge, texture and spatial distribution is integrated, the characteristics of the target image are comprehensively described, and the limitation of single feature analysis is avoided.

In an embodiment, the analyzing the first classification of the camera module based on the appearance image, the analyzing the second classification of the camera module based on the image acquisition mode, and the analyzing the third classification of the camera module based on the characteristic attribute includes:

In this embodiment, the conventional camera module classification method often relies on manual experience or single feature judgment, and has low accuracy and efficiency. The scheme aims at realizing automatic and accurate classification by using an advanced technical means.

First, a database is established which contains a plurality of appearance images of different types of camera modules. The database should cover various common and special camera module appearance patterns, including images of different shapes, sizes, materials, surface textures and other features. And labeling and classifying each appearance image in the database in detail, and determining the type of the camera module to which the appearance image belongs. The image matching algorithm can be a feature point-based matching algorithm or a depth learning-based image recognition algorithm. And extracting key characteristic points of the appearance images of the camera modules to be classified or extracting advanced characteristic representations through a neural network. And comparing and matching the characteristics with the appearance image characteristics in the database one by one. And calculating the similarity degree between the appearance image of the camera module to be classified and various types of appearance images in the database according to the matching result. A similarity measure, such as euclidean distance, cosine similarity, etc., may be used to convert the similarity into a probability value. And obtaining probability distribution of each type of the camera module, wherein a plurality of types with the highest probability value are used as a first classification result.

Secondly, a database containing image acquisition mode characteristics of various camera modules is constructed in advance. And information such as resolution setting, frame rate range, acquisition modes (such as continuous acquisition, intermittent acquisition, trigger acquisition and the like) of different types of camera modules, signal transmission parameters and the like is recorded in detail. And carrying out detailed parameter extraction and feature analysis on the image acquisition mode of the camera module to be classified. And matching and comparing the characteristics with the image acquisition mode characteristics in the database. For example, whether the frame rate is within a typical frame rate range of a certain type, whether the acquisition mode matches a certain type, etc. Based on the result of the matching, the degree of similarity is calculated and converted into a probability value. And determining probability distribution of each type of the camera module in an image acquisition mode, thereby obtaining a second classification result.

Further, a database containing characteristic attribute data of various camera modules is pre-constructed. Such characteristic attributes include, but are not limited to, color feature attributes (e.g., color mean, variance, etc.), edge feature attributes (e.g., edge length, curvature, etc.), texture feature attributes (e.g., roughness, contrast, etc.), and spatial distribution feature attributes (e.g., sub-block feature complex values, etc.) of the target image. And carrying out statistics and analysis on the characteristic attribute of each type of camera module, and establishing a typical characteristic model of the camera module. And carrying out detailed quantization and description on the characteristic attribute extracted by the camera module to be classified. And matching and comparing the characteristic attributes with the characteristic attributes of various camera modules in the database. The degree of similarity with each type is evaluated by a classification algorithm or a similarity measurement method in machine learning. And calculating probability values according to the similarity degree to obtain probability distribution of each type of the camera module in the aspect of characteristic attribute. As a result of the third classification.

In the embodiment, three key aspects of appearance images, image acquisition modes and characteristic attributes are combined for classification, various features of the camera module are comprehensively considered, and classification accuracy and reliability are improved.

In an embodiment, the performing fusion analysis on the first classification, the second classification and the third classification based on the artificial intelligence model, and identifying the target classification of the camera module includes:

judging whether the reconstruction error is in a threshold range or not;

In the present embodiment, first, the dimensional structure of the multidimensional feature space is determined according to the characteristics and properties of the first classification, the second classification, and the third classification. For example, the first classification may be associated with one dimension axis to represent an appearance feature dimension of the camera module, the second classification may be associated with another dimension axis to represent a feature dimension of the image capturing manner, and the third classification may be associated with a third dimension axis to represent a feature attribute feature dimension. The scale and the value range on each dimension axis are set according to the specific characteristic value range of the corresponding classification. And mapping the result value of the first classification onto a corresponding dimension axis of the multidimensional feature space according to a certain rule to form a coordinate point on the dimension. For example, if the first classification is based on the shape classification of the appearance image, the circle corresponds to the value 1, and the square corresponds to the value 2, then when the first classification result of a certain camera module is a circle, the position marked as the value 1 on the dimension axis of the appearance feature is marked. Similarly, the results of the second classification and the third classification are mapped onto respective corresponding dimensional axes to form a representation of a complete feature vector point in the multidimensional feature space.

