Maneuvering weak multi-target track-before-detect method based on echo energy space-time distribution characteristics, program, equipment and storage medium

Abstract

The invention discloses a maneuvering weak target pre-detection tracking method based on echo energy space-time distribution characteristics, a program, equipment and a storage medium, and belongs to the field of target detection and tracking. The method comprises the following steps: firstly, performing maximum non-coherent accumulation on multiple frames of radar target echoes, and converting accumulated radar measurement into a two-dimensional radar target echo image; and inputting the echo image into an improved YOLOv8 instance segmentation network for identification. And finally, carrying out coordinate conversion on an identification result, mapping image pixels to a radar resolution unit, and outputting detected target position information and the number of targets. Compared with an algorithm of firstly detecting and then tracking, the tracking-before-detection algorithm does not perform threshold detection on each frame of signal, but accumulates multiple frames of radar target echoes, and forms a motion track to be distinguished from noise plots, clutters and the like according to the space-time correlation characteristics of the target so as to identify the target. And the detection probability and the target number identification rate are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of target detection and tracking, and specifically relates to a method, program, device and storage medium for pre-detection tracking of maneuverable weak multi-targets based on the space-time distribution characteristics of echo energy. Background Art

[0002] With the continuous development of stealth technology, the scattering cross-section of aviation radar equipment has been significantly reduced, resulting in a significant decrease in the signal-to-noise ratio (SNR) of the target echo, which greatly increases the difficulty of target detection. Track before detection (TBD) is an important technical approach for weak target detection. This method first performs non-coherent energy accumulation to improve the target signal-to-noise ratio, then processes the accumulated results of multiple frames, and finally outputs the target detection results. The current mainstream TBD methods include Hough transform (HT), dynamic programming (DP), particle filter (PFD), etc. When using the traditional tracking before detection algorithm based on the target motion model, when the target's maneuvering method does not match the preset model, the detection effect of the algorithm will drop sharply or the amount of calculation will increase significantly. Summary of the invention

[0003] The purpose of the present invention is to fully explore the space-time distribution characteristics of target echo energy, use an improved YOLO instance segmentation network to perform target detection on multi-frame echo data based on maximum accumulation, and improve the problem that the existing pre-detection method relies on the target motion model.

[0004] The present invention provides a method for detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy, which is characterized by comprising the following steps:

[0005] Step 1: Receive multiple frames of radar echo data, obtain the maximum amplitude corresponding to each distance resolution unit in the received data, and construct a maximum value accumulation matrix with the maximum amplitude of each distance resolution unit as an element;

[0006] Step 2: Generate a radar echo image according to the maximum accumulation matrix constructed in step 1;

[0007] Step 3: Preprocess the image and input the processed image into the pre-trained model for track detection. Use the space-time correlation of the target track after multi-frame accumulation to perform instance segmentation of the target track, output the segmented image and obtain the anchor frame and segmentation mask position information;

[0008] Step 4: For each anchor frame, obtain the normalized coordinate information of all segmentation masks in the anchor frame, map the normalized coordinate information to the radar coordinate system, convert it into the corresponding radar range resolution unit position, and complete the target point trace back;

[0009] Step 5: Repeat step 4 until the mask information in all anchor boxes is completed, and output the detected target location information and target number.

[0010] Furthermore, in step 1, the receiving of multiple frames of radar echo data does not set a limit gate for each frame of data.

[0011] Furthermore, in step 1, the maximum value matrix expression is:

[0012] Z k ={z k (i,j)|i=1,…,N r ; j = 1,…,N a}

[0013] z int (i, j) = max(z 1 (i,j,z) 2 (i, j), …, z k (i, j))

[0014] Z int (i, j) = {z int (i, j)}

[0015] Among them, Z k represents the Kth frame of radar echo measurement data; z k (i, j) represents the amplitude of the (i, j) resolution unit in the K-th frame data; N r Represents the number of distance dimension resolution units; N a Indicates the number of azimuth resolution units; z int (i, j) represents the maximum value of each resolution unit; Z int (i, j) represents the maximum accumulation matrix.

[0016] Furthermore, in step 3, the image preprocessing is to convert the radar echo image generated by the maximum accumulation matrix into a grayscale image.

[0017] Furthermore, in step 3, the pre-trained model is an improved YOLOv8 instance segmentation network, including a backbone network, a neck structure, and a detection head.

