A SAR Image Target Detection and Tracking Method

Abstract

The present invention discloses a SAR image target detection and tracking method, which relates to the technical field of target detection and tracking, and includes the following steps: emitting microwave signals through synthetic aperture radar and receiving signals reflected from the ground to generate SAR images; preprocessing the generated SAR images to improve image quality and target detectability; identifying and locating selected targets from the preprocessed SAR images; and continuously acquiring each frame of SAR images in a time series. The present invention uses a convolutional neural network to evaluate the target morphological changes between adjacent frames, identify sharp and normal morphological changes, and for normal morphological changes, adopts the best constant speed tracking to ensure efficiency and accuracy; for sharp morphological changes, reduces the tracking speed to ensure continuity and stability. The method can adjust the strategy in time, reduce prediction errors, effectively prevent target loss, and provide reliable target control and monitoring in military surveillance and disaster monitoring to avoid major losses.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection and tracking, and in particular to a SAR image target detection and tracking method. Background Art

[0002] SAR image target detection and tracking refers to the detection and tracking of targets using Synthetic Aperture Radar (SAR) images. SAR is an active remote sensing technology that acquires images of ground objects by emitting microwave signals and receiving reflected signals. It has the advantages of all-weather, all-day, and penetrating clouds and smoke. Therefore, SAR images are widely used in military, disaster monitoring, and environmental protection. Target detection refers to identifying and locating selected targets, such as vehicles, ships, or buildings, from SAR images.

[0003] Target tracking is the continuous monitoring of the position of the target in the time series based on target detection. SAR image target tracking involves the use of the detected target position information in the continuously acquired SAR image sequence to continuously track the target through a specific algorithm. Common tracking methods include Kalman filtering, particle filtering, and optical flow. These methods predict the future position of the target and make corresponding corrections to maintain accurate tracking of the target. SAR image target detection and tracking is used in the military field to monitor enemy activities, in disaster monitoring to track the dynamic changes of natural disasters, and in environmental protection to monitor illegal activities. It has important practical application value. Through effective target detection and tracking, real-time monitoring and precise positioning of the target can be achieved, and the perception and response capabilities to dynamic changes of targets in complex environments can be improved.

[0004] The prior art has the following deficiencies:

[0005] When tracking a selected target identified in a SAR image, the existing technology usually tracks the selected target at a constant speed. The target may change its shape during movement. When the selected target undergoes a drastic shape change, if the target shape changes quickly, the tracking algorithm will not be able to adjust and match the new target features in time, resulting in an increase in the prediction error of the target position, and ultimately causing the tracking algorithm to be unable to continue to lock the target. Target loss is a fatal problem of the tracking system, especially in military surveillance and disaster monitoring. Target loss may mean the loss of control and surveillance of important targets, resulting in irreparable losses.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a SAR image target detection and tracking method, which ensures the accuracy of target dynamic tracking by preprocessing SAR images and acquiring continuous time series images, uses convolutional neural networks to evaluate target morphological changes between adjacent frames, identifies sudden and normal morphological changes, and adopts optimal constant speed tracking for normal morphological changes to ensure efficiency and accuracy; for sudden morphological changes, reduces the tracking speed to ensure continuity and stability. The method can adjust strategies in time, reduce prediction errors, effectively prevent target loss, provide reliable target control and monitoring in military surveillance and disaster monitoring, and avoid major losses, so as to solve the problems in the above-mentioned background technology.

[0008] In order to achieve the above object, the present invention provides the following technical solution: a SAR image target detection and tracking method, comprising the following steps:

[0009] SAR images are generated by transmitting microwave signals through synthetic aperture radar and receiving signals reflected from the ground;

[0010] Preprocess the generated SAR images to improve image quality and target detectability;

[0011] Identify and locate selected targets from preprocessed SAR images;

[0012] Continuously acquire each frame of SAR image in time series to form time series data, and provide accurate data support for dynamic tracking of targets through continuous time series images;

[0013] Extract the selected targets identified and located in two adjacent frames of SAR images, and evaluate the changes in the morphological features of the selected targets in the two frames by comparing the image data of the selected targets between the adjacent frames;

[0014] Based on the results of the morphological feature change assessment, the morphological changes of the selected targets are classified into drastic morphological changes and normal morphological changes;

[0015] For normal morphological changes, based on historical data, the optimal constant speed is set to track the selected target. For sudden morphological changes, the actual tracking speed of the selected target is adjusted to reduce the actual tracking speed and ensure the continuity and stability of target tracking.

