Feature Extraction Method, Device and Storage Medium for Super-Resolution Image Reconstruction

Abstract

This article discloses a feature extraction method for super-resolution image reconstruction, including: constructing a convolutional neural network model; the model includes an input module, a basic convolutional layer module, an attention mechanism recurrent feedback convolutional module, an upsampling module, a feature fusion module, and an output module; using the model to extract features from the original image: the input module receives the original image data; the basic convolutional layer module extracts features from the original image data to obtain a first feature map; the attention mechanism recurrent feedback convolutional module uses a spatial attention mechanism, a channel attention mechanism, and a recurrent feedback convolutional unit to extract features from the original image data to obtain a second feature map; the upsampling module upsamples the first feature map and the second feature map to obtain a third feature map and a fourth feature map; the feature fusion module performs feature fusion on the third and fourth feature maps to obtain a fifth feature map; the output module outputs the fifth feature map. The solution in this article can enhance the clarity and resolution of the image.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of machine vision, and in particular, to a feature extraction method, device, and storage medium for super-resolution image reconstruction. Background Art

[0002] With the development of computer science and technology and the increasing demand of users for driving environment safety and driving convenience, automobiles are becoming increasingly intelligent. During the process of a driver parking, it is crucial to perceive the complex surrounding environment. In order to obtain a high-quality parking assistance function, existing automobiles are equipped with a reverse camera device, which can obtain the road scene during parking in real time through an external camera sensor device and transmit it to the central control display screen to provide driving safety information for the driver in real time.

[0003] The reverse camera of an automobile is vulnerable to external interference due to the complexity of external environment information, and the video transmitted to the central control screen for imaging is blurred and has a low resolution, which poses a safety hazard for the driver to park. Summary of the Invention

[0004] The embodiments of the present application provide a feature extraction method for super-resolution image reconstruction, including:

[0005] Constructing a convolutional neural network model; wherein, the convolutional neural network model includes an input module, a basic convolutional layer module, an attention mechanism loop feedback convolutional module, an upsampling module, a feature fusion module, and an output module;

[0006] Using the convolutional neural network model to extract features from the original image: the input module receives the data of the original image; the basic convolutional layer module extracts features from the data of the original image using a basic convolutional layer to obtain a first feature map; the attention mechanism loop feedback convolutional module extracts features from the data of the original image using a spatial attention mechanism, a channel attention mechanism, and a loop feedback convolutional unit to obtain a second feature map; the upsampling module upsamples the first feature map to obtain a third feature map and upsamples the second feature map to obtain a fourth feature map; the feature fusion module performs feature fusion on the third feature map and the fourth feature map to obtain a fifth feature map; the output module outputs the fifth feature map.

[0007] The embodiments of the present application provide a feature extraction device for super-resolution image reconstruction, including: a memory and a processor, where the memory stores a computer program, and when the computer program is executed by the processor, the steps of the above feature extraction method for super-resolution image reconstruction are implemented.

[0008] An embodiment of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned feature extraction method for super-resolution image reconstruction are implemented.

[0009] The feature extraction method for super-resolution image reconstruction provided by the embodiment of the present application constructs a convolutional neural network model; wherein, the convolutional neural network model includes an input module, a basic convolutional layer module, a cyclic feedback attention mechanism module, an upsampling module, a feature fusion module, and an output module; the convolutional neural network model is used to extract features from the original image: a large amount of low-level semantic information in the original image is obtained through the basic convolutional layer, providing more feature information for the reconstruction of small targets in the image; the attention of the model to target details is improved through the spatial attention mechanism and the channel attention mechanism, suppressing relatively unimportant environmental feature information; the feature extraction ability of the deep neural network is simulated by the shallow neural network model through the cyclic feedback convolutional unit, but the weight parameters and the number of layers are less, thereby realizing a lightweight model architecture. The feature extraction method provided by the above embodiment can enhance the clarity and resolution of the image.

[0010] Other aspects can be understood after reading and understanding the drawings and the detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings are used to provide an understanding of the technical solutions of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solutions of the present application, and do not constitute a limitation to the technical solutions of the present application.

