Space debris on-orbit real-time detection method based on YOLO

Abstract

The invention discloses a space debris on-orbit real-time detection method based on YOLO, and belongs to the technical field of target detection. The method comprises the steps of collecting space debris image data of a to-be-detected area; and inputting the space debris image data into a pre-trained YOLO detection model for identification to obtain an identification result. According to the method, the balance among extremely weak target detection, multi-scale feature coupling and calculation efficiency can be realized, the detection capability and real-time monitoring performance of small fragments are remarkably improved, a high-precision and low-delay monitoring guarantee is provided for on-orbit safe operation of a spacecraft, and the limitation of a traditional space-based observation method is broken through.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, in particular to a space debris on-orbit real-time detection method, system, device, storage medium and program based on YOLO. The present application first applies a deep learning target detection algorithm to the field of space debris detection, opening up a new technical route. BACKGROUND

[0002] In recent years, space debris detection technology based on deep learning has made a breakthrough and gradually applied to the field of space safety, promoting the development of on-orbit obstacle avoidance, situation awareness and other key technologies. At the same time, with the deepening of space exploration activities of various countries and the intensification of near-earth orbital resources, the importance of space debris monitoring and collision warning is increasingly prominent. According to international aerospace standards, space debris can be divided by size: large debris (>10cm), medium debris (1-10cm), small debris (0.1-1cm) and micro debris (<0.1cm), among which the detection of small and micro debris is particularly difficult but poses a huge threat to spacecraft safety.

[0003] Therefore, research institutions have begun to focus on building on-board real-time detection systems. Such systems need to have all-weather autonomous perception capabilities to achieve rapid identification and trajectory prediction of debris in complex starry backgrounds, providing an important supplement to the traditional passive defense mode relying on ground observation, and showing broad application prospects in satellite autonomous avoidance, orbital resource maintenance and space station safety protection. Its technical advantages lie in its ability to directly process on-board optical sensor data, eliminate ground station communication delays through real-time processing, significantly improve the timeliness of space situation awareness, and provide core technical support for building a space traffic safety system.

[0004] Current space target detection mainly uses traditional methods such as difference image method, optical flow method and dynamic programming algorithm. The difference image method achieves noise suppression through background modeling, has the characteristics of simple algorithm and high computational efficiency, but is prone to false alarms in areas with high star density, especially for small debris (0.1-1cm) with low recognition rate; the optical flow method can effectively capture the motion characteristics of the target, but it is difficult to balance the computational complexity and real-time performance, making it difficult to meet the real-time monitoring needs of a large number of small debris; the dynamic programming algorithm enhances the trajectory recognition ability through energy accumulation, but has the problem of path planning combination explosion in multi-target scenarios. When faced with a large number of space debris of different sizes, the efficiency is significantly reduced. In recent years, researchers have tried to combine particle filtering, Kalman prediction and other technologies with traditional methods by introducing kinematic model constraints to reduce false detection rates, or using hierarchical processing strategies to optimize the allocation of computing resources. However, these improved methods are still limited by the expression ability of hand-designed features, making it difficult to cope with complex scenarios such as debris shape diversity and sudden changes in lighting conditions, and there are significant limitations in detection accuracy and generalization ability.

[0005] With the breakthrough of deep learning technology, the target detection algorithm based on convolutional neural network provides a new solution path for real-time detection of space debris. Single-stage detection framework realizes the joint optimization of feature learning and positioning regression through end-to-end architecture, greatly improving the algorithm efficiency. The current research focuses on the lightweight improvement of network structure, reduces the computational load by introducing cross-stage feature reuse mechanism, and improves the model generalization performance combined with adaptive data enhancement strategy. However, the existing methods still have optimization space in extreme weak target detection, multi-scale feature coupling and other aspects, especially in the task of space debris detection in deep space dark background, the traditional method is difficult to effectively distinguish the low contrast target and background noise, resulting in a significant decrease in detection rate. How to build a spaceborne detection model that balances accuracy and efficiency is still a technical difficulty that needs to be broken through in this field. SUMMARY

[0006] In view of the problem in the prior art that how to balance between extreme weak target detection, multi-scale feature coupling and computational efficiency to improve the accuracy and generalization ability of the spaceborne space debris detection model. The invention provides a YOLO-based on-orbit real-time detection method for space debris, which can balance between extreme weak target detection, multi-scale feature coupling and computational efficiency, and provide high-precision and low-delay monitoring guarantee for the on-orbit safe operation of spacecraft.

