Fine-grained remote sensing target detection method and device based on visual language instance fusion

Abstract

The present application relates to a fine-grained remote sensing target detection method and device based on visual language instance fusion. The method includes: extracting remote sensing image features according to a visual encoder, extracting category embedding features through a language encoder; inputting target instances and remote sensing image features into a visual language instance fusion module through an instance feature extractor, obtaining visual instance features and storing them in an instance feature memory area; updating the average instance features stored in the instance feature memory area, inputting category embedding features and visual average instance features into a visual language fusion layer for interaction and fusion, and updating the instance features stored in the instance feature memory area based on the interactively fused features; enhancing remote sensing image features and interactively fused features according to a visual language deep enhancement encoder, inputting the enhanced remote sensing image features and category embedding features into a detection head, and obtaining detection results. This method can achieve more accurate detection of fine-grained targets.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a fine-grained remote sensing target detection method and device based on visual language instance fusion. Background Art

[0002] Fine-grained remote sensing object detection aims to accurately identify and distinguish subtle subclasses of objects (such as aircraft models, ship types, and crop varieties) from high-resolution remote sensing imagery. Key challenges lie in the small size of objects, subtle inter-class differences, and complex background interference. This technology is widely used in precision agriculture (monitoring crop disease subtypes) and ecological conservation (tracking endangered species subpopulations). Slight inter-class differences are the core difficulty of fine-grained remote sensing object detection.

[0003] Existing methods generally introduce contrastive learning loss to increase the feature differences between different categories, but this approach has low accuracy in fine-grained classification. Summary of the Invention

[0004] Based on this, it is necessary to provide a fine-grained remote sensing target detection method and device based on visual language instance fusion, which can effectively integrate visual instance features with language category concepts to achieve more accurate detection of fine-grained targets in response to the above technical problems.

[0005] A fine-grained remote sensing target detection method based on visual language instance fusion, the method comprising:

[0006] Get the input remote sensing image and category text; the category text is a string obtained by concatenating the names of all categories to be detected;

[0007] Construct a fine-grained remote sensing image target detection network model; the fine-grained remote sensing image target detection network model includes a visual encoder, a language encoder, a visual-language instance fusion module, a visual-language deep enhancement encoder, and a detection head;

[0008] The input remote sensing image and category text are input into the fine-grained remote sensing image target detection network model, the remote sensing image features are extracted according to the visual encoder, and the category embedding features are extracted through the language encoder; the target instance and remote sensing image features are input into the visual language instance fusion module through the instance feature extractor to obtain the visual instance features and store them in the instance feature memory area; the average instance features stored in the instance feature memory area are updated to obtain the visual average instance features; the category embedding features and the visual average instance features are input into the visual language fusion layer for interaction and fusion to obtain the interactive fused features; the instance features stored in the instance feature memory area are updated according to the interactive fused features; the remote sensing image features and the interactive fused features are enhanced according to the visual language deep enhancement encoder to obtain the enhanced remote sensing image features and category embedding features; the enhanced remote sensing image features and category embedding features are input into the detection head to obtain the detection results.

[0009] A fine-grained remote sensing target detection device based on visual language instance fusion, the device comprising:

[0010] The data acquisition module is used to obtain input remote sensing images and category text; the category text is a string obtained by concatenating the names of all categories to be detected;

[0011] A model building module is used to build a fine-grained remote sensing image target detection network model; the fine-grained remote sensing image target detection network model includes a visual encoder, a language encoder, a visual-language instance fusion module, a visual-language deep enhancement encoder, and a detection head;

[0012] The fine-grained remote sensing target detection module is used to input the input remote sensing image and category text into the fine-grained remote sensing image target detection network model, extract the remote sensing image features according to the visual encoder, and extract the category embedding features through the language encoder; input the target instance and remote sensing image features into the visual language instance fusion module through the instance feature extractor to obtain the visual instance features and store them in the instance feature memory area; update the average instance features stored in the instance feature memory area to obtain the visual average instance features; input the category embedding features and the visual average instance features into the visual language fusion layer for interaction and fusion to obtain the interactive fused features; update the instance features stored in the instance feature memory area according to the interactive fused features; enhance the remote sensing image features and the interactive fused features according to the visual language deep enhancement encoder to obtain enhanced remote sensing image features and category embedding features; input the enhanced remote sensing image features and category embedding features into the detection head to obtain the detection results.

