A space target detection model and method based on a YOLOv8 improved model

Abstract

The application provides a space target detection model and method based on a YOLOv8 improved model, wherein the following improvements are made to the existing YOLOv8 model: in the backbone network, the original SPPF module is replaced with a GPF-SPPF module; a convolution and attention fusion module CAFusion is connected after the GPF-SPPF module; in the neck network, the original up-sampling module is replaced with an EMA-CARAFE module. The model can accurately identify space debris and space targets such as satellites of various types, significantly improves the space situation awareness capability, and helps correct orbit maneuver decision-making of on-orbit spacecraft. Compared with a traditional target recognition algorithm, the application uses deep learning technology to more richly extract the features of the space target, thereby significantly improving the detection accuracy of the space target.

Description

Technical Field

[0001] This invention relates to the field of machine learning, and in particular to a spatial object detection model and method based on an improved YOLOv8 model. Background Technology

[0002] With the rapid development of space technology, the number of orbiting satellites has increased significantly, and their importance in communication, navigation, surveillance, and reconnaissance has become increasingly prominent. However, with the increase in the number of satellite launches, the number of space debris and abandoned satellites has also increased, seriously threatening space security. Therefore, countries have made the utilization of outer space resources a strategic development priority, making space target detection technology a hot research area.

[0003] Traditional space target detection systems are relatively mature in the precise positioning of celestial bodies such as stars and planets. However, when detecting near-Earth space targets such as space debris, artificial satellites, and meteorites, traditional methods often perform poorly due to the complex trajectories, diverse appearances, and complex lighting conditions of these targets. These methods suffer from insufficient intelligence and poor adaptability to multiple tasks. For example, the MF02A04 satellite-borne laser terminal in the "Jilin-1" constellation successfully conducted a high-speed satellite-to-ground laser image transmission experiment, laying the foundation for the transmission of satellite-to-ground optical detection data. Space-based optical detection systems have broad application prospects in the detection of space debris, abandoned satellites, and meteorites, collision warning, and satellite integrity inspection. For instance, China's Gaofen-6 and Shijian-21 satellites possess all-weather, all-round detection capabilities.

[0004] Traditional object detection methods mainly focus on two aspects: feature extraction (enhancing feature representation and robustness) and feature classification (improving classification accuracy and speed). Researchers have proposed various feature extraction and classification methods, such as SIFT, Hough Transform, AdaBoost, and DPM. However, these hard feature-based methods suffer from three main problems: (a) insufficient target feature representation; (b) high feature similarity, leading to a high classification error rate; and (c) highly targeted feature design, with single features performing poorly in detecting multiple targets in complex scenes.

[0005] With the rise of Convolutional Neural Networks (CNNs), deep learning has been widely applied in object detection and has gradually become the mainstream method. Researching deep learning-based space object detection algorithms is of great significance for spacecraft and satellites to autonomously identify surrounding objects and avoid space collisions. Currently, deep learning object detection algorithms are mainly divided into two categories: one-stage and two-stage. Two-stage algorithms (such as Faster R-CNN) first generate candidate regions through a Region Proposal Network (RPN), and then classify and regress bounding boxes on these regions. Single-stage algorithms (such as YOLO and SSD) complete object classification and location regression simultaneously in one forward propagation, thus achieving faster detection speed. Although two-stage detectors have advantages in detection accuracy, their speed is relatively slow due to the need for two feature extractions and calculations. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing research and provide a space target detection model and method based on an improved YOLOv8 model. This invention significantly improves the detection accuracy of space debris and satellites while maintaining detection speed.

[0007] Firstly, the present invention provides a spatial target detection model based on an improved YOLOv8 model, which makes the following improvements to the existing YOLOv8 model:

[0008] In the backbone network, the original SPPF module is replaced with the GPF-SPPF module;

[0009] A convolutional and attention fusion module, CAFusion, is connected after the GPF-SPPF module.

[0010] In the neck network, the original upsampling module is replaced with the EMA-CARAFE module.

[0011] Based on the above, the GPF-SPPF module is configured with three processing paths;

[0012] The main path consists of three sequentially connected max pooling layers, which are used to generate three features containing local saliency information, respectively.

[0013] The first sub-path passes through a global average pooling layer;

[0014] The second sub-path passes through a global max pooling layer;

[0015] The features generated from the three paths are concatenated, and then channel integration and further feature extraction are performed through a CBS convolutional layer to generate the final output features.

