Image feature enhancement method and device, electronic equipment and product

Abstract

The application discloses an image feature enhancement method and device, electronic equipment and product, and relates to the technical field of image feature enhancement, and comprises the following steps: acquiring a remote sensing image; inputting the remote sensing image into an image feature enhancement model to obtain a remote sensing image feature map; the model comprises a U-Net module, a multi-direction enhancement convolution module, a gate dynamic adjustment module, an efficient attention weighting module and a cross-channel splicing module; the U-Net module extracts an original feature map of the remote sensing image; the multi-direction enhancement convolution module performs multi-direction convolution on the original feature map to obtain a multi-direction first feature map; the gate dynamic adjustment module dynamically adjusts the first feature map to obtain a multi-direction second feature map; the efficient attention weighting module extracts a multi-direction feature intensity weight, and fuses the second feature map and the feature intensity weight to obtain a multi-direction fusion feature; and the cross-channel splicing module is used for splicing the multi-direction fusion feature. The application can improve the feature extraction precision of a target.

Description

Technical Field

[0001] This invention belongs to the field of image feature enhancement technology, and specifically relates to an image feature enhancement method, device, electronic device and product. Background Technology

[0002] In remote sensing image recognition and segmentation, a commonly used approach is to extract image features from remote sensing images using a convolutional neural network (CNN).

[0003] However, when using traditional convolutional neural networks for image feature extraction, three prominent problems are encountered: insufficient directional sensitivity, weakened feature differences, and computational redundancy. It is difficult to efficiently capture the variable spatial orientation features of ground objects, and it is easy to lose key information such as target edges and textures, which seriously restricts the feature extraction accuracy of complex ground objects.

[0004] Therefore, how to provide an effective solution to improve the accuracy of target feature extraction has become an urgent problem to be solved in the existing technology. Summary of the Invention

[0005] The purpose of this invention is to provide an image feature enhancement method, apparatus, electronic device, and product to solve the aforementioned problems existing in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides an image feature enhancement method, characterized in that it includes:

[0008] Acquire the remote sensing image to be extracted;

[0009] The remote sensing image is input into a pre-trained image feature enhancement model for feature extraction to obtain a remote sensing image feature map of the remote sensing image;

[0010] The image feature enhancement model includes a U-NET module, a multi-directional enhancement convolution module, a gated dynamic adjustment module, an efficient attention weighting module, and a cross-channel stitching module.

[0011] The U-NET module is used to extract features from the remote sensing image to obtain the original feature map;

[0012] The multi-directional enhanced convolution module is used to perform convolution operations in multiple directions on the original feature map to obtain a first feature map in multiple directions;

[0013] The gated dynamic adjustment module is used to dynamically adjust the first feature map in the multiple directions to obtain the second feature map in the multiple directions;

[0014] The efficient attention weighting module is used to extract the feature intensity weights of the original feature map in multiple directions to obtain the feature intensity weights in multiple directions, and to perform weighted fusion of the second feature map in multiple directions with the feature intensity weights in multiple directions to obtain the fused features in multiple directions.

[0015] The cross-channel stitching module is used to stitch together the fused features from multiple directions to obtain the remote sensing image feature map.

[0016] In one possible design, the multi-directional enhancement convolution module is used to mirror-patch the original feature map and perform convolution operations in multiple directions on the mirror-patched original feature map to obtain a first feature map in multiple directions.

[0017] In one possible design, the multiple directions include a horizontal direction, a vertical direction, a 45° diagonal direction, and a 135° diagonal direction.

[0018] In one possible design, when the multi-directional enhancement convolution module performs convolution operations in multiple directions on the original feature map, the horizontal convolution kernel is... The vertical convolution kernel is The convolution kernel at a 45° diagonal direction is The convolution kernel in the 135° diagonal direction is , where c represents the center index of the convolution kernel, and i and j represent the horizontal and vertical indices of the pixel, respectively.

