Visual classification method and device based on adaptive multi-resolution feature fusion, electronic equipment and storage medium

Abstract

The invention provides a visual classification method and device based on adaptive multi-resolution feature fusion, electronic equipment and a storage medium, and relates to the technical field of computer vision. The method comprises the following steps: acquiring an initial feature of an image through a classification model, and processing the initial feature by adopting an adaptive pooling layer based on meta-learning in the classification model to obtain a low-resolution image feature; inputting the initial features and the low-resolution image features into a feature extractor, and executing different classification constraint tasks to respectively obtain high-resolution classification features and low-resolution classification features; in the classification model, fusing the low-resolution classification features and the high-resolution classification features to obtain fused features; and transmitting the fusion features into a fusion classifier in the classification model to obtain a visual classification result of the image. According to the method, the adaptive pooling layer is obtained by using the meta-learning strategy, the low-resolution feature most complementary to the high-resolution feature can be selected, and the discriminative feature of the fine-grained image is obtained.

Description

Technical Field

[0001] The present disclosure relates to the field of computer vision technology, and in particular to a visual classification method, device, electronic device and storage medium based on adaptive multi-resolution feature fusion. Background Art

[0002] Visual classification involves the complex task of distinguishing subcategories within a given object class. This task is particularly important in the field of computer vision, as it requires accurate recognition and classification between highly similar categories. Object classification in images has a wide range of applications, including classification tasks in birds, flowers, cars, animal species, and various other specialized fields.

[0003] In related technologies, it is necessary to collect sample data from multiple sources, which consumes a lot of time, human resources, and data labeling costs. Summary of the invention

[0004] In view of this, the purpose of the present disclosure is to propose a visual classification method, device, electronic device and storage medium based on adaptive multi-resolution feature fusion, which can solve the existing problems in a targeted manner.

[0005] Based on the above-mentioned purpose, in a first aspect, the present disclosure proposes a visual classification method based on adaptive multi-resolution feature fusion, comprising: obtaining initial features of an image through a classification model, and using an adaptive pooling layer based on meta-learning in the classification model to process the initial features to obtain low-resolution image features; inputting the initial features and the low-resolution image features into a feature extractor, and performing different classification constraint tasks to obtain high-resolution classification features and low-resolution classification features, respectively; in the classification model, fusing the low-resolution classification features and the high-resolution classification features to obtain fused features; and passing the fused features into a fusion classifier in the classification model to obtain a visual classification result of the image.

[0006] In the second aspect, a visual classification device based on adaptive multi-resolution feature fusion is also provided, including: a first processing unit, configured to obtain initial features of an image through a classification model, and use an adaptive pooling layer based on meta-learning in the classification model to process the initial features to obtain low-resolution image features; a second processing unit, configured to input the initial features and the low-resolution image features into a feature extractor, and perform different classification constraint tasks to obtain high-resolution classification features and low-resolution classification features, respectively; a fusion unit, configured to fuse the low-resolution classification features and the high-resolution classification features in the classification model to obtain fused features; and a classification unit, configured to pass the fused features into a fusion classifier in the classification model to obtain a visual classification result of the image.

[0007] According to a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first aspect.

[0008] According to a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program is executed by a processor to implement any method described in the first aspect.

[0009] In a fifth aspect, a computer program product is also provided, including a computer program, wherein the computer program is executed by a processor to implement any method described in the first aspect.

[0010] In general, the present disclosure has at least the following beneficial effects: different resolution features are used as multi-source information to perform effective multi-source information fusion. The difficulty of multi-source information collection is avoided, and the time cost, human resources, and data annotation cost of data collection are saved. In addition, the adaptive pooling layer is obtained by using the meta-learning strategy, which can realize the selection of low-resolution features that are most complementary to high-resolution features, and obtain discriminative features of fine-grained images. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in accordance with the present disclosure and should not be regarded as limiting the scope of the present disclosure.

[0012] Figure 1 A flowchart of a visual classification method based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown;

[0013] Figure 2 A schematic diagram of the structure of a classification model in a visual classification method based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown;

[0014] Figure 3a A schematic diagram of the structure of an adaptive pooling layer in a classification model in a visual classification method based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown;

[0015] Figure 3b A schematic diagram of the training process of meta-learning in a classification model in a visual classification method based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown;

[0016] Figure 4 A schematic diagram of a visual classification device based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown;

[0017] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure is shown;

[0018] Figure 6 A schematic diagram of a storage medium provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0019] The present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0020] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0021] Figure 1 The visual classification method based on adaptive multi-resolution feature fusion of the present disclosure is shown. In an embodiment of the present disclosure, the method includes:

[0022] Step S101, obtaining initial features of an image through a classification model, and using an adaptive pooling layer based on meta-learning in the classification model to process the initial features to obtain low-resolution image features.

