Spatial optical target time sequence feature extraction method and system

Abstract

The invention provides a space optical target time sequence feature extraction method and system, and the method comprises the following steps: 1, video collection: obtaining a target video shot by an optical detection device in real time through a computer and a network; step 2, video frame sequence conversion: the collected real-time tracking video stream is segmented into image frame sequences according to a frame rate; step 3, automatically detecting and positioning the position of an interested space optical target frame by frame, and obtaining a position square frame of the target in an image frame; 4, based on the target position square frame, extracting a target effective pixel region corresponding to the target actual imaging region, and calculating the area and gray value information of the target effective pixel region; 5, calculating a minimum bounding rectangular frame of the target based on the effective pixel area of the target, and calculating the position and scale information of the rectangular frame; step 6, generating a target time sequence characteristic curve graph according to the characteristic information such as the position, the posture, the scale and the gray scale, realizing visualization of the target time sequence characteristics, and analyzing a rule of the target characteristics changing along with time; the method has the advantages of automation, comprehensive feature extraction and visualization.

Description

Technical Field

[0001] The present invention relates to the field of target time series feature extraction, and in particular to a method and system for extracting time series features of a space optical target. Background Art

[0002] Optical detection equipment can track and detect targets in real time and obtain live videos of target movement. It has the characteristics of intuitive measurement results, can obtain target morphological characteristics, and is not easily interfered by electromagnetic waves. It is widely used in space situational awareness, space flight measurement and control, and scientific research. Optical detection equipment can obtain sequential images of the detected target. Each frame of the image contains characteristic information such as the position, shape, size, and brightness of the target. This information is of great significance in application scenarios such as sequential detection, tracking and identification of targets, analysis and prediction of target behavior, and visualization of target time series characteristics.

[0003] The automatic feature extraction method for spatial optical targets in videos requires first detecting and locating the target, and then calculating its feature information. At present, the extraction method for temporal features of spatial optical targets mainly relies on post-processing. The detection and positioning method has poor performance, the extracted features are single, and it is difficult to efficiently use the sequence video information. The main problems are as follows: First, the automatic extraction algorithm has poor robustness and is easily affected by factors such as background and target brightness changes and target motion, resulting in frame leakage; second, the extracted features are relatively few, generally only position and size information, lacking posture, area, and brightness features; third, the visualization effect of the extracted temporal features needs to be improved. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method and system for extracting time series features of space optical targets, which has the advantages of automation, comprehensive feature extraction and visualization. The specific technical solution is as follows:

[0005] A method for extracting time series features of an optical target comprises the following steps:

[0006] Step 1, video acquisition, real-time acquisition of target video captured by optical detection equipment through computers and networks;

[0007] Step 2: converting video into frame sequence, dividing the collected real-time tracking video stream into image frame sequence according to the frame rate;

[0008] Step 3: optical target detection and positioning: automatically detecting and positioning the optical target position of interest frame by frame, and obtaining the target position box of the target in the image frame;

[0009] Step 4, extracting the target effective pixel area, based on the target position box, extracting the target effective pixel area corresponding to the target actual imaging area, and calculating the area and gray value information of the target effective pixel area;

[0010] Step 5, calculating the minimum bounding rectangle of the target, calculating the minimum bounding rectangle of the target based on the target effective pixel area, and calculating the position and scale information of the rectangle;

[0011] Step 6, generate a multi-dimensional time series feature map, generate a target time series feature curve map with feature information such as position, posture, scale and grayscale, realize the visualization of target time series features, and analyze the law of target feature changes over time.

[0012] Furthermore, in step 3, the position of the spatial optical target of interest is automatically detected and located frame by frame, and the position box of the target in the image frame is obtained, which is completed by the YOLO target detector trained by the training test data set, specifically including:

[0013] First, we manually obtain annotated datasets using the same type of target video sequence images and divide them into training datasets and test datasets.

[0014] Use the YOLOv5 model for training in the PyTorch framework, and set the hyperparameters to: select 8 for the number of batch images (batchsize), 192×320 for the training image size, and "YOLOv5m.weights" for the pre-trained weights. Then modify the network structure parameters in the cfg network structure configuration file of the model. The trained network structure parameters are saved as the "yolo5m.pt" weight configuration file, which can be regarded as an optical video target detector. After loading it into the YOLO model framework, video targets can be detected in real time.