[0007] In order to achieve the above purpose, the invention provides the following technical scheme: In the first aspect, a YOLO-based on-orbit real-time detection method for space debris, comprising: Collecting space debris image data of the detection area; Inputting the space debris image data into a pre-trained YOLO detection model for recognition to obtain a recognition result; The training method of the pre-trained YOLO detection model, comprising: Collecting space debris image data of the detection area and pre-processing to obtain a space debris image data set; Inputting the training set in the space debris image data set into the YOLO detection model to extract features through the neural network in the YOLO detection model; Performing multi-fusion processing on the extracted features to obtain a multi-scale fusion feature map; Training the YOLO detection model based on the multi-scale fusion feature map to obtain a pre-trained YOLO detection model.

[0008] As a further improvement of the invention, the collecting of space debris image data of the detection area and the pre-processing to obtain a space debris image data set, comprising: Image acquisition of space debris in the detection area to obtain a space debris image; The space debris image is standardized to obtain a standard size space debris image data set; The standard size space debris image data set is divided into a training set, a validation set and a test set.

[0009] As a further improvement of the application, the training set in the space debris image data set is input into the YOLO detection model, and the features are extracted through the neural network in the YOLO detection model, comprising: The neural network of the YOLO detection model is improved, including constructing a CSP-Darknet53 backbone network and a FPN+PAN bidirectional pyramid structure in the neural network of the YOLO detection model, and setting an SPPF module at the end of the backbone network to obtain an improved YOLO detection model; The training set of the space debris image data set is input into the CSP-Darknet53 backbone network of the improved YOLO detection model for feature extraction.

[0010] As a further improvement of the application, the extracted features are subjected to multi-fusion processing to obtain multi-scale fusion feature maps, comprising: The SPPF module is used for multi-scale feature fusion processing of the CSP-Darknet53 backbone network feature extraction to obtain multi-scale fusion feature maps.

[0011] As a further improvement of the application, the YOLO detection model is trained based on the multi-scale fusion feature maps to obtain a pre-trained YOLO detection model, comprising: Based on the multi-scale fusion feature maps, the YOLO detection model is trained using a positioning loss and a confidence loss to obtain a trained YOLO detection model; The validation set of the space debris image data set is input into the trained YOLO detection model to obtain a detection result; The detection result is compared with the artificial annotation result, and if the comparison result meets the set requirement, a pre-trained YOLO detection model is obtained.

[0012] As a further improvement of the application, the detection result is compared with the artificial annotation result, and if the comparison result meets the set requirement, comprising: The validation set of the space debris image data set is input into the trained YOLO detection model to obtain a detection result, and the detection result is compared with the artificial annotation result to evaluate the model performance; If the performance of the trained YOLO detection model meets the set condition, a test phase is entered, the test set of the space debris image data set is input into the trained YOLO model for detection, a detection result is obtained, and the performance index of the detection result meets the requirement, and the pre-trained YOLO detection model is obtained. If the performance of the trained YOLO detection model does not meet the set condition, the trained YOLO detection model is trained based on the trained YOLO detection model.

[0013] In a second aspect, the present application provides a YOLO-based space debris on-orbit real-time detection system, comprising: A data acquisition module is configured to acquire space debris image data of a detection area; An identification result module is configured to input the space debris image data into a pre-trained YOLO detection model for identification to obtain an identification result. The training method of the pre-trained YOLO detection model comprises: Space debris image data of a detection area is acquired and preprocessed to obtain a space debris image data set; A training set in the space debris image data set is input into a YOLO detection model, and features are extracted through a neural network in the YOLO detection model; The extracted features are subjected to multi-fusion processing to obtain a multi-scale fusion feature map; The YOLO detection model is trained based on the multi-scale fusion feature map to obtain a pre-trained YOLO detection model.

[0014] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the YOLO-based space debris on-orbit real-time detection method.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the steps of the YOLO-based space debris on-orbit real-time detection method.

[0016] In a fifth aspect, the present application provides a computer program product comprising computer instructions, wherein the computer instructions are executed by a processor to implement the steps of the YOLO-based space debris on-orbit real-time detection method.

