On-orbit moving target detection method in inertial space stable imaging mode

Abstract

The invention relates to the technical field of optical remote sensing, and particularly discloses an in-orbit moving target detection method in an inertial space stable imaging mode, which comprises the following steps: inputting a control instruction and business data into a load for analysis processing to obtain image data; filtering background noise of the image data; sequentially executing target track pre-detection, further denoising processing and moving target identification and segmentation, finally obtaining a position area of a moving target with third precision, and obtaining target position information; and combining the obtained position information of the moving target with the file number and the auxiliary data, and calculating to obtain multi-dimensional information of the moving target at different moments. Complicated star catalogue data do not need to be introduced to remove interference information in the whole processing flow, only a traditional image processing algorithm is used, processing modes such as a convolutional neural network which have high requirements for the computing power of equipment are not used, the actual measurement processing speed and accuracy performance of edge equipment are excellent, and the method can be used for detecting and recognizing moving targets in on-orbit deployment.

Description

Technical Field

[0001] The present invention relates to the field of optical remote sensing technology, and particularly relates to a method for detecting on-orbit moving targets in an inertial space stable imaging mode. Background Art

[0002] The detection of space moving targets is an important technology for space situation awareness and a key link to ensure space security and promote scientific research and technological development. In the increasingly complex space environment, the types of space targets are diverse, including operating satellites, space debris, and near-Earth asteroids, etc. Their motion patterns are complex and changeable, and optical observations are easily affected by background interference, increasing the technical difficulty of detection. Precise space target detection, surveillance, tracking, and rapid target recognition and cataloging can not only timely detect potential threat targets, avoid space collisions, and ensure the safe operation of spacecraft, but also support the orbital observation of near-Earth celestial bodies and deep space exploration missions, and promote human understanding and exploration of the universe.

[0003] Space motion target detection mainly processes the captured sequence of star maps to obtain key information such as the position and shape of the moving target. The core of this technology lies in extracting the motion feature information of the target. Currently, the algorithms for space target detection are mainly divided into two categories: The first category is the method based on star catalog data. For example, in the Chinese patent with the publication number CN118293928A and the name "A Method and Device for Rapid Identification and Positioning of Space Targets in Space-based Infrared Detection", it is proposed to separate the starry sky background by extracting feature information such as the angular positions of background stars, thereby revealing the trajectory of the space target. However, this method relies on a large amount of star catalog data, which not only requires efficient extraction of star features but also rapid retrieval of their positions in the star map, and there are significant limitations in terms of data scale and processing efficiency. The second category of methods detects based on the difference in the motion characteristics between space targets and background stars. This method does not rely on star catalog data and can achieve detection only by analyzing the different motion patterns of the star background and the target, so it is more flexible and suitable for target detection in dynamic scenarios. In addition, currently, researchers have proposed various methods to solve the problem of space target detection, such as realizing the detection of moving targets based on theories such as machine learning and deep learning; specifically: in the Chinese patent with the publication number CN115131392A and the name "A Method and Device for Detecting and Tracking Space Moving Targets Based on Space-based Optical Observation Images", it is proposed to perform correlation matching on each frame of the image to realize the tracking of moving targets. In the Chinese patent with the publication number CN119091223A and the name "A Space Target Detection Model and Method Based on an Improved YOLOv8 Model", it is proposed to introduce a traditional target detection model into space target detection. These methods have relatively high detection accuracy and applicability in theory. Especially, deep learning methods can automatically extract features from a large amount of data, showing significant advantages. However, in practical applications, these methods still face some problems. In the inertial space stable imaging mode, affected by sensor noise and background star points, the redundant information and complexity of the image are relatively high, making it difficult for them to be directly applied to the on-orbit real-time detection scenario; in addition, the on-orbit environment has very strict requirements for computing resources and real-time performance, and it is difficult to implement on-orbit deployment for overly complex algorithm models. Therefore, how to simplify the algorithm complexity and improve the operation efficiency has become one of the important directions for the further development of space motion target detection technology. Currently, those skilled in the art urgently need to design an on-orbit detection method to meet the detection and recognition requirements of space motion targets. Summary of the Invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the defects in the above-mentioned prior art, so as to provide an on-orbit moving target detection method in the inertial space stable imaging mode.

