Image classification recognition model training method and device, and computer device

Abstract

The application discloses an image classification and recognition model training method and device and computer equipment, and the method comprises the following steps: obtaining an image self-encoder based on local open image training; obtaining a first image feature set by processing remote sensitive images based on the image self-encoder, and obtaining a second image feature set by processing local open images based on the image self-encoder; the first image feature set and the second image feature set form an image feature training set; training a pre-established image classification and recognition model based on the image feature training set to obtain a target image classification and recognition model; the image self-encoder obtained by training based on local open image data is used for desensitization feature extraction of remote sensitive image data, then the image feature training set is obtained by combining the desensitization feature sets of the remote and local, and the training of the image classification and recognition model is performed, finally, the image classification and recognition learning is realized without accessing original sensitive data, and meanwhile, the sample amount used for learning and the recognition accuracy are ensured.

Description

Technical Field

[0001] This invention relates to the field of computer image processing, and specifically to an image classification and recognition model training method, apparatus, and computer equipment. Background Technology

[0002] With the rapid development of big data processing and artificial intelligence (AI) technologies, information mining and classification of massive image data using traditional machine learning or deep neural networks has become a key focus across various industries. However, on the one hand, accurate classification of image data requires training image classification and recognition models based on a large number of real image data samples. On the other hand, many images obtained by actual business departments or production lines are highly private and confidential, making it impossible to fully open data access permissions to internal or third-party R&D personnel. In particular, it is required that R&D personnel cannot access the original sensitive image data. This results in the inability to obtain the original sensitive image data when acquiring the training set for the image classification and recognition model, and the limited amount of data in the training set poses a significant obstacle to the development of image classification and recognition models.

[0003] Therefore, an image classification and recognition model training method is needed that can ensure that the original sensitive image information is not leaked, while also ensuring that there are enough rich and realistic image visual features for training the image classification and recognition model. Summary of the Invention

[0004] The purpose of this invention is to provide a training method, apparatus, and computer device for an image classification and recognition model that can be trained using sensitive image data.

[0005] The technical solution of the present invention is as follows: In a first aspect, the present invention provides a method for training an image classification and recognition model, the method comprising:

[0006] Image autoencoders are obtained by training local open images;

[0007] The image autoencoder processes the local open image to obtain a first image feature set, and the image autoencoder processes the remote sensitive image to obtain a second image feature set; the first image feature set and the second image feature set together form an image feature training set.

[0008] The target image classification and recognition model is obtained by training a pre-established image classification and recognition model based on the image feature training set.

[0009] In a preferred embodiment, the image autoencoder acquired based on local open image training includes:

[0010] An encoder for converting an image into image features and a decoder for reconstructing an image based on image features are constructed based on a convolutional neural network. The image autoencoder includes the encoder and the decoder.

[0011] The encoder and decoder are trained based on local open images, and the error between the image reconstructed by the decoder and the original image converted by the encoder is minimized by a pre-built loss function.

[0012] In a preferred embodiment, before training the encoder and the decoder based on local open images and minimizing the error between the decoder-reconstructed image and the original image transformed by the encoder using a pre-built loss function, the method further includes:

[0013] Local open images are obtained based on a first user ID and a security access policy. The first user ID is an ID that has permission to access local open images but does not have permission to remotely access sensitive images.

[0014] In a preferred embodiment, before obtaining the second image feature set by processing the remotely sensitive image based on the image autoencoder, the method further includes:

[0015] Remote sensitive images are obtained based on a second user ID and a security access policy, wherein the second user ID is an ID with remote access permissions to sensitive images;

[0016] The second image feature set includes the image features of the remote sensitive image after it has been desensitized by the image autoencoder.

[0017] In a preferred embodiment, the step of processing the remotely sensitive image based on the image autoencoder to obtain the second image feature set includes:

[0018] Based on the image autoencoder, the remote sensitive image is desensitized to obtain image features that retain the structure and texture of the remote sensitive image and filter out text and facial information.