[0019] In one possible design, the efficient attention weighting module is used to weight and fuse the second feature maps in multiple directions with feature intensity weights in multiple directions according to the following formula;

[0020]

[0021] Among them, Y attn Y represents the fusion features in multiple directions. h This represents the second feature map in the horizontal direction, Y. v This represents the second feature map in the vertical direction, Y. d45 This represents the second feature map along the 45° diagonal direction, Y. d135 This represents the second feature map along the 135° diagonal direction. This represents the XOR operation, and A represents the feature intensity weights in multiple directions.

[0022] In one possible design, the gated dynamic adjustment module is used to dynamically adjust the first feature map in the plurality of directions according to the following formula;

[0023]

[0024] Among them, Y conv This represents the first feature map in the plurality of directions, and σ represents the activation function. G represents the XOR operation. k and S k The gating coefficient and scaling factor are represented in sequence, X represents the original feature map, BN represents batch normalization, ReLU represents the ReLU activation function, and Conv represents the convolution operation.

[0025] In one possible design, the efficient attention weighting module is used to extract feature intensity weights of the original feature map in multiple directions according to the following formula;

[0026]

[0027] Where A represents the feature intensity weights in multiple directions, X represents the original feature map, σ represents the activation function, BN represents batch normalization, ReLU represents the ReLU activation function, and W1 and W2 represent the convolution weights, respectively.

[0028] In a second aspect, the present invention provides an image feature enhancement apparatus, comprising:

[0029] The acquisition unit is used to acquire the remote sensing image to be extracted;

[0030] The feature extraction unit is used to input the remote sensing image into a pre-trained image feature enhancement model for feature extraction, and obtain the remote sensing image feature map of the remote sensing image;

[0031] The image feature enhancement model includes a U-NET module, a multi-directional enhancement convolution module, a gated dynamic adjustment module, an efficient attention weighting module, and a cross-channel stitching module.

[0032] The U-NET module is used to extract features from the remote sensing image to obtain the original feature map;

[0033] The multi-directional enhanced convolution module is used to perform convolution operations in multiple directions on the original feature map to obtain a first feature map in multiple directions;

[0034] The gated dynamic adjustment module is used to dynamically adjust the first feature map in the multiple directions to obtain the second feature map in the multiple directions;

[0035] The efficient attention weighting module is used to extract the feature intensity weights of the original feature map in multiple directions to obtain the feature intensity weights in multiple directions, and to perform weighted fusion of the second feature map in multiple directions with the feature intensity weights in multiple directions to obtain the fused features in multiple directions.

[0036] The cross-channel stitching module is used to stitch together the fused features from multiple directions to obtain the remote sensing image feature map.

[0037] Thirdly, the present invention provides an electronic device comprising a memory, a processor, and a transceiver connected in sequence and communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the image feature enhancement method as described in the first aspect or any possible design of the first aspect.

[0038] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform the image feature enhancement method described in the first aspect or any possible design of the first aspect.

[0039] Fifthly, the present invention provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the image feature enhancement method as described in the first aspect or any possible design of the first aspect.

[0040] Beneficial effects:

[0041] This invention extracts features from multiple directions using a multi-directional enhanced convolution module, comprehensively covering the main spatial orientations and effectively capturing the varied spatial orientation features of ground features. This solves the directional blind spot problem of traditional convolutional neural networks. Simultaneously, a gating dynamic modulation mechanism is introduced to dynamically adjust features and enhance feature diversity. Furthermore, an efficient attention weighting module is used to adjust the feature intensity weights in multiple directions, achieving intelligent direction selection and accurately focusing on the dominant direction of the target while suppressing noise interference. Then, feature fusion enhancement from multiple directions enables directional optimization of feature maps in different directions. Finally, the fused features from multiple directions are stitched together across channels to enhance image directional features and preserve key information such as texture, improving the overall completeness and extraction accuracy of structured ground features in remote sensing images.