[0013] The aforementioned method and device for fine-grained remote sensing target detection based on visual-language instance fusion constructs a fine-grained remote sensing image target detection network model comprising multiple modules. The visual encoder and language encoder are responsible for extracting remote sensing image features and category embedding features, respectively. This achieves preliminary information extraction from both visual and language modalities, providing a rich set of basic features for subsequent fusion operations. The instance feature extractor in the visual-language instance fusion module extracts visual instance features from remote sensing image features and stores them in an instance feature memory area. This design specifically captures and preserves target instance features, providing a data foundation for subsequent utilization of these features. The average instance features stored in the instance feature memory area are updated to obtain visual average instance features. This continuous updating allows the model to adapt to different input images, better capture target variations and diversity, and improve the model's ability to represent fine-grained features. The category embedding features and visual average instance features are fed into the visual-language fusion layer for interaction and fusion. Through methods such as a bidirectional multi-head attention mechanism, deep fusion of visual and language information is achieved. This fusion approach can fully utilize language category concepts to guide and supplement visual instance features, while also allowing visual instance features to refine and enhance language category embeddings, so that the fused features contain both rich visual details and accurate category concept information. The visual language deep enhancement encoder enhances remote sensing image features and interactively fused features, further improving feature quality and representation capabilities. The enhanced features can better reflect the characteristics of fine-grained targets, providing more accurate input to the detection head, thereby improving detection accuracy. Through end-to-end training, the entire model can automatically learn how to effectively fuse visual instance features with language category concepts from the input remote sensing images and category text to achieve accurate detection of fine-grained targets. During training, the model can continuously adjust the parameters of each module according to the loss function to optimize the feature extraction and fusion process to adapt to different datasets and task requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 1 is a flow chart of a fine-grained remote sensing target detection method based on visual language instance fusion in one embodiment;

[0015] Figure 2 Schematic diagram of a fine-grained remote sensing image object detection network model in one embodiment;

[0016] Figure 3 Schematic diagram of a fine-grained remote sensing target detection device based on visual language instance fusion in one embodiment;

[0017] Figure 4 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0019] In one embodiment, Figure 1 As shown in FIG, a fine-grained remote sensing target detection method based on visual language instance fusion is provided, which includes the following steps:

[0020] Step 102: Obtain an input remote sensing image and category text; the category text is a string obtained by concatenating the names of all categories to be detected.

[0021] The category text is a string obtained by concatenating the names of all categories to be detected by using ".".

[0022] Step 104, constructing a fine-grained remote sensing image target detection network model; the fine-grained remote sensing image target detection network model includes a visual encoder, a language encoder, a visual language instance fusion module, a visual language depth enhancement encoder and a detection head.

[0023] The structure of the fine-grained remote sensing image target detection network model is as follows: Figure 2 As shown, the visual language instance fusion module includes the instance extractor and bidirectional multi-head attention layer in the figure.

[0024] Step 106: input the input remote sensing image and category text into the fine-grained remote sensing image target detection network model, extract remote sensing image features according to the visual encoder, and extract category embedding features through the language encoder; input the target instance and remote sensing image features into the visual language instance fusion module through the instance feature extractor to obtain visual instance features and store them in the instance feature memory area; update the average instance features stored in the instance feature memory area to obtain visual average instance features; input the category embedding features and the visual average instance features into the visual language fusion layer for interaction and fusion to obtain interactive fused features; update the instance features stored in the instance feature memory area according to the interactive fused features; enhance the remote sensing image features and the interactive fused features according to the visual language deep enhancement encoder to obtain enhanced remote sensing image features and category embedding features; input the enhanced remote sensing image features and category embedding features into the detection head to obtain detection results.

[0025] Vision encoder through the network Extracting remote sensing image features , through the language encoder Extracting category embedding features , Indicates the maximum number of words, Represents the implicit dimension. The target instance GT annotation box and feature map Input the instance feature extractor in the visual-language instance fusion module to extract visual instance features , Indicates the number of instances, represents the number of feature channels, Indicates the instance feature size. Updates the average instance feature stored in the instance feature memory area: ,in Indicates the In the iterative process The average instance features of the categories, Representation category No. Instance features, Representation category The number of instances. In particular, ; . Embed ( For category The number of word units) and the corresponding visual average instance features Interaction and fusion are performed through bidirectional multi-head attention layers; first, and Perform layer normalization separately, and then and The corresponding key vectors are obtained by linear layer projection respectively and , value vector and , query vector and ; Calculate the attention matrix ; Get the embedding and features after interaction: ,in Represents the linear layer; finally, the interactive features are fused into the original features ,in is a learnable parameter. Update the instance features stored in the instance feature memory area again.