Furthermore, a self-encoder based on deep learning is employed as a core model. The self-encoder is an unsupervised learning model that automatically learns potential feature representations of the data. And designing and constructing a proper self-encoder structure according to the characteristics of the camera module data and the requirements of classification tasks. For example, a self-encoder of a multi-layer neural network structure may be selected, including an input layer, a hidden layer, and an output layer. The number of nodes of the input layer and the output layer is determined according to the dimension of the multidimensional feature space, and the structure and the number of nodes of the hidden layer are determined through experiments and optimization so as to realize effective extraction and compression of data features. Model training is performed using a number of known types of camera modules to point data on feature vector points in a multidimensional feature space. These data are input into the self-encoder, and the parameters of the self-encoder are continuously adjusted through a back-propagation algorithm, so that the self-encoder can learn the potential distribution pattern and rule of these data in the feature space. In the training process, various optimization algorithms and techniques, such as a random gradient descent method, self-adaptive learning rate adjustment and the like, can be adopted to improve training efficiency and model performance. Meanwhile, regularization methods, such as L1 regularization and L2 regularization, can be used to prevent the model from being over fitted.

And inputting the feature vector points of the camera module to be identified into the trained self-encoder. The self-encoder tries to reconstruct the input feature vector points according to the learned feature mapping relation, namely, the self-encoder decodes and restores the input feature vector points through the output layer after the self-encoder codes through the hidden layer. Suitable error calculation metrics such as Mean Square Error (MSE) or Mean Absolute Error (MAE) are preselected. Error values between the input feature vector points and the feature vector points reconstructed from the encoder are calculated. For example, for a mean square error, the squares of the differences in each dimension are calculated, then summed and averaged to obtain the final reconstructed error value. A reasonable reconstruction error threshold range is determined through reconstruction error analysis and experiments on a large number of camera modules of known types. This threshold range should be able to distinguish between a normal known type and a new type or an abnormal type. In an embodiment, the method can be adjusted and optimized according to different application scenes and requirements of classification accuracy.

If the reconstruction error is within the threshold range, the characteristics of the camera module are similar to the characteristic distribution mode of the known type. At this time, the nearest known type in the feature space is found as the target classification according to the feature mapping relation inside the self-encoder and the similarity measurement method with the known type. The similarity measure may be obtained by various methods, such as euclidean distance, cosine similarity, etc. And calculating similarity measurement values between the feature vector points of the camera module to be identified and the feature vector points of each known type, and selecting the known type with the smallest measurement value (namely the most similar type) as a target classification result. If the reconstruction error exceeds the threshold range, the characteristics of the camera module are greatly different from the known type, and the camera module may be of a new type or have an abnormal condition. At this point, the system determines a new type or an abnormal type and triggers a further manual intervention mechanism.

When the new type or the abnormal type is judged, the system sends out prompt information to inform related personnel to perform manual intervention. Meanwhile, the detailed information related to the camera module is automatically collected and arranged, wherein the detailed information comprises original data (such as an appearance image, an image acquisition mode parameter, a target image characteristic attribute and the like), specific numerical values of a reconstruction error, position information of feature vector points in a multidimensional feature space and the like. And the professional carries out detailed analysis and judgment on the camera module according to the information provided by the system. The method can determine whether the camera module is of a new type or not by using expert knowledge and experience and combining other auxiliary information (such as manufacturers of the camera module, application scenes and the like). If the model is of a new type, the model is classified and marked manually, and relevant information is added into a database for subsequent model learning and classification. If the type is abnormal, further analyzing the reasons of the abnormality, such as data acquisition errors, equipment faults and the like, and taking corresponding processing measures.

