Infrared target anti-interference tracking method based on feature segmentation tracking network

Abstract

The invention provides an infrared target anti-interference tracking method based on a feature segmentation tracking network, which relates to the technical field of image recognition and comprises three steps of constructing a neural network model, defining constraint conditions and judging a tracking state. Firstly, in the construction process of a feature segmentation tracking network model, key information of a target can be accurately extracted by introducing an attention mechanism and a multi-scale fusion module, and background interference can be remarkably suppressed. And secondly, a segmentation result is constrained from three dimensions of overlapping degree, energy and shape by adopting a comprehensive loss function and combining Soft-IoU loss, energy loss and layered shape loss, so that the network prediction precision is remarkably improved, and the tracking stability is kept. And finally, in the target tracking process, in combination with inter-frame continuity and shape feature matching, the system not only can accurately judge whether the target is lost, but also can quickly recover tracking after the target is lost, so that the problem of target loss in a traditional method is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, in particular to an infrared target anti-interference tracking method based on a feature segmentation tracking network. BACKGROUND

[0002] Infrared target recognition and tracking technology has important application value in many fields such as military reconnaissance, security monitoring and other fields with high requirements for real-time and accuracy. However, the core challenge of this technology is how to ensure the continuous and stable tracking of the target in a complex interference scene. Complex background interference, target occlusion and motion blur and other factors often significantly affect the performance of the tracking algorithm. Existing researches mainly focus on improving the robustness of the tracking algorithm, especially by improving the feature extraction or motion prediction module to enhance the adaptability of the algorithm in the interference environment.

[0003] However, the current tracking method still has significant limitations in practical application. For example, when the target encounters partial occlusion or complex background interference, the existing algorithm often fails due to the insufficient recognition ability of the local features of the target, resulting in tracking drift or failure. Although some methods introduce segmentation networks to improve the positioning accuracy, such methods usually bring high computational complexity, which is difficult to meet the real-time tracking requirements.

[0004] Specifically, the interference sources in the complex ground scene, such as high-intensity areas, target-like backgrounds, smoke, and ground obstructions, can significantly reduce the recognizability of the target. Traditional global feature matching methods are prone to failure in this case, resulting in incorrect target positioning. For example, although deep learning-based algorithms can extract rich semantic features, their ability to perceive the overall shape features or local details of the target is weak, which greatly reduces their tracking performance when the target is partially occluded. In order to solve these problems, some methods attempt to combine segmentation and tracking modules, but these methods mostly use serial architectures, lack deep interaction between the segmentation and tracking modules, and are difficult to balance accuracy and efficiency.

[0005] In addition, most of the existing technologies do not fully utilize the synergistic effect of spatial local features and temporal context information of the target. In the actual tracking process, the apparent changes of the target and the uncertainty of the motion increase the difficulty of stable tracking. In order to improve the infrared target tracking performance in complex scenes, it is necessary to develop a new network architecture that deeply integrates local feature segmentation and temporal tracking mechanisms, and has good anti-interference ability and real-time performance, in order to significantly improve the robustness and accuracy of target tracking in complex environments. SUMMARY

[0006] The present application aims to provide an infrared target anti-interference tracking method based on a feature segmentation tracking network to overcome the deficiencies in the above technology.

[0007] To achieve the above object, the technical scheme adopted by the present application is as follows: The present application provides an infrared target anti-interference tracking method based on a feature segmentation tracking network, comprising the following steps: S1, constructing a neural network model: inputting the previous several frames of real images containing target regions and the current frame of real image, sequentially performing inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement and spatial position enhancement on the input image to obtain the feature enhancement map of the current frame, sequentially performing down-sampling, feature perception and feature fusion on the feature enhancement map, and outputting the prediction probability map of the current frame; S2, defining the constraint condition: generating a comprehensive loss function according to the Soft-IoU loss function, the energy loss function and the hierarchical shape loss function, and the comprehensive loss function is used to constrain the prediction probability map from the three dimensions of overlap, energy and shape; S3, judging the tracking state: extracting several candidate regions in the prediction probability map, calculating the comprehensive score of each candidate region through the Soft-IoU loss function and the hierarchical shape loss function, and judging the tracking state of the target by the neural network model according to the comprehensive score of each candidate region and the feature degradation degree index.

[0008] Further, in S1, the specific steps of obtaining the feature enhancement map of the current frame by sequentially performing inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement and spatial position enhancement on the input image are as follows: Generating a blur mask based on the binary mask of the previous frame of real image; Differencing the real images of adjacent frames to generate several inter-frame difference channels; Performing gradient filtering in different directions on the current frame of real image to construct gradient channels; Normalizing the plane coordinates of each pixel in the current frame of real image to obtain the horizontal coordinate channel and the vertical coordinate channel; Performing channel splicing and convolution operation on the real image, the blur mask, the inter-frame difference channel, the gradient channel, the horizontal coordinate channel and the vertical coordinate channel to obtain the feature enhancement map of the current frame.

[0009] Further, in S1, the specific steps of sequentially performing down-sampling, feature perception and feature fusion on the feature enhancement map to output the prediction probability map of the current frame are as follows: Residual enhancement and down-sampling are performed on the feature enhancement map to generate shallow feature maps, middle layer first feature maps, middle layer second feature maps, deep layer first feature maps and deep layer second feature maps; Multi-scale hollow convolution is performed on the middle layer first feature map and the deep layer first feature map respectively to generate middle layer perception feature maps and deep layer perception feature maps; The middle layer secondary feature map and the deep layer perception feature map after upsampling are cross-gate fused to generate a first feature fusion map; The first feature fusion map and the middle layer perception feature map are cross-gate fused to generate a second feature fusion map; The second feature fusion map after upsampling and the deep layer secondary feature map after two-level upsampling are cross-gate fused to generate a third feature fusion map; The shallow layer feature map after residual enhancement and the third feature fusion map after two-level residual enhancement are cross-gate fused to generate a fourth feature fusion map; The fourth feature fusion map is subjected to two-level residual enhancement and convolution operation to obtain a prediction probability map.

[0010] Further, in S1, the residual enhancement includes residual convolution and gradient attention enhancement, the gradient attention enhancement includes channel attention enhancement and spatial attention enhancement, and the expression is:

[0011]

[0012]

[0013]

[0014] wherein, represents an enhanced object, represents a channel number, represents that is convolved and output as channels, represents a first batch normalization, represents a ReLU function activation, represents a second batch normalization, represents a residual convolution output result, represents an average pooling, represents a maximum pooling, represents a first full connection, represents a second full connection, represents a Sigmoid function activation, represents a channel attention enhancement result, represents channel splicing, represents a spatial attention enhancement result, represents a residual adjustment term, represents that is subjected to residual enhancement and output as channels.