[0018] Furthermore, in step 3, the improved YOLOv8 instance segmentation network is specifically:

[0019] Step 3.1: Integrate the BoTNet module into the backbone network of the YOLOv8 network, extract features from the input image, and output feature maps;

[0020] Step 3.2: Integrate the DSCoov module in the neck structure of the YOLOv8 network, perform feature fusion on the feature map output by the backbone network, and output a multi-scale feature map;

[0021] Step 3.3: Input the multi-scale feature map output in step 3.2 into the detection head, output the detected target track mask information and anchor frame information, and obtain the pixel coordinates of the target track.

[0022] Furthermore, the step 4 is specifically as follows:

[0023] Assume that the normalized coordinates of the target in the pixel coordinate system are (I tx , I ty ), then the coordinates of the target pixel in the radar range-range coordinate system can be expressed as:

[0024]

[0025] Among them, f tx is the coordinate of the target in the X-axis direction, f ty is the coordinate of the target in the Y-axis direction; w p is the width of the image, h p is the height of the image; R x is the detection range of the radar in the X-axis direction, R y is the detection range of the radar in the Y-axis direction; f ox and f oy is the coordinate of the image origin in the distance-distance coordinate system;

[0026] The coordinates of the target in the range-azimuth coordinate system are expressed as:

[0027]

[0028] Among them, u tr is the target distance coordinate, u ta is the target azimuth coordinate;

[0029] The target coordinates in the range-azimuth coordinate system are converted to the radar range resolution unit, expressed as:

[0030]

[0031] Among them, (v tr ,v ta ) is the distance resolution unit corresponding to the target, N r is the distance resolution, N a is the azimuth resolution.

[0032] The present invention also provides a computer device / equipment / system, comprising a memory, a processor and a computer program stored in the memory, wherein when the processor executes the computer program, the steps of the method for pre-detection and tracking of maneuvering weak targets based on the spatial-temporal distribution characteristics of echo energy are implemented as described in any one of the above-mentioned items.

[0033] The present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon, characterized in that when the computer program / instruction is executed by a processor, the steps of the method for pre-detection and tracking of maneuvering weak targets based on the spatial-temporal distribution characteristics of echo energy are implemented as described in any of the above-mentioned items.

[0034] The present invention also provides a computer program product, including a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the method for pre-detection and tracking of maneuvering weak targets based on the spatial-temporal distribution characteristics of echo energy are implemented in any of the above-mentioned items.

[0035] The beneficial effects of the present invention are:

[0036] (1) The present invention is applicable to maneuvering targets and does not rely on motion models or prior information such as the position, velocity, and acceleration of the target, while achieving a higher detection probability.

[0037] (2) The present invention is applicable to multi-target detection. When the number of targets is unknown and the target motion models are the same or different, a high detection probability can be achieved. When the signal-to-noise ratio is 8, the detection probability of multiple maneuvering targets can reach more than 85%.

[0038] (3) The present invention utilizes the instance segmentation characteristics of the algorithm network, which can not only distinguish between the target and the background, but also accurately distinguish the tracks of different targets and accurately obtain the target quantity information. Under the condition of a low signal-to-noise ratio of 8, the target quantity recognition rate can reach more than 90%.

[0039] (4) The energy of weak target echo is small, and the detection-before-tracking technology is likely to cause missed alarms. Compared with the detection-before-tracking algorithm, the detection-before-tracking algorithm does not perform threshold detection on each frame signal, but accumulates multiple frames of radar target echoes, and forms a motion track based on the target's space-time correlation characteristics to distinguish between noise point tracks and clutter, thereby identifying the target. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is the overall flow chart of the present invention;

[0041] Figure 2 It is the radar echo multi-frame synthetic image of the present invention;

[0042] Figure 3It is the improved YOLOv8 network structure diagram of the present invention;

[0043] Figure 4 is the network structure of BoTNet of the present invention;

[0044] Figure 5 This is the basic deformation process of the convolution kernel of DSConv of the present invention;

[0045] Figure 6 It is a mobile multi-target detection data graph of the present invention;

[0046] Figure 7 1 is a graph of multi-target detection results under three signal-to-noise ratios of the present invention;

[0047] Figure 8 It is a target quantity detection accuracy graph of the present invention. DETAILED DESCRIPTION

[0048] The present invention is further described below in conjunction with the accompanying drawings.

[0049] The present invention provides a method for tracking a maneuvering weak target before detection based on the space-time distribution characteristics of echo energy, wherein the detection process includes: non-coherent accumulation of multi-frame echo maximum values, radar echo image weak target track detection, and radar echo measurement coordinate conversion.

