Image recognition method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the field of artificial intelligence and the field of financial science and technology, and relates to an artificial intelligence-based image recognition method, which comprises the steps of preprocessing initial image data to obtain image training data; performing principal component analysis and sparse processing on the image training data based on the feature extraction network to obtain a target projection matrix; performing convolution operation on the image training data through the target projection matrix to obtain a first mapping feature; performing feature screening on the first mapping features based on a feature selection algorithm to obtain target mapping features; training the classification model based on the target mapping features to obtain a target image recognition model; and identifying the to-be-identified image based on the target image identification model and the target projection matrix, and generating a target image identification result. The invention further provides an image recognition device based on artificial intelligence, computer equipment and a storage medium. Based on the use of the feature extraction network and the feature selection algorithm, the training speed of the image recognition model is effectively improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and financial technology, and in particular to an image recognition method, apparatus, computer equipment and storage medium based on artificial intelligence. Background Art

[0002] Image recognition refers to the technology of identifying the content in a picture. It can be used in many fields such as face recognition, object recognition, scene recognition, etc. Through image recognition technology, the computer can automatically identify the shape, size, color and other features of the object from the picture, thereby realizing automatic classification, detection, recognition and other operations on the content in the picture.

[0003] In recent years, with the increasing popularity of artificial intelligence, the field of image recognition has gradually attracted people's attention, and image recognition technology has also been widely used in many financial businesses in financial enterprises that require image recognition, such as image recognition of insurance policy data and insurance bill data.

[0004] At present, many financial enterprises use PCANet, a deep learning network based on principal component analysis, to perform image recognition processing. However, principal component analysis is based on one-dimensional vectors, which means that the two-dimensional matrix must first be converted into a one-dimensional vector before its principal component is calculated. However, connecting a two-dimensional matrix into a one-dimensional vector usually results in a high-dimensional vector space and a large amount of storage space requirements, and calculating the eigenvector is very time-consuming. Therefore, the existing deep learning network based on principal component analysis has the problem of low training speed. Summary of the invention

[0005] The purpose of the embodiments of the present application is to propose an image recognition method, device, computer equipment and storage medium based on artificial intelligence to solve the technical problem of low training speed in the existing image recognition processing method based on principal component analysis deep learning network.

[0006] In order to solve the above technical problems, the embodiment of the present application provides an image recognition method based on artificial intelligence, which adopts the following technical solutions:

[0007] Acquire pre-constructed initial image data; wherein the initial image data is a data set that has been annotated with category labels;

[0008] Preprocessing the initial image data to obtain corresponding image training data;

[0009] Based on a preset feature extraction network, the image training data is subjected to principal component analysis and sparse processing to obtain a corresponding target projection matrix; wherein the feature extraction network is a deep subspace network based on two-dimensional principal component analysis and sparse processing;

[0010] Performing a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature;

[0011] Performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature;

[0012] Training a preset classification model based on the target mapping features to obtain a trained target image recognition model;

[0013] The image to be recognized is recognized based on the target image recognition model and the target projection matrix to generate a target image recognition result corresponding to the image to be recognized.

[0014] Furthermore, the step of performing principal component analysis and sparse processing on the image training data based on a preset feature extraction network to obtain a corresponding target projection matrix specifically includes:

[0015] Performing principal component extraction processing on the image training data based on the feature extraction network to obtain a corresponding projection matrix;

[0016] Construct sparse regularization terms based on the preset L1 norm and L2 norm;

[0017] Performing sparse processing on the projection matrix based on the sparse regularization term to obtain a corresponding sparse projection matrix;

[0018] The sparse projection matrix is ​​used as the target projection matrix.

[0019] Furthermore, the step of performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature specifically includes:

[0020] Calculating a first importance value of the first mapping feature based on a preset first feature selection algorithm; and,

[0021] Calculating a second importance value of the first mapping feature based on a preset second feature selection algorithm;

[0022] generating a comprehensive importance value of the first mapping feature based on the first importance value and the second importance value;

[0023] Sorting all the first mapping features in descending order according to the comprehensive importance values ​​to obtain a corresponding sorting list;

[0024] According to the sorting, a specified number of second mapping features are sequentially selected from the sorted list;

[0025] The second mapping feature is used as the target mapping feature.

[0026] Furthermore, the step of performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature specifically includes:

[0027] Calculating a third importance value of the first mapping feature based on a preset third feature selection algorithm; and,

[0028] Calculating a fourth importance value of the first mapping feature based on a preset fourth feature selection algorithm;

[0029] Filtering out third mapping features whose importance values ​​are greater than a first preset threshold from the first mapping features; and,

[0030] Selecting a fourth mapping feature whose importance value is greater than a second preset threshold from the first mapping features;

[0031] The target mapping feature is generated based on the third mapping feature and the fourth mapping feature.

[0032] Further, the step of generating the target mapping feature based on the third mapping feature and the fourth mapping feature specifically includes:

[0033] Integrate the third mapping feature and the fourth mapping feature to obtain a corresponding mapping feature set;

[0034] Filter out a fifth mapping feature that appears repeatedly from the mapping feature set;

[0035] The fifth mapping feature is used as the target mapping feature.