[0015] Further, in S1, the expression of residual enhancement and down-sampling of the feature enhancement graph to generate shallow feature graph, middle first-level feature graph, middle second-level feature graph, deep first-level feature graph and deep second-level feature graph is as follows:

[0016]

[0017]

[0018]

[0019]

[0020] wherein, represents the feature enhancement graph, represents residual enhancement, represents two-dimensional maximum pooling, represents the shallow feature graph, represents the shallow channel number, represents the middle first-level feature graph, represents the middle second-level feature graph, represents the middle channel number, represents the deep first-level feature graph, represents the deep second-level feature graph, represents the deep channel number.

[0021] Further, the expression of cross-gate fusion is as follows:

[0022]

[0023]

[0024]

[0025] wherein, represents the high-resolution feature graph to be fused, represents the low-resolution feature graph to be fused, represents the projection channel number, represents 1x1 convolution, represents the high-resolution feature fusion result, represents the low-resolution feature fusion result, represents Sigmoid function activation, represents the gate coefficient, represents the feature after gate screening, represents channel splicing, This represents a 1×1 convolution with batch normalization and ReLU activation. This represents a feature fusion map.

[0026] Furthermore, in S2, the expression for the comprehensive loss function is:

[0027] in, , , These represent the Soft-IoU loss weights, energy loss weights, and layered shape loss weights, respectively. Indicates Soft-IoU loss. Indicates energy loss. Indicates layered shape loss, This indicates the overall loss.

[0028] Furthermore, in S2, the expression for the energy loss function is:

[0029]

[0030]

[0031] in, Represents pixel coordinates, Indicates the current frame's true image in grayscale value at that location The binary mask representing the real image of the current frame is in The value at ∑ represents the summation of pixel values ​​at all locations in the image. Represents normalized real energy. Indicates the current frame prediction probability map in The value at that location, This represents the normalized predicted energy. Indicates the smoothing term. This indicates energy loss.

[0032] Furthermore, in S2, the expression for the layered shape loss function is:

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] wherein, represents the value of the binary mask of the real image of the current frame at or the predicted probability of the current frame Figure Two after segmentation at , represents the normalized gray value of any pixel of the target region of the real image of the current frame, represents the area of the target region, represents the horizontal coordinate of the center of the target region, represents the vertical coordinate of the center of the target region, represents the normalized maximum radius, represents the normalized horizontal coordinate of the center, represents the normalized vertical coordinate of the center, represents the normalized radius of the pixel , represents the dimension sequence number of the feature vector, represents the th feature weight of the pixel with the normalized radius of , represents the th eigenvalue extracted using , represents the feature vector of the current frame extracted using , represents the th eigenvalue calculated using , represents the th eigenvalue calculated using , represents the relative difference degree of the th feature, represents the numerical stability term, respectively represent one-layer loss weight, two-layer loss weight, three-layer loss weight, four-layer loss weight, respectively represent one-layer shape loss, two-layer shape loss, three-layer shape loss, four-layer shape loss, represent hierarchical shape loss.

[0046] Further, the specific steps of S3 are as follows: The binaryzation processing and connected domain operation are performed on the prediction probability map to extract a plurality of candidate regions, and the expression is:

[0047] wherein, represents a binaryzation threshold, represents the prediction probability map of the current frame, represents the 8-neighbor connected operation, represents the candidate region, represents the area of the candidate region, represents the area threshold, represents the candidate region set; The comprehensive score of each candidate region is calculated by the Soft-IoU loss function and the hierarchical shape loss function, and the expression of the comprehensive score is:

[0048] wherein, respectively represent the Soft-IoU loss score weight and the hierarchical shape loss score weight, represents the target region in the prediction probability map of the previous frame, represents the Soft-IoU loss of represents the binary mask of the real image of the current frame, represents the hierarchical shape loss of represents the comprehensive score of the candidate region. When the target is not occluded, the hierarchical shape loss of at least one candidate region is less than or equal to the feature degradation degree index, and the candidate region with the minimum comprehensive score value is selected as the predicted target region; When the target is occluded, the hierarchical shape loss of all candidate regions is greater than the feature degradation degree index, and the neural network model is used to re-track the target until the hierarchical shape loss of at least one candidate region is less than or equal to the degradation degree index, and the candidate region with the minimum hierarchical shape loss value is selected as the predicted target region.

[0049] The beneficial effects of the present application include: ​​The application provides an infrared target anti-interference tracking method based on a feature segmentation tracking network, which includes three steps of constructing a neural network model, defining a constraint condition and judging a tracking state. First, in the construction process of the feature segmentation tracking network model, by introducing an attention mechanism and a multi-scale fusion module, the key information of the target can be accurately extracted, and the background interference can be significantly suppressed. The operations such as inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement and spatial position enhancement enable the network to identify and extract the detailed features of the target in a complex interference environment, instead of relying on only global features. This process greatly improves the distinguishability of the infrared target, especially in the case of complex background and blurred target, effectively avoids the confusion between the target and the background, and improves the tracking accuracy.

[0050] Secondly, a comprehensive loss function is adopted, which combines a Soft-IoU loss, an energy loss and a hierarchical shape loss, to constrain the segmentation result from three dimensions of overlap, energy and shape, thereby significantly improving the shape retention ability of the target, especially in the case of complex situations such as target occlusion or background mutation, the stable shape of the target can be maintained and the shape distortion can be effectively suppressed. Each item in the comprehensive loss function accurately optimizes the positioning accuracy and shape consistency of the target from different dimensions, ensuring that the target is not affected by external interference during the segmentation tracking process.

[0051] Finally, in the target tracking process, the accurate judgment of the target tracking state is realized through a comprehensive scoring mechanism and a feature degradation degree index. By scoring and matching multiple candidate regions, the system can quickly recover the target tracking based on the historical information of the target in the case of partial occlusion or loss of the target. Combined with the inter-frame continuity and shape feature matching, the system can not only accurately judge whether the target is lost, but also quickly recover the tracking after the target is lost, avoiding the target loss problem in the traditional method.

