Method and system for close-range space target detection and special component intelligent identification

Abstract

The application discloses a close-in space target detection and special component intelligent identification method and system, which comprises the following contents: preprocessing a visible light image to obtain a binary image; extracting all suspected targets in the binary image to obtain an initial suspected target point set, sorting the suspected target points according to target feature brightness and area fusion confidence to obtain a first suspected target point set; removing stars in the field of view based on star-sensitive pointing to obtain a second suspected target point set; performing track association, removing false alarms and confirming target points through multiple frames; determining a close-in stage according to the area size of the target points, wherein the close-in stage comprises a long-distance stage and a close-distance stage, directly outputting the centroid coordinates of the target in the long-distance stage, and identifying the special components of the target and outputting the centroid coordinates of the special components in the close-distance stage. The application can realize close-in space target whole-process detection, meet the continuity requirement of space monitoring, and intelligently identify special components.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of space target detection, in particular to a close-in space target detection and special component intelligent identification method and system. BACKGROUND

[0002] With the development of space technology, the demand for space exploration of human beings is gradually increasing, but the increasingly frequent space exploration activities bring more and more space targets, and a large number of space targets are prone to collision, which seriously threatens the safety of space environment. And with the continuous development of global space resources, the outer space of the earth is gradually becoming a new field of military struggle.

[0003] In order to ensure the sustainability of space activities and maintain the safety of China's space environment, it is necessary to quickly improve the ability of space target monitoring. Space target detection is an important part of space target monitoring system, and the detection and identification of space targets are the prerequisite for space target monitoring and carrying out other activities, so the target feature detection and identification technology is the key technology for space target monitoring. For complex and special space environment, visual sensor has the advantages of simple structure, rich application scenarios, low price and strong real-time performance, and has become the most important sensor of space manipulator, which is equivalent to the eyes of the manipulator. Therefore, the research and application of space target feature detection and identification algorithm based on vision have important significance.

[0004] Optoelectronic imaging system can obtain visible light image of space target in close-in space, get its geometric shape, attitude and other feature information, and then detect, identify, monitor and track the target. However, in the process of close-in of space target, the target becomes larger, and the target with large size and high signal-to-noise ratio is easy to be detected. Smaller space targets are approximately point target imaging in image and have weak energy, which is difficult to distinguish from star points and noise in space in vision, which brings great difficulty to space target detection.

[0005] In view of this problem, scholars start from reducing false alarm and study the method of star elimination. The false alarm points appearing in the image are eliminated according to target feature method and motion information, but the detected target is not the part of the system focus monitoring. And the existing space target detection technology mostly focuses on the detection and identification of a special process, and the detection ability in the whole process of target close-in is not enough, which cannot meet the continuity requirement of space monitoring, especially the intelligent identification ability of special components is insufficient, which is still insufficient to support the current intelligent and refined identification requirements.

[0006] In summary, the existing space target detection technology still has great deficiencies in real-time detection and identification of close-in non-cooperative targets, especially in feature extraction of known structure information, unknown motion information, and non-information exchange space non-cooperative targets, which is difficult to adapt to close-in space target whole process locking and effective identification of special components. SUMMARY

[0007] In view of the deficiencies of the existing space target detection technology in real-time detection and identification of close-in non-cooperative targets, the present application provides a close-in space target detection and special component intelligent identification method and system, which can realize close-in space target whole process detection, meet the continuity requirement of space monitoring, and intelligently identify special components.

[0008] In a first aspect, the present application provides a close-in space target detection and special component intelligent identification method, comprising:

[0009] preprocessing the visible light image to obtain a binary image ;

[0010] extracting all suspected targets in the binary image to obtain an initial suspected target point set , calculating target feature values, sorting the suspected target points according to the target feature value fusion confidence, and obtaining a first suspected target point set ;

[0011] based on the star-sensitive pointing, removing the stars in the field of view corresponding to the first suspected target point set to obtain a second suspected target point set ;

[0012] using multi-frame motion information for track association, removing false alarms, and obtaining a target point through multi-frame confirmation ;

[0013] determining a plurality of close-in stages according to the area size of the target point , the close-in stages including a first close-in stage and a second close-in stage, directly outputting the centroid coordinates of the target in the first close-in stage , and identifying the special components of the target and outputting the centroid coordinates of the special components in the second close-in stage .