[0016] Preferably, the specific steps of identifying and locating the selected target from the SAR image are as follows:

[0017] Perform image segmentation on the preprocessed SAR image;

[0018] Based on the image segmentation, feature extraction is performed on each segmented area;

[0019] After extracting a large number of features, feature selection is performed to reduce the feature dimension;

[0020] Use the selected features for target detection;

[0021] After the target is detected, target positioning is performed.

[0022] Preferably, the contour curvature information and motion blur degree information of the selected target between adjacent frames are obtained, and after analyzing the contour curvature information and motion blur degree information of the selected target, a contour curvature change index and a motion blur degree change index are generated respectively, and the analyzed contour curvature change index and motion blur degree change index are input into a pre-trained convolutional neural network, and a morphological feature change coefficient is generated by the convolutional neural network, and the morphological feature change coefficient is used to evaluate the morphological feature change of the selected target in the two frames.

[0023] Preferably, the steps of obtaining the contour curvature information of the selected target between adjacent frames, analyzing the contour curvature information of the selected target, and generating the contour curvature change index are as follows:

[0024] Perform target detection on each frame of SAR image, use edge detection algorithm to extract the contour edge of the selected target, and serialize the edge points of the contour edge into contour curve representation;

[0025] For the contour edge extracted in each frame, the curvature of each edge point is calculated. The calculation expression of the curvature in the discrete edge point sequence is:

[0026]

[0027] , where κ i is the curvature of the ith edge point, A i is the area of ​​the triangle formed by the i-1th, ith, and i+1th edge points, d i-1，i is the distance between the i-1th edge point and the i-th edge point, d i，i+1 is the distance between the ith edge point and the i+1th edge point, d i+1，i-1 is the distance between the i+1th edge point and the i-1th edge point;

[0028] Use the method of minimizing the distance between edge points to align the corresponding contour points in the two frames of images;

[0029] Calculate the curvature difference of the edge points of the contour curve in adjacent frames. The calculation expression of the curvature difference is: Δκ i =κ i (t+1) -κ i (t) |, where Δκ iis the curvature difference of the ith edge point between adjacent frames, κ i (t+1) is the curvature of the ith edge point in the t+1th frame, κ i (t) is the curvature of the i-th edge point in the t-th frame;

[0030] The curvature differences of all edge points are accumulated to calculate the total curvature change of the entire target contour. The calculation expression is: Δκ total is the total curvature change of the entire contour, N is the total number of edge points, and w i is the weight of the i-th edge point;

[0031] The profile curvature change index is calculated by the total curvature change, and the calculation expression is: In the formula, Cont curva Represents the profile curvature variation index, max(Δκ total ) is the predefined maximum total curvature change used for normalization.

[0032] Preferably, the steps of obtaining the motion blur degree information of the selected target between adjacent frames, analyzing the motion blur degree information of the selected target, and generating the motion blur degree change index are as follows:

[0033] The motion blur information of the target is extracted from the SAR images of adjacent frames by analyzing the gradient change of the selected target area. The SAR images of adjacent frames are respectively t and I t+1 It is represented by, where I t and I t+1 Represent the t-th frame and the t+1-th frame SAR image respectively, and I t and I t+1 The corresponding gradient images of the selected target area are represented by G t and G t+1 To express, In the formula, is the gradient operator;

[0034] In the selected target area, the motion blur vector of each pixel is calculated. The calculation expression is: In the formula, v t (x, y) and v t+1 (x, y) are the motion blur vectors of each pixel in the selected target area in the tth frame and the t+1th frame, respectively, and (x, y) are the coordinates of the pixel;

[0035] The fuzzy degree matrix of the selected target area is constructed using the fuzzy vector. The constructed expression is: t(x, y) = Σ x′，y′ v t (x′, y′)·K(xx′, yy′), x′ and y′ are variables that traverse all pixels in the selected target area, M t+1 (x, y) = Σ x'，y′ v t+1 (x′, y′)·K(xx′, yy′), where K is a kernel function used to smooth the motion blur vector, M t and M t+1 are the blur degree matrices of the selected target area in the t-th frame and the t+1-th frame respectively;

[0036] Calculate the blur degree change matrix, which represents the blur degree change of each pixel in the selected target area between adjacent frames. The calculation expression is: ΔM(x, y) = M t+1 (x, y)-M t (x, y)|, where ΔM(x, y) is the fuzzy degree change matrix;

[0037] The motion blur change index is calculated by comprehensively considering the blur change of the entire selected target area. The calculation expression is: Motion blur =Σ x，y∈R ΔM(x, y), where Motion blur is the blur degree variation index, which represents the comprehensive measure of blur degree variation of the entire target area, and R is the selected target area, which represents the pixel area of ​​the selected target being tracked in the SAR image.