[0052] Overall, the method provided by the application adopts a lightweight network design, significantly optimizes the computing efficiency, and ensures the real-time performance of the method in practical applications. At the same time, through the hierarchical feature extraction and dynamic fusion mechanism, the loss function design and the tracking state judgment mechanism, the false alarm rate and the risk of target loss are reduced, the stability and efficiency of the network tracking are guaranteed, and the method exhibits excellent robustness and practicality in complex infrared scenes in the fields of military reconnaissance, security monitoring and the like, providing a new solution for infrared target tracking technology, and having wide practical application value and important market prospect. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure One A flowchart of an infrared target anti-interference tracking method based on a feature segmentation tracking network provided by the application; Figure TwoA framework diagram of an infrared target anti-interference tracking method based on a feature segmentation tracking network is provided in the present application. Figure Three A detailed architecture diagram of a feature segmentation tracking network is provided in the present application. Figure Four A feature segmentation tracking network training effect change diagram is provided. Figure Five An anti-suspected target interference tracking effect diagram is provided. Figure Six A target occlusion and successful re-identification recovery tracking effect diagram is provided. DETAILED DESCRIPTION

[0054] The present application is based on the following software and hardware environment and parameter configuration to fully verify the performance of the provided method in complex interference scenarios: 1) Parameter configuration Basic network parameters: the local patch size (Local_size) is set to 32 to ensure effective local feature extraction of infrared small targets; the dilate convolution kernel (dilate_kernel) is 3, which moderately expands the feature perception range while preserving the target core area; the Gaussian blur radius (blur_radius) is 3.0, which enhances the fault tolerance to small inter-frame displacements.

[0055] Training parameters: the batch size (batch_size) is 1, which is suitable for small sample infrared data training; the number of training rounds (num_epochs) is 20, the learning rate (lr) is 2e-3, which ensures that the model is fully converged and the training is stable.

[0056] Tracking parameters: the target area threshold (area_thresh) is set to 5 to filter out small interference areas at the pixel level and accurately extract effective target candidate regions.

[0057] 2) Data set description The data set used in the experiment is generated by modifying the target template of the VOT-TIR infrared tracking data set. Artificially set interference elements such as occlusion and suspected targets to simulate strong interference environments in complex battlefield, security, and other scenarios to more comprehensively test the performance of the proposed algorithm in infrared target anti-interference tracking tasks.

[0058] 3) Hardware environment The computer configuration relied on by the experiment is as follows: an 11th Gen Intel (R) Core (TM) i7-11800H @ 2.30GHz processor, 16GB of memory (15.8GB available), an NVIDIA GeForce RTX 3060Laptop GPU (6GB video memory) and Intel (R) UHD Graphics (128MB video memory), and storage composed of a 477GB SSD SAMSUNG MZVL2512HCJQ-00BL2 and a 954GB SSD Colorful CN700 1TB PLUS, which provide sufficient and high-speed storage support for data reading and writing and model training, ensuring efficient experiment implementation.

[0059] The present application provides an infrared target anti-interference tracking method based on a feature segmentation tracking network, aiming to solve the problems of target loss and tracking drift caused by target occlusion and background interference in infrared images. Figure One and Figure Two As shown, the method comprises the following steps: S1, constructing a neural network model: inputting the first several frames of real images containing the target region and the current frame of real images, sequentially performing inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement and spatial position enhancement on the input images to obtain the feature enhancement map of the current frame, and sequentially performing down-sampling, feature perception and feature fusion on the feature enhancement map to output the prediction probability map of the current frame.

[0060] Specifically, the constructed neural network model is a Feature Aggregation Segmentation Tracking Network (FASTNet) model, which includes an enhancement module, an encoding down-sampling module, a multi-scale perception module, an inter-layer fusion module, and an output module. After inputting several frames of real images containing target regions, the enhancement module performs a series of enhancement operations such as inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement, and spatial position enhancement to enhance the features of the input images, thereby solving the problems of weak target features and spatial position ambiguity in infrared images and significantly improving the target recognition ability and tracking accuracy. After feature enhancement, the encoding down-sampling module extracts low-level detail features and high-level semantic information of the enhanced feature map, realizing the integration of feature extraction, attention enhancement, and resolution reduction. Then, the multi-scale perception module extracts features of different scales to capture small-scale details and large-scale context, enhancing the network's ability to recognize targets of different sizes. Next, the inter-layer fusion module dynamically fuses features of different resolutions, enabling the network to more comprehensively understand the spatial structure of the target and maintain the continuity of information flow, ensuring stable tracking of the target under different scales and resolutions. Finally, the output module enhances the feature fusion map to output the prediction probability map of the current frame. The prediction probability map can accurately represent the position and shape of the target and provide high-quality input for subsequent segmentation and tracking.

[0061] In summary, the feature segmentation and tracking integrated model constructed in step one adapts to complex infrared scenes, realizes accurate segmentation and continuous tracking of targets, and adopts a lightweight design to ensure efficient computing performance, adapt to real-time tracking requirements, and can be widely applied in military reconnaissance, security monitoring, and other fields, and performs outstanding robustness and real-time performance in dynamic environments.

[0062] As shown in Figure Three , the specific steps of S1 are as follows: S11: After sequentially performing inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement, and spatial position enhancement on the input image, a feature enhanced map of the current frame is obtained, and the specific steps are as follows: S111, inter-frame attention enhancement: a blur mask is generated based on the binary mask of the previous frame real image as an inter-frame attention clue to construct a time sequence association. The blur mask generation process is as follows:

[0063] wherein, represents the binary mask of the previous frame real image, represents the maximum value pooling dilation, and in this application, the pooling kernel size is 3, denotes Gaussian blur, in this application, Gaussian blur radius is 3.0, denotes blur mask of previous frame real image.

[0064] S112, inter-frame difference enhancement: difference between real images of adjacent frames, generate several inter-frame difference channels as inter-frame gradient clues, further strengthen temporal correlation. Inter-frame difference channel generation process is:

[0065] wherein, and denote input real image of the first frame (previous several frames) and the first frame (current frame), in this application, real image size is 32x32, denotes frame order of input real image, denotes total frame number of input real image, denotes generated inter-frame difference channel of the first frame.

[0066] S113, gradient enhancement: gradient filtering of different directions on current frame real image, construct gradient channel as current frame gradient clue, construct spatial gradient attention perception. Gradient channel generation process is:

[0067] wherein, denotes current frame real image, denotes gradient operator direction, denotes gradient filtering of direction, denotes gradient channel of direction.

[0068] S114, spatial position enhancement: respectively normalize plane coordinates of each pixel in current frame real image to [-1,1] interval, get horizontal coordinate channel and vertical coordinate channel:

[0069] wherein, denotes pixel coordinate, H and W denote height and width of input real image, in this application, , denotes horizontal coordinate channel, denotes vertical coordinate channel.