[0017] Compared with the prior art, the present application has the following beneficial effects: The pre-trained YOLO detection model of the aspect can acutely capture extremely weak space debris target signals. In the preprocessing stage, the collected space debris image data is processed in detail, the image quality is optimized, the contrast between the target and the background is enhanced, and the weak target features that are originally difficult to detect are highlighted. In the model training process, a large number of real and diverse space debris image data sets are used for deep learning, and the model gradually masters the unique feature patterns of weak targets, so that these weak targets can be accurately identified in actual detection, greatly improving the sensitivity of detection, providing more reliable protection for the safe operation of spacecraft in orbit, and effectively reducing the collision risk caused by undetected weak debris.

[0018] Further, the present application realizes efficient and accurate fusion. Space debris has different sizes and shapes, and different scale features are crucial for accurate detection. In the YOLO detection model, the input image is processed by a neural network to extract multi-level features, and then these different scale features are processed by multi-fusion to generate multi-scale fusion feature maps. This fusion method fully considers the relevance and complementarity between different scale features, so that the model can comprehensively understand the target features from multiple angles, whether it is the overall shape of large debris or the subtle features of small debris, which can be effectively captured and integrated. This multi-scale feature coupling mechanism greatly improves the detection accuracy of the model for space debris of different scales, enhances the adaptability of the model to complex scenes, and enables the model to maintain excellent detection performance when facing space debris of various sizes and shapes. BRIEF DESCRIPTION OF DRAWINGS

[0019] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present disclosure in any way. In the drawings: Figure 1 A flowchart of a YOLO-based space debris in-orbit real-time detection method of the present application; Figure 2 A flowchart of a training method of a pre-trained YOLO detection model of the present application; Figure 3 A training set picture in a YOLO-based space debris in-orbit real-time detection method of the present application; Figure 4 A YOLOv5 main architecture diagram in a YOLO-based space debris in-orbit real-time detection method of the present application; Figure 5 An FPN+PAN structure in a YOLO-based space debris in-orbit real-time detection method of the present application; Figure 6 An artificially labeled space debris detection data in a YOLO-based space debris in-orbit real-time detection method of the present application; Figure 7 Space debris detection data for training a model in a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 8 Confusion matrix for a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 9 F1-Score curve with confidence change for a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 10 Loss function result graph for a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 11 Accuracy and confidence relationship graph for a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 12 Recall rate and confidence relationship graph for a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 13 P-R curve graph for a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 14 Space debris detection result for a YOLO-based space debris on-orbit real-time detection method of the present application; Figure 15 Structure diagram of a YOLO-based space debris on-orbit real-time detection system of the present application; Figure 16 Structure diagram of a pre-trained YOLO detection model training system of the present application; Figure 17 Electronic device schematic diagram in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order for those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application, and the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] To address the challenges of space debris detection in existing technologies, traditional space-based observation methods suffer from large orbit prediction errors, difficulty in identifying small debris, and challenges in real-time tracking of debris evolution and decomposition. This invention innovatively applies the YOLOv5 target detection algorithm to the field of space debris detection, fully leveraging YOLOv5's advantages in target detection to provide high-precision, low-latency monitoring assurance for the safe operation of spacecraft in orbit. Figure 1 As shown, this method includes: S100: Acquire spatial debris image data of the area to be detected; S200: Input spatial debris image data into a pre-trained YOLO detection model for recognition and obtain the recognition result; like Figure 2 As shown, the training methods for the pre-trained YOLO detection model include: S201: Collect spatial debris image data of the area to be detected and preprocess it to obtain a spatial debris image dataset; S202: Input the training set from the spatial debris image dataset into the YOLO detection model, and extract features through the neural network in the YOLO detection model; S203: Perform multi-scale fusion processing on the extracted features to obtain a multi-scale fused feature map; S204: The YOLO detection model is trained based on multi-scale fusion feature maps to obtain a pre-trained YOLO detection model.

[0023] This method utilizes the advanced target detection capabilities of YOLOv5 to achieve efficient identification of space debris. It strikes a balance between extremely weak target detection, multi-scale feature coupling, and computational efficiency, providing high-precision, low-latency monitoring assurance for the safe operation of spacecraft in orbit.