[0019] In a preferred embodiment, after obtaining the second image feature set by processing the remotely sensitive image based on the image autoencoder, the method further includes:

[0020] Download and store the second image feature set locally so as to obtain the second image feature set based on the first user ID and security access policy.

[0021] In a preferred embodiment, the image feature training set includes image features and corresponding annotation and classification information.

[0022] In a preferred embodiment, the step of training a pre-established image classification and recognition model based on the image feature training set to obtain a target image classification and recognition model includes:

[0023] Based on the image features and the labeled classification information, a pre-established image classification and recognition model is trained to obtain a target image classification and recognition model. The target image classification and recognition model predicts the test sample and then obtains the category label corresponding to the test sample.

[0024] Secondly, the present invention also provides an image classification and recognition model training device, the device comprising:

[0025] The first acquisition module is used to train and acquire an image autoencoder based on local open images;

[0026] The second acquisition module is used to process the local open image based on the image autoencoder to obtain a first image feature set, and to process the remote sensitive image based on the image autoencoder to obtain a second image feature set; the first image feature set and the second image feature set constitute an image feature training set;

[0027] The training module is used to train a pre-established image classification and recognition model based on the image feature training set to obtain the target image classification and recognition model.

[0028] Thirdly, the present invention also provides a computer device, comprising:

[0029] One or more processors; and

[0030] A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform any of the methods described in the first aspect.

[0031] The advantages of this invention are as follows: This invention provides an image classification and recognition model training method, apparatus, and computer equipment. The method includes: training an image autoencoder based on local open images; processing remote sensitive images based on the image autoencoder to obtain a first image feature set, and processing local open images based on the image autoencoder to obtain a second image feature set; the first image feature set and the second image feature set constitute an image feature training set; training a pre-established image classification and recognition model based on the image feature training set to obtain a target image classification and recognition model; the image autoencoder trained based on local open image data performs desensitization feature extraction on remote sensitive image data, and then combines the remote and local desensitized feature sets to obtain an image feature training set for training the image classification and recognition model, ultimately achieving image classification and recognition learning without accessing the original sensitive data, while ensuring the sample size and recognition accuracy used in learning. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of the image classification and recognition model training method provided in the embodiments of the present invention;

[0034] Figure 2 This is a flowchart illustrating the image autoencoder obtained by training based on local open images in the image classification and recognition model training method provided in this embodiment of the invention.

[0035] Figure 3 This is a flowchart of the desensitization feature extraction and recognition process based on local open images and remote sensitive images in the image classification and recognition model training method provided in this embodiment of the invention;

[0036] Figure 4 This is a structural diagram of the image classification and recognition model device provided in an embodiment of the present invention;

[0037] Figure 5 This is an architectural diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0039] As described in the background section, image classification and recognition models require training with samples after they are established. To achieve accurate classification and recognition of image data, these models need to be trained using a large number of real image data samples. However, for security reasons, a large amount of real image data contains sensitive information and is subject to access restrictions. Researchers developing image classification and recognition models cannot obtain this image data, resulting in only locally available images being usable as training samples. Consequently, the number of real image samples used to train the model is insufficient, making it difficult to achieve the required classification accuracy.

[0040] To address the aforementioned issues, this invention creatively proposes an image classification and recognition model training method, apparatus, and computer equipment. Sensitive or confidential images are pre-processed to desensitize them, ensuring that the original sensitive image information is not accessed by researchers, thus guaranteeing privacy and security. At the same time, it ensures that there are sufficiently rich and realistic image visual features available for training the image classification and recognition model.

[0041] The image classification and recognition model training method, apparatus, and computer equipment provided by the present invention will be further described below with reference to specific embodiments.

[0042] Example 1: This example provides a method for training an image classification and recognition model, referring to... Figure 1 As shown, the method includes:

[0043] S110, Image autoencoder obtained by training local open images.

[0044] Autoencoders are artificial neural networks that learn efficient representations of input data through unsupervised learning. This efficient representation of the input data is called encoding, and its dimensionality is generally much smaller than the input data, making autoencoders useful for dimensionality reduction. Furthermore, autoencoders can serve as powerful feature detectors and for randomly generating data similar to the training data.