[0070] S115, coordinate convolution: channel splicing is performed on the real image, the blur mask, the several inter-frame difference channels, the gradient channels, the horizontal coordinate channels and the vertical coordinate channels, so that the network explicitly learns the target space position features, and then the channel number is changed through a convolution operation, in the application, the channel number is changed to 16 through the convolution operation, and after batch normalization and activation, a feature enhancement map is obtained:

[0071]

[0072] wherein, represents the channel number, in the application, , represents channel splicing, represents that the convolution is performed and output as channels, represents the result after coordinate convolution, represents batch normalization, represents ReLU function activation, which avoids feature distribution deviation, represents the feature enhancement map.

[0073] Example: assuming that the total number of input real images is 3, and the 0002, 0003 and 0004 frames in the infrared sequence image are taken as the input real images, first, the target center coordinates of each frame of real image are calculated: the 0002 frame is (147, 39), the 0003 frame is (146, 38), and the 0004 frame is (145, 37), after the horizontal coordinate channel and the vertical coordinate channel of the 0004 frame real image are generated, the 0002, 0003 and 0004 frames of real image, the blur mask (based on the 0003 frame binary mask inflation + blur generation), the inter-frame difference channel (0003-0002 frame and 0004-0003 frame difference), and the four-direction gradient channel of the 0004 frame real image are spliced, and then the 0004 frame feature enhancement map with a channel number of 16 and a size of 32*32 is output through 3*3 convolution.

[0074] S12: residual enhancement and double-layer down-sampling are performed on the feature enhancement map to generate a shallow feature map, a middle-layer first feature map, a middle-layer second feature map, a deep-layer first feature map and a deep-layer second feature map, so as to realize the integration of feature extraction, attention strengthening and resolution reduction. The specific expression is:

[0075]

[0076]

[0077]

[0078]

[0079] wherein, denotes a feature enhancement map, denotes residual enhancement, denotes two-dimensional max pooling, denotes a shallow feature map, denotes a shallow channel number, denotes a middle first-level feature map, denotes a middle second-level feature map, denotes a middle channel number, denotes a deep first-level feature map, denotes a deep second-level feature map, denotes a deep channel number.

[0080] Residual enhancement includes residual convolution and gradient attention enhancement, and gradient attention enhancement includes channel attention enhancement and spatial attention enhancement, and the expression is:

[0081]

[0082]

[0083]

[0084] wherein, denotes an enhancement object, such as a feature enhancement map, a middle first-level feature map, or other feature maps that need to be enhanced by residual enhancement, denotes a channel number, denotes a convolution on and outputs channels, denotes a first batch normalization, denotes a ReLU function activation, denotes a second batch normalization, denotes a residual convolution output result, denotes average pooling, denotes max pooling, denotes a first full connection, denotes a second full connection, denotes a Sigmoid function activation, denotes a channel attention enhancement result, denotes channel splicing, denotes a spatial attention enhancement result, denotes a residual adjustment term, which is used when the input and output channels or resolutions are inconsistent, and is fitted by 1x1 convolution, denotes a convolution on Perform residual enhancement and output as One channel.

[0085] Example Description: Input the feature enhancement map with 16 channels and a size of 32×32 generated by S11 into the encoding downsampling module: First, a shallow feature map with a size of 32×32 is obtained through one residual enhancement; the shallow feature map is downsampled to 16×16 through two-dimensional max pooling, and then subjected to two residual enhancements to obtain a mid-level secondary feature map with 32 channels and a size of 16×16; the mid-level secondary feature map is downsampled to 8×8 through two-dimensional max pooling, and then subjected to two residual enhancements to obtain a deep secondary feature map with 48 channels and a size of 8×8, thus completing the encoding downsampling.

[0086] S13: The multi-scale perception module includes mid-level multi-scale dilated convolutions ( Perception module and deep multi-scale dilated convolution ( Perception module, Perception module and The perception module performs multi-scale dilated convolutions on the intermediate-level and deep-level feature maps, respectively, to generate intermediate-level and deep-level perception feature maps, capturing small-scale details and large-scale context. The specific expression is:

[0087]

[0088] in, Indicates the void ratio, use Capture small-scale target details. use Capture large-scale background context. Indicates the void ratio convolution, Indicates channel splicing. This indicates convolution with batch normalization and ReLU activation. This represents the mid-level perceptual feature map. This represents a deep perception feature map.

[0089] Example illustration: Application to the mid-level first-order feature map (32 channels, 16×16) generated by S12 The perception module generates 32-channel feature maps using 3×3 convolutions with dilation rates of 1, 2, and 4. These maps are then concatenated and projected using a 1×1 convolution to form a mid-level perception feature map (32 channels, 16×16). The deep first-level feature map (48 channels, 8×8) generated by S12 is then applied... Perception module: 32-channel feature map is generated by 3x3 convolution with hole rate 1, 2, and is projected into deep perception feature map (32 channels, 8x8) after splicing.

[0090] S14: Dynamic fusion of feature maps with different resolutions is realized based on the cross-gating fusion module. The expression of cross-gating fusion is:

[0091]

[0092]

[0093]

[0094] wherein, represents a high-resolution feature map that needs to be fused, represents a low-resolution feature map that needs to be fused, represents the number of projection channels, represents 1x1 convolution, represents a high-resolution feature fusion result, represents a low-resolution feature fusion result, represents Sigmoid function activation, used to compress the result to the interval [0, 1], represents a gating coefficient, used to reflect the correlation of feature maps with different resolutions. The closer the value of the gating coefficient to 1, the stronger the correlation of feature maps with different resolutions at this position, and the closer to 0, the weaker the correlation of feature maps with different resolutions, represents a feature after gating screening, represents channel splicing, represents 1x1 convolution plus batch normalization and ReLU activation, represents a feature fusion map.

[0095] In this application, four fusions need to be realized through the cross-gating fusion module, and the specific steps are as follows: S141: Cross-gating fusion is performed on the middle two-level feature map and the deep perception feature map after upsampling at H / 2 resolution to generate a first feature fusion map, and the expression is:

[0096] wherein, represents upsampling, represents cross-gating fusion, represents the first feature fusion map.

[0097] S142: Cross-gating fusion is performed on the first feature fusion map and the middle layer perception feature map at H / 2 resolution to generate a second feature fusion map, and the expression is:

[0098] wherein, represents the second feature fusion map.