[0023] In this embodiment, the execution subject of the visual classification method based on adaptive multi-resolution feature fusion can obtain the initial features of the image through the classification model. For example, the initial features of the image are obtained through the convolution layer in the classification model. Afterwards, the above execution subject can use the adaptive pooling layer in the classification model to process the initial features to obtain low-resolution image features.

[0024] The initial features here are image features obtained by convolution and other methods. The adaptive pooling layer is used to obtain low-resolution image features from the initial features. Low-resolution image features are image features corresponding to low-resolution images. The low resolution here refers to a resolution lower than a preset resolution threshold.

[0025] Step S102, inputting the initial features and the low-resolution image features into a feature extractor, and performing different classification constraint tasks to obtain high-resolution classification features and low-resolution classification features respectively.

[0026] In this embodiment, the execution subject may input the initial features and the low-resolution image features into the feature extractor. The feature extractor processes the initial features and the low-resolution image features and performs different classification constraint tasks respectively, thereby obtaining high-resolution classification features and low-resolution classification features. Specifically, the initial features correspond to the high-resolution classification features, and the low-resolution image features correspond to the low-resolution classification features.

[0027] Step S103: In the classification model, the low-resolution classification features and the high-resolution classification features are fused to obtain fused features.

[0028] In this embodiment, the execution subject may fuse the low-resolution classification features and the high-resolution classification features in the classification model, and the fusion result is the fusion feature.

[0029] Step S104, transferring the fused features to the fusion classifier in the classification model to obtain a visual classification result of the image.

[0030] In this embodiment, the execution entity may pass the fusion features into the fusion classifier in the classification model to classify the objects presented in the image and obtain a visual classification result of the image.

[0031] This embodiment can use features of different resolutions as multi-source information to perform effective multi-source information fusion. This avoids the difficulty of collecting multi-source information, saves the time cost, human resources, and data annotation cost of data collection. In addition, by using the meta-learning strategy to obtain an adaptive pooling layer, it is possible to select the low-resolution features that are most complementary to the high-resolution features, and obtain discriminative features for fine-grained images.

[0032] In some optional implementations of any embodiment of the present disclosure, the training step of the adaptive pooling layer includes: optimizing the parameters of the adaptive pooling layer by using a meta-learning training method through the loss function of the adaptive pooling layer, wherein the loss function of the adaptive pooling layer includes a loss function of an inner loop and a loss function of an outer loop.

[0033] Optionally, the loss function of the adaptive pooling layer is the sum of the loss function of the inner loop and the loss function of the outer loop; the loss function of the inner loop includes the loss function of the low-resolution classification feature, and the loss function of the outer loop is calculated by the loss function of the high-resolution classification feature and the mixed loss function, and the calculation is to determine the quotient of the mixed loss function and the loss function of the high-resolution classification feature.

[0034] In some optional implementations of any embodiment of the present disclosure, the training step of the classification model includes: training the feature extractor, the fusion classifier and the expert classifier for performing different classification constraint tasks in the classification model by a joint training method through a universal loss function of the classification model; the universal loss function is the sum of the loss function of the inner loop, the loss function of the outer loop and the mixed loss function.

[0035] In some optional implementations of any embodiment of the present disclosure, the initial features and the low-resolution image features are input into a feature extractor, and different classification constraint tasks are performed to obtain low-resolution classification features and high-resolution classification features, respectively, including: inputting the initial features into the feature extractor to obtain high-resolution features; inputting the low-resolution image features into the feature extractor to obtain low-resolution features; and performing different classification constraint tasks on the low-resolution features and the high-resolution features to obtain low-resolution classification features and high-resolution classification features, respectively.

