Remote sensing image target detection method, device, equipment and storage medium
By extracting and fusing the features of optical images and SAR images, the problem of lack of independent feature fusion in the prior art is solved, and a more efficient remote sensing image object detection effect is achieved, highlighting the target and fading the background.
Patent Information
- Application Number
- CN202410429782.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-10
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2044-04-10
AI Technical Summary
The lack of extraction and fusion of independent features of optical images and SAR images in the prior art, resulting in the inability to exert their respective advantages, thereby affecting the effect of remote sensing image object detection.
By obtaining the embedding vectors of optical images and SAR images, their context features are extracted respectively, and these features are combined and fused using a multi-source fusion model to generate total features for object detection.
Through the combined features, we can effectively highlight the target, fade the background, improve the target detection effect of remote sensing images, and achieve better accuracy and recall.
Smart Images
Figure CN118447387B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of target detection technology, and in particular to a method, device, equipment and storage medium for detecting a target in a remote sensing image. Background Art
[0002] Target detection in remote sensing images has broad application prospects. Its purpose is to automatically extract and identify target objects from remote sensing images, which can provide strong support for tasks such as agriculture, environmental protection, and urban planning.
[0003] In the related technologies, target detection methods based on deep learning are mainly divided into two types: convolutional neural networks and attention mechanisms. Convolutional neural networks can only perceive local information and lack the ability to understand the global image, while the attention mechanism can highlight the key areas of the perceived target and achieve good results under the premise of ensuring computational efficiency. Whether using convolution or attention mechanism target detection methods, optical images and SAR images are often processed separately to obtain detection results. However, due to the different imaging mechanisms of optical images and SAR images, their respective advantages cannot be brought into play.
[0004] Based on the above analysis of the development status of this technical field, the existing technology lacks a solution for first extracting independent features of optical images and SAR images, fusing the independent features using a multi-source fusion model, and then performing target detection based on the fused features. Summary of the invention
[0005] The purpose of the present invention is to provide a method, device, equipment and storage medium for target detection of remote sensing images, aiming to solve the above-mentioned problems in the prior art.
[0006] According to a first aspect of an embodiment of the present invention, a method for detecting an object in a remote sensing image is provided, comprising:
[0007] Obtain an optical image and a SAR image of a remote sensing image, extract an optical embedding vector from the optical image, and extract a SAR embedding vector from the SAR image;
[0008] Extracting context features of the optical embedding vector through the first single-source extraction model to obtain optical single-source features, and extracting context features of the SAR embedding vector through the second single-source extraction model to obtain SAR single-source features;
[0009] The SAR single-source feature and the important features of the optical single-source feature are fused through the first multi-source fusion model to obtain the optical fusion feature, the optical single-source feature and the important features of the SAR single-source feature are fused through the second multi-source fusion model to obtain the SAR fusion feature, and the optical fusion feature and the SAR fusion feature are spliced into the total feature;
[0010] The total features are input into the trained detection model to obtain the target detection results of the remote sensing image.
[0011] According to a second aspect of an embodiment of the present invention, there is provided a target detection device for a remote sensing image, comprising:
[0012] An initial extraction module is used to obtain optical images and SAR images of remote sensing images, extract optical embedding vectors from optical images, and extract SAR embedding vectors from SAR images;
[0013] A single-source extraction module is used to extract context features of the optical embedding vector through a first single-source extraction model to obtain an optical single-source feature, and to extract context features of the SAR embedding vector through a second single-source extraction model to obtain a SAR single-source feature;
[0014] A multi-source fusion module is used to fuse the important features of the SAR single-source features and the optical single-source features through a first multi-source fusion model to obtain an optical fusion feature, fuse the important features of the optical single-source features and the SAR single-source features through a second multi-source fusion model to obtain a SAR fusion feature, and splice the optical fusion feature and the SAR fusion feature into a total feature;
[0015] The target detection module is used to input the total features into the trained detection model to obtain the target detection results of the remote sensing image.