[0012] Figure 1 It is a flowchart of a feature extraction method for super-resolution image reconstruction according to an embodiment of the present application;

[0013] Figure 2 It is a schematic structural diagram of a convolutional neural network model according to an embodiment of the present application;

[0014] Figure 3 It is a schematic structural diagram of a sub-module of an attention mechanism cyclic feedback convolution module according to an embodiment of the present application;

[0015] Figure 4 It is a schematic structural diagram of a cyclic feedback convolutional unit according to an embodiment of the present application;

[0016] Figure 5 It is a schematic diagram of a feature extraction device for super-resolution image reconstruction according to an embodiment of the present application;

[0017] Figure 6 It is a schematic diagram of another feature extraction device for super-resolution image reconstruction according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] This application describes multiple embodiments, but the description is exemplary rather than restrictive, and it will be apparent to those of ordinary skill in the art that there can be more embodiments and implementation solutions within the scope covered by the embodiments described in this application. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically restricted, any feature or element of any embodiment can be combined with any other feature or element in any other embodiment, or can replace any other feature or element in any other embodiment.

[0019] This application includes and contemplates combinations with features and elements known to those of ordinary skill in the art. The embodiments, features, and elements disclosed in this application can also be combined with any conventional features or elements to form unique inventive solutions defined by the appended claims. Any feature or element of any embodiment can also be combined with features or elements from other inventive solutions to form another unique inventive solution defined by the appended claims. Therefore, it should be understood that any feature shown and / or discussed in this application can be implemented alone or in any suitable combination. Therefore, the embodiments are not subject to other limitations except those imposed by the appended claims and their equivalents. In addition, various modifications and changes can be made within the scope of the appended claims.

[0020] In addition, when describing representative embodiments, the specification may have presented the method and / or process as a specific sequence of steps. However, to the extent that the method or process does not depend on the specific order of the steps described herein, the method or process should not be limited to the specific order of steps described. As will be understood by those of ordinary skill in the art, other step orders are possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation on the appended claims. In addition, the claims directed to the method and / or process should not be limited to performing their steps in the order written, and those skilled in the art can easily understand that these orders can vary and still remain within the spirit and scope of the embodiments of this application.

[0021] As Figure 1 shown, an embodiment of this application provides a feature extraction method for super-resolution image reconstruction, including:

[0022] Step S10, constructing a convolutional neural network model; wherein, the convolutional neural network model includes an input module, a basic convolutional layer module, an attention mechanism recurrent feedback convolutional module, an upsampling module, a feature fusion module, and an output module;

[0023] Step S20, use the convolutional neural network model to extract features from the original image: the input module receives the data of the original image; the basic convolutional layer module uses the basic convolutional layer to extract features from the data of the original image to obtain a first feature map; the attention mechanism recurrent feedback convolutional module uses the spatial attention mechanism, the channel attention mechanism and the recurrent feedback convolutional unit to extract features from the data of the original image to obtain a second feature map; the upsampling module upsamples the first feature map to obtain a third feature map and upsamples the second feature map to obtain a fourth feature map; the feature fusion module fuses the features of the third feature map and the fourth feature map to obtain a fifth feature map; the output module outputs the fifth feature map.

[0024] The feature extraction method for super-resolution image reconstruction provided in the above embodiment constructs a convolutional neural network model; wherein, the convolutional neural network model includes an input module, a basic convolutional layer module, a recurrent feedback attention mechanism module, an upsampling module, a feature fusion module and an output module; use the convolutional neural network model to extract features from the original image: the input module receives the data of the original image; the basic convolutional layer module uses the basic convolutional layer to extract features from the data of the original image to obtain a first feature map; the attention mechanism recurrent feedback convolutional module uses the spatial attention mechanism, the channel attention mechanism and the recurrent feedback convolutional unit to extract features from the data of the original image to obtain a second feature map; the upsampling module upsamples the first feature map to obtain a third feature map and upsamples the second feature map to obtain a fourth feature map; the feature fusion module fuses the features of the third feature map and the fourth feature map to obtain a fifth feature map; the output module outputs the fifth feature map. The above basic convolutional layer module obtains more low-level semantic information in the image through the basic convolutional layer, providing more feature information for the reconstruction of small targets in the image. The above attention mechanism recurrent feedback convolutional module improves the model's attention to target details through the spatial attention mechanism and the channel attention mechanism, suppresses relatively unimportant environmental feature information, and realizes the feature extraction ability of the shallow neural network model simulating the deep neural network through the recurrent feedback convolutional unit, but with fewer weight parameters and layers, thus realizing a lightweight model architecture.