[0050] The method of the present invention inputs radar echo measurements containing targets, performs maximum non-coherent accumulation on radar signals, converts the accumulated radar echo measurements into two-dimensional radar echo images, and extracts target space-time correlation features; radar echo images of target tracks of various models are used as training samples for data annotation, so as to train the improved YOLO instance segmentation network adopted in the present scheme; the two-dimensional radar echo images to be detected are input into the trained network to extract space-time correlation features, amplitude features, etc.; the detection results are obtained and coordinate conversion is performed, the image pixels are mapped to the radar resolution unit, and the target position information and the number of targets obtained by detection are output; specifically, Figure 1 shown.

[0051] A method for tracking a maneuvering weak target before detection based on the space-time distribution characteristics of echo energy specifically comprises the following steps:

[0052] Step 1: Store multiple frames of radar echo measurement data. Directly store K frames of radar echo measurement data, and do not set a threshold for each frame of data. The kth frame of data can be expressed as:

[0053] Z k ={z k (i, j)|i=1,…,N r ; j = 1, ..., N a} (1)

[0054] Among them, Z k represents the Kth frame of radar echo measurement data; z k (i, j) represents the amplitude of the (i, j) resolution unit in the k-th frame data, N r Represents the number of distance dimension resolution units, N a Indicates the number of azimuth dimension resolution units.

[0055] Step 2: Perform joint processing of multiple frames of data and perform maximum value accumulation on the K frames of data stored above. The specific method is to compare the same resolution units of different frames and obtain the maximum value of each resolution unit, that is:

[0056] z int (i, j) = max(z 1 (i,j,z) 2 (i, j), …, z k (i, j))

[0057] i=1,…,N r ; j = 1, ..., N a (2)

[0058] Step 3: Construct a maximum cumulative matrix, combining the maximum measurement values ​​of each resolution unit obtained in the previous step into a maximum cumulative matrix, which can be expressed as:

[0059] Z int (i, j) = {z int (i, j)|i=1,…,N r ; j = 1, ..., N a} (3)

[0060] Step 4: Generate a radar echo image, that is, visualize the maximum accumulation matrix generated in the previous step to generate a radar echo multi-frame synthetic image.

[0061] Step 5: Preprocess the generated radar echo multi-frame synthetic image and input the processed image into the improved YOLOv8 instance segmentation network. The improved YOLOv8 instance segmentation network includes three parts: the backbone network, the neck structure, and the detection head. The backbone network extracts features from the input image and outputs a feature map. This part integrates the BoTNet module under the YOLOv8 framework to enhance the detection of small and insignificant features.

[0062] BoTNet is a Transformer-based network structure that integrates CNN and Transformer and uses convolution and self-attention mechanisms to enable the network to better capture global information and long-distance dependencies, and enhance the network's ability to extract spatial-temporal correlation features of target tracks. Figure 4 The network structure of BoTNet is shown, using global multi-head self-attention instead of 3×3 spatial convolution in the last three bottleneck blocks of ResNet.

[0063] Step 6: The neck structure performs feature fusion on the feature map output by the backbone network and outputs a multi-scale feature map. This part integrates the DSCoov module to enhance the detection of slender and tortuous structures.

[0064] DSConv uses the learned deformation offset to change the convolution shape and adopts an iterative strategy to process one target at a time to ensure attention continuity. This improvement makes the network model more adaptable to complex geometric features, thereby improving the accuracy and robustness of detection.

[0065] The basic principle of DSConv is discussed below, where the basic deformation process of the convolution kernel is as follows: Figure 5 shown.

[0066] First, assume that the standard two-dimensional convolution coordinate is K and its center coordinate is K i =(x i ,y i ). The 3×3 convolution kernel K with dilation rate 1 is expressed as:

[0067] K={(x-1,y-1),(x-1,y),…(x+1,y+1)} (4)

[0068] In order to give the convolution kernel more flexibility to better focus on the slender and tortuous structure of the target point trace and the complex and changeable curved structure of the target track, DSConv introduces a deformation offset Δ. To prevent the offset from causing the receptive field to deviate from the target during free learning, DSConv adopts an iterative strategy to adjust the convolution kernel in the X-axis and Y-axis directions to ensure the continuity of attention and avoid excessive diffusion of the receptive field due to large deformation offsets.

[0069] In the X-axis direction, the coordinate becomes

[0070]

[0071] Where c = {0, 1, 2, 3, 4}, represents the horizontal distance from the center grid. Δ = {δ|δ∈-1, 1} is the deformation offset.