[0024] The present invention will be further explained and described below with reference to the accompanying drawings: S1: Experimental Environment Configuration The present application is based on Pytorch1.11.0 version for experiment, using Windows 11 operating system, CPU is Intel(R) Core(TM) i7-13700F, GPU is NVIDIA GeForce GTX 4060 Ti, CUDA version is 12.7. The experiment uses Python3.10.0 environment, uses Anaconda3 to create a virtual environment for isolated configuration, configures the necessary deep learning library, including NumPy, matplotlib, OpenCV-Python, etc., for data processing and model building.

[0025] S2: Data set preparation and preprocessing Image acquisition of space debris in the detection area to obtain space debris images; Standardization processing of space debris images to obtain space debris image data set with standard size; Construction of space debris image data set containing different backgrounds and different complexities.

[0026] As shown in Figure 3 The collected space debris images are uniformly processed to 640x640 standard resolution. LabelImg tool is used to label "debris" category, generating 116 training sets and 45 validation set pictures. To ensure the reliability and practicality of detection, the data set contains different background conditions (bright starry sky and dark starry sky background), different target features (including complex background, multi-target scene and small size target) and different observation conditions (including different light conditions and different angle images).

[0027] S3: YOLOv5 model application Collect space debris image data of the detection area and preprocess: feature labeling of the image, divide the labeled image data set into training set, validation set and test set, standardization processing of the image, construction of space debris image data set containing different backgrounds and different complexities. Especially containing space debris samples of different brightness levels, among which the dark background samples simulate the weak light condition in the real space environment, to ensure that the model can learn the discriminative features of dark area targets.

[0028] As shown in Figure 3As shown, this invention uses the YOLOv5 model for spatial debris detection, employing its standard architecture, including a backbone network using CSP-Darknet53 for feature extraction and an SPPF module for feature fusion and multi-scale feature processing. The CSP-Darknet53 backbone network optimizes gradient flow through cross-stage local connections, enhancing the representation of targets in dark areas while maintaining feature diversity. The SPPF module uses a multi-scale pooling structure to fuse features from different receptive fields, effectively improving contour recognition accuracy in low-light areas. The FPN+PAN structure achieves complementary features between shallow and deep layers through a bidirectional feature pyramid, ensuring that detailed information about small debris in dark backgrounds is not lost. Figure 5 The diagram shows feature enhancement using the FPN+PAN bidirectional feature fusion network, and detection outputs for detecting targets of different sizes using feature maps at three scales: 20×20, 40×40, and 80×80.

[0029] The model training process specifically includes the following three stages: 1) Feature extraction stage: like Figure 4 As shown, the input image is processed by the CSP-Darknet53 backbone network to extract multi-level features and generate multi-level feature maps, focusing on the details and semantic information of small, medium and large targets respectively. The SPPF module further enhances the key features: the 80×80 feature map captures the details of small fragments, the 40×40 feature map identifies medium targets, and the 20×20 feature map locates large fragments; the SPPF module enhances the contour features of dark areas through multi-scale pooling fusion processing.

[0030] 2) Feature fusion stage: like Figure 5 As shown, the FPN path upsamples the deep semantic feature P5 and fuses it with the mid-level feature C4 to generate P4; the PAN path downsamples P4 and fuses it with the shallow feature C3 to generate P3; the output is a three-scale fused feature of P3 / P4 / P5 to achieve fragment scale adaptive detection.

[0031] 3) Training optimization phase: like Figure 10 As shown, the localization loss (box_loss) and confidence loss (obj_loss) are jointly optimized.

[0032] The loss function of an object detection model typically consists of three parts: box regression loss, object confidence loss, and class classification loss.

[0033] Box regression loss measures the difference in position between the predicted bounding box and the ground truth bounding box. Its formula is:

[0034] where, p is the Euclidean distance between the predicted box and the real box center point, c is the diagonal length of the minimum circumscribed rectangle, α is the weight coefficient, v measures the consistency of the aspect ratio.

[0035] The object confidence loss (obj loss) is used to determine whether the target exists, and is usually calculated using binary cross entropy (BCE), as follows:

[0036] where, is the real label (0 or 1), pi is the predicted confidence.

[0037] In this task, since only spatial debris is detected (without class distinction), the class classification loss (clsloss) is always 0.

[0038] Therefore, the positioning loss (box_loss) and the confidence loss (obj_loss) are jointly optimized, as shown in Figure 10 After each round of training, the performance is evaluated using the dark background validation set, and the parameters are dynamically adjusted.