[0036] Optionally, the different classification constraint tasks include a first classification constraint task and a second classification constraint task; the classification model includes a first expert classifier for performing the first classification constraint task, and a second expert classifier for performing the second classification constraint task; the performing different classification constraint tasks on the low-resolution features and the high-resolution features to obtain low-resolution classification features and high-resolution classification features, respectively, includes: using the first expert classifier to perform the first classification constraint task on the low-resolution features to obtain the low-resolution classification features; using the second expert classifier to perform the second classification constraint task on the high-resolution features to obtain the high-resolution classification features.

[0037] In some optional implementations of any one of the embodiments of the present disclosure, the adaptive pooling layer includes a feature selection module and an average pooling layer; the step of inputting the initial features into the adaptive pooling layer based on meta-learning in the classification model to obtain low-resolution image features comprises: inputting the initial features into the feature selection module based on a gating mechanism to obtain a vector of a specified dimension; and inputting the vector of the specified dimension and the initial features into the average pooling layer to obtain the low-resolution image features.

[0038] In some optional implementations of any embodiment of the present disclosure, the process of the visual classification method based on adaptive multi-resolution feature fusion is as follows:

[0039] Step 1. Input image x∈R W×W×H , then the obtained deep features are:

[0040]

[0041] Where W represents the width of the input image and H represents the height of the input image.

[0042] The steps of obtaining low-resolution features as described above include:

[0043] Step 2a. First, create a set of average pooling layers {AvgPool 1 ,AvgPool 2 ,…,AvgPool M}:

[0044] AvgPool m =AdaptiveAvgPool(f,k)

[0045] Where m = {1, 2, ..., M}, AdaptiveAvgPool (·) is the average pooling layer, f is the extracted deep feature, and k is the size of the feature output by the average pooling layer.

[0046] Step 2b. Design a feature selection module G(·) based on the gating mechanism:

[0047] G(·)=GumbleSoftmax(FC(EUL(BN(·))))

[0048]

[0049] Among them, BN(·) is the batch normalization function, ELU(·) is the nonlinear activation function, FC(·) is the fully connected layer with output dimension M. GumbleSoftmax(·) is the Gumble-Softmax function, and GumbleSample(·) represents Gumble sampling.

[0050] Step 2c. Input the deep features into the feature selection module based on the gating mechanism and output an M-dimensional vector α:

[0051] α=G(f)

[0052] Among them, G(·) will set the position with the largest output value of FC(EUL(BN(·))) to 1 and the other positions to 0.

[0053] Step 2d. The output of the final adaptive pooling layer is f ′ :

[0054]

[0055] The steps of obtaining the unique low-resolution features and the unique high-resolution features as described above include:

[0056] Step 3a. Through a unique feature extractor, obtain unique low-resolution features f 1 * and unique high-resolution features

[0057]

[0058]

[0059] E n (·)=ReLU(BN n (Conv n (·)))

[0060] Where n = {1, 2}, Conv n (·) is a 1×1 convolutional layer, which is used to extract low-resolution and high-resolution unique features of the image. Its input and output channels are the same as f. BN n (·) is the batch normalization function, and ReLU(·) is the nonlinear activation function.

[0061] Step 3b. Pass two expert classifiers Constrain high-resolution features and low-resolution features separately so that they have their own characteristics:

[0062]

[0063] Where y is the correct label of the input data, is the cross entropy loss, Only used during the training phase and discarded during the analysis phase.

[0064] The steps of ordinary optimization and meta-learning optimization as described above include:

[0065] Step 4a. Fuse the unique low-resolution features with the unique high-resolution features to obtain the fused features:

[0066]

[0067] Step 4b. Input the fusion features into the fusion classifier to obtain the prediction results:

[0068]

[0069] Step 4c. Use the common joint training method to train the parts other than the adaptive pooling layer, including: feature extractor Two unique feature extractors E 1 (·) and E 2 (·), fusion classifier Two expert classifiers and

[0070]

[0071] L trivial =L low +L high +L fusion

[0072] Loss = L trivial (Θ)

[0073] Here, Θ represents the parameters except for the adaptive pooling layer.

[0074] Step 4d. Use the meta-learning strategy to train the parameters of the adaptive pooling layer to obtain the most complementary combination of low-resolution features and high-resolution features. The complementarity here can refer to the complementarity of the locations of the detected features in the image. For example, the high-resolution features capture the mouth features, and the low-resolution features capture the wings features, because the captured discriminative features are different. The loss function of the inner loop is:

[0075]

[0076] Among them, θ represents the parameters of the adaptive pooling layer, and the parameters of the adaptive pooling layer are updated after optimization as β is the learning rate.