[0099] S143: cross-gate fusion is performed on the up-sampled second feature fusion map and the two-level up-sampled deep second-level feature map at full resolution to generate a third feature fusion map, and the expression is:

[0100] wherein, represents the third feature fusion map.

[0101] S144: cross-gate fusion is performed on the residual-enhanced shallow feature map and the two-level residual-enhanced third feature fusion map at full resolution to generate a fourth feature fusion map, and the expression is:

[0102] wherein, represents two-level residual enhancement, and the output channel is C 1, represents the fourth feature fusion map.

[0103] Example: 1) H / 2 resolution fusion: the deep perception feature map generated by S13 is up-sampled, the middle second-level feature map (32 channels, 16x16) generated by S12 is input into the cross-gate fusion module with the up-sampled deep perception feature map (32 channels, 16x16), to generate a gating coefficient (16x16), and after weighting, a first feature fusion map (32 channels, 16x16) is obtained; 2) full resolution fusion: the deep second-level feature map generated by S12 is two-level up-sampled, the up-sampled second feature fusion map (32 channels, 32x32) is fused with the two-level up-sampled deep second-level feature map (32 channels, 32x32) to obtain a third feature fusion map (32 channels, 32x32); the shallow feature map generated by S12 is residual-enhanced, and the two-level residual-enhanced third feature fusion map (16 channels, 32x32) is fused with the residual-enhanced shallow feature map (16 channels, 32x32) to obtain a fourth feature fusion map (16 channels, 32x32).

[0104] S15: two-level residual enhancement and convolution operation are performed on the fourth feature fusion map to obtain a prediction probability map, and the expression is:

[0105]

[0106] wherein, denotes the result of two-stage residual enhancement on the fourth feature fusion map, denotes 1x1 convolution, denotes Sigmoid function activation, denotes the prediction probability map of the current frame, the size is consistent with the input real image, , denotes the pixel confidence of the target area.

[0107] Instance description: the fourth feature fusion map (16 channels, 32x32) generated by S14 is subjected to two-stage residual enhancement, and then subjected to 1x1 convolution and Sigmoid to output the prediction probability map (32x32) of the current frame. Taking frame 0005 as an example, the prediction probability Figure Two value of frame 0005 is segmented into a target area of 30, and the global center coordinates are (143.5, 37.66666666666667), which is consistent with the segmentation and tracking result of the real image of frame 0005.

[0108] S2, define the constraint condition: generate a comprehensive loss function according to the Soft-IoU loss function, the energy loss function and the hierarchical shape loss function, and the comprehensive loss function is used to constrain the prediction probability map from the three dimensions of overlap, energy and shape.

[0109] Specifically, on the basis of the feature segmentation and tracking network (FASTNet) constructed in S1, the training constraints set by S2 constrain the entire learning process in three dimensions of "overlap- energy- shape", forming a collaborative convergence drive for the prediction probability map. The constraint system takes the comprehensive loss function as the core: reasonably weighting the Soft-IoU loss, the energy loss and the hierarchical shape loss in the same objective function, so that the network synchronously optimizes the three key indicators of region overlap, area consistency and shape consistency in each forward-backward iteration. The logical relationship of the three parallel constraints is clear: Soft-IoU faces the optimization of the intersection over union at the pixel set level, solves the root cause of "inaccurate segmentation", the energy term faces the area registration of the overall scale of the target, resolves the scale mismatch of "area drift", and the hierarchical shape term faces the shape fidelity of the geometric structure, relieves the geometric instability of "shape distortion". The three dimensions are complementary and balanced with each other, avoiding the "partial performance" convergence of the model relying on a single indicator, and jointly constraining the robust and continuous tracking of the network in complex dynamic environments.

[0110] Further, in S2, the expression of the comprehensive loss function is:

[0111] wherein, , , respectively, are the Soft-IoU loss weight, the energy loss weight, and the hierarchical shape loss weight, which are optimized by experiments as = 0.4, , , Soft-IoU loss is used to optimize the overall overlap degree, Energy loss is used to constrain the target energy, Hierarchical shape loss is used to constrain the target shape structure, Comprehensive loss is used.

[0112] Further, in S2, the Soft-IoU loss function is a commonly used segmentation loss function, which can solve the problem of discrete and non-derivable of traditional IoU loss, and directly optimize the intersection over union index. The expression of the Soft-IoU loss function is:

[0113] wherein, represents the pixel coordinate, represents the value of the binary mask of the real image of the current frame at represents the target pixel, represents the background pixel, represents the value of the predicted probability map of the current frame at ∑ represents the sum of pixel values of all positions of the image, represents the smoothing term, which avoids the denominator being zero, and in the present application, , , represents the Soft-IoU loss, The value range of is [0, 1], The smaller the value is, the higher the overlap degree between the predicted mask and the real mask is.

[0114] Example: as shown in Figure Four , the Soft-IoU loss of the FASTNet model constructed in S1 gradually decreases with iterations during the 20 rounds of training: Epoch 1 is 0.9577, Epoch 10 is reduced to 0.5517, and Epoch 20 converges to 0.0174, indicating that the overlap degree between the predicted mask and the real mask gradually increases from low to 98.26% (1-0.0174), and the segmentation accuracy is significantly optimized.

[0115] Further, in S2, the expression of the energy loss function is:

[0116]

[0117]

[0118] wherein, represents the gray value of the current frame real image at represents the normalized real energy, represents the normalized predicted energy, represents the energy loss, the value range of is [0, 1], The smaller the value is, the closer the predicted area is to the real area, which can effectively improve the area prediction accuracy in the occlusion scene.

[0119] Example: The energy loss of the FASTNet model constructed by S1 gradually converges in the training process: Epoch 1 is 0.8670 (the difference between the predicted area and the real area is large), Epoch 8 is reduced to 0.0566, Epoch 10 and later is stable at 0 (the predicted area and the real area are completely matched); In the tracking stage, the target area of 0005-0022 frames is stable at 17-30 (such as 0005 frame 30, 0010 frame 28, and 0022 frame 17), and the energy loss is less than 0.1, indicating that the area prediction accuracy meets the requirements.