[0015] The i-th image frame i The pixel size is width w and height h. The target position box obtained after detection is the rectangular image block Rect i =(x i ,y i , w i ,h i ), 1≤i≤N, where x i / y i 、w i / h i They represent the horizontal / vertical coordinates of the upper left corner of the target position box and the box width / height respectively, and the units are percentage values ​​relative to the pixel size w / h of the image frame.

[0016] Further, step 4, extracting the target effective pixel area, based on the rectangular image block in the target position box, extracting the target effective pixel area corresponding to the target actual imaging area, and calculating the area and gray value information of the target effective pixel area, specifically includes:

[0017] The target position box based on the i-th frame image, that is, the rectangular image block Rect i, the image block has a total of n i pixels, the gray value of the jth pixel is g j ∈[0, 255], first calculate its grayscale mean

[0018] Then th i =rm i +60 is the threshold value, and the pixels with grayscale values ​​below the threshold are set to 0. The retained pixel area is the target effective imaging pixel area Field i , a total of s i pixels, i.e., the target area, where the gray value of the jth pixel is g j ∈[0, 255], and then calculate the average gray value of the area and the maximum gray value fm i =max(g j ), j∈Field i .

[0019] Further, step 5, calculating the minimum bounding rectangle of the target, searching for the minimum bounding rectangle of the target based on the target effective pixel area, and calculating the position and scale information of the rectangular box, specifically includes the following steps:

[0020] First, according to the target effective pixel area Field i Calculate its minimum enclosing rectangle MORect i (cx i ,cy i , cw i ,ch i ,θ i ), where cx i ,cy i , cw i ,ch i ,θ i They represent the horizontal coordinate, vertical coordinate, width, height and tilt angle of the center pixel of the circumscribed rectangular frame, respectively, and can reflect the position, scale and posture characteristics of the target. i ,cy i The unit is the percentage of the image frame pixel size w / h, cw i ,ch i The unit is the number of pixels, which directly reflects the absolute size of the target. i It is the angle relative to the horizontal x-axis of the rectangular coordinate system. A positive value indicates the angle with the positive x-axis, and a negative value indicates the angle with the negative x-axis.

[0021] Furthermore, in step 6, a multi-dimensional time series feature graph is generated, and the feature information such as position, posture, scale and grayscale is used to generate a target time series feature curve graph to realize the visualization of the target time series features and analyze the law of the target features changing over time. The specific steps include:

[0022] The nine physical quantities of the target in all frame images calculated in the above steps are divided into four types of time series features, namely position features (including relative miss distance cx i 、cy i ), posture characteristics (inclination angle θ i ), size characteristics (length of the target i 、Width cw i and area s i ) and grayscale features (the mean grayscale value of the target block rm i , target effective grayscale mean fa i and the target effective grayscale maximum value fm i ), and then divided into three sub-graphs to draw the time series change curves of these four types of features. The target time series feature curve cluster can intuitively reflect the overall trend and characteristics of the target changing over time.

[0023] The present invention also provides a spatial optical target temporal feature extraction system, comprising:

[0024] The video acquisition module acquires the target video shot by the optical detection device in real time through the computer and the network;

[0025] The video frame sequence module divides the collected real-time tracking video stream into image frame sequences according to the frame rate;

[0026] The target detection and positioning module automatically detects and locates the target position of interest and obtains the target position rectangular box in the image frame;

[0027] The target effective pixel detection module extracts the target effective pixel area corresponding to the target imaging area based on the target position box, and calculates the area and gray value information of the target effective pixel area;

[0028] The target minimum bounding rectangle detection module calculates the target's minimum bounding rectangle based on the target's effective pixel area, and calculates the position and scale information of the rectangle;

[0029] The target time series feature visualization module generates a target time series feature map based on feature information such as position, posture, scale and grayscale, realizes the visualization of target time series features, and analyzes the law of target feature changes over time.

[0030] The present invention also 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 steps of the above method when executing the computer program.