[0072] In the Y-axis direction, the coordinates become:

[0073]

[0074] Where c = {0, 1, 2, 3, 4}, indicating the vertical distance from the center grid.

[0075] Step 7: The detection head inputs the multi-scale feature map and outputs the bounding box and corresponding mask of each instance to obtain the target detection data.

[0076] Step 8: Perform coordinate transformation on the detected data.

[0077] Assume that the normalized coordinates of the target in the pixel coordinate system are (I tx , I ty ), then the coordinates of the target pixel in the radar range-range coordinate system can be expressed as:

[0078]

[0079] Among them, f tx is the coordinate of the target in the X-axis direction, f ty is the coordinate of the target in the Y-axis direction; w p is the width of the image, h p is the height of the image; R x is the detection range of the radar in the X-axis direction, R y is the detection range of the radar in the Y-axis direction; f ox and f oy is the coordinate of the image origin in the distance-distance coordinate system;

[0080] The coordinates of the target in the range-azimuth coordinate system are expressed as:

[0081]

[0082] Among them, u tr is the target distance coordinate, u ta is the target azimuth coordinate;

[0083] The target coordinates in the range-azimuth coordinate system are converted to the radar range resolution unit, expressed as:

[0084]

[0085] Among them, (v tr , v ta ) is the distance resolution unit corresponding to the target, N r is the distance resolution, N a is the azimuth resolution.

[0086] After the above coordinate transformation, the target point trace can be completed.

[0087] Step 9: Repeat step 8 until the mask information in all anchor boxes is completed, and output the detected target location information and target number.

[0088] Embodiment 1:

[0089] The method for tracking a maneuvering weak target before detection based on the space-time distribution characteristics of echo energy of the present invention is described in detail below in conjunction with the accompanying drawings.

[0090] (1) Receive K frames of radar echo data in sequence and directly store each frame of data without setting a threshold.

[0091] (2) Each distance resolution unit of K frames of radar echo data is jointly processed, and a temporary matrix temp is set to sequentially store the amplitude information of the same distance resolution unit of K frames of data.

[0092] (3) Calculate and store the maximum value in the temp matrix.

[0093] (4) Repeat steps 2 and 3 until all the distance resolution unit data of K frames are processed.

[0094] (5) Use the data obtained in step 4 to construct the maximum value matrix.

[0095] (6) Generate a radar echo image through the maximum value matrix constructed in step 5 to show the space-time correlation characteristics of the target. The track model of the radar echo image includes a straight-line motion path, a "C"-type motion path, and an "S"-type motion path.

[0096] (7) Image preprocessing, outputting a grayscale image suitable for detection by the improved YOLOv8 instance segmentation network.

[0097] (8) The image obtained in step 7 is input into the improved YOLOv8 instance segmentation network for track detection. The improved YOLOv8 network integrates the BoTNet (Bottleneck Transformer) module and the DSConv module in the YOLOv8 framework; the instance segmentation of the target track is performed by utilizing the space-time correlation presented by the target track after multi-frame accumulation. The network structure diagram is shown in the attached figure. Figure 2 shown.

[0098] (9) Output the segmented image and obtain the anchor box and segmentation mask location information.

[0099] (10) The segmentation mask position information output from step 9 is divided according to the corresponding different anchor frames.

[0100] (11) Obtain the normalized coordinate information of all segmentation masks within an anchor frame and calculate their range-range coordinates mapped to the radar coordinate system.

[0101] (12) Process the target range-range coordinates calculated in step 11 in sequence to calculate their corresponding range-azimuth coordinates.

[0102] (13) Process the target range-azimuth coordinates calculated in step 12 in sequence and calculate the corresponding radar range resolution unit position.

[0103] (14) Repeat steps 11 to 13 until the mask information in all anchor boxes is processed.

[0104] (15) Output the detected target location information and target number.

[0105] The present invention provides the detection probability and false alarm rate of mobile multi-targets under different signal-to-noise ratios. Figure 6 This is the detection data diagram of the present invention. When the signal-to-noise ratio is 8, the detection probability of multiple maneuvering targets can reach more than 85%; Figure 7 Graphs showing the detection results of the present invention under three signal-to-noise ratio conditions; Figure 8 This is the target number detection accuracy graph. When the signal-to-noise ratio is as low as 8, the target number recognition rate can reach more than 90%.