[0045] Preferred, refer to Figure 2 As shown, this step includes:

[0046] S111. Construct an encoder for converting an image into image features and a decoder for reconstructing the image based on the image features, using a convolutional neural network. The image autoencoder includes an encoder and a decoder.

[0047] The base image uses a convolutional neural network (CNN) to construct an encoder, which encodes the original image X∈D into image features f, where: X is the original image, a three-dimensional matrix of H*W*C; H is the height of the image; W is the width of the image; C is the number of channels; and D is a local open image dataset, typically C=3 for color images and C=1 for grayscale images. f is the encoded image feature, a feature matrix of H / r*W / r*C', where: r is the scaling factor of the feature matrix's width and height relative to the original image, typically r can be 4, 8, 16, etc., meaning it is scaled down by a factor of 4 (or 8, 16), and C' is the number of channels in the feature matrix, typically C' can be 32 or 64, etc. This invention does not limit the specific design of the encoder.

[0048] The decoder is constructed based on a convolutional neural network (CNN). Unlike the encoder described above, the decoder uses deconvolutional layers to transform image features f into a reconstructed image X' similar to the original image. The reconstructed image retains the structure and main texture information of the original image, but filters out sensitive details such as text and faces. This invention does not limit the specific design of the decoder.

[0049] Preferably, prior to S112, the method further includes:

[0050] S11A. Obtain local open images based on the first user ID and security access policy. The first user ID is an ID that has permission to access local open images but does not have permission to remotely access sensitive images.

[0051] Specifically, researchers developing image classification and recognition models typically have access to locally accessible images but not sensitive images. The first user ID is the user ID of the researcher developing the image classification and recognition model. The security access policy specifies different access permissions for different user IDs. After obtaining permission from the security access policy, the first user ID accesses locally accessible images to obtain local open image data.

[0052] S112. Train the encoder and decoder based on the local open image and minimize the error between the image reconstructed by the decoder and the original image transformed by the encoder using a pre-built loss function.

[0053] Specifically, the encoder and decoder are trained using locally available images, such that after the encoder encodes an image X into image features f, the decoder reconstructs an image X' based on the image features f obtained by the encoder, minimizing the error between the reconstructed image X' and the original image X. During training, a loss function is used to minimize the error between the reconstructed image X' and the original image X; here, the mean square error function can be used as the loss function. This invention does not limit the specific design of the loss function.

[0054] Preferably, prior to S120, the method further includes:

[0055] S12A: Obtain remote sensitive images based on the second user ID and security access policy. The second user ID is an ID with remote access permissions to sensitive images.

[0056] This part of the process is executed remotely. The security access policy specifies different access permissions for different user IDs. After obtaining permission from the security access policy, the second user ID accesses and obtains remote sensitive images.

[0057] The second image feature set includes image features of remotely sensitive images after desensitization processing by an image autoencoder.

[0058] S120. A first image feature set is obtained by processing local open images based on an image autoencoder, and a second image feature set is obtained by processing remote sensitive images based on an image autoencoder; the first image feature set and the second image feature set constitute an image feature training set.

[0059] Reference Figure 3 As shown, specifically, the process of obtaining the first image feature set by processing local open images based on image autoencoders is executed locally. The local open image X1∈D1 (D1 is an open image dataset) is encoded into image features f1 using the trained encoder, i.e., f1=Encoder(X1). Let F1={f1} be the image feature set corresponding to the obtained local open image dataset D1.

[0060] The trained encoder is used to encode the remote sensitive image X2∈D2 (D2 is the remote sensitive image dataset) into image feature f2, i.e. f2=Encoder(X2). Let F2={f2} be the desensitized image feature set corresponding to the remote sensitive image set D2, i.e. the second image feature set. The desensitization process is executed remotely. F1 and F2 form a complete image feature set F, i.e. the image feature training set.