[0031] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0032] The present invention also provides a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0033] The beneficial technical effects of the present invention are:

[0034] A spatial optical target temporal feature extraction method and system of the present invention can automatically detect the position of the spatial optical target in the video frame by frame, and has good robustness to background changes, target size and brightness changes and noise;

[0035] A method and system for extracting time series features of a space optical target of the present invention can automatically extract time series features of the position, posture, scale and grayscale of the space optical target, and can more comprehensively characterize the characteristics of the space optical target changing over time;

[0036] The present invention provides a method and system for extracting time series features of a space optical target, which uses multiple sub-graphs and rich curve forms to combine and visualize nine time series feature variables and their corresponding relationships with key frames, and has comprehensive and intuitive characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flow chart of temporal feature extraction of a space optical target according to the present invention;

[0038] Figure 2 It is a schematic diagram of generating a comprehensive curve of multi-dimensional time series characteristics of a spatial optical target according to the present invention. DETAILED DESCRIPTION

[0039] In order to make the technical personnel in this technical field better understand the technical solution in this specification, the following will be combined with the appendix of the specification. Figure 1-2 The technical solution of the present invention is described in further detail.

[0040] The present invention provides a method for automatically extracting time series features of a space optical target, comprising the following steps:

[0041] Step 1, video acquisition, real-time acquisition of target video captured by optical detection equipment through computers and networks;

[0042] Step 2: converting video into frame sequence, dividing the collected real-time tracking video stream into image frame sequence according to the frame rate;

[0043] Step 3: spatial optical target detection and positioning, automatically detecting and positioning the spatial optical target position of interest frame by frame, and obtaining the target position rectangular box of the target in the image frame;

[0044] Firstly, the target video sequence of the same type is segmented into several image frames, and a labeled dataset is obtained by manual labeling method, which is divided into training dataset and test dataset in a ratio of 3:1.

[0045] Use the YOLOv5 model for training in the PyTorch framework, and set the hyperparameters to: select 8 for the number of batch images (batchsize), 192×320 for the training image size, and "YOLOv5m.weights" for the pre-trained weights. Then modify the network structure parameters in the cfg network structure configuration file of the model, change the number of categories to 1, and save the trained network structure parameters as the "yolo5m.pt" weight configuration file, which can be regarded as an optical video target detector. After loading it into the YOLO model framework, video targets can be detected in real time.

[0046] The detector reads the image frame by frame, and the i-th image frame Frame i The pixel size is width w and height h, and the target position box obtained after detection, that is, the rectangular image block is Rect i =(x i ,y i , w i ,hi),0≤i≤N-1,where x i / y i 、w i / h i They represent the horizontal / vertical coordinates of the upper left corner of the target position box and the box width / height respectively, and the units are percentage values ​​relative to the pixel size w / h of the image frame.

[0047] Step 4, extracting the target effective pixel area, based on the target position box, extracting the target effective pixel area corresponding to the target actual imaging area, and calculating the area and gray value information of the target effective pixel area;

[0048] The target position box Rect of the i-th frame image obtained based on the detector i , count the total number of image blocks is n i pixels, the gray value of the jth pixel is g j ∈[0, 255], first calculate its grayscale mean

[0049] Then th i=rm i +60 is the threshold value, and the target position rectangle Rect i Pixels with grayscale values ​​below the threshold are set to 0, and the retained pixel area is the target effective imaging pixel area Field i , count the total number of pixels in this area i pixels, i.e., the target area, where the gray value of the jth pixel is g j ∈[0, 255], and then calculate the average gray value of the area and the maximum gray value fm i =max(g j ), j∈Field i .

[0050] Step 5, calculating the minimum bounding rectangle of the target, calculating the minimum bounding rectangle of the target based on the target effective pixel area, and calculating the position and scale information of the rectangle;

[0051] According to the target effective pixel area Field i , for the target location box Rect i Perform binarization, that is, in Rect i In the effective pixel area Field i The corresponding pixels are all set to 255, and the remaining pixels are set to 0. Based on this, the search and calculation of the valid pixel area Field i The minimum bounding rectangle MORect i (cx i ,cy i , cw i ,ch i ,θ i ), where cx i ,cy i , cw i ,ch i ,θ i They represent the horizontal coordinate, vertical coordinate, width, height and tilt angle of the center pixel of the external rectangular frame, which can reflect the position, scale and posture characteristics of the target. i ,cy i The unit is the percentage of the image frame pixel size w / h, cw i ,ch i The unit is the number of pixels, which directly reflects the absolute size of the target. i It is the angle relative to the horizontal x-axis of the rectangular coordinate system. A positive value indicates the angle with the positive x-axis, and a negative value indicates the angle with the negative x-axis.