[0038] Preferably, the morphological feature change coefficient generated after evaluating the morphological feature change of the selected target is compared and analyzed with a preset morphological feature change coefficient reference threshold, and the morphological change of the selected target is dynamically divided. The specific division steps are:

[0039] If the morphological feature change coefficient is greater than the morphological feature change coefficient reference threshold, the morphological change of the selected target is classified as a sharp morphological change;

[0040] If the morphological feature variation coefficient is less than or equal to the morphological feature variation coefficient reference threshold, the morphological variation of the selected target is classified as a normal morphological variation.

[0041] Preferably, in response to the rapid morphological changes, the actual tracking speed of the selected target tracking is adjusted, and the specific steps are as follows:

[0042] When a sharp morphological change is detected, the actual tracking speed baseline is adjusted. First, according to the morphological feature change coefficient Morp fl Calculate the speed adjustment Δv, the calculation expression is: Δv = γ·(Morp fl -τthre ) opt , where Morp fl represents the coefficient of variation of morphological characteristics, τ thre represents the reference threshold of the coefficient of variation of morphological features, and Morp fl >τ thre , γ represents the speed adjustment coefficient, which is used to control the speed adjustment amplitude, v opt represents the optimum constant speed;

[0043] According to the calculated speed adjustment Δv, adjust the actual tracking speed v actual , the adjusted expression is: v actual =v opt -Δv=v opt -γ·(Morp fl -τ thre )·v opt ;

[0044] During the tracking process, the change of the morphological feature change coefficient is dynamically monitored, and the actual tracking speed is dynamically adjusted according to the real-time feedback to ensure the continuity and stability of the tracking. The expression for dynamically adjusting the actual tracking speed is: actual =v opt ·(1-γ·max(0,Morp fl -τ thre ));

[0045] Use a smoothing function to adjust the actual tracking speed after dynamic adjustment. The adjustment expression is: smooth =λ· actual +(1-λ)·v prev , v smooth represents the actual tracking speed after smoothing, λ represents the smoothing coefficient, 1>λ>0, v prev Indicates the actual tracking speed of the previous SAR image;

[0046] The actual tracking speed v after smoothing smooth Applied to the target tracking process between the current frame and the next frame to ensure the continuity of tracking.

[0047] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0048] The present invention ensures the data accuracy of target dynamic tracking by preprocessing SAR images and acquiring continuous time series images, and uses convolutional neural networks to evaluate the changes in morphological features of selected targets between adjacent frames, which can effectively identify the sudden morphological changes and normal morphological changes of the target. According to the dynamic classification mechanism of morphological changes, the best constant speed tracking is adopted for normal morphological changes to ensure the efficiency and accuracy of tracking. For sudden morphological changes, the continuity and stability of tracking are guaranteed by reducing the actual tracking speed. Not only can the tracking strategy be adjusted in time when the target changes sharply to reduce the target position prediction error, but also the target loss can be effectively prevented, thereby providing reliable target control and monitoring in key applications such as military surveillance and disaster monitoring to avoid irreparable losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0050] Figure 1 The present invention is a method flow chart of a SAR image target detection and tracking method. DETAILED DESCRIPTION

[0051] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0052] The present invention provides Figure 1 A SAR image target detection and tracking method shown includes the following steps:

[0053] Synthetic Aperture Radar (SAR) emits microwave signals and receives the signals reflected from the ground to generate SAR images.

[0054] The SAR system acquires high-resolution images by moving the antenna to synthesize a long aperture. This process involves synthesizing echo signals from different locations to ultimately generate a complete SAR image, providing basic data for subsequent target detection and tracking.

[0055] Preprocess the generated SAR images to improve image quality and target detectability;

[0056] The preprocessing step includes techniques such as denoising, filtering and image enhancement. Commonly used methods include Gaussian filtering, mean filtering, adaptive filtering, etc. The purpose is to reduce speckle noise and other interference in the image, improve the contrast between the target and the background, and make the target easier to identify and locate in subsequent steps.

[0057] Identify and locate selected targets from preprocessed SAR images;

[0058] The specific steps to identify and locate the selected target from the SAR image are as follows:

[0059] Perform image segmentation on the preprocessed SAR image;

[0060] The purpose of image segmentation is to decompose an image into different regions to facilitate target identification. Common image segmentation methods include threshold segmentation, region growing, and watershed algorithm. Threshold segmentation divides an image into two parts, the target and the background, by setting a gray value threshold; the region growing method starts from a seed point and gradually expands the region according to the similarity criterion; the watershed algorithm uses the gradient information of the image to divide the image into several connected regions.