[0061] More preferably, obtaining a second image feature set by processing remotely sensitive images based on an image autoencoder includes:

[0062] Image features derived from remotely sensitive images using an image autoencoder (AAE) are obtained by preserving the structure and texture of the remotely sensitive images while filtering out text and facial information. Specifically, the encoder in the AAE transforms the remotely sensitive image into remotely sensitive image features, and the decoder filters the parts of the remotely sensitive image features that correspond to text and facial sensitive information, reconstructing desensitized image features that retain the structure and main texture information of the original remotely sensitive image. All desensitized image features constitute a second image feature set.

[0063] Preferably, after S120, the method further includes:

[0064] S12B: Download and store the second image feature set locally so as to obtain the second image feature set based on the first user ID and security access policy.

[0065] Specifically, the desensitized image feature set, also known as the second image feature set F2, is downloaded locally.

[0066] Because the security access policy restricts the ID of the R&D personnel developing the image classification and recognition model to access only local data, the second user ID, which has the right to access and process remote sensitive data, needs to download and save the second image feature set obtained by de-identifying the remote sensitive image to the local machine. This allows the first user ID to access and use the second image feature set to obtain the real de-identified remote sensitive image for training the image classification and recognition model.

[0067] S130. Train a pre-established image classification and recognition model based on the image feature training set to obtain the target image classification and recognition model.

[0068] Specifically, a traditional machine learning or deep convolutional network is selected, and an image classification and recognition model is trained based on the image feature training set F obtained above to obtain the target image classification and recognition model, i.e., the image classifier, denoted as Classifier. The traditional machine learning mentioned above can be models such as Support Vector Machine (SVM), Logistic Regression (LR), Random Forest, etc., and the deep convolutional network can be a regular network such as ResNet, DenseNet, etc., or a lightweight network such as MobileNet, ShuffleNet, etc., and the present invention does not limit the specific network.

[0069] Preferably, the image feature training set includes image features and corresponding annotation and classification information. That is, the ground truth used for training comes from the annotation information of local open image data and the annotation information of remote sensitive image data, wherein the annotation of remote sensitive image data is completed based on a second user ID with access qualifications to sensitive images.

[0070] Preferably, this step includes:

[0071] The target image classification and recognition model is obtained by training a pre-established image classification and recognition model based on image features and labeled classification information. The target image classification and recognition model then predicts the test sample to obtain the corresponding category label.

[0072] The image classification and recognition model training method provided in this embodiment includes: training an image autoencoder based on local open images; processing remote sensitive images based on the image autoencoder to obtain a first image feature set, and processing local open images based on the image autoencoder to obtain a second image feature set; the first image feature set and the second image feature set constitute an image feature training set; training a pre-established image classification and recognition model based on the image feature training set to obtain a target image classification and recognition model; the image autoencoder trained based on local open image data performs desensitization feature extraction on remote sensitive image data, and then combines the remote and local desensitized feature sets to obtain an image feature training set for training the image classification and recognition model, ultimately achieving image classification and recognition learning without accessing the original sensitive data, while ensuring the sample size and recognition accuracy used for learning.

[0073] Furthermore, this invention illustrates the application of desensitization features based on image classification tasks. In practice, desensitization features can also be applied to specific tasks such as image segmentation, object detection, and face recognition.

[0074] Example 2: This example provides an image classification and recognition model training device, referring to... Figure 4 As shown, the device includes:

[0075] The first acquisition module 410 is used to train an image autoencoder based on local open images;

[0076] The second acquisition module 420 is used to process the local open image based on the image autoencoder to obtain a first image feature set, and to process the remote sensitive image based on the image autoencoder to obtain a second image feature set; the first image feature set and the second image feature set constitute an image feature training set;

[0077] The training module 430 is used to train a pre-established image classification and recognition model based on the image feature training set to obtain a target image classification and recognition model.

[0078] Preferably, the first acquisition module 410 includes:

[0079] The construction unit 411 is used to construct an encoder for converting an image into image features and a decoder for reconstructing an image based on the image features, wherein the image autoencoder includes the encoder and the decoder;

[0080] Training unit 412 is used to train the encoder and the decoder based on local open images and minimize the error between the image reconstructed by the decoder and the original image converted by the encoder using a pre-built loss function.