In an embodiment, after the fusion analysis is performed on the first classification, the second classification, and the third classification based on the artificial intelligence model, the identifying the target classification of the camera module includes:

Encoding the target classification to obtain a first code;

In this embodiment, after completing the target classification and identification of the camera module, the classification result needs to be encoded and stored, so as to facilitate subsequent data query, management and application. The lack of efficient coding and data association mechanisms in conventional approaches results in confusion and retrieval of data. The present solution solves this problem by means of encoding and hashing. For a large amount of image data collected by camera modules, a reliable identification method is needed to distinguish between images collected by different modules and different stages, so as to facilitate tracking, analysis and arrangement of image data. The existing identification method is not flexible or difficult to effectively correlate with the classification information of the camera module, and the scheme aims to establish an image identification system based on the classification information.

First, a set of encoding rules for target classification is formulated in advance. The coding rules can be designed according to the hierarchical structure, type characteristics and application requirements of the classification. For example, numerical coding, alphabetical coding, or hybrid coding may be employed. If the classification of the camera module includes different application fields (such as security, medical treatment, traffic, etc.) and different performance levels (such as high, medium, low), a coding system including field codes and performance level codes can be designed. The coding rules should be unique and scalable to accommodate new classifications and refined classifications that may occur in the future. And according to the designed coding rule, converting the target classification of the camera module obtained by recognition into a corresponding code, namely a first code. The code may be a short string or sequence of numbers for ease of storage and processing. For example, if a camera module is identified as a high performance type in the security domain, a code similar to "a18C977" may be generated according to the coding rules.

Then, a hash algorithm, such as MD5, SHA-256, etc., is selected to perform hash calculation on the appearance image of the camera module. The hash algorithm converts the apparent image data into a hash value of a fixed length, i.e., a first hash value. This hash value is unique and even if there is a small change in the appearance image, a different hash value is generated. Therefore, the difference between different appearance images can be quickly identified and compared, and the appearance images can be stored and retrieved in the form of hash values, so that the storage space is saved and the processing efficiency is improved. And similarly, carrying out hash calculation on the related data of the image acquisition mode of the camera module by using the selected hash algorithm. The data includes information such as resolution setting, frame rate parameters, acquisition modes, etc. And converting the information into a second hash value for uniquely identifying the image acquisition mode characteristics of the camera module. Similarly, the characteristic attribute data of the camera module is hashed, including color feature attributes, edge feature attributes, texture feature attributes, spatial distribution feature attributes and the like of the target image. The complex data is converted into a compact third hash value through a hash algorithm so as to facilitate subsequent data processing and association.

The matrix can be two-dimensional or multidimensional, and the specific dimension can be adjusted according to the complexity degree and the data volume of the camera module. For example, a two-dimensional matrix may be designed in which a first row is used to store fragments of a first hash value, a second row is used to store fragments of a second hash value, and a third row is used to store fragments of a third hash value. Each hash value may be partitioned into multiple segments according to certain rules to facilitate reasonable distribution and storage in the matrix. And reconstructing the first hash value, the second hash value and the third hash value according to the designed matrix structure. For example, the first hash value can be divided into a plurality of equal-length sub-segments and sequentially filled into corresponding positions of the first row of the matrix, and the second hash value and the third hash value can be filled into the second row and the third row of the matrix in the same way. During the reconstruction process, some check information or padding characters can be added to ensure the integrity and consistency of the matrix.

An association relationship between the first code and the hash value matrix is established in advance. This may be accomplished by adding pointers, indexes, or other associated fields in the data structure. When the image which is acquired by the camera module subsequently needs to be identified, a corresponding hash value matrix can be quickly found according to the first code.

A screening rule based on the first code is preconfigured for extracting key information from the hash value matrix to generate an identification code. The screening rules can be customized according to application requirements and data characteristics. For example, some hash value segments at specific positions in the matrix can be selected for combination according to the specific bit values of the first code, or the hash values in the matrix can be weighted according to the classification information of the code, so as to obtain a comprehensive identification code. And extracting relevant information from the hash value matrix according to the screening rule, and combining, converting or calculating to generate an identification code for identifying the subsequent acquired image of the camera module. The identification code may be a combination of numbers, letters or symbols, with uniqueness and recognizability. For example, in one embodiment, if the first encoding represents a particular type of camera module, the information extracted from the hash matrix may be combined and converted to generate an identification code similar to "CAM-001-A18C977", where "CAM-001" represents the serial number of the camera module and "A18C977" represents its target classification encoding.