[0042] Furthermore, by mirroring the original feature map, image edge information can be preserved, further improving the overall recognition integrity and boundary extraction accuracy of structured features in remote sensing images. This is especially suitable for scenarios with obvious directional features, such as road networks and building complexes, providing reliable technical support for high-precision geographic information feature extraction and facilitating practical application and promotion. Attached Figure Description

[0043] Figure 1 A flowchart of the image feature enhancement method provided in the embodiments of this application;

[0044] Figure 2 A flowchart illustrating feature extraction from remote sensing images using the U-NET module provided in this application embodiment;

[0045] Figure 3 A schematic diagram of a local structure of the image feature enhancement model provided in the embodiments of this application;

[0046] Figure 4 This is a block diagram of the image feature enhancement device provided in the embodiments of this application;

[0047] Figure 5 This is a block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is 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. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0049] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0050] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0051] To improve the accuracy of target feature extraction, embodiments of this application provide an image feature enhancement method, apparatus, electronic device, and product. This image feature enhancement method, apparatus, electronic device, and product can significantly improve the overall recognition integrity and boundary extraction accuracy of structured features in remote sensing images.

[0052] like Figure 1 As shown, the first aspect of the present application provides an image feature enhancement method, which may include, but is not limited to, the following steps S101-S102.

[0053] Step S101. Obtain the remote sensing image to be extracted.

[0054] Step S102. Input the remote sensing image into the pre-trained image feature enhancement model for feature extraction to obtain the remote sensing image feature map.

[0055] The image feature enhancement model includes the U-NET module, the multi-directional enhancement convolution module, the gated dynamic adjustment module, the efficient attention weighting module, and the cross-channel stitching module.

[0056] The U-NET module is used to extract features from the remote sensing image to obtain the original feature map;

[0057] The multi-directional enhanced convolution module is used to perform convolution operations in multiple directions on the original feature map to obtain a first feature map in multiple directions.

[0058] The gating dynamic adjustment module is used to dynamically adjust the first feature map in multiple directions to obtain the second feature map in multiple directions.

[0059] The efficient attention weighting module is used to extract the feature intensity weights of the original feature map in multiple directions to obtain the feature intensity weights in multiple directions, and to perform weighted fusion of the second feature map in multiple directions with the feature intensity weights in multiple directions to obtain the fused features in multiple directions.

[0060] The cross-channel stitching module is used to stitch together the fused features from multiple directions to obtain the remote sensing image feature map.

[0061] In this embodiment, the U-NET module can extract image features through the encoder and map the features back to the image space through skip connections and the decoder to obtain the original feature map. Figure 2 The diagram shows a flowchart of the U-NET module performing feature extraction on remote sensing images.

[0062] The multi-directional enhancement convolution module is used to perform convolution operations in multiple directions on the original feature map to obtain first feature maps in multiple directions. For example... Figure 3 The diagram shows a partial structure of the image feature enhancement model excluding the U-NET module. The multi-directional enhancement convolution module includes convolution kernels in multiple directions and performs convolution operations on the original feature map in multiple directions through these kernels.

[0063] In one or more embodiments, the multiple directions include a horizontal direction, a vertical direction, a 45° diagonal direction, and a 135° diagonal direction. When the original feature map is convolved in multiple directions using a multi-directional enhanced convolution module, the horizontal convolution kernel can be represented as follows: The vertical convolution kernel can be represented as A convolution kernel with a 45° diagonal direction can be represented as A convolution kernel with a 135° diagonal direction can be represented as , where c represents the center index of the convolution kernel, and i and j represent the horizontal and vertical indices of the pixel, respectively.

[0064] In one or more embodiments, when performing convolution operations in multiple directions, the original feature map can be mirror-filled using a multi-directional enhanced convolution module. Then, the mirror-filled original feature map is subjected to convolution operations in multiple directions to obtain first feature maps in multiple directions. For example, when the multiple directions include horizontal, vertical, 45° diagonal, and 135° diagonal directions, the original feature map can be mirror-filled in the water direction, vertical direction, 45° diagonal direction, and 135° diagonal direction respectively to preserve image edge information.

[0065] In one or more embodiments, before image feature enhancement, convolution kernels in multiple directions can be normalized to maintain feature stability. The normalization formula for the convolution kernels can be as follows:

[0066]

[0067] Where K represents the convolution kernel in any direction, and k represents the kernel size. K represents the small constant added for numerical stability. i，j This represents the convolution kernel for the pixel with pixel coordinates (i, j).

[0068] The gating dynamic adjustment module is used to dynamically adjust the first feature map in multiple directions to obtain the second feature map in multiple directions.