[0077] Step 4e. The loss function of the outer loop is:

[0078]

[0079] Step 4f. The inner loop and the outer loop are optimized simultaneously, and the adaptive pooling layer is trained using the meta-learning loss function. The meta-learning loss function is:

[0080]

[0081] Loss = L meta (θ)

[0082] Through the method of the present invention, a fine-grained visual classification model based on adaptive multi-resolution feature fusion is obtained, which can use the image itself to construct multi-source information and reduce the difficulty of data collection. And through the meta-learning strategy, the best low-resolution and high-resolution feature combination is selected to obtain more discriminative fine-grained features. It promotes the application of fine-grained visual classification algorithms based on multi-source information in practice.

[0083] Figure 2 A structural schematic diagram of a classification model in a visual classification method based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown.

[0084] Figure 3a A schematic diagram of the structure of an adaptive pooling layer in a classification model in a visual classification method based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown.

[0085] Figure 3b A schematic diagram of the training process of meta-learning in a classification model in a visual classification method based on adaptive multi-resolution feature fusion according to an embodiment of the present disclosure is shown.

[0086] The embodiment of the present disclosure provides a visual classification device based on adaptive multi-resolution feature fusion, which is used to execute the visual classification method based on adaptive multi-resolution feature fusion described in the above embodiment, such as Figure 4 As shown, the device includes: a first processing unit 401, configured to obtain initial features of an image through a classification model, and use an adaptive pooling layer based on meta-learning in the classification model to process the initial features to obtain low-resolution image features; a second processing unit 402, configured to input the initial features and the low-resolution image features into a feature extractor, and perform different classification constraint tasks to obtain high-resolution classification features and low-resolution classification features respectively; a fusion unit 403, configured to fuse the low-resolution classification features and the high-resolution classification features in the classification model to obtain fused features; a classification unit 404, configured to pass the fused features into a fusion classifier in the classification model to obtain a visual classification result of the image.

[0087] The visual classification device based on adaptive multi-resolution feature fusion provided by the above-mentioned embodiments of the present disclosure and the visual classification method based on adaptive multi-resolution feature fusion provided by the embodiments of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0088] The disclosed embodiment also provides an electronic device corresponding to the visual classification method based on adaptive multi-resolution feature fusion provided in the above embodiment, so as to execute the above visual classification method based on adaptive multi-resolution feature fusion. The disclosed embodiment is not limited.

[0089] Please refer to Figure 5 , which shows a schematic diagram of an electronic device provided by some embodiments of the present disclosure. Figure 5 As shown, the electronic device 50 includes: a processor 500, a memory 501, a bus 502 and a communication interface 503, and the processor 500, the communication interface 503 and the memory 501 are connected via the bus 502; the memory 501 stores a computer program that can be run on the processor 500, and the processor 500 executes the method provided in any of the aforementioned embodiments of the present disclosure when running the computer program.

[0090] The memory 501 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 503 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0091] The bus 502 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 501 is used to store programs, and the processor 500 executes the programs after receiving the execution instructions. The visual classification method based on adaptive multi-resolution feature fusion disclosed in any implementation of the aforementioned embodiments of the present disclosure may be applied to the processor 500, or implemented by the processor 500.

[0092] The processor 500 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 500 or the instruction in the form of software. The above processor 500 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present disclosure can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present disclosure can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 501, and the processor 500 reads the information in the memory 501 and completes the steps of the above method in combination with its hardware.

[0093] The electronic device provided by the embodiment of the present disclosure and the visual classification method based on adaptive multi-resolution feature fusion provided by the embodiment of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented by them.

[0094] The present disclosure also provides a computer-readable storage medium corresponding to the visual classification method based on adaptive multi-resolution feature fusion provided in the above embodiment. Figure 6 The computer-readable storage medium shown is a CD 60 on which a computer program (ie, a program product) is stored. When the computer program is run by a processor, the visual classification method based on adaptive multi-resolution feature fusion provided in any of the aforementioned embodiments will be executed.

[0095] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0096] The computer-readable storage medium provided by the above-mentioned embodiments of the present disclosure and the visual classification method based on adaptive multi-resolution feature fusion provided by the embodiments of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0097] It should be noted that:

[0098] In the above text, the term "comprises", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be noted that the scope of the method and device in the embodiments of the present disclosure is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0099] 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 disclosure, 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, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present disclosure.