[0014] In a possible implementation manner of the first aspect, the preprocessing adopts a filtering algorithm to perform noise reduction processing on the visible light image , and adopts a global threshold segmentation algorithm to perform binaryzation processing on the filtered noise reduction image to obtain a binary image ; the specific steps are as follows:​​

[0015] computing a global brightness mean value of the visible light image ;

[0016] subtracting the global brightness mean value from each pixel point of the visible light image to obtain a mean-filtered denoised image , i and j represent the row and column respectively where the pixel point

[0017] setting a gray threshold value ; denotes an adjustment coefficient, denotes the global brightness mean value, denotes the global brightness standard deviation;

[0018] for each pixel point of the denoised image , comparing the gray value of the pixel point with the gray threshold value ; if the gray value of the pixel point is greater than , the pixel point is retained and the binarization result is set to 1, otherwise the pixel point is deleted and the binarization result is set to 0; after traversing all the pixel points, outputting all the points with the binarization result of 1 to obtain a binary image .

[0019] In a possible implementation manner of the first aspect, the extracting all the suspected targets in the binary image , calculating target feature values, and sorting the suspected targets according to the target feature values and the confidence level to obtain a first suspected target point set The specific steps are as follows:

[0020] extracting all the suspected targets in the binary image according to a connected domain extraction rule to obtain an initial suspected target point set ;

[0021] setting a confidence level , ; denotes the brightness value of the current target point, denotes the area value of the current target point, denotes the maximum brightness value among all the target points, denotes the maximum area value among all the target points, , denotes an adjustment coefficient;

[0022] ​​​​​​Based on confidence level For the initial set of suspected target points Sort all suspected target points in the range from highest to lowest to obtain the first set of suspected target points. .

[0023] In one possible implementation of the first aspect, the star-sensitive pointing pair is used to define the first set of suspected target points. Stars within the corresponding field of view are eliminated to obtain a second set of suspected target points. The specific steps are as follows:

[0024] Calculate the transformation matrix from star-sensitive coordinates to inertial coordinates. ;

[0025] According to the transformation matrix And star-aware visual axis pointing vector Calculate the camera line-of-sight pointing vector in the inertial coordinate system. :

[0026] ;

[0027] definition It is a unit vector, based on the camera's line-of-sight pointing vector. Calculate the transformation matrix from the camera coordinate system to the inertial coordinate system. :

[0028] ;

[0029] Filter the first set of suspected target points For all star points within the corresponding field of view, obtain the coordinates of the star points in the inertial coordinate system. ;

[0030] Based on the coordinates of the star point in the inertial coordinate system Transformation matrix from camera coordinate system to inertial coordinate system Calculate the coordinates of the star points in the camera coordinate system. :

[0031] ;

[0032] Calculate the coordinates of the star points in the image coordinate system based on the camera parameters. ;

[0033] The binary image f th First suspected target point set Coordinates of the star points in the image coordinate system By comparing the results and eliminating stars based on the nearest neighbor principle, a second set of suspected target points is obtained. .

[0034] In a possible implementation manner of the first aspect, the screening of the first set of suspected target points The screening criterion for all star points in the corresponding field of view range is:

[0035] The cosine value of the angle between the star point and the camera visual axis is calculated; if the cosine value of the angle between the star point and the camera visual axis is greater than or equal to the cosine value of half of the field of view angle, it is represented that the star point will appear in the field of view; if the cosine value of the angle between the star point and the camera visual axis is less than the cosine value of half of the field of view angle, it is represented that the star point will not appear in the field of view.

[0036] In a possible implementation manner of the first aspect, a conversion matrix from a star sensor coordinate system to an inertial coordinate system is calculated by using a quaternion calculated in real time by the star sensor :

[0037] ;

[0038] wherein, , , , represents a quaternion.

[0039] In a possible implementation manner of the first aspect, the track correlation is performed by using multi-frame motion information, false alarms are eliminated, and a target point is obtained through multi-frame confirmation The specific steps are as follows:

[0040] The track correlation is performed by using multi-frame motion information, and the standard deviation of the displacement of each suspected target point in a second set of suspected target points from the average value of the displacements of all suspected target points is calculated; the suspected target point with the maximum standard deviation is the fastest moving point, and the point is selected as the target point ;

[0041] After the target point is detected , the continuous multi-frame images are confirmed to determine whether the target point exists in the continuous multi-frame images ;

[0042] If the target point exists in the continuous multi-frame images , the target point is taken as the final target point; if the target point does not exist in the continuous multi-frame images , the visible light image is reacquired , and the above-mentioned preprocessing of the visible light image is repeated to obtain a binary image ; all suspected targets in the binary image are extracted to obtain an initial set of suspected target points , calculate a target feature value, sort the suspected target points according to the target feature value and the confidence level to obtain a first suspected target point set ; remove stars in a corresponding field of view range from the first suspected target point set to obtain a second suspected target point set ; perform track association using multi-frame motion information, remove false alarms, and obtain a target point through multi-frame confirmation ; and repeat the above steps until a final target point is found.