[0036] Furthermore, the step of performing recognition processing on the image to be recognized based on the target image recognition model and the target projection matrix to generate a target image recognition result corresponding to the image to be recognized specifically includes:

[0037] Preprocessing the image to be identified to obtain a corresponding designated image;

[0038] Performing a convolution operation on the designated image through the target projection matrix to obtain a corresponding sixth mapping feature;

[0039] Performing recognition processing on the sixth mapping feature based on the target image recognition model to obtain a corresponding designated image recognition result;

[0040] The designated image recognition result is used as the target image recognition result.

[0041] Furthermore, the step of preprocessing the initial image data to obtain corresponding image training data specifically includes:

[0042] Performing image resizing processing on the initial image data to obtain first image data;

[0043] Performing grayscale processing on the first image data to obtain second image data;

[0044] Performing histogram equalization processing on the second image data to obtain third image data;

[0045] The third image data is used as the image training data.

[0046] In order to solve the above technical problems, the embodiment of the present application further provides an image recognition device based on artificial intelligence, which adopts the following technical solution:

[0047] An acquisition module, used to acquire pre-constructed initial image data; wherein the initial image data is a data set that has been annotated with category labels;

[0048] A preprocessing module, used for preprocessing the initial image data to obtain corresponding image training data;

[0049] A processing module, used to perform principal component analysis and sparse processing on the image training data based on a preset feature extraction network to obtain a corresponding target projection matrix; wherein the feature extraction network is a deep subspace network based on two-dimensional principal component analysis and sparse processing;

[0050] An operation module, configured to perform a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature;

[0051] A screening module, used to perform feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature;

[0052] A training module, used to train a preset classification model based on the target mapping features to obtain a trained target image recognition model;

[0053] The recognition module is used to perform recognition processing on the image to be recognized based on the target image recognition model and the target projection matrix, and generate a target image recognition result corresponding to the image to be recognized.

[0054] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0055] The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the artificial intelligence-based image recognition methods proposed in the embodiments of the present application are implemented.

[0056] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0057] The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of any one of the artificial intelligence-based image recognition methods proposed in the embodiments of the present application are implemented.

[0058] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0059] The present application first obtains pre-constructed initial image data; then pre-processes the initial image data to obtain corresponding image training data; then performs principal component analysis and sparse processing on the image training data based on a preset feature extraction network to obtain a corresponding target projection matrix; wherein the feature extraction network is a deep subspace network based on two-dimensional principal component analysis and sparse processing; subsequently, a convolution operation is performed on the image training data through the target projection matrix to obtain a corresponding first mapping feature; further, based on a preset feature selection algorithm, feature screening is performed on the first mapping feature to obtain a corresponding target mapping feature; and based on the target mapping feature, a preset classification model is trained to obtain a trained target image recognition model; finally, the image to be recognized is recognized based on the target image recognition model and the target projection matrix to generate a target image recognition result corresponding to the image to be recognized. After preprocessing the pre-constructed initial image data to obtain image training data, the present application performs principal component analysis and sparse processing on the image training data using a feature extraction network based on two-dimensional principal component analysis and sparse processing to obtain a target projection matrix, and performs a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature. Then, the target mapping feature obtained by performing feature screening processing on the first mapping feature based on a feature selection algorithm is used to train the preset classification model to obtain a target image recognition model, and finally the target image recognition model and the target projection matrix are used to perform recognition processing on the image to be recognized to generate a corresponding target image recognition result. The present application performs principal component analysis and sparse processing on the image training data by using a deep subspace network based on two-dimensional principal component analysis and sparse processing, so that the obtained target projection matrix has a smaller dimension. Compared with the principal components obtained by the traditional principal component analysis algorithm, the feature extraction network has less overhead in storage and feature calculation. In addition, based on the use of feature selection algorithm, the first mapping features obtained will be further screened to retain the target mapping features that are most discriminative and important for the image classification task, so that when the target mapping features with smaller data dimensions are subsequently used to train the image recognition model, the workload of model training can be further reduced, thereby effectively improving the training speed of the image recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0061] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0062] Figure 2 A flowchart of an embodiment of an artificial intelligence-based image recognition method according to the present application;

[0063] Figure 3 is a schematic structural diagram of an embodiment of an artificial intelligence-based image recognition device according to the present application;

[0064] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0066] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0067] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0068] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0069] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0070] The terminal devices 101, 102, and 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, and the like.

[0071] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0072] It should be noted that the artificial intelligence-based image recognition method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based image recognition device is generally set in the server / terminal device.

[0073] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0074] Continue to refer Figure 2 , showing a flow chart of an embodiment of an artificial intelligence-based image recognition method according to the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. The artificial intelligence-based image recognition method provided in the embodiment of the present application can be applied to any scenario where image recognition is required, and the artificial intelligence-based image recognition method can be applied to products in these scenarios, for example, the recognition and processing of financial image data in the field of finance and insurance. Financial image data may include policy data, insurance bill data, and the like. The artificial intelligence-based image recognition method comprises the following steps:

[0075] Step S201, obtaining pre-constructed initial image data; wherein the initial image data is a data set that has been annotated with category labels.