[0005] An on-orbit moving target detection method in the inertial space stable imaging mode includes the following steps: Input the control instructions and service data into the payload, and perform parsing. Process the service data based on the parsing results to obtain multiple frames of 8-bit image data; Perform three noise filtering operations on the image data to filter out background noise and obtain an image highlighting the trajectory information of the moving target; among them, the three noise filtering operations are preprocessing, filtering processing, and segmentation processing in sequence; Based on the line segment detection method, pre-detect the target trajectory of the image with background noise filtered, and obtain the position area where the moving target is located with the first accuracy; Use the morphological processing method to further denoise the image data after pre-detection processing, and obtain the position area where the moving target is located with the second accuracy; Perform moving target recognition and segmentation on the image data processed by the morphological processing method, confirm the position area where the moving target is located with the third accuracy, and obtain the target position information; among them, the accuracy of the position area where the moving target is located with the third accuracy > the accuracy of the position area where the moving target is located with the second accuracy > the accuracy of the position area where the moving target is located with the first accuracy; Combine the obtained moving target position information with the file number and auxiliary data, and calculate to obtain the multi-dimensional information of the moving target at different times.

[0006] Preferably, the service data is multiple frames of original panchromatic data taken by a remote sensing satellite in the inertial space stable imaging mode.

[0007] Preferably, the preprocessing is specifically: use linear transformation to initially filter out the background noise interference in the image data, and use the frame difference method to filter out the sensor fixed pattern noise and amplify the position information of the moving target.

[0008] Preferably, the segmentation processing is specifically: use conditional threshold segmentation and adaptive threshold segmentation to process the image to highlight the trajectory information of the moving target.

[0009] Preferably, based on the line segment detection method, pre-detect the target trajectory of the image with background noise filtered, and obtain the position area where the moving target is located with the first accuracy, specifically: use the Hough transform to pre-detect the target trajectory, filter out the elongated dot targets, and obtain the position area where the moving target is located with the first accuracy.

[0010] Preferably, use the morphological processing method to further denoise the image data after pre-detection processing, and obtain the position area where the moving target is located with the second accuracy. The specific operation is: erode first and then dilate; Among them, the erosion operation: erode the star point interference and the moving target boundary noise; The dilation operation: obtain the actual coverage area information of the position area where the moving target is located through dilation.

[0011] Preferably, perform moving target recognition and segmentation on the image data processed by the morphological processing means, confirm the position area of the moving target with the third precision, and obtain the target position information. Specifically: perform connected component detection on the image data processed by the morphological processing means, and design corresponding detection thresholds as detection indicators according to the target features to confirm the position area of the moving target with the third precision and obtain the target position information.

[0012] The technical solution of the present invention has the following advantages: The method of the present invention can achieve fast, accurate, and low-complexity detection, and can be directly applied to the in-orbit recognition of space moving targets of optical remote sensing satellites; moreover, the entire processing process does not need to introduce complex star catalog data to remove interference information, only uses traditional image processing algorithms, and does not use processing methods with high computing power requirements for devices such as convolutional neural networks. The measured processing speed and accuracy performance are excellent on edge devices and can be used for in-orbit deployment for the detection and recognition of moving targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0014] Figure 1 is a schematic flowchart of the method of the present invention; Figure 2 is a schematic diagram of the original panchromatic image in the inertial space stable imaging mode; Figure 3 is an 8-bit image after bit depth conversion; Figure 4 is a schematic diagram of the stretched result of the image after data preprocessing; Figure 5 is a schematic diagram of the stretched result of the image after filtering and denoising; Figure 6 is a schematic diagram of the image after threshold denoising; Figure 7 is a schematic diagram of the image after pre-detection of moving targets; Figure 8 is a schematic diagram of the image after erosion operation; Figure 9 is a schematic diagram of the image after dilation operation; Figure 10 is a schematic diagram of the labeled result after detection and recognition. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention. In addition, the terms "first", "second", "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0017] In the description of the present invention, it should be noted that unless otherwise clearly specified and limited, the terms "installation", "connection", "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0018] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0019] Embodiment 1 The detection of space moving targets is not only an important technical means to ensure space safety, but also provides technical support for scientific research and space situation awareness construction. The conventional method of relying on star catalogs for positioning requires the introduction of complex cache data and matching algorithms, which is not conducive to the rapid extraction of target information in the on-orbit mode. Currently, most use the direct recognition method to filter out the interference of background noise and star points to achieve the detection of space moving targets. In recent years, with the development of machine learning and deep learning, the extraction of the feature dimension of target information by algorithms has gradually deepened, and the recognition ability has also gradually increased. However, in the inertial space stable imaging mode, the image background noise is complex and diverse, and the overall appearance of key targets is in the form of line segments or arcs. Using deep learning and other methods will undoubtedly increase the algorithm complexity and there are also certain difficulties in the annotation of the sample data set. In addition, the limited power consumption and computing power of the on-orbit platform are also an important limiting factor.