[0043] In a possible implementation manner of the first aspect, the target point is determined according to an area size of the target point , and the area size is used to determine a plurality of approaching stages, the approaching stages including a first approaching stage and a second approaching stage, in which the first approaching stage directly outputs a centroid coordinate of the target , and the second approaching stage identifies a special component of the target and outputs a centroid coordinate of the special component .

[0044] When the number of pixel points of the target point is less than or equal to a pixel point threshold, the approaching stage is determined as the first approaching stage; and when the number of pixel points of the target point is greater than the pixel point threshold, the approaching stage is determined as the second approaching stage .

[0045] In the first approaching stage, the centroid coordinate of the target point is directly outputted .

[0046] In the second approaching stage, a target detection algorithm is used to identify the special component of the target and output the centroid coordinate of the special component .

[0047] In a possible implementation manner of the first aspect, in the second approaching stage, a YOLO model is used to identify the special component of the target and output the centroid coordinate of the special component .

[0048] In a second aspect, the present application provides a system for performing the approaching space target detection and special component intelligent identification method, comprising:

[0049] An optoelectronic imaging unit is configured to acquire a visible light image .

[0050] An image preprocessing unit is configured to preprocess the visible light image and output a binary image .

[0051] A target detection and feature extraction unit is configured to extract the binary image All suspected targets, get the initial suspected target point set , calculate the target characteristic value, and sort the suspected target points according to the target characteristic value and the confidence level to obtain a first suspected target point set ; based on the star sensitive pointing, the first suspected target point set Corresponding to the field of view range of the star is removed to obtain a second suspected target point set ; adopt multi-frame motion information for track association, remove false alarm and obtain target point through multi-frame confirmation

[0052] Stage confirmation and special component identification unit, according to the area size of the target point Determine multiple approaching stages, including a first approaching stage and a second approaching stage, directly output the centroid coordinates of the target in the first approaching stage , and identify the special components of the target in the second approaching stage and output the centroid coordinates of the special components .

[0053] Compared with the prior art, the present application has the following beneficial effects:

[0054] 1. Compared with the traditional space target detection method which can only adapt to a special stage, the present application adopts a continuous perception mechanism based on hierarchical vision, which can realize both long-distance small target detection and near-distance special component detection and identification in different approaching stages. In the approaching process, the number of pixels of the target in the image is smooth and monotonically increasing with time, which realizes the continuous detection effect of the whole approaching process of the target;

[0055] 2. The present application constructs a multi-level, multi-dimensional filtering and confirmation system through star elimination and track association, gradually peels off the interference, and finally locks the real target, which can reliably detect and track weak space targets in the background of stars and noise; for special components, the present application adopts a target detection algorithm, which can accurately identify the special components of space non-cooperative targets with known structure information, unknown motion information and no information exchange, and the final output result has high confidence;

[0056] 3. The present application can realize the overall detection of space targets (such as failed satellites, space debris and unknown spacecraft) and the automatic identification and positioning of special components, provide key decision information for on-orbit service, space situation awareness and asset confirmation, and is an important supplement to the existing space target detection technology, which has practical engineering application value. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 The embodiment of the present application provides a flow chart of a close-in space target detection and special component intelligent identification method.

[0058] Figure 2 A starry sky background space visible light target simulation scene graph provided for an embodiment of the present application;

[0059] Figure 3 A binary image obtained after mean filtering and global threshold segmentation preprocessing of a visible light target simulation scene graph provided for an embodiment of the present application;

[0060] Figure 4 A flowchart for star rejection based on star sensitivity pointing provided by an embodiment of the present application;

[0061] Figure 5 A small target detection result graph in the long-distance stage of a close-in space target detection process provided for an embodiment of the present application;

[0062] Figure 6(a) is a detection result graph of the YOLOv8 recognizing the special component (visible light detector) in the first direction in the close-in stage of a close-in space target detection process provided for an embodiment of the present application;

[0063] Figure 6(b) is a detection result graph of the YOLOv8 recognizing the special component (visible light detector) in the second direction in the close-in stage of a close-in space target detection process provided for an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be described in further detail below with reference to the drawings.