[0025] Figure 2The structure of a convolutional neural network model is shown. The convolutional neural network model includes an input module, a basic convolutional layer module, an attention mechanism recurrent feedback convolutional module, an upsampling module, a feature fusion module, and an output module. The input module is configured to receive the data of the original image; the basic convolutional layer module is configured to extract features from the data of the image using a basic convolutional layer to obtain a first feature map; the attention mechanism recurrent feedback convolutional module is configured to extract features from the data of the image using a spatial attention mechanism, a channel attention mechanism, and a recurrent feedback convolutional unit to obtain a second feature map; the upsampling module is configured to upsample the first feature map to obtain a third feature map and upsample the second feature map to obtain a fourth feature map; the feature fusion module is configured to perform feature fusion on the third feature map and the fourth feature map to obtain a fifth feature map; the output module is configured to output the fifth feature map.

[0026] In some exemplary embodiments, the attention mechanism recurrent feedback convolutional module includes a plurality of serially connected sub-modules, and each sub-module includes an input unit, a first convolutional layer unit, a spatial attention mechanism unit, a channel attention mechanism unit, a recurrent feedback convolutional unit, a fusion unit, and an output unit;

[0027] The input unit receives the data of the original image or the feature map output by the previous sub-module; wherein, the input unit of the first sub-module receives the data of the original image, and the input units of the remaining sub-modules receive the feature map output by the previous sub-module;

[0028] The first convolutional layer unit performs convolutional processing on the input feature map or the data of the original image to generate a sixth feature map;

[0029] The spatial attention mechanism unit selectively compresses the spatial dimension of the sixth feature map using a spatial attention mechanism to generate a seventh feature map;

[0030] The channel attention mechanism unit selectively compresses the channel dimension of the sixth feature map using a channel attention mechanism to generate an eighth feature map;

[0031] The recurrent feedback convolutional unit performs recurrent feedback convolutional processing on the sixth feature map to generate a ninth feature map;

[0032] The fusion unit performs feature fusion on the seventh feature map, the eighth feature map, and the ninth feature map output by the recurrent feedback convolutional unit for the last time to generate a second feature map;

[0033] The output unit outputs the second feature map.

[0034] The above attention mechanism loop feedback convolution module selectively compresses the spatial dimension of the model through the spatial attention mechanism, improves the model's attention to target details through the channel attention mechanism, suppresses relatively unimportant environmental feature information, and realizes the feature extraction ability of a shallow neural network model by simulating that of a deep neural network through multiple serially connected loop feedback convolution units. However, the weight parameters and the number of layers are small, thus realizing a lightweight model architecture.

[0035] Figure 3 Fig. shows a sub-module of an attention mechanism loop feedback convolution module. The sub-module of the attention mechanism loop feedback convolution module includes an input unit, a first convolution layer unit, a spatial attention mechanism unit, a channel attention mechanism unit, a loop feedback convolution unit, a fusion unit, and an output unit.

[0036] In some exemplary embodiments, the basic convolution layer includes a convolution kernel of size a; a is greater than or equal to 4.