[0039] The YOLO detection model in this application adopts the YOLOv5 architecture, which has the following advantages: first, a gray-scale adaptive processing mechanism: directly processing single-channel gray images, optimizing dark gradient flow through CSP-Darknet53 backbone network cross-stage local connection, enhancing the edge response ability of low-contrast targets; second, a multi-scale feature fusion structure: combining the multi-scale pooling of the SPPF module and the FPN+PAN bidirectional pyramid, realizing the deep-shallow layer complementation of 20×20, 40×40, 80×80 three-scale features, and improving the detection sensitivity of small debris; third, a lightweight real-time inference design: through grid prediction, the three-scale detection results are output in parallel, and the depth ratio (0.67) and channel ratio (0.75) are compressed to meet the millisecond-level response requirements of spaceborne devices.

[0040] S4: Training parameter configuration The YOLO detection model is selected as the basic model, the training parameters are set, the training set is used for model training, the validation set is used to evaluate the model performance and adjust the parameters, and finally the test set is used to test the performance of the trained model to obtain the final YOLO detection model.

[0041] The following key parameter settings are used for training, as shown in Table 1.

[0042] Table 1 Hyperparameter settings

[0043] S5: System Validation and Performance Evaluation S5.1: Comparison of Detection Results like Figure 6 and Figure 7 As shown, the results of manual annotation are compared with the results of model detection. The confidence level of the model detection results generally reaches above 0.8, verifying its effectiveness in the space debris detection task.

[0044] S5.2: Accuracy Assessment like Figure 8 As shown, the model performance was analyzed using a confusion matrix, and the recognition accuracy for the "debris" category reached 0.71. Figure 9 The F1-Score curve shown peaks at 0.71 with a confidence threshold of 0.5, indicating that the model achieves a good balance between precision and recall.

[0045] S5.3: Training Process Analysis like Figure 10 As shown, the model training process exhibits excellent convergence characteristics. The localization loss (box_loss) on the training set significantly decreased from an initial 0.12 to 0.04, a reduction of 66.7%, while the box_loss on the validation set also steadily decreased from 0.10 to 0.058, indicating that the model's ability to locate spatial debris continuously improves. The confidence loss (obj_loss) also shows a good decreasing trend, decreasing from 0.075 to 0.045 on the training set and from 0.066 to 0.060 on the validation set, indicating that the model's judgment of the existence of targets is becoming increasingly accurate. It is particularly noteworthy that, since the task only involves a single "debris" category, the classification loss (cls_loss) remains near the ideal value of 0, perfectly matching the expected design. In terms of evaluation metrics, both precision and recall steadily increase with the number of training epochs and eventually converge, demonstrating the model's excellent detection performance.

[0046] S5.4: Detection Performance Analysis like Figure 11 , Figure 12 and Figure 13 As shown, in the accuracy-confidence relationship, the higher the confidence threshold, the higher the detection accuracy; in the recall-confidence relationship, the lower the confidence threshold, the more comprehensive the detection coverage; the PR curve maintains high precision and recall under different thresholds, confirming the stability of the model.

[0047] S5.5: Practical Application Results like Figure 14As shown, the model shows good adaptability in actual space debris detection, is effective for large debris (confidence > 0.8) and small debris, can adapt to the detection requirements under different background conditions, and meets the performance requirements of real-time monitoring.

[0048] In the task of dark background debris detection, the method has the following advantages: first, gray adaptability: the ability to directly process single-channel images avoids information loss during RGB conversion; second, weak light robustness: the multi-scale feature fusion mechanism ensures feature fidelity under extremely low illumination; third, real-time efficiency: the lightweight network structure meets the real-time processing requirements under the resource constraints of spaceborne devices.

[0049] The application applies the YOLOv5 target detection algorithm to the field of space debris detection, realizes debris detection through image recognition without complex orbit prediction calculation, realizes real-time detection under limited computing resources by virtue of the lightweight characteristics of YOLOv5, effectively identifies and tracks small space debris by virtue of the small target detection capability of YOLOv5, and adopts a strict model training, verification and testing process to ensure the reliability and generalization ability of the detection model.