[0052] Step 6, generate a multi-dimensional time series feature map, generate a target time series feature curve map with feature information such as position, posture, scale and grayscale, realize the visualization of target time series features, and analyze the law of target feature changes over time.

[0053] The nine physical quantities of the target in all frame images calculated in the above steps are divided into four types of time series features, namely position features (including relative miss distance cx i 、cy i ), posture characteristics (inclination angle θ i ), size characteristics (length of the target i 、Width cw i and area s i ) and grayscale features (the mean grayscale value of the target block rm i , target effective grayscale mean fa i and the target effective grayscale maximum value fm i ), and then divided into three sub-graphs to draw the time series change curves of these four types of features. The automatically extracted target time series feature curve cluster can intuitively reflect the overall trend and characteristics of the target changing over time. Figure 2 Schematic diagram for generating a comprehensive curve of multi-dimensional timing characteristics of a space optical target.

[0054] The present invention also provides a spatial optical target temporal feature extraction system, comprising:

[0055] The video acquisition module acquires the target video shot by the optical detection device in real time through the computer and the network;

[0056] The video frame sequence module divides the collected real-time tracking video stream into image frame sequences according to the frame rate;

[0057] The target detection and positioning module automatically detects and locates the position of the spatial optical target of interest frame by frame, and obtains the target position rectangular box in the image frame;

[0058] The target effective pixel region extraction module extracts the target effective pixel region corresponding to the target actual imaging region based on the target position box, and calculates the area and gray value information of the target effective pixel region;

[0059] The target minimum bounding rectangle calculation module calculates the minimum bounding rectangle of the target based on the target effective pixel area, and calculates the position and scale information of the rectangle;

[0060] The target time series feature visualization module generates a target time series feature map based on feature information such as position, posture, scale and grayscale, realizes the visualization of target time series features, and analyzes the law of target feature changes over time.

[0061] Furthermore, the target detection and positioning module automatically detects and positions the spatial optical target position of interest frame by frame, and obtains the target position rectangular box of the target in the image frame, which specifically includes the following steps:

[0062] Firstly, the target video sequence of the same type is segmented into several image frames, and a labeled dataset is obtained by manual labeling method, which is divided into training dataset and test dataset in a ratio of 3:1.

[0063] Use the YOLOv5 model for training in the PyTorch framework, and set the hyperparameters to: select 8 for the number of batch images (batchsize), 192×320 for the training image size, and "YOLOv5m.weights" for the pre-trained weights. Then modify the network structure parameters in the cfg network structure configuration file of the model, change the number of categories to 1, and save the trained network structure parameters as the "yolo5m.pt" weight configuration file, which can be regarded as an optical video target detector. After loading it into the YOLO model framework, video targets can be detected in real time.

[0064] The detector reads the image frame by frame, and the i-th image frame Frame i The pixel size is width w and height h, and the target position box obtained after detection, that is, the rectangular image block is Rect i =(x i ,y i , w i ,hi),0≤i≤N-1,where x i / y i 、w i / h i They represent the horizontal / vertical coordinates of the upper left corner of the target position box and the box width / height respectively, and the units are percentage values ​​relative to the pixel size w / h of the image frame.

[0065] Furthermore, the target effective pixel area extraction module extracts the target effective pixel area corresponding to the target actual imaging area based on the target position box, and calculates the area and gray value information of the target effective pixel area, which specifically includes the following steps:

[0066] The target position box Rect of the i-th frame image obtained based on the detector i , count the total number of image blocks is n i pixels, the gray value of the jth pixel is g j ∈[0, 255], first calculate its grayscale mean

[0067] Then th i =rm i +60 is the threshold value, and the target position rectangle Recti Pixels with grayscale values ​​below the threshold are set to 0, and the retained pixel area is the target effective imaging pixel area Field i , count the total number of pixels in this area i pixels, i.e., the target area, where the gray value of the jth pixel is g j ∈[0, 255], and then calculate the average gray value of the area and the maximum gray value fm i =max(g j ), j∈Field i .