[0081] More preferably, the device further includes:

[0082] The first access module 440 is used to acquire the local open image based on a first user ID and a security access policy before the training unit 412 trains the encoder and the decoder based on the local open image and minimizes the error between the image reconstructed by the decoder and the original image converted by the encoder using a pre-built loss function. The first user ID is an ID that has the right to access the local open image but does not have the right to remotely access sensitive images.

[0083] More preferably, the device further includes:

[0084] The second access module 450 is used to acquire the remote sensitive image based on the second user ID and a security access policy before the second acquisition module 420 processes the remote sensitive image based on the image autoencoder to obtain the second image feature set. The second user ID is an ID with remote access permission for the sensitive image.

[0085] The second image feature set includes the image features of the remote sensitive image after it has been desensitized by the image autoencoder.

[0086] More preferably, the second acquisition module 420 includes:

[0087] The processing unit 421 is used to perform desensitization processing on the remote sensitive image based on the image autoencoder to obtain image features that retain the structure and texture of the remote sensitive image and filter out text and facial information.

[0088] More preferably, the device further includes:

[0089] The storage module 460 is used to download and store the second image feature set to the local machine after the second acquisition module 420 obtains the second image feature set by processing the remote sensitive image based on the image autoencoder, so as to obtain the second image feature set based on the first user ID and security access policy.

[0090] Preferably, the training module 430 includes:

[0091] Training unit 431 is used to train a pre-established image classification and recognition model based on the image features and the labeled classification information to obtain a target image classification and recognition model. The target image classification and recognition model predicts the test sample and obtains the category label corresponding to the test sample.

[0092] The image classification and recognition model training device provided in this embodiment is used to execute the image classification and recognition model training method provided in Embodiment 1. Its beneficial effects are the same as those of the image classification and recognition model training method executed, and will not be described in detail here.

[0093] It should be noted that the image classification and recognition model training device provided in the above embodiments is only illustrated by the division of the above functional modules when triggering the image classification and recognition model training service. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the image classification and recognition model training device provided in the above embodiments and the image classification and recognition model training method provided in Embodiment 1 belong to the same concept, that is, the device is based on the method, and its specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0094] Example 3: This example provides a computer device, including:

[0095] One or more processors; and

[0096] A memory associated with the one or more processors, the memory storing program instructions, which, when read and executed by the one or more processors, perform the following steps:

[0097] Image autoencoders are obtained by training local open images;

[0098] The image autoencoder processes the local open image to obtain a first image feature set, and the image autoencoder processes the remote sensitive image to obtain a second image feature set; the first image feature set and the second image feature set together form an image feature training set.

[0099] The target image classification and recognition model is obtained by training a pre-established image classification and recognition model based on the image feature training set.

[0100] in, Figure 5 An exemplary computer system architecture is shown, which may include a processor 510, a video display adapter 511, a disk drive 512, an input / output interface 513, a network interface 514, and a memory 520. The processor 510, video display adapter 511, disk drive 512, input / output interface 513, network interface 514, and memory 520 can communicate with each other via a communication bus 530.

[0101] The processor 510 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to achieve the technical solution provided in this application.

[0102] The memory 520 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 520 can store the operating system 521 for controlling the operation of the computer device 500, and the basic input / output system (BIOS) 522 for controlling the low-level operations of the computer device 500. Additionally, it can store a web browser 523, a data storage management system 524, and an icon font processing system 525, etc. The aforementioned icon font processing system 525 can be the application program that specifically implements the aforementioned steps in this embodiment. In summary, when the technical solution provided in this application is implemented through software or firmware, the relevant program code is stored in the memory 520 and is called and executed by the processor 510.

[0103] Input / output interface 513 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0104] Network interface 514 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0105] Bus 530 includes a pathway for transmitting information between various components of the device, such as processor 510, video display adapter 511, disk drive 512, input / output interface 513, network interface 514, and memory 520.

[0106] In addition, the computer device 500 can also obtain information on specific claim conditions from the virtual resource object claim condition information database for use in condition judgment, etc.

[0107] It should be noted that although the above-described device only shows the processor 510, video display adapter 511, disk drive 512, input / output interface 513, network interface 514, memory 520, bus 530, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the solution of this application, and does not necessarily include all the components shown in the figures.