[0050] The application realizes accurate detection of space debris by improving the YOLOv5 architecture, combines adaptive feature fusion and multi-scale optimization strategy, and meets the real-time processing demand while ensuring the detection accuracy. Any equivalent modification based on the core architecture of the application, such as adjusting the pyramid network level and optimizing the residual component structure. In summary, the application introduces the YOLO algorithm into the field of space debris detection and establishes a new detection technology route, breaking through the limitations of traditional ground-based observation methods. Compared with the traditional ground-based observation method, the application does not need complex orbit prediction calculation, but directly realizes detection through image recognition, reducing the influence of orbit prediction error. At the same time, the lightweight characteristics of the YOLO algorithm enable it to realize real-time detection under limited computing resources, and it has a significant advantage in small target detection, which can effectively identify and track small space debris. In addition, the application adopts a scientific division method of training set, validation set and test set, and through a strict model training, verification and test process, the reliability and generalization ability of the detection model are ensured, and the real-time performance is good, which can timely discover and track the evolution and decomposition process of debris; significantly improve the detection capability of small debris; realize real-time monitoring. At the same time, YOLOv5 has a unique advantage in processing grayscale images: considering that space debris images are mostly single-channel grayscale images, the end-to-end convolutional architecture of YOLOv5 can directly process grayscale input, and its multi-level feature extraction mechanism can adaptively learn dark texture features. Through the grayscale adaptive feature extraction mechanism and dark target enhancement strategy, the inherent defects of traditional methods in low-contrast space debris detection are solved. In the design of the backbone network, CSP-Darknet53 gradually enhances the target edge response in the dark background by stacking convolutional layers, effectively solving the missing detection problem of traditional methods in low-contrast scenes.

[0051] The second object of the application is to provide a YOLO-based space debris on-orbit real-time detection system, as shown in Figure 15 , comprising: The data acquisition module 100 is used for acquiring space debris image data of the detection area; The recognition result module 200 inputs the space debris image data into the pre-trained YOLO detection model for recognition to obtain the recognition result.

[0052] As shown in Figure 16 , a training system of a pre-trained YOLO detection model, comprising: The image acquisition module 201 is used for acquiring space debris image data of the detection area and performing preprocessing to obtain a space debris image data set; The feature extraction module 202 is used for inputting the training set in the space debris image data set into the YOLO detection model to extract features through the neural network in the YOLO detection model; Feature fusion module 203: used to perform multi-fusion processing on the extracted features to obtain a multi-scale fused feature map; Training Model Module 204: Trains the YOLO detection model based on multi-scale fusion feature maps to obtain a pre-trained YOLO detection model.

[0053] like Figure 17 As shown, a third objective of this invention is to provide an electronic device comprising a processor 301, a memory 302, and a display screen 303. The memory 302 and the display screen 303 are both connected to the processor 301, such as via a bus 304. Optionally, the electronic device may further include a transceiver 305. It should be noted that in practical applications, the transceiver 305 is not limited to one type, and the structure of this electronic device does not constitute a limitation on the embodiments of this application.

[0054] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0055] Bus 304 may include a pathway for transmitting information between the aforementioned components. Bus 304 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 304 can be divided into address bus, data bus, control bus, etc.

[0056] The memory 302 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or it may be an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0057] The memory 302 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 302 to implement the content shown in the foregoing method embodiments.

[0058] Figure 17 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0059] A fourth objective of this invention is to provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, performs the aforementioned functions. Figures 1-2 The illustrated method embodiments include various processes. For example, a memory may include instructions that can be executed by a processor of an electronic device to perform the described method.

[0060] A computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. A computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, a computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), staging random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory stick, floppy disk, optical disk, magnetic disk, mechanical encoding device, or any combination thereof.

[0061] A fifth objective of this invention is to provide a computer program product, including computer instructions that, when executed by a processor, implement the above-described objectives. Figures 1-2 The various processes of the method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0062] Many embodiments and applications beyond the examples provided will be apparent to those skilled in the art upon reading the foregoing description. Therefore, the scope of this teaching should not be determined by reference to the foregoing description, but rather by reference to the foregoing claims and the full scope of their equivalents. For purposes of completeness, all articles and references, including patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the foregoing claims is not intended as a waiver of that subject matter, nor should it be construed as an indication that the applicant has not considered that subject matter as part of the disclosed inventive subject matter.

[0063] The above content provides a further detailed description of the present invention. It should not be construed that the specific embodiments of the present invention are limited to this. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention as defined by the submitted claims.