[0020] Therefore, in this embodiment, by analyzing the image information and moving target features in the inertial space stable imaging mode, an on-orbit moving target detection method in the inertial space stable imaging mode is proposed. This method filters out the background noise through frame difference and linear transformation; highlights the moving target trajectory information through denoising algorithms such as median filtering, adaptive threshold segmentation, and binarization processing; then uses double Huffman transform to perform pre-detection processing on the target trajectory; afterwards, uses morphological processing to further erode the interference noise points around the target, and restores the target position information through dilation processing; then uses multi-dimensional information such as the compactness of the target area, the aspect ratio of the minimum oblique rectangle, and the area to further realize the recognition and detection of the target; finally, outputs label information such as target coordinates, categories, and auxiliary data. This algorithm program is robust, has low complexity, fast execution speed, and the multi-parameter algorithm is adjustable, suitable for on-orbit deployment applications. The following details the specific execution process.

[0021] An on-orbit moving target detection method in the inertial space stable imaging mode includes the following steps: Step 1: Input control instructions and service data into the payload, and perform parsing. Based on the parsing results, process the service data to obtain multiple frames of 8-bit image data; Perform three noise filtering operations on the image data to filter out background noise and obtain an image highlighting the moving target trajectory information; among them, the three noise filtering operations are Step 2: Preprocessing, Step 3: Filtering processing, and Step 4: Segmentation processing in sequence; Step 5: Based on the line segment detection means, perform pre-detection of the target trajectory on the image after filtering out background noise, and obtain the position area where the moving target is located with the first accuracy; Step 6: Use morphological processing means to further denoise the image data after pre-detection processing, and obtain the position area where the moving target is located with the second accuracy; Step 7: Perform moving target recognition and segmentation on the image data processed by the morphological processing means, confirm the position area where the moving target is located with the third accuracy, and obtain the target position information; among them, the accuracy of the position area where the moving target is located with the third accuracy > the accuracy of the position area where the moving target is located with the second accuracy > the accuracy of the position area where the moving target is located with the first accuracy; Step 8: Combine the obtained moving target position information with the file number and auxiliary data to calculate and obtain the multi-dimensional information of the moving target at different times.

[0022] The following details the above eight steps in this embodiment in combination with specific implementation scenarios and experimental data: Step 1: Input control instructions and service data into the payload; in this embodiment, multiple frames of original panchromatic data taken by the "Jilin-1" Gaofen 06 optical remote sensing satellite in the inertial space stable imaging mode are used, such asFigure 2 As shown, the acquired image data has a storage quantization bit depth of 16 bits, denoted as Image_16, and the image size is 9520 * 3520 pixels.

[0023] Through instruction parsing and algorithm parameter configuration, and performing a loading operation on the original panchromatic data, bit splicing is performed on the original panchromatic data; due to limited hardware overhead of the actual on-orbit platform, bit-depth conversion processing is required. In order to retain the effective information of the original panchromatic data, global normalization is used for processing to obtain 8-bit image data for loading, denoted as Image_8; ; In the formula: represents the original 16-bit panchromatic image, represents the maximum DN value of the original panchromatic image, the minimum DN value of the original panchromatic image, represents the converted 8-bit image data, and the processing result is as Figure 3 shown.

[0024] Step Two: The preprocessing means are as follows: Using linear transformation to preliminarily filter out background noise interference in the image data, and using the inter-frame difference method to filter out sensor fixed pattern noise and amplify the position information of moving targets.