[0016] Based on the above, the convolution and attention fusion module CAFusion includes a first normalization module LayerNorm, a convolutional attention information module CAIM, a second normalization module LayerNorm, and a multi-scale feedforward network module MFFN connected in sequence.

[0017] The convolutional attention information module CAIM is used to capture long-range dependencies and neighborhood spectral correlations, and the multi-scale feedforward network module MFFN is used to extract features at different scales and suppress multi-scale features.

[0018] The input of the second normalization module LayerNorm is also connected to the input of the first normalization module LayerNorm;

[0019] The output of the convolutional attention information module CAIM is also connected to the output of the multi-scale feedforward network module MFFN.

[0020] Based on the above, the Convolutional Attention Information Module (CAIM) includes global branches and local branches for calculating convolutional attention information;

[0021] The convolutional attention information of local branches is represented as follows:

[0022] F CONV =C 3×3×3 (C shuffle (C 1×1 (x)))

[0023] F CONV This is for local branch output, C 1×1 For a 1×1 convolution, C 3×3×3 For a 3×3×3 convolution, C shuffle For channel shuffling operations, x is the input feature;

[0024] In the global branch, a 1×1 convolution and three 3×3 depthwise convolutions are first used to generate the query Q, key K, and value, respectively, producing three shapes. tensor;

[0025] Next, reconstruct Q as K is reconstructed as And then according to and The interaction between them calculates the attention mechanism. To reduce computation and obtain a lightweight model, the size of the model is not calculated. Attention map;

[0026] Define the convolutional attention information of the global branch as F att :

[0027]

[0028]

[0029] α is a learnable scaling parameter used to control the size of the matrix multiplication of K and Q before applying the softmax function; Y represents the original input, which is added to the output of the attention mechanism to form a residual connection. The representative value vector represents the actual input data, which will be weighted according to the attention score;

[0030] Finally, the calculation result of the convolutional attention information module CAIM is: F out =F att +F conv .

[0031] Based on the above, the multi-scale feedforward network module MFFN uses two 1×1 convolutions to expand the feature channels into two feature channels. The input features are processed in two parallel feature channels, and a gating mechanism is introduced. The element-wise product of the two feature channels is used as the output to enhance the nonlinear transformation.

[0032] In one feature channel, the input features are processed by two parallel dilated convolutional paths and then merged for output. In the other feature channel, the input features are processed by a 3×3 dilated convolution and then output.

[0033] The gating mechanism is introduced in another feature channel.

[0034] Based on the above, the EMA-CARAFE module includes a feature compression module, an upsampling kernel prediction module, and a feature recombination module;

[0035] The feature compression module is used to perform a 1×1 convolution on the input feature map of shape H×W×C to compress the number of channels, resulting in H×W×C. m Feature map;

[0036] The upsampling kernel prediction module is connected to the feature compression module and is used to perform H×W×C analysis. m Content encoding and upsampling kernel prediction are performed on the feature maps; among them, K... encoder ×K encoder The convolutional layer predicts the upsampling kernel, with C input channels. m The output is σ 2 K up 2 The channel is then expanded in spatial dimensions to obtain a shape of σH×σW×K. up 2 The upsampling kernel;

[0037] The normalization module is connected to the upsampling kernel prediction module and is used to perform a normalization operation on the upsampling kernel so that the sum of the convolution kernel weights is 1.

[0038] The feature recombination module, connected to the normalization module, is used to perform the following operations:

[0039] Map each position in the normalized feature map back to the input feature map, and take K as the center of each position mapped back to the input feature map. up ×K up The region is multiplied by the normalized upsampling kernel to obtain the output value; different channels at the same position share the same upsampling kernel, and the final output feature map is σH×σW×C.

[0040] H is the height of the input feature map, W is the width of the input feature map, and C is the number of channels in the input feature map; C m K represents the number of channels in the compressed feature map. encoder ×K encoder K represents the size of a convolutional kernel used for upsampling prediction, σ represents the upsampling factor, and K... up 2 This indicates the size of the upsampling kernel.

[0041] Secondly, the present invention provides a space target detection method, comprising:

[0042] Step 1: Use STK software to simulate real space scenes, collect images of near-target scenes in the starry sky, and create an image dataset; use labeling tools to label the images and form a labeled dataset;

[0043] Step 2: Integrate the image dataset and the label dataset to create a complete training dataset; divide the dataset into training set, validation set and test set in an 8:1:1 ratio to create the training dataset.