[0076] In this embodiment, the electronic device (eg, Figure 1 The server / terminal device shown in the figure) can obtain the initial image data through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, Wi-Fi connection, Bluetooth connection, Wi MAX connection, Zigbee connection, UWB (Ultra Wi-Fi Band) connection, and other wireless connection methods currently known or to be developed in the future. The above-mentioned initial image data is an image data set containing category labels of the images. Each image contained in the image data set is associated with its corresponding category label. The above-mentioned labeling processing of identity information can be carried out by manual labeling or automatic machine labeling.

[0077] Step S202: pre-process the initial image data to obtain corresponding image training data.

[0078] In this embodiment, the specific implementation process of preprocessing the initial image data to obtain the corresponding image training data will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0079] Step S203, performing principal component analysis and sparse processing on the image training data based on a preset feature extraction network to obtain a corresponding target projection matrix.

[0080] In the present embodiment, the above-mentioned feature extraction network based on the preset performs principal component analysis and sparse processing on the image training data to obtain the specific implementation process of the corresponding target projection matrix. This application will further describe this in detail in the subsequent specific embodiments, and will not be elaborated on here. In addition, the feature extraction network is a deep subspace network based on two-dimensional principal component analysis and sparse processing. The above-mentioned feature extraction network performs sparse solution on the original principal components of the image training data through an elastic network, and adjusts the principal component load with less influence to zero to achieve the effect of sparseness. While reducing the number of principal components, it improves the accuracy of data dimensionality reduction and the robustness of the feature extraction network.

[0081] Principal Component Analysis (PCA) is a commonly used unsupervised dimensionality reduction algorithm that aims to transform high-dimensional data into a low-dimensional space through linear transformation while retaining as much information as possible from the original data. The PCA algorithm used to learn filters is based on one-dimensional vectors, which means that the two-dimensional matrix must first be converted into a one-dimensional vector before its principal components are calculated. However, concatenating a two-dimensional matrix into a one-dimensional vector usually results in a high-dimensional vector space and a large amount of storage space requirements, and calculating the eigenvectors is very time-consuming. To make matters worse, the inherent two-dimensional structure of the image matrix is ​​removed.

[0082] In this embodiment, the feature extraction network has the same overall structure as PCANet (Principal Component Analysis Network), except that the feature extraction network uses the 2DPCA algorithm to replace the PCA algorithm in PCANet to perform principal component extraction processing, and after the 2D PCA algorithm learns the principal components, the step of sparseness of the principal components is added, and the constraints combined with L1 and L2 norms are used to sparse the principal components extracted by deep subspace learning, remove redundant parts, highlight the main factors, make the projection matrix more sparse and have stronger learning ability, and achieve a better balance between calculation intensity and accuracy. Among them, the target projection matrix obtained by sparseness processing retains the most important feature information and removes minor features. When learning the subspace, the feature extraction network can focus more on learning the most discriminative and representative features, improving learning ability and generalization performance. By enhancing the expressive power of features, it can better adapt to complex data distribution and changes. In addition, the sparse principal components have smaller dimensions, and compared with the principal components obtained by the traditional PCA algorithm, the feature extraction network has less overhead in storage and calculation. When processing large-scale data sets or real-time applications, feature extraction networks have higher efficiency and scalability. In addition, sparse principal components are more robust to noise and outliers. Since redundant information is removed, sparse principal components are more focused on the real data distribution, reducing the interference of noise and outliers. This makes the feature extraction network more robust when facing complex data scenarios.

[0083] Step S204: performing a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature.

[0084] In this embodiment, by using the sparsely processed target projection matrix to perform a convolution operation on the image training data, mapping features of the image training data can be extracted. These mapping features capture local texture and feature information in the image training data.

[0085] Step S205: performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature.

[0086] In this embodiment, the above-mentioned specific implementation process of performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain the corresponding target mapping feature will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0087] Step S206: training a preset classification model based on the target mapping features to obtain a trained target image recognition model.

[0088] In this embodiment, the above classification model can specifically adopt any one of the classifiers such as support vector machine or multi-layer perceptron, K nearest neighbor algorithm, naive Bayes, etc. Among them, the above classification model can be trained by using the target mapping features so that the classification model can learn the mapping relationship between the input target mapping features and the target output (category label), thereby obtaining a trained classification model and using it as a target image recognition model.

[0089] Step S207 , performing recognition processing on the image to be recognized based on the target image recognition model and the target projection matrix, and generating a target image recognition result corresponding to the image to be recognized.

[0090] In this embodiment, the above-mentioned specific implementation process of performing recognition processing on the image to be recognized based on the target image recognition model and the target projection matrix to generate a target image recognition result corresponding to the image to be recognized will be further described in detail in the subsequent specific embodiments of the present application and will not be elaborated in detail here.

[0091] After preprocessing the pre-constructed initial image data to obtain image training data, the present application performs principal component analysis and sparse processing on the image training data using a feature extraction network based on two-dimensional principal component analysis and sparse processing to obtain a target projection matrix, and performs a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature. Then, the target mapping feature obtained by performing feature screening processing on the first mapping feature based on a feature selection algorithm is used to train the preset classification model to obtain a target image recognition model, and finally the target image recognition model and the target projection matrix are used to perform recognition processing on the image to be recognized to generate a corresponding target image recognition result. The present application performs principal component analysis and sparse processing on the image training data by using a deep subspace network based on two-dimensional principal component analysis and sparse processing, so that the obtained target projection matrix has a smaller dimension. Compared with the principal components obtained by the traditional principal component analysis algorithm, the feature extraction network has less overhead in storage and feature calculation. In addition, based on the use of feature selection algorithm, the first mapping features obtained will be further screened to retain the target mapping features that are most discriminative and important for the image classification task, so that when the target mapping features with smaller data dimensions are subsequently used to train the image recognition model, the workload of model training can be further reduced, thereby effectively improving the training speed of the image recognition model.