[0016] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the target detection method for remote sensing images provided in the first aspect of the present disclosure are implemented.
[0017] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which a program for implementing information transmission is stored. When the program is executed by a processor, the steps of the target detection method for remote sensing images provided in the first aspect of the present disclosure are implemented.
[0018] The technical solution provided by the embodiment of the present invention includes the following beneficial effects: the remote sensing image target detection task is divided into two branches, optical and SAR; in one branch, the SAR single source feature is combined with the important features of the optical single source feature; in the other branch, the optical single source feature is combined with the important features of the SAR single source feature, so that the combined SAR feature or optical feature combines the advantages of the two images, and the final fusion feature can more effectively highlight the target and weaken the background.
[0019] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0021] Figure 1 is a flow chart of a method for detecting a target in a remote sensing image according to an embodiment of the present invention;
[0022] Figure 2 is a schematic diagram of a target detection design framework according to an embodiment of the present invention;
[0023] Figure 3 is a schematic diagram of a target detection device for remote sensing images according to an embodiment of the present invention;
[0024] Figure 4 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0025] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.
[0026] Method Embodiment
[0027] According to an embodiment of the present invention, a method for detecting a target in a remote sensing image is provided. Figure 1 is a flow chart of a method for detecting a target in a remote sensing image according to an embodiment of the present invention. Figure 1 As shown, the target detection method of remote sensing images according to an embodiment of the present invention specifically includes:
[0028] In step S110, an optical image and a SAR image of a remote sensing image are acquired, and an optical embedding vector is extracted from the optical image, and a SAR embedding vector is extracted from the SAR image, specifically including:
[0029] The optical embedding vector is extracted through the backbone network of the optical image, and the SAR embedding vector is extracted through the backbone network of the SAR image. The backbone network consists of the ResNet residual network and the FPN feature pyramid. When the network layer is deep, ResNet can still stably train the model to avoid gradient disappearance. ResNet is used to extract the primary features of the image and pass the primary features to FTN. FPN is used to extract the multi-scale features of the image.
[0030] The extracted vectors are respectively fed into the corresponding single-source extraction models.
[0031] In step S120, the context features of the optical embedding vector are extracted by the first single-source extraction model to obtain the optical single-source features, and the context features of the SAR embedding vector are extracted by the second single-source extraction model to obtain the SAR single-source features, which specifically includes:
[0032] The first single-source extraction model specifically includes: a multi-head attention layer and a feedforward layer. Preferably, each layer has a corresponding implementation module and a residual connection and layer standardization module, that is, the multi-head attention layer has a multi-head attention module and a residual connection and layer standardization module, and the feedforward layer has a feedforward neural network module and a residual connection and layer standardization module.
[0033] Multiple groups of first single-source extraction models can be stacked, and multiple multi-head attention layers can be stacked inside each group of first single-source extraction models. An embodiment of the present invention stacks two layers of first single-source extraction models, and two multi-head attention layers are stacked inside.
[0034] The important vectors of the optical embedding vector are extracted through the multi-head attention layer of the first single-source extraction model, and the important vectors of the optical embedding vector are converted into optical single-source features through the feed-forward layer of the first single-source extraction model.
[0035] The second single-source extraction model specifically includes: a multi-head attention layer and a feedforward layer. The neural network structure setting in the second unit extraction model is exactly the same as that of the first single-source extraction model. Both are for extracting single-source features, but the objects of extraction are different, and the second single-source extraction model can also be stacked.
[0036] The important vectors of the SAR embedding vector are extracted through the multi-head attention layer of the second single-source extraction model, and the important vectors of the SAR embedding vector are converted into SAR single-source features through the feed-forward layer of the second single-source extraction model.
[0037] The embodiment of the present invention stacks two layers of second single-source extraction models, and superimposes two multi-head attention layers internally.