[0026] The image features and category embedding features Input the visual language global fusion encoder to obtain enhanced image features and category embedding features The visual language global fusion encoder consists of 6 network blocks, each of which is composed of a cross-modal multi-head attention module, a DyHead module, and a BERT layer. The cross-modal multi-head attention module is used to realize cross-modal feature interaction, and the DyHead module is used to encode visual features and extract image region features. , BERT layer is used to encode language features. Image region features and category embedding features Input to the detection head, the detection head includes two branches: positioning and classification. The positioning branch consists of a layer The convolutional layer is composed of the classification branch, which involves the similarity calculation between the category embedding features and the image region features; the regional features The box prediction is obtained by the positioning branch, and the similarity between the category embedding features and the image region features is calculated in the classification branch: , the similarity is the classification score of each category. During the model training process, the classification branch output results are calculated by combining the true value Focal Loss , calculate the output result of the positioning branch IoU Loss , to optimize network parameters.

[0027] In the above process, the single-stage detector GLIP is based on a parallel visual encoder and a language encoder, which are used to extract the input image feature map and category text embedding, respectively; a visual-language instance fusion module is used to extract visual instance features from the feature map, continuously store and update the average instance features of each category, and perform bidirectional multi-head attention interaction fusion with the corresponding category embedding; a visual-language global fusion encoder is used to fuse the global feature map and all category embeddings; and a detection head is used to generate detection results, including classification and localization. Specifically, the visual-language instance fusion module consists of an instance feature extractor for extracting instance features from feature maps, an instance feature memory area for storing category average instance features, and a visual-language fusion layer for fusing visual instance features with language category embeddings. Since the classification result is obtained by calculating the similarity between category embedding and image region features, this application extracts visual instance features and continuously stores and updates them, and fuses the visual instance features with language category embedding, thereby learning rich fine-grained information from the conceptual level and the instance level. Furthermore, in the classification process, the similarity between the fused language category embedding and the image region features is calculated to realize the utilization of the fine-grained visual instance information introduced in the category embedding, thereby improving the accuracy of fine-grained remote sensing target detection.

[0028] In the aforementioned fine-grained remote sensing target detection method based on visual-language instance fusion, this application constructs a fine-grained remote sensing image target detection network model comprising multiple modules. The visual encoder and language encoder are responsible for extracting remote sensing image features and category embedding features, respectively. This achieves preliminary information extraction from both visual and language modalities, providing a rich set of basic features for subsequent fusion operations. The instance feature extractor in the visual-language instance fusion module extracts visual instance features from remote sensing image features and stores them in an instance feature memory area. This design specifically captures and preserves target instance features, providing a data foundation for subsequent utilization of these features. The average instance features stored in the instance feature memory area are updated to obtain visual average instance features. This continuous updating allows the model to adapt to different input images, better capture target variations and diversity, and improve the model's ability to represent fine-grained features. The category embedding features and visual average instance features are fed into the visual-language fusion layer for interaction and fusion. Through methods such as a bidirectional multi-head attention mechanism, deep fusion of visual and language information is achieved. This fusion approach can fully utilize language category concepts to guide and supplement visual instance features, while also allowing visual instance features to refine and enhance language category embeddings, so that the fused features contain both rich visual details and accurate category concept information. The visual language deep enhancement encoder enhances remote sensing image features and interactively fused features, further improving feature quality and representation capabilities. The enhanced features can better reflect the characteristics of fine-grained targets, providing more accurate input to the detection head, thereby improving detection accuracy. Through end-to-end training, the entire model can automatically learn how to effectively fuse visual instance features with language category concepts from the input remote sensing images and category text to achieve accurate detection of fine-grained targets. During training, the model can continuously adjust the parameters of each module according to the loss function to optimize the feature extraction and fusion process to adapt to different datasets and task requirements.

[0029] In one embodiment, updating the average instance feature stored in the instance feature memory area to obtain the visual average instance feature includes:

[0030] Update the average instance features stored in the instance feature memory area to obtain the visual average instance features:

[0031]

[0032] in, Indicates the In the iterative process The average instance features of the categories, Representation category No. Instance features, Representation category Number of instances, ; .