[0065] It should be clear that 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 of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0066] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0067] In a first aspect, as Figure 1 indicated, the present application provides a close-in space target detection and special component intelligent identification method, which includes the following contents:

[0068] S1: Collecting a visible light image , preprocessing the visible light image to obtain a binary image ;

[0069] Exemplarily, the preprocessing adopts a filtering algorithm to perform noise reduction processing on the visible light image , and performs binaryzation processing on the filtered noise reduction image to obtain a binary image ; the filtering algorithm includes but is not limited to mean filtering, median filtering, gradient filtering, and curvature filtering.

[0070] In a preferred embodiment, step S1 adopts mean filtering to perform noise reduction processing on the visible light image , and further adopts a global threshold segmentation algorithm to perform binaryzation processing on the filtered noise reduction image; specifically including the following contents:

[0071] S11: calculating a global brightness mean value of the visible light image ;

[0072] S12: subtracting the global brightness mean value from each pixel point of the visible light image to obtain a mean filtered noise reduction image , i and j are indexes of each pixel position in the visible light image, representing the row (vertical position) and column (horizontal position) of the pixel point , respectively;

[0073] S13: setting a gray threshold value for threshold segmentation; wherein, represents an adjustment coefficient, and an experience value 0.1 can be selected, represents the global brightness mean value, represents the global brightness standard deviation;

[0074] S14: comparing the gray value of each pixel point of the noise reduction image with the gray threshold value ; if the gray value of the pixel point is greater than the gray threshold value , the pixel point is retained, and the binaryzation result is set to 1, otherwise the pixel point is deleted, and the binaryzation result is set to 0; after traversing all the pixel points, outputting all the points with the binaryzation result of 1 to obtain a binary image ;

[0075] Figure 2 Fig. 1 shows a starry sky background space visible light target simulation scene diagram provided by the embodiment of the present application, as Figure 3 is a binary image after mean filtering and global threshold segmentation preprocessing of the visible light image Figure 2 .

[0076] S2: Extract the binary image This involves identifying all suspected targets within the target area, resulting in an initial set of suspected target points. Calculate the target feature values, and sort the suspected target points according to the confidence scores of the target feature values ​​to obtain the first set of suspected target points. ;

[0077] In a preferred embodiment, step S2 specifically includes the following:

[0078] S21: Process the binary image according to connected component extraction rules, such as octal connected component extraction rules, quad connected component extraction rules, etc. An initial set of suspected target points is obtained by extracting all suspected targets within the area. ;

[0079] S22: Calculate the target feature values, including the brightness and area values ​​of the target point; set the confidence level. , ;in, This represents the brightness value of the current target point. This represents the area value of the current target point. This represents the maximum brightness value among all target points. This represents the largest area value among all target points. , Indicates the adjustment coefficient; preferably, The value is 0.5. The value is 0.5;

[0080] S23: Based on confidence level For the initial set of suspected target points Sort all suspected target points in the range from highest to lowest to obtain the first set of suspected target points. .

[0081] S3: Based on the star-sensitive pointing pair, the first set of suspected target points Stars within the corresponding field of view are quickly eliminated to obtain a second set of suspected target points. ;

[0082] like Figure 4 As shown, in a preferred embodiment, step S3 specifically includes the following:

[0083] S31: Calculate the transformation matrix from star-sensitive coordinates to inertial coordinates. ;

[0084] Optionally, the transformation matrix can be calculated using quaternions transmitted in real time via a star-sensor. The transformation matrix The expression is as follows:

[0085] ;

[0086] in, , , , Represent quaternions;

[0087] S32: Based on the transformation matrix And star-aware visual axis pointing vector Calculate the camera line-of-sight pointing vector in the inertial coordinate system. The camera's line-of-sight pointing vector The expression is as follows:

[0088] ;

[0089] S33: Definition It is a unit vector, based on the unit vector pointing to the camera's line of sight. Calculate the transformation matrix from the camera coordinate system to the inertial coordinate system. The transformation matrix The expression is as follows:

[0090] ;

[0091] S34: Filter the first set of suspected target points For all star points (including stars) within the field of view, the selection criterion is to determine whether the cosine of the angle between the star point and the camera's line of sight is greater than or equal to the cosine of half the field of view angle. The judgment formula is as follows:

[0092] ;

[0093] in, This represents the coordinates of the star point in the inertial coordinate system. Indicates the field of view. This represents the unit vector pointing to the camera's line of sight.