[0038] In step S130, the SAR single source feature and the important features of the optical single source feature are fused through the first multi-source fusion model to obtain the optical fusion feature, and the optical single source feature and the important features of the SAR single source feature are fused through the second multi-source fusion model to obtain the SAR fusion feature. The optical fusion feature and the SAR fusion feature are spliced into the total feature, which specifically includes:
[0039] The first multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer and a feedforward layer. Preferably, there are corresponding implementation modules and residual connection and layer normalization modules in each layer. Other settings are consistent with the single-source extraction model. The fused multi-head attention layer includes a fused multi-head attention module and a residual connection and layer normalization module.
[0040] The important features in the optical single-source features are extracted through the multi-head attention layer of the first multi-source fusion model, the important features in the optical single-source features and the SAR single-source features are combined through the fusion multi-head attention layer of the first multi-source fusion model to obtain optical combination features, and the optical combination features are converted into optical fusion features through the feedforward layer of the first multi-source fusion model.
[0041] A plurality of first multi-source fusion models may be stacked. The embodiment of the present invention stacks two layers of first multi-source fusion models.
[0042] The second multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer, and a feedforward layer.
[0043] The important features of SAR single-source features are extracted through the multi-head attention layer of the second multi-source fusion model, the important features in the SAR single-source features and the optical single-source features are combined through the fusion multi-head attention layer of the second multi-source fusion model to obtain SAR combined features, and the SAR combined features are converted into SAR fusion features through the feedforward layer of the second multi-source fusion model.
[0044] The embodiment of the present invention stacks two layers of the second multi-source fusion model.
[0045] The multi-head attention mechanism can combine different types of content into knowledge, and the feedforward neural network can output a one-dimensional feature vector from the combined knowledge.
[0046] The one-dimensional optical fusion features and the one-dimensional SAR fusion features are concatenated into two-dimensional total features through the fully connected layer, in order to facilitate the subsequent detection convolution processing.
[0047] In step S140, the total features are input into the trained detection model to obtain the target detection results of the remote sensing image, which specifically include:
[0048] The detection model consists of RPN and region of interest feature extraction. The function of RPN is to generate candidate regions of several targets for optical images and SAR images. The candidate regions are all potential bounding boxes that may contain the true bounding box of the target. When training the model of this application, the training process will continuously adjust the position of the candidate area to minimize the deviation from the true bounding box.
[0049] Region of interest extraction is used to map the candidate regions generated by RPN to the feature map. It is achieved through the RoI pooling operation, which normalizes the candidate regions of different sizes and aspect ratios into feature maps of fixed sizes. The feature maps are then sent to the Faster R-CNN classification model to complete the target classification and positioning tasks.
[0050] The target detection results of the embodiment of the present invention are shown in Table 1:
[0051] Table 1. Comparison of target detection results
[0052]
[0053] As shown in Table 1, the performance of the ToNet of the present invention and the other fifteen models on the VIS, SAR, and VIS+SAR data sets are listed. The VIS column represents the ToNet results when the sources are both optical images, that is, two identical optical images are input, and the SAR column represents the ToNet results when the sources are both SAR images; the VIS+SAR column represents that two backbone networks simultaneously input optical images and SAR images of the same scene, that is, the scheme designed in full accordance with the present invention; the results of the VIS and SAR columns show that although the two inputs are homologous images and there is information redundancy, the single-source extraction model and the multi-source fusion model enhance the context feature extraction capability and the two-way context feature fusion capability respectively; for the VIS+SAR column, the ToNet that comprehensively uses optical images and SAR multi-source images achieves the best results in both precision AP and recall AR.
[0054] The above technical solutions of the embodiments of the present invention are illustrated with reference to the following drawings.