[0044] Step 3: Input the training dataset into the spatial object detection model based on the improved YOLOv8 model for iterative training. When the model's performance index on the validation set reaches the preset accuracy threshold, stop training, save the model's parameter weights at this time, and perform starry sky near object detection on the test image data to complete the model training.

[0045] Step 4: Input the image to be detected into the trained model, and the model outputs the detection result.

[0046] Thirdly, the present invention provides a computer device comprising:

[0047] One or more processors;

[0048] Memory, used to store one or more programs.

[0049] When the one or more programs are executed by the one or more processors, the one or more processors perform the steps of the space target detection method as described above.

[0050] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the space target detection method as described.

[0051] Fifthly, the present invention provides a computer program product comprising a computer program / instructions, characterized in that, when the computer program / instructions are executed by a processor, they implement the steps of the space target detection method as described above.

[0052] This invention has outstanding substantive features and significant progress compared to the prior art, specifically:

[0053] 1. The model of this invention can accurately identify space debris and various types of satellites and other space targets, significantly improving space situational awareness and helping to make correct orbital maneuver decisions for spacecraft in orbit.

[0054] 2. Compared with traditional target recognition algorithms, this invention utilizes deep learning technology to extract more features of spatial targets, thereby significantly improving the detection accuracy of spatial targets.

[0055] 3. By optimizing the model structure, this invention effectively reduces the number of model parameters and lowers the computational cost. Moreover, the model can be deployed on resource-constrained terminal devices, ensuring both detection accuracy and improving detection speed. Attached Figure Description

[0056] Figure 1 This is the network structure of Context Sensing-YOLOv8 in Embodiment 1 of the present invention.

[0057] Figure 2 This is a network structure diagram of the GPF-SPPF module in Embodiment 1 of the present invention.

[0058] Figure 3 This is a network structure diagram of the CAFusion module in Embodiment 1 of the present invention.

[0059] Figure 4 This is a structural diagram of the CAIM module in Embodiment 1 of the present invention.

[0060] Figure 5 This is a structural diagram of the multi-scale feedforward network (MFFN) in Embodiment 1 of the present invention.

[0061] Figure 6 This is a structural diagram of the EMA-CARAFE in Embodiment 1 of the present invention.

[0062] Figure 7 This is a graph of evaluation indicators in Embodiment 1 of the present invention.

[0063] Figure 8 The image shows the detection results of the Context Sensing-YOLOv8 model and the original YOLOv8n model in a near-object scene in the starry sky. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] Example 1

[0066] This embodiment proposes a spatial object detection model based on an improved YOLOv8 model – Context Sensing-YOLOv8, such as... Figure 1 As shown. Context Sensing-YOLOv8 is an improvement on the existing YOLOv8 model made as follows:

[0067] In the backbone network, the original SPPF module is replaced with the GPF-SPPF module;

[0068] A convolutional and attention fusion module, CAFusion, is connected after the GPF-SPPF module.

[0069] In the neck network, the original upsampling module is replaced with the EMA-CARAFE module.

[0070] Specifically, such as Figure 2 As shown, the GPF-SPPF module is configured with three processing paths;

[0071] The main path consists of three sequentially connected max pooling layers, which are used to generate three features containing local saliency information, respectively.

[0072] The first sub-path passes through a global average pooling layer;

[0073] The second sub-path passes through a global max pooling layer;

[0074] The features generated from the three paths are concatenated, and then channel integration and further feature extraction are performed through a CBS convolutional layer to generate the final output features.

[0075] The GPF-SPPF module in this embodiment can add some global background information and edge information to help the feature extraction network extract target features better.

[0076] To enhance the network's ability to model global and local features, this embodiment designs a convolutional and attention fusion module, CAFusion. For example... Figure 3 As shown, CAFusion includes a first normalization module LayerNorm, a convolutional attention information module CAIM, a second normalization module LayerNorm, and a multi-scale feedforward network module MFFN, which are connected in sequence.

[0077] The convolutional attention information module CAIM is used to capture long-range dependencies and neighborhood spectral correlations, and the multi-scale feedforward network module MFFN is used to extract features at different scales and suppress multi-scale features.