[0120] Further, in S2, the hierarchical shape loss function first extracts the feature vector of the infrared target in the discrete digital image through the steps of gray normalization, center distance normalization and feature vector extraction, and different dimensions of the feature vector represent different center level quantization features of the target; Then, the relative difference measurement method is used to calculate the feature difference; Finally, according to the different center levels of the target, a hierarchical weight structure is designed, and the hierarchical shape loss is calculated. The hierarchical shape loss combined with the pyramid weight distribution mechanism can maintain the overall structure of the shape while considering the local detail features, effectively improving the target shape preservation ability in complex scenes. The expression is:

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128] ​

[0129]

[0130]

[0131]

[0132]

[0133] wherein, represents the value of the binary mask of the real image of the current frame at or the predicted probability of the current frame Figure Two after segmentation at , represents the normalized gray value of any pixel of the target region of the real image of the current frame, represents the area of the target region, represents the horizontal coordinate of the center of the target region, represents the vertical coordinate of the center of the target region, represents the normalized maximum radius, represents the normalized horizontal coordinate of the center, represents the normalized vertical coordinate of the center, represents the normalized radius of the pixel , represents the dimension sequence number of the feature vector, represents the th feature weight of the pixel with the normalized radius of , represents the th eigenvalue extracted using , represents the feature vector of the current frame extracted using , represents the th eigenvalue calculated using , represents the th eigenvalue calculated using , represents the relative difference degree of the th feature, represents the numerical stability term, in this application, , respectively represent the loss weight of the first layer, the loss weight of the second layer, the loss weight of the third layer, and the loss weight of the fourth layer, respectively represent the shape loss of the first layer, describing the overall uniform feature difference of the target, the shape loss of the second layer, describing the internal feature difference under the quadratic function distribution of the target, the shape loss of the third layer, describing the peripheral feature difference under the cubic function distribution of the target, and the shape loss of the fourth layer, describing the edge feature difference under the quartic function distribution of the target, The layered shape loss is represented. The smaller the value is, the higher the consistency of the predicted shape with the real shape is.

[0134] Example: Taking frame 0076 as an example, the feature vector extracted from the binary mask of the real image of frame 0076 is:

[0135] The predicted probability of frame 0076 is Figure Two The feature vector extracted after value segmentation is:

[0136] The relative difference degree is calculated: Further, the layered shape loss of each layer is calculated: The total layered shape loss indicates that the predicted shape is highly consistent with the real shape. During the training process, the layered shape loss decreases from 1.0802 at Epoch 1 to 0.0288 at Epoch 10, and finally converges to 0, verifying the effectiveness of the shape constraint.

[0137] S3, determining the tracking state: extracting a plurality of candidate regions in the predicted probability map, calculating the comprehensive score of each candidate region through the Soft-IoU loss function and the layered shape loss function, and determining the tracking state of the target by the neural network model according to the comprehensive score of each candidate region and the feature degradation degree index.

[0138] Specifically, based on the predicted probability map output in S1, through the closed loop of “candidate generation-reliability measurement-state determination-history update / re-identification”, the optimal target of the current frame can be stably given under the conditions of multiple candidates, occlusion and strong interference, and the temporal prior is updated accordingly. The core idea is: to establish the posterior comprehensive score with Soft-IoU and layered shape loss in S2 as the main axis, and to determine the state in combination with the area constraint, so that the spatial appearance consistency, geometric shape consistency and temporal continuity can work together in the same evaluation framework, and under the premise of maintaining real-time, higher target distinguishability and trajectory consistency can be obtained. The performance improvement of the four key problems of “inaccurate segmentation, area drift, shape distortion and occlusion loss” is systematic, interpretable and verifiable.

[0139] The specific steps of S3 are as follows: ​​​​​​​​​​S31: As shown in Figure Five , the prediction probability map is binarized and connected domain operation is performed to extract several candidate regions. When the area of a candidate region is smaller than the area threshold, it is judged as low confidence interference and directly removed. When the area of a candidate region is larger than the area threshold, the candidate region is retained to enter the subsequent scoring and measurement link, which can reduce the scoring calculation overhead and avoid introducing obvious false targets into the decision. The expression is:

[0140] , wherein, represents the binarization threshold, represents the prediction probability map of the current frame, represents the 8-neighbor connected operation, represents the candidate region, represents the area of the candidate region, represents the area threshold, represents the candidate region set.

[0141] Example: 1) normal tracking frame (such as 0005-0022 frame): the prediction probability Figure Two of the 0005-0022 frame is binarized to extract only one candidate region with an area of 17-30 (all larger than the area threshold), which is directly used as the predicted target region; 2) occlusion frame (such as 0023 frame): the prediction probability Figure Two of the 0023 frame is binarized to obtain a candidate region with an area of 3 (smaller than the area threshold), and after filtering, there is no effective candidate region, triggering the occlusion judgment of S33; 3) multiple interference frame (such as 0076 frame): the prediction probability Figure Two of the 0076 frame is binarized and connected domain analysis can extract three candidate regions with areas of 24, 8, and 10. After filtering the interference with an area smaller than 5, three candidate regions ( ) are retained to enter the posterior scoring stage.

[0142] S32: The comprehensive score of each candidate region is calculated by the Soft-IoU loss function and the hierarchical shape loss function, and the matching degree of each candidate region with the historical target is evaluated. The expression of the comprehensive score is:

[0143] , wherein, respectively represent the Soft-IoU loss score weight and the hierarchical shape loss score weight, represents the target region in the prediction probability map of the previous frame tracking success, represents the Soft-IoU loss of , which is used to measure the inter-frame continuity, and the calculation process is shown in S2, represents the binary mask of the real image of the current frame, denotes the hierarchical shape loss of , for evaluating shape consistency, see S2, denotes the comprehensive score of the candidate region, with a value range of [0, 2], The smaller the value, the higher the matching degree of the candidate region with the historical target. Example: Take frame 0076 as an example, the posterior comprehensive score of the three candidate regions is calculated as follows: Candidate region (area 24, target center coordinates (138.625, 96.043)): , , comprehensive score ; Candidate region (area 8, target center coordinates (139.210, 95.112)): , , comprehensive score ; Candidate region (area 10, target center coordinates (137.982, 96.875)): , , comprehensive score .

[0144] The candidate region with the lowest comprehensive score is selected as the predicted target region of frame 0076, i.e. "Selected candidate 1 with score 0.8504", ensuring the continuity and accuracy of the tracking target.

[0145] S33: According to the comprehensive score and the feature degradation degree index, judge the current tracking state and select the optimal candidate region. As shown in Figure Five , in the normal tracking state, when there are , among them, is the feature degradation threshold, the candidate region with the smallest value is selected as the tracking result of the current frame , and the historical target information is updated.