In this embodiment, an efficient data management and association mechanism is established by encoding the target class and associating it with the hash value matrix. Corresponding camera module data and image identification information can be conveniently and rapidly found according to codes, and the data retrieval and application efficiency is improved. The method for reconstructing the generation matrix by using the hash values innovatively organically integrates the hash values of different types to form a unified data structure. The matrix structure not only can save storage space, but also can conveniently compare, analyze and process data, and provides abundant information sources for subsequent image identification.

In an embodiment, the reconstructing the first hash value, the second hash value, and the third hash value to generate a hash value matrix includes:

Extracting the digital characters in the first hash value, and adding the digital characters into a matrix of three rows and three columns to obtain a first matrix;

extracting the digital characters in the second hash value, and adding the digital characters into a matrix of three rows and three columns to obtain a second matrix;

Extracting English characters in the third hash value, and adding the English characters into a matrix of three rows and three columns to obtain a third matrix;

Adding the digital characters at the corresponding positions in the first matrix and the second matrix to obtain a fourth matrix;

multiplying the third matrix with the fourth matrix to obtain the hash value matrix.

In the present embodiment, the digital characters therein are accurately recognized and extracted for the first hash value. The digital characters represent certain feature codes or feature values associated with the camera module look image. For example, the size parameter of the appearance image after the specific processing, the digital part in the color coding, and the like may be mentioned. According to a preset rule and sequence, the extracted digital characters are sequentially added into an initial blank matrix with three rows and three columns to form a first matrix. For example, the locations of the matrix may be filled one by one in a left to right, top to bottom order.

The digital characters in the second hash value are extracted using a method similar to the first hash value. And adding the extracted digital characters into another blank matrix with three rows and three columns according to the same rule to obtain a second matrix.

And for the third hash value, identifying and extracting English characters from the third hash value. And adding the extracted English characters into a blank matrix of three rows and three columns according to a preset sequence and a preset rule to form a third matrix. For example, the filling may be performed according to the importance of the characteristic attribute represented by the english character or some preset classification order.

In order to further fuse the information in the first matrix and the second matrix, a simple and efficient addition operation is devised. And adding the digital characters at the same positions in the first matrix and the second matrix according to the corresponding positions of the matrices. For example, if the number at a certain position of the first matrix is "5", and the number at a corresponding position of the second matrix is "3", then "8" is obtained at a corresponding position of the fourth matrix. If the added result exceeds a certain numerical range (for example, exceeds a representable range of digital characters or a reasonable range set according to practical application requirements), modulo operation or other suitable processing modes can be adopted to ensure that the result is within a reasonable range and can accurately reflect the fusion of information.

And adding all corresponding positions in the first matrix and the second matrix one by one, and obtaining a new three-row three-column matrix, namely a fourth matrix after finishing. The matrix fuses digital information related to the appearance image and the image acquisition mode of the camera module.

Finally, the third matrix and the fourth matrix are multiplied. For example, the multiplication may be directly performed by a matrix-based operation method, or a multiplication rule may be defined based on a combination of character encoding and digital operation. For each element at the corresponding position, the english characters in the third matrix are converted into digital values according to a preset transcoding rule (for example, corresponding digital values are given according to the order of the letters in the alphabet), and then multiplication is performed with the numbers at the corresponding positions in the fourth matrix.

After multiplication operation, a new three-row three-column matrix is obtained, namely a final hash value matrix. The matrix contains fusion information from the first hash value, the second hash value and the third hash value, so that fusion of digital characteristics is reflected, and characteristic attribute information represented by English characters is also contained. The camera module has unique structure and information representation capability, can be used as a comprehensive characteristic representation of the camera module, and is used for subsequent applications such as data analysis, identification generation, safety management and the like.

In this embodiment, by adding the first matrix and the second matrix and multiplying the first matrix and the third matrix, multi-level fusion of hash value information of different sources is realized, a hash value matrix with rich connotation and unique structure is formed, and a more comprehensive and accurate information basis is provided for subsequent applications. The generated hash value matrix can be flexibly interpreted and processed according to different application requirements. For example, in the aspect of image identification, a unique identification code can be generated according to characters or numbers at specific positions in a matrix, encryption and decryption operations can be performed by utilizing the complexity of the matrix in the field of data security, and in the data analysis, the characteristics and the performance of the camera module can be deeply known by analyzing the characteristics and the modes of the matrix.