[0094] If the above condition is met, it means that the star will appear in the field of view; if it is not met, it means that the star will not appear in the field of view.

[0095] Obtain the coordinates of the star points in the inertial coordinate system under the current field of view. ;

[0096] S35: Based on the coordinates of the star point in the inertial coordinate system Transformation matrix from camera coordinate system to inertial coordinate system Calculate the coordinates of the star points in the camera coordinate system. The The expression is as follows:

[0097] ;

[0098] S36: According to the camera parameters, the coordinates of the star points in the image coordinate system are calculated ;

[0099] S37: The binary image The first set of suspected target points The coordinates of the star points in the image coordinate system are compared, and the stars are removed according to the nearest neighbor principle to obtain the second set of suspected target points .

[0100] S4: Track association is performed using multi-frame motion information, false alarms are removed, and target points are obtained through multi-frame confirmation ;

[0101] In a preferred embodiment, step S4 specifically includes the following contents:

[0102] S41: Track association is performed using multi-frame motion information, and the standard deviation of the displacement of each suspected target point in the second set of suspected target points from the average displacement of all suspected target points is calculated. The suspected target point with the largest standard deviation is the fastest moving point, and this point is selected as the target point ;

[0103] S42: After detecting the target point , the consecutive n frames of images are confirmed to determine whether the target point exists in all the consecutive n frames of images ;

[0104] S43: If the target point exists in all the consecutive n frames of images , it is taken as the final target point; if the target point does not exist in all the consecutive n frames of images , return to step S1 to reacquire the visible light image and detect the image; the frame number n is generally selected to be 3-5.

[0105] S5: According to the area size of the target point , a plurality of approaching stages are determined, the approaching stages include a first approaching stage and a second approaching stage, in the first approaching stage, the centroid coordinates of the target are directly output , in the second approaching stage, the special components of the target are identified and the centroid coordinates of the special components are output ;

[0106] In a preferred embodiment, step S5 specifically includes the following contents:

[0107] S51: Set a pixel point threshold N, when the target point ​​the number of pixel points of the target point is less than or equal to a pixel point threshold N, it is determined that the approaching stage is a first approaching stage, i.e. a long-distance stage; when the number of pixel points of the target point is greater than the pixel point threshold N, it is determined that the approaching stage is a second approaching stage, i.e. a close-distance stage

[0108] S52: directly outputting the centroid coordinates of the target point in the first approaching stage ;

[0109] S53: in the second approaching stage, a target detection algorithm is used to identify the target and output the centroid coordinates of the special component ; preferably, the target detection algorithm is a convolutional neural network module, such as a YOLO model.

[0110] As shown in the drawings, Figure 5 is a small target detection result graph in the long-distance stage of the embodiment, Figure 6(a) , 6(b) are two detection result graphs of different orientations of the special component (visible light detector in this embodiment) of the small target in the close-distance stage; the parameter symbol of the confidence level defined in this embodiment is LEN, and LEN0.91 and LEN0.91 in the graph represent the confidence level indication.

[0111] The results of the above embodiments show that, compared with the traditional space target detection method which can only adapt to a certain special stage, the present application can realize continuous detection of the whole approaching process of the target, accurately lock the real target, and effectively identify the special component of the space non-cooperative target with known structure information, unknown motion information and no information exchange, and the output result has high confidence level.

[0112] In a second aspect, the present application provides a system for performing the approaching space target detection and special component intelligent identification method, comprising:

[0113] an optoelectronic imaging unit for acquiring a visible light image ;

[0114] an image preprocessing unit for preprocessing the visible light image and outputting a binary image ;

[0115] a target detection and feature extraction unit for extracting all suspected targets in the binary image to obtain an initial suspected target point set , calculating a target feature value, sorting the suspected target points according to the target feature value fusion confidence level to obtain a first suspected target point set ; based on the star sensor pointing, the stars in the field of view are quickly removed to obtain a second suspected target point set​ Multi-frame motion information is used for track association, false alarms are eliminated, and the target point is obtained through multi-frame confirmation. ;

[0116] Phase confirmation and special component identification unit, based on target point The area size determines multiple approach stages, including a first approach stage and a second approach stage. In the first approach stage, the centroid coordinates of the target are directly output. During the second approach phase, specific components of the target are identified and the centroid coordinates of these components are output. .