[0069] In one or more embodiments, the gated dynamic adjustment module can adjust the intensity of features in each direction by introducing learned gating coefficients and scaling factors. This gating design can adaptively adjust the contribution of each direction according to the input content, accurately focusing on the dominant direction of the target. The gated dynamic adjustment module can dynamically adjust the first feature map in the multiple directions according to the following formula:

[0070]

[0071] Among them, Y conv This represents the first feature map in the plurality of directions, and σ represents the activation function. G represents the XOR operation. k and S k The gating coefficient and scaling factor are represented in sequence, X represents the original feature map, BN represents batch normalization, ReLU represents the ReLU activation function, and Conv represents the convolution operation.

[0072] The efficient attention weighting module is used to extract the feature intensity weights of the original feature map in multiple directions to obtain the feature intensity weights in multiple directions, and to perform weighted fusion of the second feature map in multiple directions with the feature intensity weights in multiple directions to obtain the fused features in multiple directions.

[0073] In one or more embodiments, the efficient attention weighting module can be used to extract feature intensity weights of the original feature map in multiple directions according to the following formula:

[0074]

[0075] Where A represents the feature intensity weights in multiple directions, X represents the original feature map, σ represents the activation function, BN represents batch normalization, ReLU represents the ReLU activation function, and W1 and W2 represent the convolution weights of the attention mechanism layer, respectively.

[0076] In one or more embodiments, the efficient attention weighting module can be used to perform weighted fusion of second feature maps in multiple directions with feature intensity weights in multiple directions according to the following formula;

[0077]

[0078] Among them, Y attn Y represents the fusion features in multiple directions. h This represents the second feature map in the horizontal direction, Y. v This represents the second feature map in the vertical direction, Y. d45 This represents the second feature map along the 45° diagonal direction, Y. d135This represents the second feature map along the 135° diagonal direction, and A represents the feature intensity weights in multiple directions.

[0079] The cross-channel stitching module is used to stitch together the fused features from multiple directions to obtain the remote sensing image feature map.

[0080] In one or more embodiments, the cross-channel stitching module can first stitch together the fused features from multiple directions along the channel dimension, and then perform grouped convolution, batch normalization, and ReLU activation to finally output a remote sensing image feature map, the formula of which can be as follows:

[0081]

[0082] Where Output represents the output remote sensing image feature map, W3 represents the convolution weights across the fusion mechanism, and Y... fused This represents the result of splicing fused features from multiple directions along the channel dimension.

[0083] In summary, the image feature enhancement method provided by this invention extracts features from multiple directions through a multi-directional enhancement convolution module, comprehensively covering the main spatial orientation and effectively capturing the varied spatial orientation features of ground features. This solves the directional blind spot problem of traditional convolutional neural networks. Furthermore, a gated dynamic modulator is introduced to dynamically adjust features and enhance feature diversity. In addition, an efficient attention weighting module is used to adjust the feature intensity weights in multiple directions, achieving intelligent direction selection and accurately focusing on the dominant direction of the target while suppressing noise interference. Then, feature fusion enhancement from multiple directions enables directional optimization of feature maps in different directions. Finally, the fused features from multiple directions are stitched together across channels to enhance image directional features and preserve key information such as texture, improving the overall recognition integrity and extraction accuracy of structured ground features in remote sensing images. Furthermore, by mirroring the original feature map, image edge information can be preserved, further improving the overall recognition integrity and boundary extraction accuracy of structured ground features in remote sensing images. This method is particularly suitable for scenarios with obvious directional features, such as road networks and building complexes, providing reliable technical support for high-precision geographic information feature extraction and facilitating practical application and promotion.

[0084] Please see Figure 4 The second aspect of this application provides an image feature enhancement apparatus, which includes:

[0085] The acquisition unit is used to acquire the remote sensing image to be extracted;

[0086] The feature extraction unit is used to input the remote sensing image into a pre-trained image feature enhancement model for feature extraction, and obtain the remote sensing image feature map of the remote sensing image;

[0087] The image feature enhancement model includes a U-NET module, a multi-directional enhancement convolution module, a gated dynamic adjustment module, an efficient attention weighting module, and a cross-channel stitching module.