[0068] Furthermore, the target minimum bounding rectangle calculation module calculates the minimum bounding rectangle of the target based on the target effective pixel area, and calculates the position and scale information of the rectangular frame, which specifically includes the following steps:

[0069] According to the target effective pixel area Field i , for the target location box Rect i Perform binarization, that is, in Rect i In the effective pixel area Field i The corresponding pixels are all set to 255, and the remaining pixels are set to 0. Based on this, the search and calculation of the valid pixel area Field i The minimum bounding rectangle MORect i (cx i ,cy i , cw i ,ch i ,θ i ), where cx i ,cy i , cw i ,ch i ,θ i They represent the horizontal coordinate, vertical coordinate, width, height and tilt angle of the center pixel of the external rectangular frame, which can reflect the position, scale and posture characteristics of the target. i ,cy i The unit is the percentage of the image frame pixel size w / h, cw i ,ch i The unit is the number of pixels, which directly reflects the absolute size of the target. i It is the angle relative to the horizontal x-axis of the rectangular coordinate system. A positive value indicates the angle with the positive x-axis, and a negative value indicates the angle with the negative x-axis.

[0070] Furthermore, the target time series feature visualization module generates a target time series feature map based on the feature information of position, posture, scale and grayscale, realizes the visualization of the target time series features, and analyzes the law of target feature changes over time. Specifically, the following steps are included:

[0071] The nine physical quantities of the target in all frame images calculated in the above steps are divided into four types of time series features, namely position features (including relative miss distance cx i 、cy i ), posture characteristics (inclination angle θ i ), size characteristics (length of the target i 、Width cw i and area s i ) and grayscale features (the mean grayscale value of the target block rm i , target effective grayscale mean fa i and the target effective grayscale maximum value fm i ), and then divided into three sub-graphs to draw the time series change curves of these four types of features. The automatically extracted target time series feature curve cluster can intuitively reflect the overall trend and characteristics of the target changing over time. Figure 2 Schematic diagram for generating a comprehensive curve of multi-dimensional timing characteristics of a space optical target.

[0072] The present invention also 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 steps of the above method when executing the computer program.

[0073] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0074] The present invention also provides a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present application. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for extracting temporal features of a spatial optical target, characterized in that: The steps include: Step 1, video acquisition, real-time acquisition of target video captured by optical detection equipment through computers and networks; Step 2: converting video into frame sequence, dividing the collected real-time tracking video stream into image frame sequence according to the frame rate; Step 3: spatial optical target detection and positioning, automatically detecting and positioning the spatial optical target position of interest frame by frame, and obtaining the target position rectangular box of the target in the image frame; Step 4, extracting the target effective pixel area, based on the target position box, extracting the target effective pixel area corresponding to the target actual imaging area, and calculating the area and gray value information of the target effective pixel area; Step 5, calculating the minimum bounding rectangle of the target, calculating the minimum bounding rectangle of the target based on the target effective pixel area, and calculating the position and scale information of the rectangle; Step 6, generate a multi-dimensional time series feature map, generate a target time series feature curve map with feature information such as position, posture, scale and grayscale, realize the visualization of target time series features, and analyze the law of target feature changes over time.

2. A method for extracting temporal features of a spatial optical target according to claim 1, characterized in that: In step 3, the spatial optical target detection and positioning automatically detects and locates the spatial optical target position of interest frame by frame, and obtains the target position box of the target in the image frame. This is completed by the YOLO target detector trained by the training and test data set, which specifically includes: First, we manually obtain annotated datasets using the same type of target video sequence images and divide them into training datasets and test datasets. Use the YOLOv5 model for training in the PyTorch framework, and set the hyperparameters to: select 8 for the number of batch images, 192×320 for the training image size, and "YOLOv5m.weights" for the pre-trained weights. Then modify the network structure parameters in the cfg network structure configuration file of the model. The trained network structure parameters are saved as the "yolo5m.pt" weight configuration file, which can be regarded as an optical video target detector. After loading it into the YOLO model framework, video targets can be detected in real time. The i-th image frame i The pixel size is width w and height h. The target position box obtained after detection is the rectangular image block Rect i =(x i ,y i ,w i ,h i ), 1≤i≤N, where x i / y i 、w i / h i They represent the upper left corner horizontal coordinate / vertical coordinate and the width / height of the target position box respectively, and the unit is the percentage value relative to the pixel size w / h of the image frame.