[0117] It should be understood that the division of processing units in the above system is only a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, the processing units in the system can be implemented by a processor calling software; for example, the system includes a processor connected to memory, which stores instructions. The processor calls the instructions stored in memory to implement any of the above methods or to realize the functions of each processing unit in the system. Here, the processor is a general-purpose processor, such as a central processing unit or a microprocessor, and the memory is either internal or external to the system.

[0118] The above embodiments are only for illustrating the technical concept and features of this application, and are intended to enable those skilled in the art to understand the content of this application and implement it accordingly. They should not be used to limit the scope of protection of this application. For those skilled in the art, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for close-range space target detection and intelligent identification of special components, characterized in that, include: Visible light images Preprocessing is performed to obtain a binary image. ; Extracting binary images This involves identifying all suspected targets within the target area, resulting in an initial set of suspected target points. Calculate the target feature value, and sort the suspected target points according to the confidence score based on the target feature value to obtain the first set of suspected target points. ; Based on the star-pointing pair of the first suspected target point set Stars within the corresponding field of view are eliminated to obtain a second set of suspected target points. ; The trajectory is correlated using multi-frame motion information, false alarms are eliminated, and the target point is obtained through multi-frame confirmation. ; According to the target point The area size determines multiple approach stages, including a first approach stage and a second approach stage. In the first approach stage, the centroid coordinates of the target are directly output. During the second approach phase, specific components of the target are identified and the centroid coordinates of these components are output. .

2. The method for close-range space target detection and intelligent identification of special components according to claim 1, characterized in that, The preprocessing uses a filtering algorithm on the visible light image. Noise reduction is performed, and a global threshold segmentation algorithm is used to binarize the filtered, denoised image to obtain a binary image. The specific steps are as follows: Calculate visible light images global average brightness ; Visible light images Each pixel Subtract the global average brightness Obtain the noise-reduced image after mean filtering i and j represent pixels respectively. The row and column it belongs to; Set grayscale threshold ; This represents the adjustment coefficient. This represents the global average brightness. Indicates the global brightness standard deviation; For noise reduction images Each pixel Compare its grayscale value with the grayscale threshold. Size; if pixels The gray value is greater than If the pixel is true, retain it and set the binarization result to 1; otherwise, delete the pixel and set the binarization result to 0. After traversing all pixels, output all points whose binarization result is 1 to obtain the binary image. .

3. The method for close-range space target detection and intelligent identification of special components according to claim 1, characterized in that, Extracting binary images This involves identifying all suspected targets within the target area, resulting in an initial set of suspected target points. Calculate the target feature value, and sort the suspected target points according to the confidence score based on the target feature value to obtain the first set of suspected target points. The specific steps are as follows: Binary images are processed according to connected component extraction rules. An initial set of suspected target points is obtained by extracting all suspected targets within the area. ; Set confidence level , ; This represents the brightness value of the current target point. This represents the area value of the current target point. This represents the maximum brightness value among all target points. This represents the largest area value among all target points. , Indicates the adjustment coefficient; Based on confidence level For the initial set of suspected target points Sort all suspected target points in the range from highest to lowest to obtain the first set of suspected target points. .

4. The method for close-range space target detection and intelligent identification of special components according to claim 1, characterized in that, The first set of suspected target points based on star-sensor pointing Stars within the corresponding field of view are eliminated to obtain a second set of suspected target points. The specific steps are as follows: Calculate the transformation matrix from star-sensitive coordinates to inertial coordinates. ; According to the transformation matrix And star-aware visual axis pointing vector Calculate the camera line-of-sight pointing vector in the inertial coordinate system. : ； definition It is a unit vector, based on the camera's line-of-sight pointing vector. Calculate the transformation matrix from the camera coordinate system to the inertial coordinate system. : ； Filter the first set of suspected target points For all star points within the corresponding field of view, obtain the coordinates of the star points in the inertial coordinate system. ; Based on the coordinates of the star point in the inertial coordinate system Transformation matrix from camera coordinate system to inertial coordinate system Calculate the coordinates of the star points in the camera coordinate system. : ； Calculate the coordinates of the star points in the image coordinate system based on the camera parameters. ; Binary image First suspected target point set Coordinates of the star points in the image coordinate system By comparing the results and eliminating stars based on the nearest neighbor principle, a second set of suspected target points is obtained. .