[0033] In one embodiment, the category embedding feature and the visual average instance feature are input into a bidirectional multi-head attention layer for interaction and fusion, and the interactive fusion feature obtained includes:

[0034] Embed categories into features and visual average instance features Perform layer normalization separately and then embed the category into features and visual average instance features The corresponding key vectors are obtained by linear layer projection respectively and , value vector and , query vector and ,in, Indicates the categories;

[0035] Calculate the attention matrix , and then calculate the interactive embedding and features based on the attention matrix; fuse the interactive features into the original features to obtain the interactive fused features.

[0036] In one embodiment, the interaction embedding and features are calculated based on the attention matrix, including:

[0037] The embedding and features after interaction are calculated based on the attention matrix.

[0038] ;

[0039] in, represents a linear layer.

[0040] In one embodiment, the interactive features are fused into the original features to obtain interactive fused features, including:

[0041] The interactive features are fused into the original features, and the interactive fusion features are obtained as follows:

[0042] ;

[0043] in, are learnable parameters.

[0044] In one embodiment, the visual language deep enhancement encoder includes 6 network blocks, each of which is composed of a cross-modal multi-head attention module, a DyHead module and a BERT layer. The cross-modal multi-head attention module is used to realize cross-modal feature interaction, the DyHead module is used to encode visual features and extract image area features, and the BERT layer is used to encode language features.

[0045] In one embodiment, the detection head includes two branches: positioning and classification. The positioning branch consists of a layer The classification branch is composed of convolutional layers. It involves calculating the similarity between the category embedding features and the image region features. The enhanced remote sensing image features and the category embedding features are input into the detection head to obtain the detection results, including:

[0046] Enhanced remote sensing image features Get box prediction and category embedding features by positioning the branch Similarity calculation is performed with the enhanced remote sensing image features in the classification branch: , the similarity is the classification score of each category, and the category with the highest similarity is taken as the classification result.

[0047] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0048] In one embodiment, Figure 3 As shown, a fine-grained remote sensing target detection device based on visual language instance fusion is provided, including: a data acquisition module 302, a model construction module 304 and a fine-grained remote sensing target detection module 306, wherein:

[0049] The data acquisition module 302 is used to obtain input remote sensing images and category text; the category text is a string obtained by concatenating the names of all categories to be detected;

[0050] A model building module 304 is used to build a fine-grained remote sensing image target detection network model; the fine-grained remote sensing image target detection network model includes a visual encoder, a language encoder, a visual language instance fusion module, a visual language depth enhancement encoder, and a detection head;

[0051] The fine-grained remote sensing target detection module 306 is used to input the input remote sensing image and category text into the fine-grained remote sensing image target detection network model, extract the remote sensing image features according to the visual encoder, and extract the category embedding features through the language encoder; input the target instance and remote sensing image features into the visual language instance fusion module through the instance feature extractor to obtain the visual instance features and store them in the instance feature memory area; update the average instance features stored in the instance feature memory area to obtain the visual average instance features; input the category embedding features and the visual average instance features into the visual language fusion layer for interaction and fusion to obtain the interactive fused features; update the instance features stored in the instance feature memory area according to the interactive fused features; enhance the remote sensing image features and the interactive fused features according to the visual language deep enhancement encoder to obtain enhanced remote sensing image features and category embedding features; input the enhanced remote sensing image features and category embedding features into the detection head to obtain the detection results.