[0061] Based on the image segmentation, feature extraction is performed on each segmented area;

[0062] Feature extraction is an important step in identifying and distinguishing targets, and usually includes shape features, texture features, and grayscale features. Shape features can include the area, perimeter, and outline of the target; texture features can be extracted through methods such as gray-level co-occurrence matrix (GLCM) or wavelet transform; and grayscale features directly use the grayscale value statistics of the target area, such as mean and variance.

[0063] After extracting a large number of features, feature selection is performed to reduce the feature dimension;

[0064] Feature selection is performed to reduce feature dimensions in order to improve the efficiency and accuracy of the algorithm. Feature selection methods include principal component analysis (PCA), linear discriminant analysis (LDA) and feature selection algorithms (such as recursive feature elimination (RFE), etc.). PCA extracts the most representative features through dimensionality reduction; LDA selects the most discriminative features by maximizing the between-class variance and minimizing the within-class variance; RFE selects the optimal feature set by recursively removing the least important features.

[0065] Use the selected features for target detection;

[0066] The task of target detection is to distinguish the target from the background in an image. Commonly used methods include support vector machine (SVM), decision tree, random forest and deep learning model. SVM distinguishes the target from the background by finding the optimal segmentation hyperplane; decision tree and random forest improve detection accuracy by building multiple classification trees; deep learning models (such as convolutional neural network CNN) automatically learn image features through a multi-layer network structure to achieve efficient target detection.

[0067] After the target is detected, target positioning is performed;

[0068] Target localization is to determine the specific location of the target in the image, usually through a bounding box or mask to represent the location and shape of the target. The positioning method can be a simple geometric calculation (such as calculating the minimum enclosing rectangle) or predicting the precise location of the target through a regression model. The result of target localization will be used for subsequent tracking and monitoring.

[0069] Continuously acquire each frame of SAR image in time series to form time series data, and provide accurate data support for dynamic tracking of targets through continuous time series images;

[0070] Among them, each frame of image acquired goes through the same generation and preprocessing process to ensure the consistency of each frame of image. The continuous time series images provide the necessary data support for the dynamic tracking of the target, allowing the system to monitor the movement and changes of the target at different time points.

[0071] Extract the selected targets identified and located in two adjacent frames of SAR images, and evaluate the changes in the morphological features of the selected targets in the two frames by comparing the image data of the selected targets between the adjacent frames;

[0072] The contour curvature information and motion blur degree information of the selected target between adjacent frames are obtained. After analyzing the contour curvature information and motion blur degree information of the selected target, a contour curvature change index and a motion blur degree change index are generated respectively. The analyzed contour curvature change index and motion blur degree change index are input into a pre-trained convolutional neural network, and a morphological feature change coefficient is generated through the convolutional neural network. The morphological feature change coefficient is used to evaluate the morphological feature change of the selected target in these two frames.

[0073] Large changes in the contour curvature of the selected target between adjacent frames indicate that the morphological features of the selected target are changing rapidly. Contour curvature reflects the curvature of the target edge, and large changes usually mean that the shape of the target has changed significantly in a short period of time, such as a prominent bump, bend, or twist on a part of the target. This change may be due to morphological changes caused by the rotation, deformation, partial occlusion, or rapid movement of the target. When the contour curvature changes greatly, the contrast between the edge features of the target and the background changes significantly, affecting the visual features of the target, making it difficult for traditional constant speed tracking algorithms to accurately predict the target position and maintain continuous tracking. Therefore, large changes in contour curvature are an important indicator of rapid changes in target morphological features, and need to be specially processed in the tracking algorithm to improve the robustness and accuracy of target tracking.

[0074] The steps of obtaining the contour curvature information of the selected target between adjacent frames and analyzing the contour curvature information of the selected target to generate the contour curvature change index are as follows:

[0075] Perform target detection on each frame of SAR image, use edge detection algorithm to extract the contour edge of the selected target, and serialize the edge points of the contour edge into contour curve representation;

[0076] For the contour edge extracted in each frame, the curvature of each edge point is calculated. The calculation expression of the curvature in the discrete edge point sequence is:

[0077]

[0078] , where k i is the curvature of the ith edge point, A i is the area of ​​the triangle formed by the i-1th, ith, and i+1th edge points, d i-1，i is the distance between the i-1th edge point and the i-th edge point, d i，i+1 is the distance between the ith edge point and the i+1th edge point, d i+1，i-1 is the distance between the i+1th edge point and the i-1th edge point;

[0079] Use the method of minimizing the distance between edge points to align the corresponding contour points in the two frames of images;

[0080] Its main purpose is to compare the curvature between adjacent frames;

[0081] Minimizing the distance between edge points to achieve the alignment of corresponding contour points in two frames means that in two adjacent frames, by matching the contour points of each frame, the sum of the Euclidean distances between the matching points in adjacent frames is minimized. This usually involves finding the closest point pairs in the previous frame and the next frame to ensure that these points represent the same physical location or feature. In this way, corresponding contour point pairs can be found in the two frames.