[0078] The input of the second normalization module LayerNorm is also connected to the input of the first normalization module LayerNorm;

[0079] The output of the convolutional attention information module CAIM is also connected to the output of the multi-scale feedforward network module MFFN.

[0080] Preferably, the convolutional attention information module (CAIM) includes global branches and local branches, such as... Figure 4 As shown, this is used to calculate convolutional attention information;

[0081] The convolutional attention information of local branches is represented as follows:

[0082] F CONV =C 3×3×3 (C shuffle (C 1×1 (x)))

[0083] F CONV This is for local branch output, C 1×1 For a 1×1 convolution, C 3×3×3 For a 3×3×3 convolution, C shuffle For channel shuffling operations, x is the input feature;

[0084] In the global branch, the query Q, key K, and value are first generated using a 1×1 convolution and three 3×3 depthwise convolutions, respectively, resulting in three shapes. tensor;

[0085] Next, reconstruct Q as K is reconstructed as And then according to and The interaction between them calculates the attention mechanism. To reduce computation and obtain a lightweight model, the size of the model is not calculated. Attention map;

[0086] Define the convolutional attention information of the global branch as F att :

[0087]

[0088]

[0089] α is a learnable scaling parameter used to control the size of the matrix multiplication of K and Q before applying the softmax function; Y represents the original input, which is added to the output of the attention mechanism to form a residual connection. The representative value vector represents the actual input data, which will be weighted according to the attention score;

[0090] Finally, the calculation result of the convolutional attention information module CAIM is: F out =F att +F conv .

[0091] Preferably, the multi-scale feedforward network module (MFFN) is as follows: Figure 5 As shown, two 1×1 convolutions are used to expand the feature channels into two feature channels. The input features are processed in two parallel feature channels, and a gating mechanism is introduced. The element-wise product of the two feature channels is used as the output to enhance the nonlinear transformation.

[0092] In one feature channel, the input features are processed by two parallel dilated convolutional paths and then merged for output. In the other feature channel, the input features are processed by a 3×3 dilated convolution and then output. A gating mechanism is introduced in the other feature channel.

[0093] The formula for the gated output of the multi-scale feedforward network module (MFFN) is as follows:

[0094]

[0095]

[0096] in, Indicates nonlinearity. Represents unit multiplication. This represents a 3×3 dilated convolution with a dilation rate of 3. It is represented as a 3×3 dilated convolution with a dilation rate of 2.

[0097] Traditional upsampling modules ignore the semantic information following the feature image, resulting in the loss of detail information in images with dense targets. To solve this problem, this embodiment also provides an EMA-CARAFE module. Figure 6 As shown, the EMA-CARAFE module includes a feature compression module, an upsampling kernel prediction module, and a feature recombination module;

[0098] The feature compression module is used to perform a 1×1 convolution on the input feature map of shape H×W×C to compress the number of channels, resulting in H×W×C. m Feature map;

[0099] The upsampling kernel prediction module is connected to the feature compression module and is used to perform H×W×C analysis. m Content encoding and upsampling kernel prediction are performed on the feature maps; among them, K... encoder ×K encoder The convolutional layer predicts the upsampling kernel, with C input channels. m The output is σ 2 K up 2 The channel is then expanded in spatial dimensions to obtain a shape of σH×σW×K. up 2 The upsampling kernel;

[0100] The normalization module is connected to the upsampling kernel prediction module and is used to perform a normalization operation on the upsampling kernel so that the sum of the convolution kernel weights is 1.

[0101] The feature recombination module, connected to the normalization module, is used to perform the following operations:

[0102] Map each position in the normalized feature map back to the input feature map, and take K as the center of each position mapped back to the input feature map. up ×K up The region is multiplied by the normalized upsampling kernel to obtain the output value; different channels at the same position share the same upsampling kernel, and the final output feature map is σH×σW×C.

[0103] H is the height of the input feature map, W is the width of the input feature map, and C is the number of channels (also known as the depth) of the input feature map.

[0104] C m To represent the number of channels in the compressed feature map, in the upsampling prediction module, the input feature map is compressed to C after a 1×1 convolution. m To reduce the amount of subsequent calculations;

[0105] K encoder ×K encoderThis represents the size of a convolutional kernel used for upsampling kernel prediction;

[0106] σH×σW: This is the size of the output feature map after upsampling, σ 2 K up 2 This represents the number of channels in the upsampling kernel, where σ represents the upsampling factor, and K... up 2 This indicates the size of the upsampling kernel.