[0146] As shown in Figure Six , when all candidate regions , it is determined that the target is occluded or lost. At this time, the scanning and re-identification mechanism is triggered, and the center coordinates of the last successful tracking target are recorded. And target verification and accurate positioning: after triggering the re-identification mechanism, take as the center, and preset the search radius sliding window scan is performed in the range, and each scan window size is the same as the network training input real image. For each scan window , the input FASTNet network is segmented for calculation to obtain the segmentation mask of each scan window , the current frame feature vector extracted using is (method same as S2), the feature vector extracted using the binary mask of the current frame real image is (method same as S2), and the hierarchical shape loss (method same as S2) is calculated. After each scan calculation, re-identification target selection is performed: among all the scan window segmentation results, the segmentation mask with the smallest hierarchical shape loss is selected as the re-identification target:

[0147] and needs to meet . If there is a candidate area meeting the requirements, the re-identification is successful, the system updates the tracking state to normal tracking, sets , and calculates the global coordinates according to the position in the scan window to complete target relocalization. If no re-identification target meeting the conditions is found in the search range, it is determined that the target is lost continuously, and the next frame continues to execute the re-identification process.

[0148] The re-identification mechanism performs fine search in the target lost area through shape feature matching, effectively solves the problem of re-capture after temporary occlusion or disappearance of the target, and significantly improves the robustness and persistence of the tracking system in complex scenes.

[0149] Example: 1) occlusion trigger: the area of the candidate region in frame 0023 is 3<5, and , it is determined that the target is occluded, the center coordinates of the target in the previous frame prediction probability map (frame 0022 target center coordinates) are recorded, and the scan re-identification is triggered; frames 0024-0035 do not find a target meeting the conditions (the areas of the candidate regions in the scan window are all <5 or ), and output “area=0, pos=None”; 2) re-identification success: the scan window offset by 36 pixels (center (62.837, 24.705)) in frame 0036, the segmentation mask area is 17>5, , and the hierarchical shape loss is the smallest, which is the re-identification target global coordinates (62.837, 24.705), output "(scanning) Found at offset 36: area=17, global=(62.83747276688449, 24.705045665775877)"; 3) Recovery of tracking: from frame 0037 and later, the tracking results are stable (e.g., frame 0037 candidate area 27, target center coordinates (60.356, 24.372), frame 0038 area 28, center (56.856, 25.943)), All <0.8, return to normal tracking state.

[0150] Based on the above, the specific data and performance of the training and tracking process in the examples, the infrared target anti-jamming tracking method based on the feature segmentation tracking network (FASTNet) provided in the application exhibits excellent performance in the three core indicators of precision, robustness, and real-time performance, fully meeting the needs of infrared target tracking in complex interference scenarios.

[0151] In terms of training accuracy, each component loss of the comprehensive loss function achieves stable convergence: the Soft-IoU loss gradually decreases from 0.9577 in the initial Epoch 1 to 0.0174 in Epoch 20, meaning that the overlap degree of the predicted mask and the real mask increases from less than 5% to 98.26%; the energy loss is optimized from 0.8670 in Epoch 1 to 0 in Epoch 10 and later, indicating that the target area prediction error is completely eliminated; the hierarchical shape feature loss converges from 1.0802 in Epoch 1 to 0, proving that the network accurately learns the shape structure features of the target (e.g., the difference between the predicted shape and the real shape in frame 0076 is only 0.00022). The extreme convergence of the three types of losses fully verifies the network's full-coverage accurate segmentation capability for the "overlap-area-shape" of the infrared target.

[0152] From the perspective of tracking robustness, the network performs particularly well in complex scenarios: in the normal tracking stage (frames 0005-0022), the target area is stable at 17-30 pixels (matching the true target area of 30), and the global center coordinates smoothly transition from (143.5, 37.67) to (111.6, 26.23), fully adhering to the actual target motion trajectory; in the occlusion scenario (occlusion triggered at frame 0023 and recovered at frame 0036), even though there are 12 frames without valid targets (output area = 0), the scanning and re-identification mechanism can still accurately locate the target at a 36-pixel offset (area = 17, global coordinates (62.84, 24.71)); in the multi-interference scenario (frame 0076), the true target is accurately selected from 3 candidate regions (scores 0.8504, 3.2158, 2.7276), avoiding tracking drift caused by background interference.

[0153] In terms of real-time performance, relying on lightweight network design (input 32x32 local patch, channel number 16-48, no redundant modules), the actual running speed on the experimental hardware (i7-11800H + RTX 3060 Laptop GPU) is above 25fps, meeting the engineering requirements of the infrared tracking system for "real-time response" and avoiding the lag or target loss problems caused by the computational complexity of traditional segmentation and tracking algorithms.

[0154] In summary, the infrared target anti-interference tracking method based on the segmentation and tracking network architecture designed in this application has excellent performance, not only solving the pain points of infrared target occlusion sensitivity and drift in complex interference scenarios, but also ensuring real-time performance, fully verifying its practicality and superiority in military reconnaissance, security monitoring and other practical applications, and providing an efficient and reliable technical solution for infrared target anti-interference tracking.

[0155] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An infrared target anti-jamming tracking method based on a feature segmentation tracking network, characterized in that, Includes the following steps: S1, Construct a neural network model: Input the real images of the previous few frames containing the target region and the real image of the current frame. After performing inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement and spatial location enhancement on the input image in sequence, the feature enhancement map of the current frame is obtained. The feature enhancement map is then subjected to downsampling, feature perception and feature fusion in sequence, and the predicted probability map of the current frame is output. S2, Define constraints: Generate a comprehensive loss function based on the Soft-IoU loss function, energy loss function and hierarchical shape loss function. The comprehensive loss function is used to constrain the prediction probability map from three dimensions: overlap, energy and shape. S3, Determine the tracking status: Extract several candidate regions from the predicted probability map, calculate the comprehensive score of each candidate region using the Soft-IoU loss function and the hierarchical shape loss function, and determine the tracking status of the neural network model on the target based on the comprehensive score of each candidate region and the feature degradation index.

2. The method according to claim 1, characterized in that, In step S1, the specific steps for obtaining the feature enhancement map of the current frame by sequentially performing inter-frame attention enhancement, inter-frame difference enhancement, gradient enhancement, and spatial location enhancement on the input image are as follows: Generate a blur mask based on the binary mask of the previous frame's real image; Subtract the real images from adjacent frames to generate several inter-frame difference channels; Gradient filtering is applied to the real image of the current frame in different directions to construct gradient channels; The planar coordinates of each pixel in the current frame's real image are normalized to obtain the horizontal and vertical coordinate channels. The feature enhancement map of the current frame is obtained by performing channel concatenation and convolution operations on the real image, the blur mask, the inter-frame difference channel, the gradient channel, the horizontal axis channel, and the vertical axis channel.