[0037] In some exemplary embodiments, the loop feedback convolution unit performs loop feedback convolution processing on the sixth feature map to generate a ninth feature map, including:

[0038] Performing loop feedback convolution operations on the sixth feature map until the number of feedback times reaches the number threshold, and the result output by the last round of convolution operations is output as the ninth feature map to the fusion unit; wherein, each round of convolution operations includes performing the following processing:

[0039] Inputting the sixth feature map into the second convolution layer for feature extraction to obtain a tenth feature map, inputting the tenth feature map into the first BN (Batch Normalization) layer for regularization operation of convolution kernel parameters, and then outputting the regularization result to the activation layer. The activation layer uses an activation function to obtain an eleventh feature map; the eleventh feature map and the sixth feature map are fused through a feature fusion unit to generate a twelfth feature map;

[0040] Inputting the twelfth feature map into the third convolution layer for feature extraction to obtain a thirteenth feature map, inputting the thirteenth feature map into the second BN layer for regularization operation of convolution kernel parameters, and taking the result of the regularization operation as the output result of this round of convolution operations to be fed back to the input of the next round of convolution operations and become the sixth feature map of the next round of convolution operations.

[0041] Figure 4 Fig. shows the structure of a loop feedback convolution unit. The loop feedback convolution unit includes: a second convolution layer, a first BN layer, an activation layer, a feature fusion unit, a third convolution layer, and a second BN layer.

[0042] In some exemplary embodiments, the number threshold is a hyperparameter, and the value of the number threshold can be determined by balancing the inference speed and accuracy requirements of the model.

[0043] In some exemplary embodiments, the attention mechanism recurrent feedback convolution module includes m serially connected sub-modules; the value of m can be determined by balancing the inference speed and accuracy requirements of the model.

[0044] In some exemplary embodiments, the activation function includes the ReLU function.

[0045] In some exemplary embodiments, the resolution of the third feature map is higher than that of the first feature map; the resolution of the fourth feature map is higher than that of the second feature map; the resolution of the fifth feature map is higher than that of the original image.

[0046] As Figure 5 shown, an embodiment of the present disclosure provides a feature extraction device for super-resolution image reconstruction, including:

[0047] A model construction module 100 configured to construct a convolutional neural network model; wherein, the convolutional neural network model includes an input module, a basic convolutional layer module, an attention mechanism recurrent feedback convolution module, an upsampling module, a feature fusion module, and an output module;

[0048] A feature extraction module 200 configured to extract features from an original image by using the convolutional neural network model: the input module receives data of the original image; the basic convolutional layer module extracts features from the data of the original image by using a basic convolutional layer to obtain a first feature map; the attention mechanism recurrent feedback convolution module extracts features from the data of the original image by using a spatial attention mechanism, a channel attention mechanism, and a recurrent feedback convolution unit to obtain a second feature map; the upsampling module upsamples the first feature map to obtain a third feature map and upsamples the second feature map to obtain a fourth feature map; the feature fusion module performs feature fusion on the third feature map and the fourth feature map to obtain a fifth feature map; the output module outputs the fifth feature map.

[0049] The feature extraction device for super-resolution image reconstruction in the above embodiments includes a model construction module and a feature extraction module. The model construction module constructs a convolutional neural network model. Among them, the convolutional neural network model includes an input module, a basic convolutional layer module, an attention mechanism recurrent feedback convolutional module, an upsampling module, a feature fusion module, and an output module. The feature extraction module uses the convolutional neural network model to extract features from the original image: the input module receives the data of the original image; the basic convolutional layer module uses the basic convolutional layer to extract features from the data of the original image to obtain a first feature map; the attention mechanism recurrent feedback convolutional module uses a spatial attention mechanism, a channel attention mechanism, and a recurrent feedback convolutional unit to extract features from the data of the original image to obtain a second feature map; the upsampling module upsamples the first feature map to obtain a third feature map and upsamples the second feature map to obtain a fourth feature map; the feature fusion module performs feature fusion on the third feature map and the fourth feature map to obtain a fifth feature map; the output module outputs the fifth feature map. The feature extraction device provided in the above embodiments can enhance the clarity and resolution of the image.

[0050] In some exemplary embodiments, the attention mechanism recurrent feedback convolutional module includes a plurality of serially connected sub-modules. Each sub-module includes an input unit, a first convolutional layer unit, a spatial attention mechanism unit, a channel attention mechanism unit, a recurrent feedback convolutional unit, a fusion unit, and an output unit.