[0092] In some optional implementations, step S203 includes the following steps:

[0093] Based on the feature extraction network, principal component extraction processing is performed on the image training data to obtain a corresponding projection matrix.

[0094] In this embodiment, the traditional PCA (principal component analysis) algorithm usually needs to convert the two-dimensional image into a one-dimensional vector when processing the image, which not only has a large amount of calculation, but also may destroy the two-dimensional structural information of the image, and the principal components obtained by the traditional PCA algorithm are usually dense and contain a large amount of redundant information. Therefore, the 2DPCA (two-dimensional principal component analysis) algorithm is used in the feature extraction network to perform principal component extraction processing on the image training data and extract the corresponding principal components, specifically by calculating the covariance matrix, solving the eigenvalues ​​and eigenvectors, and selecting the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components, and then constructing the corresponding projection matrix from the extracted principal components. The 2DPCA algorithm can directly calculate the covariance matrix and eigenvector on the two-dimensional image matrix, and extract the optimal projection axis to obtain the above-mentioned projection matrix by performing PCA dimensionality reduction processing in both the row and column directions. The 2DPCA algorithm can retain the two-dimensional structural information of the image, which helps to better capture the spatial relationship in the image, thereby improving the effect and recognition rate of feature extraction. Compared with the method of converting a two-dimensional image into a one-dimensional vector and then performing PCA dimensionality reduction processing, the 2DPCA algorithm has a smaller computational complexity in dimensionality reduction processing because it operates directly on a two-dimensional matrix and can achieve a higher recognition rate than the traditional PCA algorithm.

[0095] A sparse regularization term is constructed based on the preset L1 norm and L2 norm.

[0096] In the present embodiment, although the 2DPCA algorithm avoids the problem of structural information loss and increased computational complexity caused by converting a two-dimensional image into a one-dimensional vector in traditional PCA. However, even if the principal component extracted by the 2DPCA algorithm can reduce the data dimension to a certain extent, it may still contain more non-zero elements, which is not conducive to subsequent feature selection and model training. Therefore, it is a meaningful step to sparse the principal component. Wherein, the L1 norm refers to the sum of the absolute values ​​of each element in the vector. In the sparse process, the L1 norm can cause many elements of the model weight vector to become zero, thereby producing a sparse solution. This sparsity helps to reduce the complexity of the model, improve interpretability, and may reduce the risk of overfitting. In the sparseness of the principal component extracted by the 2DPCA algorithm, the introduction of the L1 norm constraint can make the extracted principal component more sparse, that is, the coefficients of some principal components (i.e., the elements in the projection vector) are zero or close to zero. The L2 norm refers to the square root of the sum of the squares of each element of the vector. The L2 norm is used as a regularization term to prevent the model from overfitting. In the sparsification of the principal components extracted by 2DPCA, although the L2 norm does not directly produce a sparse solution, it can smooth the weight vector to avoid excessive or too small weights, thereby enhancing the stability and generalization ability of the model. In addition, the above-mentioned sparse regularization term refers to the weighted sum of the L1 norm and the L2 norm. The weight parameters of the L1 norm and the L2 norm can be set according to actual business needs.

[0097] The projection matrix is ​​sparsely processed based on the sparse regularization term to obtain a corresponding sparse projection matrix.

[0098] In this embodiment, the sparse regularization term is used to perform sparse processing on the projection matrix, which is extracted from the projection matrix containing the principal component. The sparse regularization term, i.e., the weighted sum of the L1 norm and the L2 norm, is introduced into the objective function of the 2DPCA algorithm, and then the sparse projection matrix is ​​obtained by using an iterative optimization algorithm. The iterative optimization algorithm may adopt a coordinate descent method or a minimum angle regression method. The coordinate descent method refers to iteratively updating each parameter (i.e., the principal component coefficient), optimizing only one parameter at a time and fixing other parameters until the convergence condition is met. The minimum angle regression method refers to advancing in the direction most relevant to the objective function (i.e., the gradient direction) at each iteration until a new corner point is encountered, and then updating the direction and continuing the iteration. Specifically, according to the selected iterative optimization algorithm, the corresponding code is written to solve the optimization problem. In the solution process, the parameters need to be continuously updated iteratively until the convergence condition is met (e.g., the change in the objective function value is less than a certain threshold) or the maximum number of iterations is reached, and then the corresponding sparse projection matrix is ​​obtained.

[0099] The sparse projection matrix is ​​used as the target projection matrix.

[0100] In this embodiment, by adding a sparse step after learning the principal components using the 2DPCA algorithm, and by utilizing the constraints of the L1 norm and the L2 norm to perform sparse processing on the projection matrix, redundant information can be effectively removed, and the main factors can be highlighted, so that the obtained sparse projection matrix is ​​more sparse and has a stronger learning ability, and can improve the interpretability of the feature extraction network.