[0088] The U-NET module is used to extract features from the remote sensing image to obtain the original feature map;

[0089] The multi-directional enhanced convolution module is used to perform convolution operations in multiple directions on the original feature map to obtain a first feature map in multiple directions;

[0090] The gated dynamic adjustment module is used to dynamically adjust the first feature map in the multiple directions to obtain the second feature map in the multiple directions;

[0091] The efficient attention weighting module is used to extract the feature intensity weights of the original feature map in multiple directions to obtain the feature intensity weights in multiple directions, and to perform weighted fusion of the second feature map in multiple directions with the feature intensity weights in multiple directions to obtain the fused features in multiple directions.

[0092] The cross-channel stitching module is used to stitch together the fused features from multiple directions to obtain the remote sensing image feature map.

[0093] The working process, working details and technical effects of the image feature enhancement device provided in the second aspect of this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0094] like Figure 5 As shown, a third aspect of this application provides an electronic device, including a memory, a processor, and a transceiver that are sequentially and communicatively connected, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the image feature enhancement method as described in the first aspect of the embodiment.

[0095] Specifically, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or last-in-first-out (FILO) memory, etc.; the processor may not be limited to microprocessors of the STM32F105 series, ARM (Advanced RISC Machines), x86 architecture processors, or processors with integrated NPU (neural-network processing units); the transceiver may be, but is not limited to, WiFi (Wireless Fidelity) wireless transceivers, Bluetooth wireless transceivers, General Packet Radio Service (GPRS) wireless transceivers, ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard), 3G transceivers, 4G transceivers, and / or 5G transceivers, etc.

[0096] This fourth aspect of the embodiment provides a computer-readable storage medium storing instructions containing the image feature enhancement method described in the first aspect of the embodiment. Specifically, the computer-readable storage medium stores instructions that, when executed on a computer, perform the image feature enhancement method as described in the first aspect. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0097] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the image feature enhancement method as described in the first aspect of the embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0098] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of image feature enhancement, characterized by, The method comprises: obtaining a remote sensing image to be extracted; inputting the remote sensing image into a pre-trained image feature enhancement model for feature extraction to obtain a remote sensing image feature map of the remote sensing image; wherein the image feature enhancement model comprises a U-NET module, a multi-directional enhancement convolution module, a gated dynamic adjustment module, an efficient attention weighting module and a cross-channel splicing module; the U-NET module is used for feature extraction on the remote sensing image to obtain an original feature map; the multi-directional enhancement convolution module is used for convolution operation on the original feature map in multiple directions to obtain first feature maps in multiple directions; the gated dynamic adjustment module is used for dynamic adjustment on the first feature maps in multiple directions to obtain second feature maps in multiple directions; the efficient attention weighting module is used for extracting feature intensity weights of the original feature map in multiple directions to obtain the feature intensity weights in multiple directions, and performing weighted fusion of the second feature maps in multiple directions and the feature intensity weights in multiple directions to obtain fused features in multiple directions; the cross-channel splicing module is used for splicing the fused features in multiple directions to obtain the remote sensing image feature map; The plurality of directions include a horizontal direction, a vertical direction, a 45° diagonal direction, and a 135° diagonal direction, and when the multi-directional enhancement convolution module performs convolution operation on the original feature map in a plurality of directions, the convolution kernel in the horizontal direction is , the convolution kernel in the vertical direction is , the convolution kernel in the 45° diagonal direction is , and the convolution kernel in the 135° diagonal direction is , wherein c represents a convolution kernel center index, and i and j represent the horizontal direction index and the vertical direction index of the pixel, respectively.

2. The image feature enhancement method of claim 1, wherein, the multi-directional enhancement convolution module is used for mirror filling on the original feature map, and convolution operation on the original feature map after mirror filling in multiple directions to obtain first feature maps in multiple directions.

3. The image feature enhancement method of claim 1, wherein, the efficient attention weighting module is used for weighted fusion of the second feature maps in multiple directions and the feature intensity weights in multiple directions according to the following formula; where Y attn represents fusion features in multiple directions, Y h represents a second feature map in a horizontal direction, Y v represents a second feature map in a vertical direction, Y d45 represents a second feature map in a 45° diagonal direction, Y d135 represents a second feature map in a 135° diagonal direction, represents an XOR operation, and A represents feature intensity weights in multiple directions.