Claims

1. A YOLO-based method for real-time on-orbit detection of space debris, characterized in that, include: Acquire spatial debris image data of the area to be detected; Spatial debris image data is input into a pre-trained YOLO detection model for identification, and the identification results are obtained. The training method for the pre-trained YOLO detection model includes: Spatial debris image data of the area to be detected are collected and preprocessed to obtain a spatial debris image dataset; The training set from the spatial debris image dataset is input into the YOLO detection model, and features are extracted through the neural network in the YOLO detection model. The extracted features are subjected to multi-scale fusion processing to obtain a multi-scale fused feature map; The YOLO detection model is trained based on multi-scale fusion feature maps to obtain a pre-trained YOLO detection model.

2. The method for real-time on-orbit detection of space debris based on YOLO according to claim 1, characterized in that, The spatial debris image data of the area to be detected is collected and preprocessed to obtain a spatial debris image dataset, including: Image acquisition is performed on the spatial debris in the area to be detected to obtain spatial debris images; The spatial debris images are standardized to obtain a standard-sized spatial debris image dataset; The standard-sized spatial fragment image dataset was divided into training, validation, and test sets.

3. The method for real-time on-orbit detection of space debris based on YOLO according to claim 1, characterized in that, The step of inputting the training set from the spatial debris image dataset into the YOLO detection model and extracting features through the neural network in the YOLO detection model includes: The neural network of the YOLO detection model is improved by constructing a CSP-Darknet53 backbone network and an FPN+PAN bidirectional pyramid structure in the neural network of the YOLO detection model, and setting an SPPF module at the end of the backbone network to obtain the improved YOLO detection model. The training set of the spatial debris image dataset is input into the improved YOLO detection model CSP-Darknet53 backbone network for feature extraction.

4. The method for real-time on-orbit detection of space debris based on YOLO according to claim 3, characterized in that, The step of performing multi-scale fusion processing on the extracted features to obtain a multi-scale fused feature map includes: The SPPF module was used to perform multi-scale feature fusion processing on the CSP-Darknet53 backbone network features to obtain a multi-scale fused feature map.

5. The method for real-time on-orbit detection of space debris based on YOLO according to claim 1, characterized in that, The process of training the YOLO detection model based on multi-scale fusion feature maps to obtain a pre-trained YOLO detection model includes: Based on the multi-scale fusion feature map, the YOLO detection model is trained using localization loss and confidence loss to obtain the trained YOLO detection model. The validation set of the spatial debris image dataset is input into the trained YOLO detection model to obtain the detection results; The detection results are compared with the manually labeled results. If the comparison results meet the set requirements, the pre-trained YOLO detection model is obtained.

6. The method for real-time on-orbit detection of space debris based on YOLO according to claim 5, characterized in that, The step of comparing the detection results with the manually labeled results, whereby the comparison results meet the set requirements, includes: The validation set of the spatial debris image dataset is input into the trained YOLO detection model to obtain the detection results. The detection results are then compared with the manually labeled results to evaluate the model performance. If the performance of the trained YOLO detection model meets the set conditions, it will enter the testing phase. The test set of the spatial fragment image dataset will be input into the trained YOLO model for detection to obtain the detection results. If the performance index of the detection results meets the requirements, it will be used to train the pre-trained YOLO detection model. If the performance of the trained YOLO detection model does not meet the set conditions, then training will be performed based on the trained YOLO detection model.

7. A YOLO-based real-time on-orbit space debris detection system, characterized in that, include: Data acquisition module: used to acquire spatial debris image data of the area to be detected; Recognition Result Module: Inputs spatial debris image data into a pre-trained YOLO detection model for recognition and obtains the recognition result; The training method for the pre-trained YOLO detection model includes: Spatial debris image data of the area to be detected are collected and preprocessed to obtain a spatial debris image dataset; The training set from the spatial debris image dataset is input into the YOLO detection model, and features are extracted through the neural network in the YOLO detection model. The extracted features are subjected to multi-scale fusion processing to obtain a multi-scale fused feature map; The YOLO detection model is trained based on multi-scale fusion feature maps to obtain a pre-trained YOLO detection model.

8. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the YOLO-based real-time on-orbit detection method for space debris as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the YOLO-based real-time on-orbit detection method for space debris as described in any one of claims 1-6.

10. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of a YOLO-based real-time on-orbit detection method for space debris as described in any one of claims 1-6.

Images