[0052] Regarding the specific limitations of the fine-grained remote sensing target detection device based on visual language instance fusion, please refer to the limitations of the fine-grained remote sensing target detection method based on visual language instance fusion above, which will not be repeated here. Each module in the above-mentioned fine-grained remote sensing target detection device based on visual language instance fusion can be implemented in whole or in part by software, hardware, and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0053] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a fine-grained remote sensing target detection method based on visual language instance fusion is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0054] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0055] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0056] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A fine-grained remote sensing target detection method based on visual language instance fusion, characterized in that: The method comprises: Obtain an input remote sensing image and a category text; the category text is a string obtained by concatenating the names of all categories to be detected; Constructing a fine-grained remote sensing image target detection network model; the fine-grained remote sensing image target detection network model includes a visual encoder, a language encoder, a visual language instance fusion module, a visual language deep enhancement encoder and a detection head; The input remote sensing image and category text are input into the fine-grained remote sensing image target detection network model, the remote sensing image features are extracted according to the visual encoder, and the category embedding features are extracted through the language encoder; the target instance and remote sensing image features are input into the visual language instance fusion module through the instance feature extractor to obtain the visual instance features and store them in the instance feature memory area; the average instance features stored in the instance feature memory area are updated to obtain the visual average instance features; the category embedding features and the visual average instance features are input into the visual language fusion layer for interaction and fusion to obtain the interactively fused features; the instance features stored in the instance feature memory area are updated according to the interactively fused features; the remote sensing image features and the interactively fused features are enhanced according to the visual language deep enhancement encoder to obtain enhanced remote sensing image features and category embedding features; the enhanced remote sensing image features and category embedding features are input into the detection head to obtain the detection results.

2. The method according to claim 1, characterized in that Update the average instance features stored in the instance feature memory area to obtain the visual average instance features, including: Update the average instance features stored in the instance feature memory area to obtain the visual average instance features: ； in, Indicates the In the iterative process The average instance features of the categories, Representation category No. Instance features, represents the number of feature channels, Representation category Number of instances, ; .

3. The method according to claim 1, characterized in that The category embedding features and the visual average instance features are input into the bidirectional multi-head attention layer for interaction and fusion, and the interactive fusion features are obtained, including: Embed categories into features and visual average instance features Perform layer normalization separately and then embed the category into features and visual average instance features The corresponding key vectors are obtained by linear layer projection respectively and , value vector and , query vector and ,in, Indicates the categories; Calculate the attention matrix , and then calculate the interactive embedding and features according to the attention matrix; fuse the interactive features into the original features to obtain the interactive fused features.

4. The method according to claim 3, characterized in that The post-interaction embedding and features are calculated based on the attention matrix, including: The embedding and features after interaction calculated according to the attention matrix are: ； in, represents a linear layer.

5. The method according to claim 4, characterized in that The interactive features are fused into the original features to obtain the interactive fusion features, including: The interactive features are fused into the original features, and the interactive fusion features are obtained as follows: ； in, are learnable parameters.

6. The method according to claim 1, characterized in that The visual language deep enhancement encoder includes 6 network blocks, each of which is composed of a cross-modal multi-head attention module, a DyHead module and a BERT layer. The cross-modal multi-head attention module is used to realize cross-modal feature interaction, the DyHead module is used to encode visual features and extract image area features, and the BERT layer is used to encode language features.

7. The method according to claim 1, characterized in that The detection head includes two branches: positioning and classification. The positioning branch consists of a layer The convolutional layer is composed of a classification branch that involves calculating the similarity between the category embedding features and the image region features. The enhanced remote sensing image features and the category embedding features are input into the detection head to obtain the detection results, including: The enhanced remote sensing image features Get box prediction and category embedding features by positioning the branch Similarity calculation is performed with the enhanced remote sensing image features in the classification branch: The similarity is the classification score of each category, and the category with the highest similarity is taken as the classification result.

8. A fine-grained remote sensing target detection device based on visual language instance fusion, characterized in that: The device comprises: A data acquisition module is used to obtain input remote sensing images and category text; the category text is a string obtained by concatenating the names of all categories to be detected; A model building module for building a fine-grained remote sensing image target detection network model; the fine-grained remote sensing image target detection network model includes a visual encoder, a language encoder, a visual language instance fusion module, a visual language deep enhancement encoder, and a detection head; A fine-grained remote sensing target detection module is used to input the input remote sensing image and category text into a fine-grained remote sensing image target detection network model, extract remote sensing image features according to the visual encoder, and extract category embedding features through the language encoder; input the target instance and remote sensing image features into the visual language instance fusion module through the instance feature extractor to obtain visual instance features and store them in the instance feature memory area; update the average instance features stored in the instance feature memory area to obtain visual average instance features; input the category embedding features and the visual average instance features into the visual language fusion layer for interaction and fusion to obtain interactively fused features; update the instance features stored in the instance feature memory area according to the interactively fused features; enhance the remote sensing image features and the interactively fused features according to the visual language deep enhancement encoder to obtain enhanced remote sensing image features and category embedding features; input the enhanced remote sensing image features and category embedding features into the detection head to obtain detection results.

Images