[0082] Calculate the curvature difference of the edge points of the contour curve in adjacent frames. The calculation expression of the curvature difference is: Δκ i =κ i (t+1) -κ i (t) |, where Δκ i is the curvature difference of the ith edge point between adjacent frames, κ i (t+1) is the curvature of the ith edge point in the t+1th frame, k i (t) is the curvature of the i-th edge point in the t-th frame;

[0083] The curvature differences of all edge points are accumulated to calculate the total curvature change of the entire target contour. The calculation expression is: Δκ total is the total curvature change of the entire contour, N is the total number of edge points, and w i is the weight of the i-th edge point (the weight can be assigned according to the importance or credibility of the edge point);

[0084] The profile curvature change index is calculated by the total curvature change, and the calculation expression is: In the formula, Cont curva Represents the profile curvature variation index, max(Δκ total ) is the predefined maximum total curvature change, used for normalization;

[0085] The predefined maximum total curvature change is a maximum possible total curvature change value set in advance, which is used to standardize the calculation results. This predefined value is usually determined based on historical data or empirical values ​​in a specific application scenario, and represents the curvature change of the target contour under the most extreme conditions. This predefined value is used to standardize the actual calculated total curvature change, so that the range of the curvature change index is limited to [0, 1], thereby facilitating comparison and analysis between different images.

[0086] The contour curvature information of the selected target between adjacent frames is obtained. The larger the performance value of the contour curvature change index generated after analyzing the contour curvature information of the selected target, the more significant the change of the contour curvature of the selected target between adjacent frames, thereby indicating that the morphological characteristics of the selected target change faster. Conversely, if the performance value of the contour curvature change index is small, it means that the contour curvature of the selected target between adjacent frames does not change significantly, indicating that the morphological characteristics of the target change slowly.

[0087] The significant increase in the degree of motion blur of the selected target between adjacent frames indicates that the morphological features of the selected target are changing rapidly. This change is usually caused by the rapid movement of the target within the camera exposure time, which makes the target appear as a blurred track in the image. The significant increase in the degree of motion blur reflects the rapid morphological transformation of the target in a short period of time. This rapid movement will cause the edge, shape and detail features of the target to change significantly between different frames. For tracking algorithms, the motion blur of the rapid morphological changes increases the difficulty of identifying and matching target features, because traditional static features may no longer be applicable and must rely on more dynamic and robust feature extraction and matching methods. It can be seen that the significant increase in the degree of motion blur is an important indicator of rapid changes in target morphology.

[0088] The steps of obtaining the motion blur degree information of the selected target between adjacent frames and analyzing the motion blur degree information of the selected target to generate the motion blur degree change index are as follows:

[0089] The motion blur information of the target is extracted from the SAR images of adjacent frames by analyzing the gradient change of the selected target area. The SAR images of adjacent frames are respectively t and I t+1 It is represented by, where I t and I t+1 Represent the t-th frame and the t+1-th frame SAR image respectively, and I t and I t+1 The corresponding gradient images of the selected target area are represented by G t and G t+1 To express, In the formula, is the gradient operator;

[0090] In the selected target area, the motion blur vector of each pixel is calculated. The calculation expression is: In the formula, v t (x, y) and v t+1 (x, y) are the motion blur vectors of each pixel in the selected target area in the tth frame and the t+1th frame, respectively, and (x, y) are the coordinates of the pixel;

[0091] The motion blur vector quantifies the degree of blur at each pixel.

[0092] The fuzzy degree matrix of the selected target area is constructed using the fuzzy vector. The constructed expression is: t (x, y) = ∑ x′，y′ v t (x′, y′)·K(xx′, yy′), x′ and y′ are variables that traverse all pixels in the selected target area, M t+1 (x, y) = ∑ x′，y′ v t+1 (x′, y′)·K(xx′, yy′), where K is a kernel function used to smooth the motion blur vector, M t and M t+1 are the blur degree matrices of the selected target area in the t-th frame and the t+1-th frame respectively;

[0093] The blur matrix quantifies the blur degree of each pixel in the selected target area.