[0106] In particular, in some preferred embodiments of the present invention, a computer device is also provided, including a memory and a processor and a computer program stored in the memory, and when the processor executes the computer program, the steps of the method for pre-detection and tracking of maneuvering weak targets based on the spatial-temporal distribution characteristics of echo energy described in any of the above-mentioned embodiments are implemented.

[0107] In some other preferred embodiments of the present invention, a computer-readable storage medium is provided, on which a computer program / instructions are stored. When the computer program is executed by a processor, the steps of the method for detecting and tracking maneuvering weak targets based on the spatial-temporal distribution characteristics of echo energy described in any of the above embodiments are implemented.

[0108] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the process of the embodiment of the method for detecting and tracking a maneuvering weak target based on the spatial-temporal distribution characteristics of the echo energy, which will not be repeated here.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, 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 detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy, characterized in that: The following steps are involved: Step 1: Receive multiple frames of radar echo data, obtain the maximum amplitude corresponding to each distance resolution unit in the received data, and construct a maximum value accumulation matrix with the maximum amplitude of each distance resolution unit as an element; Step 2: Generate a radar echo image according to the maximum accumulation matrix constructed in step 1; Step 3: Preprocess the image and input the processed image into the pre-trained model for track detection. Use the space-time correlation of the target track after multi-frame accumulation to perform instance segmentation of the target track, output the segmented image and obtain the anchor frame and segmentation mask position information; Step 4: For each anchor frame, obtain the normalized coordinate information of all segmentation masks in the anchor frame, map the normalized coordinate information to the radar coordinate system, convert it into the corresponding radar range resolution unit position, and complete the target point trace back; Step 5: Repeat step 4 until the mask information in all anchor boxes is completed, and output the detected target location information and target number.

2. The method for detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy according to claim 1, characterized in that: In step 1, multiple frames of radar echo data are received without setting a limit gate for each frame of data.

3. A method for detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy according to any one of claims 1 or 2, characterized in that: In step 1, the maximum value matrix expression is: Z k ={z k (i,j)|i=1,…,N r ;j=1,…,N a } z int (i,j)=max(z1(i,j),z2(i,j),…,z k (i,j)) Z int (i,j)={z int (i,j)} Among them, Z k represents the Kth frame of radar echo measurement data; z k (i, j) represents the amplitude of the (i, j) resolution unit in the Kth frame data; N r Represents the number of distance dimension resolution units; N a Indicates the number of azimuth resolution units; z int (i, j) represents the maximum value of each resolution unit; Z int (i,j) represents the maximum cumulative matrix.

4. The method for detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy according to claim 1, characterized in that: In step 3, the image preprocessing is to convert the radar echo image generated by the maximum accumulation matrix into a grayscale image.

5. The method for detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy according to claim 1, characterized in that: In step 3, the pre-trained model is an improved YOLOv8 instance segmentation network, including a backbone network, a neck structure, and a detection head.

6. The method for detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy according to claim 5, characterized in that: In step 3, the improved YOLOv8 instance segmentation network is specifically: Step 3.1: Integrate the BoTNet module into the backbone network of the YOLOv8 network, extract features from the input image, and output feature maps; Step 3.2: Integrate the DSCoov module in the neck structure of the YOLOv8 network, perform feature fusion on the feature map output by the backbone network, and output a multi-scale feature map; Step 3.3: Input the multi-scale feature map output in step 3.2 into the detection head, output the detected target track mask information and anchor frame information, and obtain the pixel coordinates of the target track.

7. The method for detecting and tracking a maneuvering weak target based on the space-time distribution characteristics of echo energy according to claim 1, characterized in that: The step 4 is specifically as follows: Assume that the normalized coordinates of the target in the pixel coordinate system are (I tx ,I ty ), then the coordinates of the target pixel in the radar range-range coordinate system are expressed as: Among them, f tx is the coordinate of the target in the X-axis direction, f ty is the coordinate of the target in the Y-axis direction; w p is the width of the image, h p is the height of the image; R x is the detection range of the radar in the X-axis direction, R y is the detection range of the radar in the Y-axis direction; f ox and f oy is the coordinate of the image origin in the distance-distance coordinate system; The coordinates of the target in the range-azimuth coordinate system are expressed as: Among them, u tr is the target distance coordinate, u ta is the target azimuth coordinate; The target coordinates in the range-azimuth coordinate system are converted to the radar range resolution unit, expressed as: Among them, (v tr ,v ta ) is the distance resolution unit corresponding to the target, N r is the distance resolution, N a is the azimuth resolution.

8. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images