[0051] The input unit receives the data of the original image or the feature map output by the previous sub-module. Among them, the input unit of the first sub-module receives the data of the original image, and the input units of the remaining sub-modules receive the feature map output by the previous sub-module.

[0052] The first convolutional layer unit performs convolutional processing on the input feature map or the data of the original image to generate a sixth feature map.

[0053] The spatial attention mechanism unit uses the spatial attention mechanism to selectively compress the spatial dimension of the sixth feature map to generate a seventh feature map.

[0054] The channel attention mechanism unit uses the channel attention mechanism to selectively compress the channel dimension of the sixth feature map to generate an eighth feature map.

[0055] The recurrent feedback convolutional unit performs recurrent feedback convolutional processing on the sixth feature map to generate a ninth feature map.

[0056] The fusion unit performs feature fusion on the seventh feature map, the eighth feature map, and the ninth feature map output by the recurrent feedback convolutional unit for the last time to generate a second feature map.

[0057] The output unit outputs the second feature map.

[0058] In some exemplary embodiments, the cyclic feedback convolution unit is configured to perform cyclic feedback convolution processing on the sixth feature map to generate the ninth feature map in the following manner: perform cyclic feedback convolution operations on the sixth feature map until the number of feedback times reaches a threshold number of times, and the result output by the last round of convolution operations is output as the ninth feature map to the fusion unit; wherein, each round of convolution operations includes performing the following processing: input the sixth feature map into the second convolutional layer for feature extraction to obtain the tenth feature map, input the tenth feature map into the first BN layer for regularization operation of convolutional kernel parameters, and then output the regularization result to the activation layer, and the activation layer uses an activation function to obtain the eleventh feature map; perform feature fusion on the eleventh feature map and the sixth feature map through a feature fusion unit to generate the twelfth feature map; input the twelfth feature map into the third convolutional layer for feature extraction to obtain the thirteenth feature map, input the thirteenth feature map into the second BN layer for regularization operation of convolutional kernel parameters, and use the result of the regularization operation as the output result of this round of convolution operations to be fed back to the input of the next round of convolution operations and become the sixth feature map of the next round of convolution operations.

[0059] In some exemplary embodiments, the threshold number of times is a hyperparameter, and the value of the threshold number of times is determined by balancing the inference speed and accuracy requirements of the model.

[0060] In some exemplary embodiments, the attention mechanism cyclic feedback convolution module includes m serially connected sub-modules; the value of m is determined by balancing the inference speed and accuracy requirements of the model.

[0061] In some exemplary embodiments, the activation function includes the ReLU function.

[0062] In some exemplary embodiments, the resolution of the third feature map is higher than that of the first feature map; the resolution of the fourth feature map is higher than that of the second feature map; the resolution of the fifth feature map is higher than that of the original image.

[0063] In some exemplary embodiments, the basic convolutional layer includes a convolutional kernel with a convolutional kernel size of a; a is greater than or equal to 4.

[0064] As Figure 6 shown, an embodiment of the present disclosure provides a feature extraction device for super-resolution image reconstruction, including: a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the steps of the above-mentioned feature extraction method for super-resolution image reconstruction are implemented.

[0065] An embodiment of the present disclosure provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned feature extraction method for super-resolution image reconstruction are implemented.

[0066] Those of ordinary skill in the art can understand that all or some of the steps in the above-disclosed methods, and the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, and appropriate combinations thereof. In the hardware implementation, the division of the functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or one function or step may be executed by several physical components in cooperation. Some or all of the components may be implemented as software executed by a processor, such as a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and may include any information delivery medium.