[0100] The embodiments of the present disclosure are described above in conjunction with the accompanying drawings, which are only specific implementation modes of the present disclosure. However, the present disclosure is not limited to the above-mentioned specific implementation modes. The above-mentioned specific implementation modes are merely illustrative and not restrictive. Under the guidance of the present disclosure, ordinary technicians in this field can also make many forms without departing from the purpose of the present disclosure and the scope of protection of the claims, all of which are within the protection of the present disclosure.

Claims

1. A visual classification method based on adaptive multi-resolution feature fusion, characterized in that: include: Acquire initial features of the image through a classification model, and use an adaptive pooling layer based on meta-learning in the classification model to process the initial features to obtain low-resolution image features; Inputting the initial features and the low-resolution image features into a feature extractor, and performing different classification constraint tasks to obtain high-resolution classification features and low-resolution classification features respectively; In the classification model, the low-resolution classification features and the high-resolution classification features are fused to obtain fused features; The fused features are passed into the fusion classifier in the classification model to obtain the visual classification result of the image.

2. The method according to claim 1, characterized in that The training steps of the adaptive pooling layer include: The parameters of the adaptive pooling layer are optimized by using the loss function of the adaptive pooling layer and a meta-learning training method, wherein the loss function of the adaptive pooling layer includes a loss function of an inner loop and a loss function of an outer loop.

3. The method according to claim 2, characterized in that The loss function of the adaptive pooling layer is the sum of the loss function of the inner loop and the loss function of the outer loop; The loss function of the inner loop includes the loss function of the low-resolution classification feature, and the loss function of the outer loop is calculated by the loss function of the high-resolution classification feature and the mixed loss function, and the calculation is to determine the quotient of the mixed loss function and the loss function of the high-resolution classification feature.

4. The method according to claim 1, characterized in that: The training steps of the classification model include: By using the universal loss function of the classification model, a joint training method is adopted to train the feature extractor, the fusion classifier and the expert classifier for performing different classification constraint tasks in the classification model; The universal loss function is the sum of the loss function of the inner loop, the loss function of the outer loop, and the mixed loss function.

5. The method according to claim 1, characterized in that: The initial features and the low-resolution image features are input into a feature extractor, and different classification constraint tasks are performed to obtain low-resolution classification features and high-resolution classification features, respectively, including: Inputting the initial features into the feature extractor to obtain high-resolution features; Inputting the low-resolution image features into the feature extractor to obtain low-resolution features; Different classification constraint tasks are performed on the low-resolution features and the high-resolution features to obtain low-resolution classification features and high-resolution classification features respectively.

6. The method according to claim 5, characterized in that The different classification constraint tasks include a first classification constraint task and a second classification constraint task; the classification model includes a first expert classifier for performing the first classification constraint task, and a second expert classifier for performing the second classification constraint task; The performing different classification constraint tasks on the low-resolution features and the high-resolution features to obtain low-resolution classification features and high-resolution classification features respectively includes: Using the first expert classifier, executing the first classification constraint task on the low-resolution feature to obtain the low-resolution classification feature; The second expert classifier is used to perform the second classification constraint task on the high-resolution feature to obtain the high-resolution classification feature.

7. The method according to claim 1, characterized in that The adaptive pooling layer includes a feature selection module and an average pooling layer; the initial features are input into the adaptive pooling layer based on meta-learning in the classification model to obtain low-resolution image features, including: Inputting the initial features into the feature selection module based on the gating mechanism to obtain a vector of a specified dimension; The vector of the specified dimension and the initial feature are input into the average pooling layer to obtain the low-resolution image feature.

8. A visual classification device based on adaptive multi-resolution feature fusion, characterized in that: include: A first processing unit is configured to obtain initial features of an image through a classification model, and process the initial features using an adaptive pooling layer based on meta-learning in the classification model to obtain low-resolution image features; A second processing unit is configured to input the initial features and the low-resolution image features into a feature extractor and perform different classification constraint tasks to obtain high-resolution classification features and low-resolution classification features respectively; A fusion unit is configured to fuse the low-resolution classification features and the high-resolution classification features in the classification model to obtain a fusion feature; The classification unit is configured to pass the fusion features into the fusion classifier in the classification model to obtain the visual classification result of the image.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor runs the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to any one of claims 1 to 7.

Images