[0094] Calculate the blur degree change matrix, which represents the blur degree change of each pixel in the selected target area between adjacent frames. The calculation expression is: ΔM(x, y) = M t+1 (x, y)-M t (x, y)|, where ΔM(x, y) is the fuzzy degree change matrix;

[0095] The motion blur change index is calculated by comprehensively considering the blur change of the entire selected target area. The calculation expression is: Motion blur =Σ x，y∈R ΔM(x, y), where Motion blur is the blur degree variation index, which represents the comprehensive measure of blur degree variation of the entire target area, and R is the selected target area, which represents the pixel area of ​​the selected target being tracked in the SAR image.

[0096] The motion blur degree information of the selected target between adjacent frames is obtained. The larger the performance value of the motion blur degree change index generated after analyzing the motion blur degree information of the selected target, the faster the morphological characteristics of the target change between adjacent frames. Conversely, the smaller the motion blur degree change index, the slower the target morphological characteristics change. This is because a high motion blur degree change index reflects the rapid changes in the position and shape of the target in the image, while a lower motion blur degree change index indicates that the target changes are relatively smooth and stable.

[0097] The convolutional neural network is not specifically limited here, and can realize the contour curvature change index Cont curva and blur degree change index Motion blurComprehensive analysis is performed to generate the morphological characteristic variation coefficient Morp fl In order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0098] Morphological characteristic variation coefficient Morp fl The resulting calculation formula is:

[0099]

[0100] , where μ1 and μ2 are the profile curvature variation index Cont curva and blur degree change index Motion blur The preset proportional coefficient, and μ1 and μ2 are both greater than 0.

[0101] It can be seen from the morphological feature variation coefficient that the contour curvature information of the selected target between adjacent frames is obtained, and the larger the expression value of the contour curvature variation index generated after analyzing the contour curvature information of the selected target is, the larger the expression value of the motion blur degree variation index generated after obtaining the motion blur degree information of the selected target between adjacent frames and analyzing the motion blur degree information of the selected target is, indicating that the morphological features of the selected target change faster, and vice versa, it indicates that the morphological features of the selected target change slower.

[0102] Based on the results of the morphological feature change assessment, the morphological changes of the selected targets are classified into drastic morphological changes and normal morphological changes;

[0103] The morphological feature change coefficient generated after evaluating the morphological feature change of the selected target is compared and analyzed with the pre-set morphological feature change coefficient reference threshold, and the morphological change of the selected target is dynamically divided. The specific division steps are:

[0104] If the morphological feature change coefficient is greater than the morphological feature change coefficient reference threshold, the morphological change of the selected target is classified as a sharp morphological change;

[0105] If the morphological feature variation coefficient is less than or equal to the morphological feature variation coefficient reference threshold, the morphological variation of the selected target is classified as a normal morphological variation.

[0106] For normal morphological changes, the optimal constant speed is set based on historical data to track the selected target. For sudden morphological changes, the actual tracking speed of the selected target is adjusted to reduce the actual tracking speed to ensure the continuity and stability of target tracking.

[0107] The optimal constant speed refers to the most suitable tracking speed set based on historical tracking data under the condition of normal morphological changes of the selected target to ensure the accuracy and stability of tracking. The process of setting based on historical data includes analyzing the characteristics of the target's motion trajectory, speed, acceleration, etc. over a period of time in the past, and determining a speed value that can effectively balance tracking accuracy and computational efficiency through statistical methods or machine learning models. This speed value can maintain accurate tracking when the target does not change drastically, while avoiding excessive calculation and waste of resources.

[0108] In response to the rapid shape change, the actual tracking speed of the selected target tracking is adjusted. The specific steps are as follows:

[0109] When a sharp morphological change is detected, the actual tracking speed baseline is adjusted. First, according to the morphological feature change coefficient Morp fl Calculate the speed adjustment Δv, the calculation expression is: Δv = γ·(Morp fl -τ thre )·v opt , where Morp fl represents the coefficient of variation of morphological characteristics, τ thre represents the reference threshold of the coefficient of variation of morphological features, and Morp fl >τ thre , γ represents the speed adjustment coefficient, which is used to control the speed adjustment amplitude, v opt represents the optimum constant speed;

[0110] According to the calculated speed adjustment Δv, adjust the actual tracking speed v actual , the adjusted expression is: v actual =v opt -Δv=v opt -γ·(Morp fl -τ thre )·v opt ;

[0111] During the tracking process, the change of the morphological feature change coefficient is dynamically monitored, and the actual tracking speed is dynamically adjusted according to the real-time feedback to ensure the continuity and stability of the tracking. The expression for dynamically adjusting the actual tracking speed is: actual =v opt ·(1-γ·max(0,Morp fl -τ th)e ));

[0112] Use a smoothing function to adjust the actual tracking speed after dynamic adjustment (to avoid instability caused by drastic changes in speed). The adjustment expression is: smooth =λ·v actual +(1-λ)·v prev, v smooth represents the actual tracking speed after smoothing, λ represents the smoothing coefficient, 1>λ>0, v prev Indicates the actual tracking speed of the previous SAR image;

[0113] The actual tracking speed v after smoothing smooth Applied to the target tracking process between the current frame and the next frame to ensure the continuity of tracking.