[0108] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer device, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from memory, or installed from ROM. When the computer program is executed by a processor, it performs the functions defined in the methods of embodiments of this application.

[0109] It should be noted that the computer device in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the embodiments of this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the embodiments of this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer device other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer device can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.

[0110] The aforementioned computer device may be included in the aforementioned server; or it may exist independently and not be assembled into the server. The aforementioned computer device carries one or more programs, which, when executed by the server, cause the server to: in response to detecting that the peripheral mode of the terminal is not activated, acquire the frame rate of the application on the terminal; when the frame rate meets the screen-off condition, determine whether the user is acquiring the terminal's screen information; and in response to the determination that the user is not acquiring the terminal's screen information, control the screen to enter an immediate dimming mode.

[0111] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0112] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0113] The image classification and recognition model training method, apparatus, and computer equipment provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and its core ideas. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

[0114] It should also be noted that the terms "first" and "second" in this invention are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include one or more of that feature.

[0115] All the above-mentioned optional technical solutions can be combined in any way to form optional embodiments of the present invention, and will not be described in detail here. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention. Of course, the above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and should not be used to limit the scope of protection of the present invention. All modifications made according to the spirit and essence of the main technical solutions of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for training an image classification and recognition model, characterized in that, The method includes: Image autoencoders are obtained by training local open images; Obtain local open images based on the first user ID and security access policy; The image autoencoder processes the local open image to obtain a first image feature set; a remote sensitive image is obtained based on a second user ID and a security access policy, and the image autoencoder processes the remote sensitive image to obtain a second image feature set; the first user ID is an ID that has permission to access the local open image but not permission to access the remote sensitive image. The first image feature set and the second image feature set together form an image feature training set; The target image classification and recognition model is obtained by training a pre-established image classification and recognition model based on the image feature training set. The image autoencoder includes an encoder and a decoder, and the step of processing remotely sensitive images based on the image autoencoder to obtain a second image feature set includes: The encoder in the image autoencoder transforms the remotely sensitive image into remotely sensitive image features. The decoder in the image autoencoder filters the parts of the remotely sensitive image features that correspond to text and facial sensitive information, and reconstructs desensitized image features that retain the structure and main texture information of the original remotely sensitive image. All desensitized image features constitute the second image feature set. The second image feature set obtained by the desensitized remotely sensitive image is downloaded and saved locally by the second user ID, so that the second image feature set can be obtained based on the first user ID and the security access policy. The second user ID is an ID with remote access permission for sensitive images.

2. The image classification and recognition model training method according to claim 1, characterized in that, The image autoencoder obtained by training on local open images includes: An encoder is built based on a convolutional neural network to convert images into image features and a decoder is built to reconstruct images based on image features. The encoder and decoder are trained based on local open images, and the error between the image reconstructed by the decoder and the original image converted by the encoder is minimized by a pre-built loss function.

3. The image classification and recognition model training method according to claim 1, characterized in that, The image feature training set includes image features and corresponding annotation and classification information.

4. The image classification and recognition model training method according to claim 3, characterized in that, The step of training a pre-established image classification and recognition model based on the image feature training set to obtain a target image classification and recognition model includes: Based on the image features and the labeled classification information, a pre-established image classification and recognition model is trained to obtain a target image classification and recognition model. The target image classification and recognition model predicts the test sample and then obtains the category label corresponding to the test sample.

5. An image classification and recognition model training apparatus for implementing the image classification and recognition model training method as described in any one of claims 1-4, characterized in that, The device includes: The first acquisition module is used to train and acquire an image autoencoder based on local open images; The second acquisition module is used to process the local open image based on the image autoencoder to obtain a first image feature set, and to process the remote sensitive image based on the image autoencoder to obtain a second image feature set; the first image feature set and the second image feature set constitute an image feature training set; The training module is used to train a pre-established image classification and recognition model based on the image feature training set to obtain the target image classification and recognition model.

6. A computer device, characterized in that, include: One or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the method according to any one of claims 1-4.

Images