4. The image feature enhancement method of claim 1, wherein, the gated dynamic adjustment module is used for dynamic adjustment on the first feature maps in multiple directions according to the following formula; wherein Y conv represents the first feature map in the plurality of directions, and σ represents an activation function, represents an exclusive or operation, and G k and S k represent a gating coefficient and a scaling factor in sequence, X represents an original feature map, BN represents a batch normalization process, and ReLU represents a ReLU activation function, and Conv represents a convolution operation.

5. The image feature enhancement method of claim 1, wherein, the efficient attention weighting module is used for extracting feature intensity weights of the original feature map in multiple directions according to the following formula; wherein A represents the feature intensity weights in multiple directions, X represents the original feature map, sigma represents an activation function, BN represents batch normalization processing, ReLU represents a ReLU activation function, W1 and W2 represent convolution weights.

6. An image feature enhancement apparatus characterized by comprising: The method comprises: an acquisition unit configured to acquire a remote sensing image to be extracted; a feature extraction unit configured to input the remote sensing image into a pre-trained image feature enhancement model for feature extraction to obtain a remote sensing image feature map of the remote sensing image; wherein the image feature enhancement model comprises a U-NET module, a multi-directional enhancement convolution module, a gated dynamic adjustment module, an efficient attention weighting module and a cross-channel splicing module; the U-NET module is used for feature extraction on the remote sensing image to obtain an original feature map; the multi-directional enhancement convolution module is used for convolution operation on the original feature map in multiple directions to obtain first feature maps in multiple directions; the gated dynamic adjustment module is used for dynamic adjustment on the first feature maps in multiple directions to obtain second feature maps in multiple directions; the efficient attention weighting module is used for extracting feature intensity weights of the original feature map in multiple directions to obtain the feature intensity weights in multiple directions, and performing weighted fusion of the second feature maps in multiple directions and the feature intensity weights in multiple directions to obtain fused features in multiple directions; the cross-channel splicing module is used for splicing the fused features in multiple directions to obtain the remote sensing image feature map; the multi-directional enhancement convolution module is used for mirror filling on the original feature map, and convolution operation on the original feature map after mirror filling in multiple directions to obtain first feature maps in multiple directions. the efficient attention weighting module is used for weighted fusion of the second feature maps in multiple directions and the feature intensity weights in multiple directions according to the following formula; the gated dynamic adjustment module is used for dynamic adjustment on the first feature maps in multiple directions according to the following formula; the efficient attention weighting module is used for extracting feature intensity weights of the original feature map in multiple directions according to the following formula; wherein A represents the feature intensity weights in multiple directions, X represents the original feature map, sigma represents an activation function, BN represents batch normalization processing, ReLU represents a ReLU activation function, W1 and W2 represent convolution weights. The high-efficiency attention weighting module is configured to extract feature intensity weights of the original feature map in multiple directions, obtain the feature intensity weights in the multiple directions, and perform weighted fusion on the second feature map in the multiple directions and the feature intensity weights in the multiple directions to obtain fused features in the multiple directions. The cross-channel splicing module is configured to splice the fused features in the multiple directions to obtain the remote sensing image feature map. The plurality of directions include a horizontal direction, a vertical direction, a 45° diagonal direction, and a 135° diagonal direction, and when the multi-directional enhancement convolution module performs convolution operation on the original feature map in a plurality of directions, the convolution kernel in the horizontal direction is , the convolution kernel in the vertical direction is , the convolution kernel in the 45° diagonal direction is , and the convolution kernel in the 135° diagonal direction is , wherein c represents a convolution kernel center index, and i and j represent the horizontal direction index and the vertical direction index of the pixel, respectively.

7. An electronic device, comprising: The computer program or the instruction, when executed by a computer, implements the image feature enhancement method according to any one of claims 1-5.

8. A computer program product comprising computer programs or instructions, characterized in that, The computer program or the instruction, when executed by a computer, implements the image feature enhancement method according to any one of claims 1-5.

Images