[0101] The present application performs principal component extraction processing on the image training data based on the feature extraction network to obtain a corresponding projection matrix; then constructs a sparse regularization term based on the preset L1 norm and L2 norm; subsequently performs sparse processing on the projection matrix based on the sparse regularization term to obtain a corresponding sparse projection matrix, and uses the sparse projection matrix as the target projection matrix. After the feature extraction network performs principal component extraction processing on the image training data to obtain a projection matrix, the present application performs sparse processing on the projection matrix by using a sparse regularization term constructed by combining the L1 norm and the L2 norm to obtain a target projection matrix, which can effectively reduce data redundancy to reduce feature dimensions, highlight main factors, improve the expressiveness and discrimination of features, and effectively ensure the accuracy of the obtained target projection matrix.

[0102] In some optional implementations of this embodiment, step S205 includes the following steps:

[0103] A first importance value of the first mapping feature is calculated based on a preset first feature selection algorithm.

[0104] In this embodiment, the first feature selection algorithm may specifically adopt an information gain algorithm. Information gain is an entropy-based evaluation method used to measure the information contribution of a feature to a classification task. The importance of each first mapping feature is evaluated by calculating the change in entropy before and after the data set is divided using the information gain algorithm, thereby obtaining a first importance value of the first mapping feature.

[0105] A second importance value of the first mapping feature is calculated based on a preset second feature selection algorithm.

[0106] In this embodiment, the second feature selection algorithm may specifically adopt a correlation coefficient algorithm, such as a Pearson correlation coefficient or a Spearman rank correlation coefficient. The correlation coefficient algorithm is used to calculate the correlation between each of the first mapping features and the classification target to evaluate the feature's discrimination ability, thereby obtaining the second importance value of the first mapping feature.

[0107] A comprehensive importance value of the first mapping feature is generated based on the first importance value and the second importance value.

[0108] In this embodiment, for any first mapping feature, the average value of the first importance value and the second importance value respectively matching the first mapping feature is calculated, and the average value is used as the comprehensive importance value of the first mapping feature.

[0109] All the first mapping features are sorted in descending order according to the comprehensive importance values ​​to obtain a corresponding sorting list.

[0110] In this embodiment, after all the first mapping features are sorted in descending order according to the comprehensive importance values, the sorted mapping feature data may be filled into a preset blank list to obtain the above sorted list.

[0111] According to the sorting, a specified number of second mapping features are screened out from the sorted list in sequence.

[0112] In this embodiment, there is no specific limitation on the value of the above-specified number, which can be set according to the actual requirement of the number of dimensions of feature dimensionality reduction.

[0113] The second mapping feature is used as the target mapping feature.

[0114] The present application calculates the first importance value and the second importance value corresponding to the first mapping feature respectively based on the preset first feature selection algorithm and the second feature selection algorithm, and then generates the comprehensive importance value of the first mapping feature based on the first importance value and the second importance value, and then sorts all the first mapping features in order from large to small according to the comprehensive importance value to obtain a corresponding sorting list, and finally selects a specified number of second mapping features from the sorting list in turn according to the sorting to serve as the most discriminative and important target mapping features in the image classification task. Since the second mapping feature is a plurality of mapping features with the highest comprehensive importance value among all the first mapping features, the accuracy of determining the target mapping feature is effectively improved. In addition, when the target mapping feature with a smaller data dimension is used to train the image recognition model in the future, the workload of model training can be further reduced, thereby effectively improving the training speed of the image recognition model.

[0115] In some optional implementations, step S205 includes the following steps:

[0116] A third importance value of the first mapping feature is calculated based on a preset third feature selection algorithm.

[0117] In this embodiment, the third feature selection algorithm may specifically adopt an information gain algorithm. The implementation steps of calculating the third importance value of the first mapping feature based on the preset third feature selection algorithm may refer to the aforementioned processing steps of calculating the first importance value of the first mapping feature based on the preset first feature selection algorithm, and will not be described in detail here.

[0118] A fourth importance value of the first mapping feature is calculated based on a preset fourth feature selection algorithm.

[0119] In this embodiment, the fourth feature selection algorithm may specifically adopt a correlation coefficient algorithm, for example, any one of the Pearson correlation coefficient and the Spearman rank correlation coefficient may be adopted. The implementation step of calculating the fourth importance value of the first mapping feature based on the preset fourth feature selection algorithm may refer to the aforementioned processing step of calculating the second importance value of the first mapping feature based on the preset second feature selection algorithm, and no further details are given here.

[0120] A third mapping feature whose importance value is greater than a first preset threshold is selected from the first mapping features.

[0121] In this embodiment, the value of the first preset threshold is not specifically limited and can be set according to actual importance determination requirements. If the importance value of the first mapping feature calculated by the third feature selection algorithm is greater than the first preset threshold, the first mapping feature is preliminarily determined to be an important feature.

[0122] A fourth mapping feature whose importance value is greater than a second preset threshold is selected from the first mapping features.

[0123] In this embodiment, the value of the second preset threshold is not specifically limited and can be set according to actual importance determination requirements. If the importance value of the first mapping feature calculated by the fourth feature selection algorithm is greater than the second preset threshold, the first mapping feature is preliminarily determined to be an important feature.

[0124] The target mapping feature is generated based on the third mapping feature and the fourth mapping feature.