3. The method according to claim 1 or 2, characterized in that, In step S1, the specific steps of performing downsampling, feature perception, and feature fusion on the feature enhancement map in sequence to output the prediction probability map of the current frame are as follows: Residual enhancement and downsampling are performed on the feature enhancement map to generate shallow feature map, mid-level first-class feature map, mid-level second-class feature map, deep first-class feature map, and deep second-class feature map; Multi-scale dilated convolutions are performed on the intermediate-level first-class feature maps and the deep-level first-class feature maps respectively to generate intermediate-level perceptual feature maps and deep-level perceptual feature maps; Cross-gated fusion is performed on the intermediate level two feature map and the upsampled deep perception feature map to generate the first feature fusion map; The first feature fusion map and the middle-layer perception feature map are cross-gated and fused to generate the second feature fusion map. Cross-gated fusion is performed on the upsampled second feature fusion map and the deep second-level feature map after two levels of upsampling to generate the third feature fusion map; Cross-gated fusion is performed on the shallow feature map after residual enhancement and the third feature fusion map after two-level residual enhancement to generate a fourth feature fusion map. Two-stage residual enhancement and convolution operations are performed on the fourth feature fusion map to obtain the prediction probability map.

4. The method according to claim 3, characterized in that, In S1, residual enhancement includes residual convolution and gradient attention enhancement. Gradient attention enhancement includes channel attention enhancement and spatial attention enhancement, expressed as follows: in, Indicates an enhanced object, Indicates the number of channels. Indicates to Perform convolution and output as One channel, This indicates the first batch of normalization. Indicates ReLU function activation. This indicates the second batch of normalization. This represents the output result of the residual convolution. Indicates average pooling. This indicates max pooling. This indicates the first full connection. This indicates a second full connection. This indicates that the Sigmoid function is activated. This indicates the results of channel attention enhancement. Indicates channel splicing. This indicates the results of spatial attention enhancement. This represents the residual adjustment term. Indicates to Perform residual enhancement and output as One channel.

5. The method according to claim 3, characterized in that, In S1, the expression for generating shallow feature maps, intermediate first-level feature maps, intermediate second-level feature maps, deep first-level feature maps, and deep second-level feature maps by performing residual enhancement and downsampling on the feature enhancement map is as follows: in, Represents the feature enhancement map, This indicates that the residual has been enhanced. This represents two-dimensional max pooling. This represents a shallow feature map. Indicates the number of shallow channels. This represents the first-level feature map in the middle layer. This represents the second-level feature map of the middle layer. Indicates the number of middle-layer channels. This represents the deep first-level feature map. This represents a deep secondary feature map. This indicates the number of deep channels.

6. The method according to claim 3, characterized in that, The expression for cross-gated fusion is: in, This represents the high-resolution feature maps that need to be fused. This represents the low-resolution feature maps that need to be fused. Indicates the number of projection channels. Represents a 1×1 convolution. This indicates the result of high-resolution feature fusion. This represents the result of low-resolution feature fusion. This indicates that the Sigmoid function is activated. Indicates the gating coefficient. This indicates the features after gating. Indicates channel splicing. This represents a 1×1 convolution with batch normalization and ReLU activation. This represents a feature fusion map.

7. The method according to claim 1 or 2, characterized in that, In S2, the expression for the comprehensive loss function is: in, , , These represent the Soft-IoU loss weights, energy loss weights, and layered shape loss weights, respectively. Indicates Soft-IoU loss. Indicates energy loss. Indicates layered shape loss, This indicates the overall loss.

8. The method according to claim 1 or 2, characterized in that, In S2, the expression for the energy loss function is: in, Represents pixel coordinates, Indicates the current frame's true image in grayscale value at that location The binary mask representing the real image of the current frame is in The value at ∑ represents the summation of pixel values ​​at all locations in the image. Represents normalized real energy. Indicates the current frame prediction probability map in The value at that location, This represents the normalized predicted energy. Indicates the smoothing term. This indicates energy loss.

9. The method according to claim 8, characterized in that, In S2, the expression for the layered shape loss function is: in, The binary mask representing the real image of the current frame is in The value at that location or the current frame prediction probability map after binary segmentation The value at that location, This represents the normalized grayscale value of any pixel in the target region of the current frame's real image. Indicates the area of ​​the target region. Indicates the x-coordinate of the center of the target area. Represents the ordinate of the center of the target area. Represents the normalized maximum radius. Represents the center-normalized x-coordinate, Represents the center-normalized ordinate. Represents pixels normalized radius, Indicates the feature vector dimension index. The normalized radius is The first pixel Each feature weight, Indicates use The extracted first eigenvalues ​​of order 1 Indicates use Extracted feature vector of the current frame, Indicates use The calculated first eigenvalues ​​of order 1 Indicates use The calculated first eigenvalues ​​of order 1 Indicates the first Relative difference of each feature Represents the numerically stable term. These represent the loss weights for the first, second, third, and fourth layers, respectively. These represent shape loss at one layer, two layers, three layers, and four layers, respectively. This indicates the loss of layered shape.

10. The method according to claim 9, characterized in that, The specific steps of S3 are as follows: Binarization and connected component operations are performed on the predicted probability map to extract several candidate regions, as shown in the expression: in, Indicates the binarization threshold. This represents the predicted probability map for the current frame. This indicates an 8-neighbor connectivity operation. Indicates the first Candidate regions, Indicates the area of ​​the candidate region. Indicates the area threshold. Represents the set of candidate regions; The comprehensive score for each candidate region is calculated using the Soft-IoU loss function and the hierarchical shape loss function. The expression for the comprehensive score is as follows: in, These represent the weights of the Soft-IoU loss score and the weights of the hierarchical shape loss score, respectively. This represents the target region in the prediction probability map of the previous frame. express Soft-IoU loss, Represents the binary mask of the real image in the current frame. express Layered shape loss, This represents the overall score of the candidate region; When the target is not occluded, the hierarchical shape loss of at least one candidate region is less than or equal to the feature degradation index, and the candidate region with the smallest comprehensive score is selected as the predicted target region. When the target is occluded, the layered shape loss of all candidate regions is greater than the feature degradation index. A neural network model is used to re-track the target until the layered shape loss of at least one candidate region is less than or equal to the degradation index. The candidate region with the smallest layered shape loss value is selected as the predicted target region.

Images