[0114] The present invention ensures the data accuracy of target dynamic tracking by preprocessing SAR images and acquiring continuous time series images, and uses convolutional neural networks to evaluate the changes in morphological features of selected targets between adjacent frames, which can effectively identify the sudden morphological changes and normal morphological changes of the target. According to the dynamic classification mechanism of morphological changes, the best constant speed tracking is adopted for normal morphological changes to ensure the efficiency and accuracy of tracking. For sudden morphological changes, the continuity and stability of tracking are guaranteed by reducing the actual tracking speed. Not only can the tracking strategy be adjusted in time when the target changes sharply to reduce the target position prediction error, but also the target loss can be effectively prevented, thereby providing reliable target control and monitoring in key applications such as military surveillance and disaster monitoring to avoid irreparable losses.

[0115] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0116] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0117] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A SAR image target detection and tracking method, characterized in that: The following steps are involved: SAR images are generated by transmitting microwave signals through synthetic aperture radar and receiving signals reflected from the ground; Preprocess the generated SAR images to improve image quality and target detectability; Identify and locate selected targets from preprocessed SAR images; Continuously acquire each frame of SAR image in time series to form time series data, and provide accurate data support for dynamic tracking of targets through continuous time series images; Extract the selected targets identified and located in two adjacent frames of SAR images, and evaluate the changes in the morphological features of the selected targets in the two frames by comparing the image data of the selected targets between the adjacent frames; Based on the results of the morphological feature change assessment, the morphological changes of the selected targets are classified into drastic morphological changes and normal morphological changes; For normal morphological changes, based on historical data, the optimal constant speed is set to track the selected target. For sudden morphological changes, the actual tracking speed of the selected target is adjusted to reduce the actual tracking speed and ensure the continuity and stability of target tracking.

2. A SAR image target detection and tracking method according to claim 1, characterized in that: The specific steps to identify and locate the selected target from the SAR image are as follows: Perform image segmentation on the preprocessed SAR image; Based on the image segmentation, feature extraction is performed on each segmented area; After extracting a large number of features, feature selection is performed to reduce the feature dimension; Use the selected features for target detection; After the target is detected, target positioning is performed.

3. The SAR image target detection and tracking method according to claim 1, characterized in that: The contour curvature information and motion blur degree information of the selected target between adjacent frames are obtained. After analyzing the contour curvature information and motion blur degree information of the selected target, a contour curvature change index and a motion blur degree change index are generated respectively. The analyzed contour curvature change index and motion blur degree change index are input into a pre-trained convolutional neural network, and a morphological feature change coefficient is generated through the convolutional neural network. The morphological feature change coefficient is used to evaluate the morphological feature change of the selected target in these two frames.

4. The SAR image target detection and tracking method according to claim 3, characterized in that: The steps of obtaining the contour curvature information of the selected target between adjacent frames and analyzing the contour curvature information of the selected target to generate the contour curvature change index are as follows: Perform target detection on each frame of SAR image, use edge detection algorithm to extract the contour edge of the selected target, and serialize the edge points of the contour edge into contour curve representation; For the contour edge extracted in each frame, the curvature of each edge point is calculated. The calculation expression of the curvature in the discrete edge point sequence is: , where κ i is the curvature of the ith edge point, A i is the area of ​​the triangle formed by the i-1th, ith, and i+1th edge points, d i-1，i is the distance between the i-1th edge point and the i-th edge point, d i，i+1 is the distance between the ith edge point and the i+1th edge point, d i+1，i-1 is the distance between the i+1th edge point and the i-1th edge point; Use the method of minimizing the distance between edge points to align the corresponding contour points in the two frames of images; Calculate the curvature difference of the edge points of the contour curve in adjacent frames. The calculation expression of the curvature difference is: Δκ i =|κ i (t+1) -κ i (t) }, where Δκ i is the curvature difference of the ith edge point between adjacent frames, κ i (t+1) is the curvature of the ith edge point in the t+1th frame, κ i (t) is the curvature of the i-th edge point in the t-th frame; The curvature differences of all edge points are accumulated to calculate the total curvature change of the entire target contour. The calculation expression is: Δκ total is the total curvature change of the entire contour, N is the total number of edge points, and w i is the weight of the i-th edge point; The profile curvature change index is calculated by the total curvature change, and the calculation expression is: In the formula, Cont curva Represents the profile curvature variation index, max(Δκ total ) is the predefined maximum total curvature change used for normalization.