[0125] In this embodiment, the specific implementation process of generating the target mapping feature based on the third mapping feature and the fourth mapping feature will be further described in detail in the subsequent specific embodiments of this application, and will not be elaborated on here. Alternatively, the third mapping feature and the fourth mapping feature can also be processed by taking the union of the third mapping feature and the fourth mapping feature to obtain a seventh mapping feature, and the seventh mapping feature can be used as the above-mentioned target mapping feature. By using the important features selected by the third feature selection algorithm and the important features selected by the fourth feature selection algorithm as the most discriminative and important target mapping features in the image classification task, the data richness of the obtained target mapping features can be guaranteed.

[0126] The present application calculates the third importance value and the fourth importance value corresponding to the first mapping feature based on a preset third feature selection algorithm and a fourth feature selection algorithm respectively, and then selects the third mapping features whose importance values ​​are greater than the first preset threshold value from the first mapping features and selects the fourth mapping features whose importance values ​​are greater than the second preset threshold value from the first mapping features, and then generates the target mapping features based on the third mapping features and the fourth mapping features, so as to generate the most discriminative and important target mapping features in the image classification task by using the important features selected by the third feature selection algorithm and the important features selected by the fourth feature selection algorithm, thereby effectively improving the generation intelligence and generation accuracy of the target mapping features.

[0127] In some optional implementations, generating the target mapping feature based on the third mapping feature and the fourth mapping feature comprises the following steps:

[0128] The third mapping feature and the fourth mapping feature are integrated to obtain a corresponding mapping feature set.

[0129] In this embodiment, the mapping feature set is a constructed feature data set storing the third mapping feature and the fourth mapping feature.

[0130] A fifth mapping feature that appears repeatedly is screened out from the mapping feature set.

[0131] In this embodiment, a statistical result can be obtained by performing statistical processing on the number of occurrences of all mapping features stored in the mapping feature set, and based on the statistical result, a specific mapping feature with a number of occurrences of 2 among all mapping features is used as the above-mentioned recurring fifth mapping feature.

[0132] The fifth mapping feature is used as the target mapping feature.

[0133] In this embodiment, since the fifth mapping feature is both an important feature selected by the third feature selection algorithm and an important feature selected by the fourth feature selection algorithm, the fifth mapping feature is used as the most discriminative and important target mapping feature in the image classification task, thereby improving the accuracy of determining the target mapping feature.

[0134] The present application integrates the third mapping feature and the fourth mapping feature to obtain a corresponding mapping feature set; then screens out the fifth mapping feature that appears repeatedly from the mapping feature set; and subsequently uses the fifth mapping feature as the target mapping feature. The present application effectively improves the accuracy of determining the target mapping feature by screening out the fifth mapping feature that appears repeatedly in the mapping feature set, and using the fifth mapping feature that meets both the important features screened out by the third feature selection algorithm and the important features screened out by the fourth feature selection algorithm as the most discriminative and important target mapping feature in the final image classification task. In addition, when the target mapping feature with a smaller data dimension is subsequently used to train the image recognition model, the workload of model training can be further reduced, thereby effectively improving the training speed of the image recognition model.

[0135] In some optional implementations of this embodiment, step S207 includes the following steps:

[0136] The image to be identified is preprocessed to obtain a corresponding designated image.

[0137] In this embodiment, the implementation process of preprocessing the image to be identified to obtain the corresponding designated image can refer to the aforementioned step of preprocessing the initial image data, and will not be described in detail here.

[0138] A convolution operation is performed on the designated image through the target projection matrix to obtain a corresponding sixth mapping feature.

[0139] In this embodiment, the above-mentioned convolution operation is performed on the specified image through the target projection matrix to obtain the corresponding sixth mapping feature. The implementation process can refer to the aforementioned step of performing the convolution operation on the image training data through the target projection matrix, and no further details are given here.

[0140] The sixth mapping feature is recognized based on the target image recognition model to obtain a corresponding designated image recognition result.

[0141] In this embodiment, by inputting the sixth mapping feature into the target image recognition model and using the target image recognition model to perform recognition processing on the sixth mapping feature, the processing efficiency and accuracy of the image recognition processing can be effectively improved, thereby ensuring the accuracy of the generated specified image recognition result.

[0142] The designated image recognition result is used as the target image recognition result.

[0143] The present application obtains the corresponding designated image by preprocessing the image to be identified; then obtains the corresponding sixth mapping feature by convolution operation on the designated image through the target projection matrix; and then recognizes and processes the sixth mapping feature based on the target image recognition model, so as to quickly generate a target image recognition result corresponding to the image to be identified. The present application recognizes and processes the sixth mapping feature obtained by convolution operation on the designated image through the target projection matrix using the target image recognition model. Since the sixth mapping feature is an important feature with strong expression ability and high discrimination obtained after sparse processing, the recognition accuracy can be effectively improved by using the target image recognition model to recognize and process the sixth mapping feature, thereby improving the accuracy of the generated target image recognition result.

[0144] In some optional implementations of this embodiment, step S202 includes the following steps:

[0145] The initial image data is subjected to image resizing processing to obtain first image data.

[0146] In this embodiment, the image size adjustment process is performed on the initial image data to ensure that the initial image data is data with a uniform image size, and the first image data is obtained.

[0147] The first image data is gray-scaled to obtain second image data.