In an embodiment, the screening the hash value matrix based on the first code, generating an identification code includes:

Sequentially adding characters in the first code into a matrix one by one to generate a code matrix, wherein the number of rows and the number of columns of the code matrix are smaller than those of the hash value matrix;

the coding matrix is overlapped to the upper left corner of the hash value matrix, and the number meeting the preset condition in the overlapped position of the coding matrix and the hash value matrix is calculated to be used as the target number, so that the preset condition is that the types of characters at the corresponding positions are the same;

Sequentially translating the coding matrixes, when the target quantity reaches the maximum for the first time, acquiring a region overlapped with the coding matrixes in the current hash value matrix as a target region, and combining characters in the target region to obtain a target character combination;

and processing the target character combination according to rules to obtain the identification code.

In this embodiment, the conventional method for generating the identification code often lacks close association with the core classification information (i.e., the first code) of the camera module, so that it is difficult for the identification code to directly reflect the key features and classification conditions of the camera module. According to the scheme, the first code and the hash value matrix are organically combined, so that accurate identification generation based on the characteristics of the camera module is achieved. According to the scheme, the identification codes with high differentiation can be rapidly generated through an innovative matrix superposition and screening mechanism.

Specifically, characters are first extracted one by one from the first code. The first code is typically a code sequence with a specific meaning generated according to the target classification of the camera module, which contains character representations of key information about the type, performance, etc. of the camera module. For example, a35CT682S734.

According to a preset arrangement rule, the extracted characters are orderly arranged in a new matrix to generate a coding matrix. For example, characters may be padded in a small matrix with rows and columns smaller than the hash matrix in a left-to-right, top-to-bottom order. The size of the small matrix is designed to allow for efficient interaction with the hash matrix without excessive computation resources and memory space.

Then, a corresponding data structure is allocated to each position of the coding matrix for storing character information and intermediate data and marking information required in a subsequent calculation process. For example, a data structure containing the character itself may be provided for each location.

Secondly, the generated encoding matrix is accurately superimposed and placed at the upper left corner of the hash value matrix. This is the initial superimposed position, followed by a translation operation as required. At this time, each position of the encoding matrix and the corresponding position of the hash value matrix form a correspondence. The preset condition is defined as the same type of the characters on the corresponding positions. The character types herein may be classified according to practical application requirements, for example, digital types, letter types (further subdivided into uppercase letters and lowercase letters), special symbol types, and the like. The types of the corresponding characters on the superposition positions of the coding matrix and the hash value matrix are compared one by one. If the types are the same, the matching count of the position is increased by one, and the position information meeting the preset condition is recorded. And counting the number of all the positions meeting the condition as the target number.

And sequentially translating the coding matrix on the hash value matrix according to a preset translation rule. For example, one unit location (e.g., width or height of one matrix location) may be translated at a time in a left-to-right, top-to-bottom order. In the translation process, the character type comparison and the target quantity statistics steps are repeated continuously, and the value of the target quantity is updated in real time. The change in the target number is continuously monitored. When the target number reaches the maximum value for the first time, the position of the coding matrix on the hash value matrix is recorded. And determining the region overlapped with the coding matrix in the hash value matrix as a target region. The target area contains the hash value matrix part with the maximum matching degree and correlation with the first code, and is a key information source area for generating the identification code.

All characters are extracted from the target area and combined according to the position sequence of the characters in the target area to form a target character combination. And further processing the target character combination according to a pre-designed rule to obtain a final identification code. The processing rules may include, but are not limited to, the following:

Character transcoding, namely, transcoding the characters in the target character combination, such as converting letters into corresponding numerical codes, or converting special symbols into specific character sequences according to mapping rules.

The encryption algorithm is applied to encrypt the target character combination by adopting an encryption algorithm so as to enhance the security and the uniqueness of the identification code. For example, each character in the target character combination may be shifted by a certain offset using a simple Kaiser password encryption method.

Check bit addition to ensure the accuracy and integrity of the identification code, a check bit may be added after the target character combination is processed. The check bits may be obtained by performing some calculation (e.g., hash calculation, parity calculation, etc.) on the processed character and appended to the end of the identification code.