3. The method for extracting temporal features of a spatial optical target according to claim 1, characterized in that: Step 4, extracting the target effective pixel area, based on the rectangular image block in the target position box, extracting the target effective pixel area corresponding to the target actual imaging area, and calculating the area and gray value information of the target effective pixel area, specifically including: The target position box based on the i-th frame image, that is, the rectangular image block Rect i , the image block has a total of n i pixels, the gray value of the jth pixel is g j ∈[0,255], first calculate its grayscale mean Then th i =rm i +60 is the threshold value, and the pixels with grayscale values ​​below the threshold are set to 0. The retained pixel area is the target effective imaging pixel area Field i , total s i pixels, i.e., the target area, where the gray value of the jth pixel is g j ∈[0,255], then calculate the average gray value of the area and the maximum gray value fm i =max(g j ),j∈Field i .

4. The method for extracting temporal features of a spatial optical target according to claim 1, characterized in that: Step 5, calculating the minimum bounding rectangle of the target, searching for the minimum bounding rectangle of the target based on the target effective pixel area, and calculating the position and scale information of the rectangle, specifically including the following steps: First, according to the target effective pixel area Field i Calculate its minimum enclosing rectangle MORect i (cx i ,cy i ,cw i ,ch i ,θ i ), where cx i ,cy i ,cw i ,ch i ,θ i They represent the horizontal coordinate, vertical coordinate, width, height and tilt angle of the center pixel of the circumscribed rectangular frame, respectively, and can reflect the position, scale and posture characteristics of the target. i ,cy i The unit is the percentage of the image frame pixel size w / h, cw i ,ch i The unit is the number of pixels, which directly reflects the absolute size of the target. i It is the angle relative to the horizontal x-axis of the rectangular coordinate system. A positive value indicates the angle with the positive x-axis, and a negative value indicates the angle with the negative x-axis.

5. The method for extracting temporal features of a spatial optical target according to claim 1, characterized in that: Step 6: Generate a multi-dimensional time series feature graph, generate a target time series feature curve graph based on the feature information of position, posture, scale and grayscale, realize the visualization of the target time series features, and analyze the law of target feature changes over time. The specific steps include: The nine physical quantities of the target in all frame images calculated in the above steps are divided into four types of time series features, namely, position features including relative miss distance cx i 、cy i , the posture features include the inclination angle θ i , the size features include the length of the target i 、Width cw i and area s i , and grayscale features include the target block grayscale mean rm i , target effective grayscale mean fa i and the target effective grayscale maximum value fm i , and then divided into three sub-graphs to draw the time series change curves of these four types of features. The target time series feature curve cluster can intuitively reflect the overall trend and characteristics of the target changing over time.

6. A spatial optical target temporal feature extraction system, characterized in that: include: The video acquisition module acquires the target video shot by the optical detection device in real time through the computer and the network; The video frame sequence module divides the collected real-time tracking video stream into image frame sequences according to the frame rate; The target detection and positioning module automatically detects and locates the position of the spatial optical target of interest frame by frame, and obtains the target position rectangular box in the image frame; The target effective pixel region extraction module extracts the target effective pixel region corresponding to the target actual imaging region based on the target position box, and calculates the area and gray value information of the target effective pixel region; The target minimum bounding rectangle calculation module calculates the target's minimum bounding rectangle based on the target's effective pixel area, and calculates the position and scale information of the rectangle; The target time series feature visualization module generates a target time series feature map based on the feature information such as position, posture, scale and grayscale, realizes the visualization of the target time series features, and analyzes the law of target feature changes over time.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method described in any one of claims 1 to 5 are executed.

8. A computer-readable storage medium storing a computer program, wherein the computer program implements the steps of the method according to any one of claims 1 to 5 when executed by a processor.

9. A computer program product, comprising a computer program, which implements the steps of the method according to any one of claims 1 to 5 when executed by a processor.

Images