Claims

1. A feature extraction method for super-resolution image reconstruction, including: Constructing a convolutional neural network model; wherein, the convolutional neural network model includes an input module, a basic convolutional layer module, an attention mechanism loop feedback convolutional module, an upsampling module, a feature fusion module, and an output module; Using the convolutional neural network model to extract features from the original image: The input module receives the data of the original image; the basic convolutional layer module extracts features from the data of the original image using a basic convolutional layer to obtain a first feature map; the attention mechanism loop feedback convolutional module uses a spatial attention mechanism unit, a channel attention mechanism unit, and a loop feedback convolutional unit to extract features from the data of the original image respectively and perform feature fusion to obtain a second feature map; the upsampling module upsamples the first feature map to obtain a third feature map and upsamples the second feature map to obtain a fourth feature map; the feature fusion module performs feature fusion on the third feature map and the fourth feature map to obtain a fifth feature map; the output module outputs the fifth feature map; wherein, the spatial attention mechanism unit adopts a spatial attention mechanism, the channel attention mechanism unit adopts a channel attention mechanism, and the loop feedback convolutional unit adopts a loop feedback convolutional mechanism.

2. The method according to claim 1, characterized in that: The attention mechanism loop feedback convolutional module includes a plurality of serially connected sub-modules, and each sub-module includes an input unit, a first convolutional layer unit, a spatial attention mechanism unit, a channel attention mechanism unit, a loop feedback convolutional unit, a fusion unit, and an output unit; The input unit receives the data of the original image or the feature map output by the previous sub-module; wherein, the input unit of the first sub-module receives the data of the original image, and the input units of the remaining sub-modules receive the feature map output by the previous sub-module; The first convolutional layer unit performs convolutional processing on the input feature map or the data of the original image to generate a sixth feature map; The spatial attention mechanism unit adopts a spatial attention mechanism to selectively compress the spatial dimension of the sixth feature map to generate a seventh feature map; The channel attention mechanism unit adopts a channel attention mechanism to selectively compress the channel dimension of the sixth feature map to generate an eighth feature map; The loop feedback convolutional unit performs loop feedback convolutional processing on the sixth feature map to generate a ninth feature map; The fusion unit performs feature fusion on the seventh feature map, the eighth feature map, and the ninth feature map output by the loop feedback convolutional unit for the last time to generate a second feature map; The output unit outputs the second feature map.

3. The method according to claim 2, characterized in that: The loop feedback convolutional unit performing loop feedback convolutional processing on the sixth feature map to generate a ninth feature map includes: Performing loop feedback convolutional operations on the sixth feature map until the number of feedback times reaches a threshold number of times, and the result output by the last round of convolutional operations is output as the ninth feature map to the fusion unit; wherein, each round of convolutional operations includes performing the following processing: The sixth feature map is input into the second convolutional layer for feature extraction to obtain the tenth feature map. The tenth feature map is input into the first batch normalization (BN) layer for the regularization operation of the convolutional kernel parameters, and then the regularization result is output to the activation layer. The activation layer uses the activation function to obtain the eleventh feature map; the eleventh feature map and the sixth feature map are fused by the feature fusion unit to generate the twelfth feature map; The twelfth feature map is input into the third convolutional layer for feature extraction to obtain the thirteenth feature map. The thirteenth feature map is input into the second BN layer for the regularization operation of the convolutional kernel parameters. The result of the regularization operation is fed back as the output result of this round of convolutional operation to the input of the next round of convolutional operation and becomes the sixth feature map of the next round of convolutional operation.

4. The method according to claim 3, wherein: the number threshold is a hyperparameter, and the value of the number threshold is determined by balancing the inference speed and accuracy requirements of the model.

5. The method according to claim 2, wherein: the attention mechanism cyclic feedback convolutional module includes m serially connected sub-modules; the value of m is determined by balancing the inference speed and accuracy requirements of the model.

6. The method according to claim 3, wherein: the activation function includes the ReLU function.

7. The method according to claim 1, wherein: the resolution of the third feature map is higher than that of the first feature map; the resolution of the fourth feature map is higher than that of the second feature map; the resolution of the fifth feature map is higher than that of the original image.

8. The method according to claim 1, wherein: the basic convolutional layer includes a convolutional kernel with a size of a; a is greater than or equal to 4.

9. A feature extraction device for super-resolution image reconstruction, comprising: a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the steps of the feature extraction method for super-resolution image reconstruction according to any one of claims 1-8 above are implemented.

10. A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the feature extraction method for super-resolution image reconstruction according to any one of claims 1-8 above are implemented.

Images