[0025] Specifically: Since the acquired 8-bit image data for loading has complex background noise and is affected by sensor fixed pattern noise, with light band distribution interference, preprocessing operations need to be performed on the 8-bit image data. In this embodiment, linear transformation is used to preliminarily filter out part of the background noise interference in the image, and at the same time, the inter-frame difference method is used for reverse saturation operation to filter out sensor fixed pattern noise and amplify the position information of moving targets. The processing result is as Figure 4 shown. For easy display, the image has been stretched.

[0026] Step Three: After completing the preprocessing of the 8-bit image data, the processing result is that most of the superimposed mixed noise has been removed, and the remaining are mostly independently distributed noise and residual star point interference; therefore, in this embodiment, adaptive filtering is used for sub-region processing; In practical applications, to ensure processing speed, common fast filtering means such as median filtering and Gaussian filtering are selected to process the noise, or common adaptive local noise reduction filtering means are selected to process the noise to further remove background noise interference; in this case, median filtering is used for processing, and the result is as Figure 5 shown. For easy display, the image has been stretched.

[0027] Step Four: The segmentation processing is specifically as follows: Conditional threshold segmentation and adaptive threshold segmentation are used to process the image to highlight the moving target trajectory information.

[0028] Specifically: After completing the filtering operation in Step 3, the star map target information gradually becomes clear. Due to the large size of the full map and the inconsistent DN value levels in different regions, in Step 4, conditional threshold segmentation is first used to perform regional filtering on the low DN value noise belonging to the background noise; then, adaptive threshold segmentation is used to adaptively process the images in different brightness regions to remove the low DN value background noise dots and extract the effective information, realizing the binary processing of the images. The processing results are as Figure 6 shown.

[0029] Step 5: Based on the line segment detection method, pre-detection of the target trajectory is performed on the image with background noise removed to obtain the position area where the moving target is located with the first precision; Specifically: The Hough transform is used to pre-detect the target trajectory, filter the elongated dot targets, and obtain the position area where the moving target is located with the first precision.

[0030] It should be noted that after removing the background noise, the pixel distribution in the position area where the moving target is located is basically clear. At this time, in this embodiment, the Hough transform is used to predict the potential trajectory of the moving target, and multiple pairs of superimposed target area line segment information are obtained; In order to fuse the interrupted moving trajectories and expand the target area range, in this embodiment, line segment detection is used again to select the locally connected target areas. After two line segment detection processes, the line segment areas in the pre-detected image are connected. The processing results are as Figure 7 shown.

[0031] Step 6: Perform morphological processing on the entire image after the pre-detection processing in Step 5 to further remove noise and obtain the position area where the moving target is located with the second precision. The specific operation is: erode first and then dilate; Perform the erosion operation. The convolution kernel selects a 3*3 cross kernel and iterates 1 time to erode the star point interference and the boundary noise of the moving target. The processing results are as Figure 8 shown; Perform the dilation operation. The convolution kernel selects a 3*3 rectangular kernel and iterates 1 time to obtain the actual coverage area information of the position area where the moving target is located through dilation. The processing results are as Figure 9 shown.

[0032] Step 7: Perform moving target recognition and segmentation on the image data processed by morphological processing means, confirm the position area of the moving target with the third precision, and obtain the target position information. Specifically: perform connected component detection on the image data processed by morphological processing means, and design corresponding detection thresholds as detection indicators according to target features to confirm the position area of the moving target with the third precision and obtain the target position information.

[0033] Among them, the parsed set threshold parameters are as follows: the area limit Area_th is 10000 pixel, the moving target limit Dec_th is 100 pixel, the noise target limit Noise_th is 10 pixel, the upper limit of detected targets Num_th is 5100, the aspect ratio threshold Asp_th is 4, and the compactness threshold Comp_th is 100. The specific operation process is shown in the following table:

[0034] In the table: num represents the total number of connected component targets; FAILED is used to determine the status flag bit, indicating that the condition does not meet, specifically that the number of connected components is large, the background noise interference is large, the detection task fails, and the detection of this frame is exited; i represents a single connected region; area_i represents the area of a single connected region; length_i represents the perimeter of a single connected region; L_i represents the length of the smallest inclined rectangle; W_i represents the width of the smallest inclined rectangle; asp_ratio represents the aspect ratio of the smallest inclined rectangle; comp_ratio represents the target compactness; Step Eight: Output the target numbers, labels, position information together with the image number and longitude and latitude coordinate information as a whole, and obtain multi-dimensional information such as the position information and movement trajectory of the moving target at different times through calculation; perform detection and annotation results on the background image as Figure 10 shown.