[0148] In this embodiment, the grayscale processing refers to converting the first image data into a grayscale image, that is, obtaining the second image data.

[0149] The second image data is subjected to histogram equalization processing to obtain third image data.

[0150] In this embodiment, the histogram equalization process refers to a process of enhancing the contrast of an image.

[0151] The third image data is used as the image training data.

[0152] The present application accurately completes the preprocessing of the initial image data by performing image resizing, grayscale processing, and histogram equalization processing on the initial image data. By preprocessing the initial image data, key information in the initial image data can be effectively extracted, and the performance impact of illumination and noise on the subsequent use of the model for image recognition processing can be effectively reduced.

[0153] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0154] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned target image recognition results, the above-mentioned target image recognition results can also be stored in a node of a blockchain.

[0155] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0156] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0157] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0158] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0159] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0160] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of an image recognition device based on artificial intelligence, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0161] like Figure 3 As shown, the artificial intelligence-based image recognition device 300 described in this embodiment includes: an acquisition module 301, a preprocessing module 302, a processing module 303, an operation module 304, a screening module 305, a training module 306 and a recognition module 307. Among them:

[0162] The acquisition module 301 is used to acquire pre-constructed initial image data; wherein the initial image data is a data set that has been annotated with category labels;

[0163] A preprocessing module 302 is used to preprocess the initial image data to obtain corresponding image training data;

[0164] The processing module 303 is used to perform principal component analysis and sparse processing on the image training data based on a preset feature extraction network to obtain a corresponding target projection matrix; wherein the feature extraction network is a deep subspace network based on two-dimensional principal component analysis and sparse processing;

[0165] An operation module 304 is used to perform a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature;

[0166] A screening module 305 is used to perform feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature;

[0167] A training module 306 is used to train a preset classification model based on the target mapping features to obtain a trained target image recognition model;

[0168] The recognition module 307 is used to perform recognition processing on the image to be recognized based on the target image recognition model and the target projection matrix, and generate a target image recognition result corresponding to the image to be recognized.

[0169] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based image recognition method in the aforementioned implementation, and will not be repeated here.

[0170] In some optional implementations of this embodiment, the processing module 303 includes:

[0171] An extraction submodule, used for performing principal component extraction processing on the image training data based on the feature extraction network to obtain a corresponding projection matrix;

[0172] A construction submodule is used to construct a sparse regularization term based on the preset L1 norm and L2 norm;

[0173] A sparse submodule, used for performing sparse processing on the projection matrix based on the sparse regularization term to obtain a corresponding sparse projection matrix;

[0174] The first determination submodule is configured to use the sparse projection matrix as the target projection matrix.

[0175] In some optional implementations of this embodiment, the screening module 305 includes:

[0176] A first calculation submodule is configured to calculate a first importance value of the first mapping feature based on a preset first feature selection algorithm; and

[0177] A second calculation submodule, used for calculating a second importance value of the first mapping feature based on a preset second feature selection algorithm;

[0178] A first generating submodule, configured to generate a comprehensive importance value of the first mapping feature based on the first importance value and the second importance value;

[0179] A sorting submodule, used to sort all the first mapping features in descending order according to the comprehensive importance values ​​to obtain a corresponding sorting list;

[0180] A first screening submodule, configured to screen out a specified number of second mapping features from the sorted list in sequence according to the sorting;

[0181] The second determining submodule is configured to use the second mapping feature as the target mapping feature.

[0182] In some optional implementations of this embodiment, the screening module 305 includes:

[0183] a third calculation submodule, configured to calculate a third importance value of the first mapping feature based on a preset third feature selection algorithm; and

[0184] a fourth calculation submodule, configured to calculate a fourth importance value of the first mapping feature based on a preset fourth feature selection algorithm;

[0185] A second screening submodule is used to screen out third mapping features whose importance values ​​are greater than a first preset threshold from the first mapping features; and

[0186] A third screening submodule, configured to screen out fourth mapping features whose importance values ​​are greater than a second preset threshold from the first mapping features;

[0187] The second generating submodule is used to generate the target mapping feature based on the third mapping feature and the fourth mapping feature.

[0188] In some optional implementations of this embodiment, the second generating submodule includes:

[0189] a processing unit, configured to integrate the third mapping feature and the fourth mapping feature to obtain a corresponding mapping feature set;

[0190] a screening unit, configured to screen out a fifth mapping feature that appears repeatedly from the mapping feature set;

[0191] A determining unit is configured to use the fifth mapping feature as the target mapping feature.

[0192] In some optional implementations of this embodiment, the identification module 307 includes:

[0193] A preprocessing submodule, used for preprocessing the image to be identified to obtain a corresponding designated image;

[0194] An operation submodule, configured to perform a convolution operation on the designated image through the target projection matrix to obtain a corresponding sixth mapping feature;

[0195] A recognition submodule, configured to perform recognition processing on the sixth mapping feature based on the target image recognition model to obtain a corresponding designated image recognition result;

[0196] The third determination submodule is used to use the designated image recognition result as the target image recognition result.

[0197] In some optional implementations of this embodiment, the preprocessing module 302 includes:

[0198] A first processing submodule, configured to perform image resizing processing on the initial image data to obtain first image data;

[0199] A second processing submodule, configured to perform grayscale processing on the first image data to obtain second image data;

[0200] A third processing submodule, configured to perform histogram equalization processing on the second image data to obtain third image data;

[0201] The fourth determining submodule is configured to use the third image data as the image training data.