And (3) adjusting the length, namely adjusting the length of the generated identification code according to the actual application requirement. The length of the identification code may be adapted to a particular standard or specification by truncating or filling in the particular character.

In the embodiment, through converting the first code into the code matrix and performing interactive operation with the hash value matrix, accurate screening and identification generation based on camera module classification information are realized. The combination mode fully utilizes the encoded semantic information and the structural information of the matrix, and provides rich data sources and unique processing methods for generating the identification codes. By adopting a translation encoding matrix method, searching the best matching area on the hash value matrix, and the dynamic processing mode improves the flexibility and adaptability of generating the identification code. The most representative target area can be automatically found according to different first codes and hash value matrix distribution conditions, so that the identification codes with high differentiation are generated. The whole technical scheme designs an efficient identification code generation flow from the generation of the coding matrix to the determination of the target area and then to the processing of the final identification code, and each step is carefully designed and optimized to ensure that the identification code can be quickly and accurately generated when a large amount of camera module data are processed, thereby improving the efficiency of data processing and the performance of a system.

Referring to fig. 2, in another embodiment of the present invention, there is further provided an apparatus for identifying a camera module based on artificial intelligence, including:

In this embodiment, for specific implementation of each unit in the above embodiment of the apparatus, please refer to the description in the above embodiment of the method, and no further description is given here.

Referring to fig. 3, in an embodiment of the present invention, there is further provided a computer device, which may be a server, and an internal structure thereof may be as shown in fig. 3. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected by a system bus. Wherein the computer is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used for communicating with an external terminal through a network connection. Which computer program, when being executed by a processor, carries out the above-mentioned method.

It will be appreciated by those skilled in the art that the architecture shown in fig. 3 is merely a block diagram of a portion of the architecture in connection with the present inventive arrangements and is not intended to limit the computer devices to which the present inventive arrangements are applicable.

An embodiment of the present invention also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the above method. It is understood that the computer readable storage medium in this embodiment may be a volatile readable storage medium or a nonvolatile readable storage medium.

In summary, the method for identifying the camera module based on the artificial intelligence provided by the embodiment of the invention comprises the steps of collecting an appearance image of the camera module, obtaining an image collection mode of the camera module, obtaining a target image collected by the camera module, extracting characteristic attributes of the target image, analyzing a first classification of the camera module based on the appearance image, analyzing a second classification of the camera module based on the image collection mode, analyzing a third classification of the camera module based on the characteristic attributes, and performing fusion analysis on the first classification, the second classification and the third classification based on an artificial intelligence model to identify the target classification of the camera module. In the invention, the appearance of the camera module, the image acquisition mode and the characteristic attribute of the target image are comprehensively considered, and the precise identification and classification of the camera module are realized through a fusion analysis technology.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium provided by the present invention and used in embodiments may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (SYNCHLINK) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising one does not exclude the presence of other like elements in a process, apparatus, article, or method that comprises the element.

The foregoing description is only of the preferred embodiments of the present invention and is not intended to limit the scope of the invention, and all equivalent structures or equivalent processes using the descriptions and drawings of the present invention or direct or indirect application in other related technical fields are included in the scope of the present invention.

Claims

1. A method for identifying a camera module based on artificial intelligence, characterized in that it comprises the following steps:

Acquire the appearance image of the camera module; obtain the image acquisition method of the camera module; obtain the target image acquired by the camera module, and extract the characteristic attributes of the target image;

Based on the appearance image, analyzing the first classification of the camera module; based on the image acquisition method, analyzing the second classification of the camera module; based on the characteristic attributes, analyzing the third classification of the camera module;

The first classification, the second classification and the third classification are fused and analyzed based on the artificial intelligence model to identify the target classification of the camera module.

2. The method for identifying a camera module based on artificial intelligence according to claim 1, wherein the image acquisition method of the camera module comprises:

Detecting the hardware interface information of the camera module, comparing it with preset standard interface parameters, determining the acquisition mode category corresponding to the connection interface type, and obtaining interface-related acquisition mode information;

Analyze the signal transmission mode of the camera module in the working state, determine the relevant acquisition mode characteristics according to the signal transmission mode, and obtain signal mode acquisition mode information;

Parse the driver of the camera module, extract the driver features including the configuration parameters and instruction sets of image acquisition, and determine the driver feature associated acquisition mode information according to the preset correspondence library between the driver features and the acquisition mode;

The image acquisition method of the camera module is obtained by combining the interface-related acquisition method information, the signal mode acquisition method information and the drive feature-related acquisition method information and applying a preset fusion algorithm.