[0035] The detection method adopted in this embodiment is for the detection of on-orbit moving targets in the inertial space stable imaging mode. Therefore, the detection thresholds for moving targets, such as the area threshold, aspect ratio threshold, and compactness threshold, are related to the exposure time of a single imaging; in addition, the positioning of the star point position is also achieved through target area parameters such as area and compactness, which is used to combine star catalog information for further matching verification to assist in confirming the spatial position of the moving target.

[0036] In the above-described embodiments, various algorithms can be configured with different parameters; during the specific execution process, a combined configuration mode of multiple algorithms can also be selected for image filtering and denoising. For the sake of concise description, not all possible parameter configurations and combination processes are described. However, as long as there are no contradictions in the combination of these processing methods and the configuration of parameters, and the thinking process of the processing is consistent with the expression of the block diagram thinking, it should be considered to be within the scope described in the present invention. It should be noted that for those of ordinary skill in the art, without departing from the basic processing flow of the present invention, various deformations and improvements can be made, and these changes and improvements still fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent shall be subject to the appended claims.

Claims

1. A method for detecting on-orbit moving targets in an inertial space stabilized imaging mode, characterized in that: The following steps are involved: Input control instructions and business data into the payload and parse them. Process the business data based on the parsing results to obtain multiple frames of 8-bit image data. Perform three noise filtering operations on the image data to filter out background noise and obtain an image that highlights the trajectory information of the moving target; the three noise filtering operations are preprocessing, filtering processing and segmentation processing in sequence; Perform target trajectory pre-detection on the image after background noise is filtered out based on line segment detection to obtain the location area of ​​the moving target with first accuracy; The image data after the pre-detection processing is further denoised by using morphological processing means to obtain the position area of ​​the moving target with second precision; Performing moving target recognition and segmentation on the image data processed by the morphological processing means, confirming the position area of ​​the moving target with the third precision, and obtaining the target position information; wherein the accuracy of the position area of ​​the moving target with the third precision is greater than the accuracy of the position area of ​​the moving target with the second precision and greater than the accuracy of the position area of ​​the moving target with the first precision; The acquired moving target position information is combined with the file number and auxiliary data to calculate and obtain the multi-dimensional information of the moving target at different times.

2. The target detection method according to claim 1, characterized in that: The business data are multiple frames of original full-color data taken by remote sensing satellites in the inertial space stabilized imaging mode.

3. The target detection method according to claim 1, characterized in that: The preprocessing is specifically as follows: using linear transformation to preliminarily filter out background noise interference in image data, and using frame difference method to filter out sensor fixed pattern noise, and amplify the position information of the moving target.

4. The target detection method according to claim 1, characterized in that: The segmentation process specifically includes: using conditional threshold segmentation and adaptive threshold segmentation to process the image, and highlighting the moving target trajectory information.

5. The target detection method according to claim 1, characterized in that: The target trajectory is pre-detected on the image with background noise filtered out based on line segment detection to obtain the first precision location area of ​​the moving target. Specifically, the target trajectory is pre-detected using Huffman transform, the elongated dot targets are filtered out, and the first precision location area of ​​the moving target is obtained.

6. The target detection method according to claim 1, characterized in that: The image data after pre-detection processing is further denoised by using morphological processing methods to obtain the second-precision location area of ​​the moving target. The specific operation is: first erosion and then expansion; Among them, the erosion operation: erosion of star point interference and moving target boundary noise; Dilation operation: The actual coverage area information of the moving target’s location area is obtained through dilation.

7. The target detection method according to claim 1, characterized in that: The image data processed by the morphological processing means is subjected to moving target recognition and segmentation, the location area of ​​the moving target with the third precision is confirmed, and the target position information is obtained. Specifically, connected domain detection is performed on the image data processed by the morphological processing means, and a corresponding detection threshold is designed as a detection index according to the target characteristics, the location area of ​​the moving target with the third precision is confirmed, and the target position information is obtained.

Images