[0055] Figure 2 is a schematic diagram of a target detection design framework of an embodiment of the present invention, which is called a ToNet model. Figure 2 As shown, the basic framework of the dual-branch of the present invention is demonstrated. A single-source extraction model is designed in the encoder part, and a multi-source fusion model is designed in the decoder part. The K and V vectors use the output of the single-source extraction model, and Q uses the output vector of the multi-head attention layer to participate in the calculation. M, N, and L all represent the number of superimposed layers.
[0056] To sum up, in response to the existing problems, the target detection method of remote sensing images invented this time divides the remote sensing image target detection task into two branches, optical and SAR. On one branch, the SAR single source feature is combined with the important features of the optical single source feature, and the generated optical combination feature is combined with the features of the SAR image. On the other branch, the optical single source feature is combined with the important features of the SAR single source feature, and the generated SAR combination feature is combined with the features of the optical image. The optical combination feature and the SAR combination feature are then fully connected and spliced to complete the fusion. The fused total feature can more effectively highlight the target, dilute the background, and improve the target detection effect of the remote sensing image.
[0057] Device Embodiment
[0058] According to an embodiment of the present invention, a remote sensing image target detection device is provided. Figure 3 Schematic diagram of a target detection device for remote sensing images according to an embodiment of the present invention. Figure 3 As shown, the target detection device of the remote sensing image according to the embodiment of the present invention specifically includes:
[0059] The initial extraction module 30 is used to obtain the optical image and the SAR image of the remote sensing image, extract the optical embedding vector from the optical image, and extract the SAR embedding vector from the SAR image, specifically for:
[0060] The optical embedding vector is extracted through the backbone network of the optical image, and the SAR embedding vector is extracted through the backbone network of the SAR image, where the backbone network is composed of a ResNet residual network and an FPN feature pyramid.
[0061] The single-source extraction module 32 is used to extract the context features of the optical embedding vector through the first single-source extraction model to obtain the optical single-source features, and to extract the context features of the SAR embedding vector through the second single-source extraction model to obtain the SAR single-source features, specifically for:
[0062] The first single-source extraction model specifically includes: a multi-head attention layer and a feed-forward layer,
[0063] The important vectors of the optical embedding vector are extracted through the multi-head attention layer of the first single-source extraction model, and the important vectors of the optical embedding vector are converted into optical single-source features through the feed-forward layer of the first single-source extraction model.
[0064] The second single-source extraction model specifically includes: a multi-head attention layer and a feed-forward layer,
[0065] The important vectors of the SAR embedding vector are extracted through the multi-head attention layer of the second single-source extraction model, and the important vectors of the SAR embedding vector are converted into SAR single-source features through the feed-forward layer of the second single-source extraction model.
[0066] The multi-source fusion module 34 is used to fuse the important features of the SAR single-source feature and the optical single-source feature through the first multi-source fusion model to obtain the optical fusion feature, fuse the important features of the optical single-source feature and the SAR single-source feature through the second multi-source fusion model to obtain the SAR fusion feature, and splice the optical fusion feature and the SAR fusion feature into the total feature, which is specifically used for:
[0067] The first multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer, and a feedforward layer.
[0068] The important features in the optical single-source features are extracted through the multi-head attention layer of the first multi-source fusion model, the important features in the optical single-source features and the SAR single-source features are combined through the fusion multi-head attention layer of the first multi-source fusion model to obtain optical combination features, and the optical combination features are converted into optical fusion features through the feedforward layer of the first multi-source fusion model.
[0069] The second multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer, and a feedforward layer.
[0070] The important features of SAR single-source features are extracted through the multi-head attention layer of the second multi-source fusion model, the important features in the SAR single-source features and the optical single-source features are combined through the fusion multi-head attention layer of the second multi-source fusion model to obtain SAR combined features, and the SAR combined features are converted into SAR fusion features through the feedforward layer of the second multi-source fusion model.
[0071] The one-dimensional optical fusion features and the one-dimensional SAR fusion features are concatenated into two-dimensional total features through a fully connected layer.
[0072] The target detection module 36 is used to input the total features into the trained detection model to obtain the target detection results of the remote sensing image.