[0202] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0203] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0204] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0205] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of an image recognition method based on artificial intelligence, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0206] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run the computer-readable instructions stored in the memory 41 or process data, such as running the computer-readable instructions of the artificial intelligence-based image recognition method.

[0207] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0208] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned artificial intelligence-based image recognition method.

[0209] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0210] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. An image recognition method based on artificial intelligence, characterized in that: The steps include: Acquire pre-constructed initial image data; wherein the initial image data is a data set that has been annotated with category labels; Preprocessing the initial image data to obtain corresponding image training data; Based on a preset feature extraction network, the image training data is subjected to principal component analysis and sparse processing to obtain a corresponding target projection matrix; wherein the feature extraction network is a deep subspace network based on two-dimensional principal component analysis and sparse processing; Performing a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature; Performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature; Training a preset classification model based on the target mapping features to obtain a trained target image recognition model; The image to be recognized is recognized based on the target image recognition model and the target projection matrix to generate a target image recognition result corresponding to the image to be recognized.

2. The image recognition method based on artificial intelligence according to claim 1, characterized in that: The step of performing principal component analysis and sparse processing on the image training data based on a preset feature extraction network to obtain a corresponding target projection matrix specifically includes: Performing principal component extraction processing on the image training data based on the feature extraction network to obtain a corresponding projection matrix; Construct sparse regularization terms based on the preset L1 norm and L2 norm; Performing sparse processing on the projection matrix based on the sparse regularization term to obtain a corresponding sparse projection matrix; The sparse projection matrix is ​​used as the target projection matrix.

3. The image recognition method based on artificial intelligence according to claim 1, characterized in that: The step of performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature specifically includes: Calculating a first importance value of the first mapping feature based on a preset first feature selection algorithm; and, Calculating a second importance value of the first mapping feature based on a preset second feature selection algorithm; generating a comprehensive importance value of the first mapping feature based on the first importance value and the second importance value; Sorting all the first mapping features in descending order according to the comprehensive importance values ​​to obtain a corresponding sorting list; According to the sorting, a specified number of second mapping features are sequentially selected from the sorted list; The second mapping feature is used as the target mapping feature.

4. The image recognition method based on artificial intelligence according to claim 1, characterized in that: The step of performing feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature specifically includes: Calculating a third importance value of the first mapping feature based on a preset third feature selection algorithm; and, Calculating a fourth importance value of the first mapping feature based on a preset fourth feature selection algorithm; Filtering out third mapping features whose importance values ​​are greater than a first preset threshold from the first mapping features; and, Selecting a fourth mapping feature whose importance value is greater than a second preset threshold from the first mapping features; The target mapping feature is generated based on the third mapping feature and the fourth mapping feature.

5. The image recognition method based on artificial intelligence according to claim 4, characterized in that: The step of generating the target mapping feature based on the third mapping feature and the fourth mapping feature specifically includes: Integrate the third mapping feature and the fourth mapping feature to obtain a corresponding mapping feature set; Filter out a fifth mapping feature that appears repeatedly from the mapping feature set; The fifth mapping feature is used as the target mapping feature.

6. The image recognition method based on artificial intelligence according to claim 1, characterized in that: The step of performing recognition processing on the image to be recognized based on the target image recognition model and the target projection matrix to generate a target image recognition result corresponding to the image to be recognized specifically includes: Preprocessing the image to be identified to obtain a corresponding designated image; Performing a convolution operation on the designated image through the target projection matrix to obtain a corresponding sixth mapping feature; Performing recognition processing on the sixth mapping feature based on the target image recognition model to obtain a corresponding designated image recognition result; The designated image recognition result is used as the target image recognition result.

7. The image recognition method based on artificial intelligence according to claim 1, characterized in that: The step of preprocessing the initial image data to obtain corresponding image training data specifically includes: Performing image resizing processing on the initial image data to obtain first image data; Performing grayscale processing on the first image data to obtain second image data; Performing histogram equalization processing on the second image data to obtain third image data; The third image data is used as the image training data.

8. An image recognition device based on artificial intelligence, characterized in that: include: An acquisition module, used to acquire pre-constructed initial image data; wherein the initial image data is a data set that has been annotated with category labels; A preprocessing module, used for preprocessing the initial image data to obtain corresponding image training data; A processing module, used to perform principal component analysis and sparse processing on the image training data based on a preset feature extraction network to obtain a corresponding target projection matrix; wherein the feature extraction network is a deep subspace network based on two-dimensional principal component analysis and sparse processing; An operation module, configured to perform a convolution operation on the image training data through the target projection matrix to obtain a corresponding first mapping feature; A screening module, used to perform feature screening processing on the first mapping feature based on a preset feature selection algorithm to obtain a corresponding target mapping feature; A training module, used to train a preset classification model based on the target mapping features to obtain a trained target image recognition model; The recognition module is used to perform recognition processing on the image to be recognized based on the target image recognition model and the target projection matrix, and generate a target image recognition result corresponding to the image to be recognized.

9. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the artificial intelligence-based image recognition method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the artificial intelligence-based image recognition method according to any one of claims 1 to 7.

Images