5. The SAR image target detection and tracking method according to claim 4, characterized in that: The steps of obtaining the motion blur degree information of the selected target between adjacent frames and analyzing the motion blur degree information of the selected target to generate the motion blur degree change index are as follows: The motion blur information of the target is extracted from the SAR images of adjacent frames by analyzing the gradient change of the selected target area. The SAR images of adjacent frames are respectively t and I t+1 It is represented by, where I t and I t+1 Represent the t-th frame and the t+1-th frame SAR image respectively, and I t and I t+1 The corresponding gradient images of the selected target area are represented by G t and G t+1 To express, In the formula, is the gradient operator; In the selected target area, the motion blur vector of each pixel is calculated. The calculation expression is: In the formula, v t (x, y) and v t+1 (x, y) are the motion blur vectors of each pixel in the selected target area in the tth frame and the t+1th frame, respectively, and (x, y) are the coordinates of the pixel; The fuzzy degree matrix of the selected target area is constructed using the fuzzy vector. The constructed expression is: t (x, y) = Σ x′，y′ v t (x′, y′·K(xx′, yy′), x′ and y′ are variables that traverse all pixels in the selected target area, M t+1 (x, y) = Σ x′，y′ v t+1 (x′, y′)·K(xx′, yy′), where K is a kernel function used to smooth the motion blur vector, M t and M t+1 are the blur degree matrices of the selected target area in the t-th frame and the t+1-th frame respectively; Calculate the blur degree change matrix, which represents the blur degree change of each pixel in the selected target area between adjacent frames. The calculation expression is: ΔM(x, y) = |M t+1 (x, y)-M t (x, y)}, where ΔM(x, y) is the fuzzy degree change matrix; The motion blur change index is calculated by comprehensively considering the blur change of the entire selected target area. The calculation expression is: Motion blur =Σ x，y∈R ΔM(x, y), where Motion blur is the blur degree variation index, which represents the comprehensive measure of blur degree variation of the entire target area, and R is the selected target area, which represents the pixel area of ​​the selected target being tracked in the SAR image.

6. The SAR image target detection and tracking method according to claim 3, characterized in that: The morphological feature change coefficient generated after evaluating the morphological feature change of the selected target is compared and analyzed with the pre-set morphological feature change coefficient reference threshold, and the morphological change of the selected target is dynamically divided. The specific division steps are: If the morphological feature change coefficient is greater than the morphological feature change coefficient reference threshold, the morphological change of the selected target is classified as a sharp morphological change; If the morphological feature variation coefficient is less than or equal to the morphological feature variation coefficient reference threshold, the morphological variation of the selected target is classified as a normal morphological variation.

7. A SAR image target detection and tracking method according to claim 6, characterized in that: In response to the sudden change in shape, the actual tracking speed of the selected target is adjusted. The specific steps are as follows: When a sharp morphological change is detected, the actual tracking speed baseline is adjusted. First, according to the morphological feature change coefficient Morp fl Calculate the speed adjustment Δv, the calculation expression is: Δv = γ·(Morp fl -τ thre )·v opt , where Morp fl represents the coefficient of variation of morphological characteristics, τ thre represents the reference threshold of the coefficient of variation of morphological features, and Morp fl >τ thre , γ represents the speed adjustment coefficient, which is used to control the speed adjustment amplitude, v opt represents the optimum constant speed; According to the calculated speed adjustment Δv, adjust the actual tracking speed v actual , the adjusted expression is: v actual =v opt -Δv=v opt -γ·(Morp fl -τ thre )·v opt ; During the tracking process, the change of the morphological feature change coefficient is dynamically monitored, and the actual tracking speed is dynamically adjusted according to the real-time feedback to ensure the continuity and stability of the tracking. The expression for dynamically adjusting the actual tracking speed is: actual =v opt ·(1-γ·max(0,Morp fl -τ thre )); Use a smoothing function to adjust the actual tracking speed after dynamic adjustment. The adjustment expression is: smooth =λ·v actual +(1-λ)·v prev , v smooth represents the actual tracking speed after smoothing, λ represents the smoothing coefficient, 1>λ>0, v prev Indicates the actual tracking speed of the previous frame of SAR image; The actual tracking speed v after smoothing smooth Applied to the target tracking process between the current frame and the next frame to ensure the continuity of tracking.

Images