[0073] To sum up, in response to the existing problems, the target detection device of remote sensing images invented this time divides the remote sensing image target detection task into two branches, optical and SAR. On one branch, the SAR single source feature is combined with the important features of the optical single source feature, and the generated optical combination feature is combined with the features of the SAR image. On the other branch, the optical single source feature is combined with the important features of the SAR single source feature, and the generated SAR combination feature is combined with the features of the optical image. The optical combination feature and the SAR combination feature are then fully connected and spliced to complete the fusion. The fused total feature can more effectively highlight the target, dilute the background, and improve the target detection effect of the remote sensing image.
[0074] Electronic device embodiment
[0075] Figure 44 is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device 400 may include at least one processor 410 and a memory 420. The processor 410 may execute instructions stored in the memory 420. The processor 410 is connected to the memory 420 through a data bus. In addition to the memory 420, the processor 410 may also be connected to an input device 430, an output device 440, and a communication device 450 through a data bus.
[0076] The processor 410 may be any conventional processor, such as a commercially available CPU. The processor may also include a graphics processor (Graphic Process Unit, GPU), a field programmable gate array (Field Programmable Gate Array, FPGA), a system on chip (System on Chip, SOC), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC) or a combination thereof.
[0077] Memory 420 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.
[0078] In the embodiment of the present disclosure, executable instructions are stored in the memory 420, and the processor 410 can read the executable instructions from the memory 420 and execute the instructions to implement all or part of the steps of the target detection method of the remote sensing image in any of the above exemplary embodiments.
[0079] Computer Readable Storage Medium Embodiments
[0080] In addition to the above-mentioned methods and devices, the exemplary embodiments of the present disclosure may also be a computer program product or a computer-readable storage medium storing the computer program product, wherein the computer product includes computer program instructions that can be executed by a processor to implement all or part of the steps described in the target detection method for remote sensing images in any of the above-mentioned exemplary embodiments.
[0081] The computer program product may be written in any combination of one or more programming languages to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages such as Java, C++, etc., and also conventional procedural programming languages such as "C" language or similar programming languages and scripting languages (e.g., Python). The program code may be executed entirely on the user computing device, partially on the user computing device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.
[0082] Computer readable storage media can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of readable storage media include: a static random access memory (SRAM) with one or more wires electrically connected, an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk, or any suitable combination of the above.
[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for detecting a target in a remote sensing image, characterized in that: include: Acquire an optical image and a SAR image of a remote sensing image, extract an optical embedding vector from the optical image, and extract a SAR embedding vector from the SAR image; Extracting the context features of the optical embedding vector by a first single-source extraction model to obtain an optical single-source feature, and extracting the context features of the SAR embedding vector by a second single-source extraction model to obtain a SAR single-source feature, specifically includes: The first single-source extraction model specifically includes: a multi-head attention layer and a feedforward layer, extracting the important vectors of the optical embedding vector through the multi-head attention layer of the first single-source extraction model, and converting the important vectors of the optical embedding vector into optical single-source features through the feedforward layer of the first single-source extraction model; The second single-source extraction model specifically includes: a multi-head attention layer and a feed-forward layer, extracting the important vectors of the SAR embedding vector through the multi-head attention layer of the second single-source extraction model, and converting the important vectors of the SAR embedding vector into SAR single-source features through the feed-forward layer of the second single-source extraction model; The SAR single source feature and the important features of the optical single source feature are fused by a first multi-source fusion model to obtain an optical fusion feature, the optical single source feature and the important features of the SAR single source feature are fused by a second multi-source fusion model to obtain a SAR fusion feature, and the optical fusion feature and the SAR fusion feature are spliced into a total feature, specifically including: The first multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer and a feedforward layer, the multi-head attention layer of the first multi-source fusion model is used to extract important features in the optical single-source features, the fused multi-head attention layer of the first multi-source fusion model is used to combine the important features in the optical single-source features and the SAR single-source features to obtain optical combination features, and the feedforward layer of the first multi-source fusion model is used to convert the optical combination features into the optical fusion features; The second multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer and a feedforward layer, the multi-head attention layer of the second multi-source fusion model is used to extract the important features of the SAR single-source features, the fused multi-head attention layer of the second multi-source fusion model is used to combine the important features in the SAR single-source features and the optical single-source features to obtain SAR combined features, and the feedforward layer of the second multi-source fusion model is used to convert the SAR combined features into the SAR fused features; The total features are input into a trained detection model to obtain target detection results of the remote sensing image.