5. The method for close-range space target detection and intelligent identification of special components according to claim 4, characterized in that, The first set of suspected target points was selected. The selection criteria for all stars within the corresponding field of view are: Calculate the cosine of the angle between the star point and the camera's line of sight. If the cosine of the angle between the star point and the camera's line of sight is greater than or equal to the cosine of half the field of view, the star point will appear in the field of view. If the cosine of the angle between the star point and the camera's line of sight is less than the cosine of half the field of view, the star point will not appear in the field of view.

6. The method for close-range space target detection and intelligent identification of special components according to claim 4, characterized in that, The transformation matrix from the star-sensor coordinate system to the inertial coordinate system is calculated using quaternions for real-time star-sensor transmission. : ； in, , , , This represents a quaternion.

7. The method for close-range space target detection and intelligent identification of special components according to claim 1, characterized in that, The process involves using multi-frame motion information for trajectory association, eliminating false alarms, and obtaining the target point through multi-frame confirmation. The specific steps are as follows: Multi-frame motion information is used to correlate flight paths and calculate the second set of suspected target points. The standard deviation of the displacement of each suspected target point within the range and the average displacement of all suspected target points is used to determine the fastest-moving suspected target point, which is then selected as the target point. ; Target point detected Then, multiple consecutive frames of images are examined to confirm whether the target point exists in all of them. ; If the target point exists in multiple consecutive frames of images If so, then take it as the final target point; If the target point does not exist in all consecutive frames of images Then re-acquire visible light images. And repeat the above for visible light images. Preprocessing is performed to obtain a binary image. Extracting binary images This involves identifying all suspected targets within the target area, resulting in an initial set of suspected target points. Calculate the target feature value, and sort the suspected target points according to the confidence score based on the target feature value to obtain the first set of suspected target points. Based on star-aware pointing to the first set of suspected target points Stars within the corresponding field of view are eliminated to obtain a second set of suspected target points. Multi-frame motion information is used for track association, false alarms are eliminated, and the target point is obtained through multi-frame confirmation. The steps continue until the final target point is found.

8. The method for close-range space target detection and intelligent identification of special components according to claim 1, characterized in that, According to the target point The area size determines multiple approach stages, including a first approach stage and a second approach stage. In the first approach stage, the centroid coordinates of the target are directly output. During the second approach phase, specific components of the target are identified and the centroid coordinates of these components are output. The specific steps are as follows: Set a threshold for the number of pixels, when the target point When the number of pixels is less than or equal to the pixel count threshold, the approach phase is determined as the first approach phase; when the target point When the number of pixels is greater than the pixel count threshold, the approaching phase is determined to be the second approaching phase; Output the centroid coordinates of the target point directly during the first approach phase. ; In the second approach phase, a target detection algorithm is used to identify special components of the target and output the centroid coordinates of these special components. .

9. The method for close-range space target detection and intelligent identification of special components according to claim 8, characterized in that, In the second approach phase, the YOLO model is used to identify specific components of the target and output the centroid coordinates of these components. .

10. A system for performing the close-range space target detection and intelligent identification method for special components as described in any one of claims 1-9, characterized in that, include: Photoelectric imaging unit, used to acquire visible light images ; Image preprocessing unit, used for processing the visible light image Preprocessing is performed to output a binary image. ; The object detection and feature extraction unit extracts binary images. This involves identifying all suspected targets within the target area, resulting in an initial set of suspected target points. Calculate the target feature value, and sort the suspected target points according to the confidence score based on the target feature value to obtain the first set of suspected target points. Based on star-aware pointing to the first set of suspected target points Stars within the corresponding field of view are eliminated to obtain a second set of suspected target points. Multi-frame motion information is used for track association, false alarms are eliminated, and the target point is obtained through multi-frame confirmation. ; Phase confirmation and special component identification unit, based on target point The area size determines multiple approach stages, including a first approach stage and a second approach stage. In the first approach stage, the centroid coordinates of the target are directly output. During the second approach phase, specific components of the target are identified and the centroid coordinates of these components are output. .

Images