[0107] The output formula of the EMA-CARAFE module is as follows:

[0108]

[0109] In the formula: For upsampling prediction module, It is the sigmoid activation function. Indicates the input feature map The region centered at position i is a k*k region;

[0110] The result of the upsampling kernel prediction module is multiplied by the result of the feature reconstruction module to obtain the output result image and the input feature map. The upsampling result corresponding to position i in the diagram:

[0111]

[0112] In the formula: This is a feature recombination module.

[0113] In object detection algorithms, recall, precision, mAP, and inference speed are commonly chosen as evaluation metrics for network models. Recall P represents the proportion of correctly identified positive classes out of the total number of positive classes, while precision R reflects the proportion of positive classes correctly classified by the model.

[0114]

[0115]

[0116] mAP integrates the Precision and Recall of all classes by calculating the area under the PR curve for each class. A higher value indicates better multi-class performance of the model.

[0117]

[0118] Evaluation indicators such as Figure 7 As shown, the model proposed in this embodiment improves the mAP value by 0.8%, P by 0.8%, and R by 0.7% compared to the original model. Since the detection speed is 108 frames / second, which is greater than 25 frames per second, it meets the requirements of real-time monitoring.

[0119] Example 2

[0120] This embodiment provides a space target detection method, including:

[0121] Step 1: Use STK software to simulate a real space scene, collect images of near-target scenes in the starry sky, and create an image dataset; use a labeling tool to annotate the images to form a labeled dataset, such as... Figure 7 As shown;

[0122] Step 2: Integrate the image dataset and the label dataset to create a complete training dataset; divide the dataset into training set, validation set and test set in an 8:1:1 ratio to create the training dataset.

[0123] Step 3: Input the training dataset into the spatial target detection model based on the improved YOLOv8 model described in Example 1 for iterative training. When the performance index of the model on the validation set reaches the preset accuracy threshold, stop training, save the parameter weights of the model at this time, and perform star-sky near target detection on the test image data to complete the model training.

[0124] Step 4: Input the image to be detected into the trained model, and the model outputs the detection result.

[0125] Figure 8 The images show the detection results of the Context Sensing-YOLOv8 model and the original YOLOv8n model in a near-object scene in the starry sky. It can be seen that the Context Sensing-YOLOv8 model has a significant improvement in the accuracy of spatial object detection compared to the original YOLOv8n model.

[0126] Example 3

[0127] Each module in the method of Example 2 can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in the processor of a computer device in hardware form or independent of the processor, or it can be stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0128] In an exemplary embodiment, a computer device is provided, which may be a terminal. The computer device further includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the steps of the spatial target detection method described in Embodiment 2. The display unit of the computer device is used to form a visually visible image and may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0129] Those skilled in the art will understand that the structure of the computer device described above is only a partial structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components, or combine certain components, or have different component arrangements.

[0130] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the space target detection method described in Embodiment 2.

[0131] In one exemplary embodiment, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the space target detection method as described in Embodiment 2.

[0132] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0133] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0134] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A spatial target detection model based on an improved YOLOv8 model, used for spatial target detection on input near-object scene images of starry skies, characterized in that, The following improvements were made to the existing YOLOv8 model: In the backbone network, the original SPPF module is replaced with the GPF-SPPF module; The GPF-SPPF module is configured with three processing paths; The main path consists of three sequentially connected max pooling layers, which are used to generate three features containing local saliency information, respectively. The first sub-path passes through a global average pooling layer; The second sub-path passes through a global max pooling layer; The features generated from the three paths are concatenated, and then channel integration and further feature extraction are performed through a CBS convolutional layer to generate the final output features. A convolutional and attention fusion module, CAFusion, is connected after the GPF-SPPF module. The convolution and attention fusion module CAFusion includes a first normalization module LayerNorm, a convolutional attention information module CAIM, a second normalization module LayerNorm, and a multi-scale feedforward network module MFFN, which are connected in sequence. The convolutional attention information module CAIM is used to capture long-range dependencies and neighborhood spectral correlations, and the multi-scale feedforward network module MFFN is used to extract features at different scales and suppress multi-scale features. The input of the second normalization module LayerNorm is also connected to the input of the first normalization module LayerNorm; The output of the convolutional attention information module CAIM is also connected to the output of the multi-scale feedforward network module MFFN; In the neck network, the original upsampling module is replaced with the EMA-CARAFE module; The EMA-CARAFE module includes a feature compression module, an upsampling kernel prediction module, and a feature recombination module.