2. The method according to claim 1, characterized in that The extracting of the optical embedding vector from the optical image and the extracting of the SAR embedding vector from the SAR image specifically include: The optical embedding vector is obtained by extracting the backbone network of the optical image, and the SAR embedding vector is obtained by extracting the backbone network of the SAR image, wherein the backbone network is composed of a ResNet residual network and an FPN feature pyramid.
3. The method according to claim 1, characterized in that The step of splicing the optical fusion feature and the SAR fusion feature into a total feature specifically includes: splicing a one-dimensional optical fusion feature and a one-dimensional SAR fusion feature into a two-dimensional total feature through a fully connected layer.
4. A target detection device for remote sensing images, characterized in that: include: An initial extraction module, used for acquiring an optical image and a SAR image of a remote sensing image, extracting an optical embedding vector from the optical image, and extracting a SAR embedding vector from the SAR image; The single-source extraction module is used to extract the context features of the optical embedding vector through a first single-source extraction model to obtain an optical single-source feature, and to extract the context features of the SAR embedding vector through a second single-source extraction model to obtain a SAR single-source feature, specifically for: The first single-source extraction model specifically includes: a multi-head attention layer and a feedforward layer, extracting the important vectors of the optical embedding vector through the multi-head attention layer of the first single-source extraction model, and converting the important vectors of the optical embedding vector into optical single-source features through the feedforward layer of the first single-source extraction model; The second single-source extraction model specifically includes: a multi-head attention layer and a feed-forward layer, extracting the important vectors of the SAR embedding vector through the multi-head attention layer of the second single-source extraction model, and converting the important vectors of the SAR embedding vector into SAR single-source features through the feed-forward layer of the second single-source extraction model; The multi-source fusion module is used to fuse the important features of the SAR single-source feature and the optical single-source feature through a first multi-source fusion model to obtain an optical fusion feature, fuse the important features of the optical single-source feature and the SAR single-source feature through a second multi-source fusion model to obtain a SAR fusion feature, and splice the optical fusion feature and the SAR fusion feature into a total feature, specifically used for: The first multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer and a feedforward layer, the multi-head attention layer of the first multi-source fusion model is used to extract important features in the optical single-source features, the fused multi-head attention layer of the first multi-source fusion model is used to combine the important features in the optical single-source features and the SAR single-source features to obtain optical combination features, and the feedforward layer of the first multi-source fusion model is used to convert the optical combination features into the optical fusion features; The second multi-source fusion model specifically includes: a multi-head attention layer, a fused multi-head attention layer and a feedforward layer, the multi-head attention layer of the second multi-source fusion model is used to extract the important features of the SAR single-source features, the fused multi-head attention layer of the second multi-source fusion model is used to combine the important features in the SAR single-source features and the optical single-source features to obtain SAR combined features, and the feedforward layer of the second multi-source fusion model is used to convert the SAR combined features into the SAR fused features; The target detection module is used to input the total features into the trained detection model to obtain the target detection results of the remote sensing image.
5. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the target detection method for remote sensing images as described in any one of claims 1 to 3 are implemented.
6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an implementation program for information transmission, and when the program is executed by a processor, the steps of the target detection method for remote sensing images according to any one of claims 1 to 3 are implemented.