3. The method for recognizing a camera module based on artificial intelligence according to claim 1, wherein extracting the characteristic attributes of the target image comprises:

Performing color space conversion processing on the target image, converting it from RGB color space to HSV color space, and calculating the mean, variance, and peak value of each channel in the converted color space to obtain color feature attributes;

Using an edge detection algorithm to extract edges of the target image, calculate the length, density, and curvature of the edges, and obtain edge feature attributes;

Using a texture analysis algorithm to analyze the texture of the target image, extracting the roughness, contrast, and orientation of the texture, and obtaining texture feature attributes;

Divide the target image into multiple sub-blocks, calculate the gray level co-occurrence matrix of each sub-block, extract energy, entropy, and correlation features from them, and obtain the spatial distribution feature attributes of the target image by combining the features of each sub-block;

The color feature attributes, edge feature attributes, texture feature attributes and space distribution feature attributes are fused to obtain characteristic attributes of the target image.

4. The method for identifying camera modules based on artificial intelligence according to claim 1, characterized in that the first classification of the camera module is analyzed based on the appearance image; the second classification of the camera module is analyzed based on the image acquisition method; and the third classification of the camera module is analyzed based on the characteristic attributes, including:

Matching the appearance image with appearance images of various types of camera modules stored in a database to obtain the probability that the camera module is of various types as a first classification;

Matching the image acquisition mode with the image acquisition modes of various types of camera modules stored in the database to obtain the probability that the camera module is of various types as the second classification;

The characteristic attributes are matched with the characteristic attributes of each type of camera module stored in the database to obtain the probability that the camera module is of each type as the third classification.

5. The method for identifying camera modules based on artificial intelligence according to claim 1, characterized in that the first classification, the second classification and the third classification are fused and analyzed based on the artificial intelligence model to identify the target classification of the camera module, comprising:

Constructing a multidimensional feature space, mapping the first category, the second category, and the third category to different dimensional axes of the multidimensional feature space to form a feature vector point;

Obtaining an artificial intelligence model; wherein the artificial intelligence model is an autoencoder based on deep learning, which pre-learns the distribution of feature vector points of a known type of camera module in a multidimensional feature space to obtain potential distribution patterns and laws of the feature space;

Inputting the feature vector points of the camera module into the artificial intelligence model, and calculating the reconstruction error of the feature vector points in the feature space based on the autoencoder;

Determining whether the reconstruction error is within a threshold range;

If the reconstruction error is within the threshold range, the closest known type is found as the target classification based on the feature mapping relationship inside the autoencoder and the similarity measurement with the known type;

If the reconstruction error exceeds the threshold range, it is judged as a new type or an abnormal type, and further manual intervention mechanism is triggered.

6. The method for identifying camera modules based on artificial intelligence according to claim 1, characterized in that after the first classification, the second classification and the third classification are fused and analyzed based on the artificial intelligence model and the target classification of the camera module is identified, the method further comprises:

Encoding the target classification to obtain a first code;

Performing hash operations on the appearance image, the image acquisition method, and the characteristic attributes to obtain corresponding first hash values, second hash values, and third hash values;

Reconstructing the first hash value, the second hash value, and the third hash value to generate a hash value matrix;

The hash value matrix is screened based on the first code to generate an identification code for identifying images subsequently captured by the camera module.

7. A device for identifying camera modules based on artificial intelligence, comprising:

An acquisition unit is used to acquire an appearance image of the camera module; acquire an image acquisition method of the camera module; acquire a target image acquired by the camera module, and extract characteristic attributes of the target image;

An analysis unit, configured to analyze a first classification of the camera module based on the appearance image; analyze a second classification of the camera module based on the image acquisition method; and analyze a third classification of the camera module based on the characteristic attribute;

An identification unit is used to perform a fusion analysis on the first classification, the second classification and the third classification based on an artificial intelligence model to identify the target classification of the camera module.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein the processor implements the steps of the method according to any one of claims 1 to 6 when executing the computer program.