2. The spatial target detection model based on the improved YOLOv8 model according to claim 1, characterized in that, The Convolutional Attention Information Module (CAIM) includes global and local branches for calculating convolutional attention information. The convolutional attention information of local branches is represented as follows: F CONV =C 3×3×3 (C shuffle (C 1×1 (x))) F CONV This is for local branch output, C 1×1 For a 1×1 convolution, C 3×3×3 For a 3×3×3 convolution, C shuffle For channel shuffling operations, x is the input feature; In the global branch, the query Q, key K, and value are first generated using a 1×1 convolution and three 3×3 depthwise convolutions, respectively, resulting in three shapes. tensor; Next, reconstruct Q as K is reconstructed as And then according to and The interaction between them calculates the attention mechanism. To reduce computation and obtain a lightweight model, the size of the model is not calculated. Attention map; Define the convolutional attention information of the global branch as F att : α is a learnable scaling parameter used to control the size of the matrix multiplication of K and Q before applying the softmax function; Y represents the original input, which is added to the output of the attention mechanism to form a residual connection. The representative value vector represents the actual input data, which will be weighted according to the attention score; Finally, the calculation result of the convolutional attention information module CAIM is: F out =F att +F conv .

3. The spatial target detection model based on the improved YOLOv8 model according to claim 2, characterized in that: The multi-scale feedforward network module MFFN uses two 1×1 convolutions to expand the feature channels into two feature channels. The input features are processed in two parallel feature channels, and a gating mechanism is introduced. The element-wise product of the two feature channels is used as the output to enhance the nonlinear transformation. In one feature channel, the input features are processed by two parallel dilated convolutional paths and then merged for output. In the other feature channel, the input features are processed by a 3×3 dilated convolution and then output. The gating mechanism is introduced in another feature channel.

4. The spatial target detection model based on the improved YOLOv8 model according to claim 1, characterized in that: The feature compression module is used to perform a 1×1 convolution on the input feature map of shape H×W×C to compress the number of channels, resulting in H×W×C. m Feature map; The upsampling kernel prediction module is connected to the feature compression module and is used to perform H×W×C analysis. m Content encoding and upsampling kernel prediction are performed on the feature maps; among them, K... encoder ×K encoder The convolutional layer predicts the upsampling kernel, with C input channels. m The output is σ 2 K up 2 The channel is then expanded in spatial dimensions to obtain a shape of σH×σW×K. up 2 The upsampling kernel; The normalization module, connected to the upsampling kernel prediction module, is used to perform a normalization operation on the upsampling kernel so that the sum of the convolution kernel weights is 1. The feature recombination module, connected to the normalization module, is used to perform the following operations: Map each position in the normalized feature map back to the input feature map, and take K as the center of each position mapped back to the input feature map. up ×K up The region is multiplied by the normalized upsampling kernel to obtain the output value; different channels at the same position share the same upsampling kernel, and the final output feature map is σH×σW×C. H is the height of the input feature map, W is the width of the input feature map, and C is the number of channels in the input feature map; C m K represents the number of channels in the compressed feature map. encoder ×K encoder K represents the size of a convolutional kernel used for upsampling prediction, σ represents the upsampling factor, and K... up 2 This indicates the size of the upsampling kernel.

5. A method for detecting space targets, characterized in that, include: Step 1: Use STK software to simulate real space scenes, collect images of near-target scenes in the starry sky, and create an image dataset; use labeling tools to label the images and form a labeled dataset; Step 2: Integrate the image dataset and the label dataset to create a complete training dataset; divide the dataset into training set, validation set and test set in an 8:1:1 ratio to create the training dataset. Step 3: Input the training dataset into the spatial target detection model based on the YOLOv8 improved model as described in any one of claims 1-4 for iterative training. When the performance index of the model on the validation set reaches the preset accuracy threshold, stop training, save the parameter weights of the model at this time, and perform star-sky near target detection on the test image data to complete the model training. Step 4: Input the image to be detected into the trained model, and the model outputs the detection result.

6. A computer device, characterized in that, include: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the steps of the space target detection method as described in claim 5.

7. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by the processor, it implements the steps of the space target detection method as described in claim 5.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the space target detection method as described in claim 5.

Images