| As transportation in the Yangtze River Basin and nearshore areas of China increases,ship target monitoring has become increasingly important.In the field of target tracking,different sensors produce different forms of data.Compared to other sensors,radar has the advantages of long detection range and less susceptibility to environmental factors.Ship monitoring using radar data has always been a hot research topic and has high research value.This article conducts experiments on real radar X-band ship data and carries out research on multiexpansion target detection and tracking methods.The following work has been done:1.To address the problem of track drift or loss in the case of long-term target occlusion,close proximity to the target,and high clutter radar data using multi-kernel correlation filtering,a multi-extended target optimization tracking algorithm based on multi-kernel correlation filtering is proposed.The algorithm uses a Kalman filter that optimizes speed information for prediction,proposes a fusion intersection-over-union(IOU)correlation detection method,and completes adaptive recognition of target states,followed by three-step trajectory estimation based on the target state.In addition,in the case of high clutter radar data,where it is difficult to create a good target template using multi-kernel correlation filtering,a template optimization method is proposed.The template creation time is considered as an adaptive weight,and the template fusion is optimized from the perspective of maximum likelihood,thereby improving tracking accuracy and re-identifying lost targets.Experimental results show that the proposed algorithm has good tracking performance and robustness for X-band radar targets.2.The traditional handcrafted feature extraction methods have significant limitations as they need to consider numerous invariance factors,and the discriminative power of handcrafted features is weak.Additionally,the original siamese network with only one output layer is not suitable for tracking small targets in X-band radar,nor is it suitable for extended target tracking.To address these issues,an improved siamese network X-band radar multi-extended target optimization tracking algorithm is proposed.The algorithm proposes a bidirectional velocity prediction that combines microscopic and macroscopic velocities,which,along with distance correlation,identifies the target state.An adaptive feature extraction module is proposed to adaptively complete feature extraction from candidate regions.The output is processed through three convolution layers,and the results of three cross-correlations are fused to obtain the optimal target position prediction.Additionally,a criterion for judging the intersection-overunion of the width-to-height ratio is proposed,which,combined with the detection algorithm,estimates the shape of the target and updates the template.The experimental results show that the proposed algorithm has good tracking accuracy,can handle target tracking problems in Xband radar,and has good appearance accuracy.3.Aiming at the problems of weak targets in radar data that are difficult to detect,easy to miss,and the traditional detection method is not suitable for radar weak targets and low tracking accuracy of weak targets,an optimized detection and tracking algorithm based on Fourier transform radar is proposed.A threshold detection algorithm combined with two-dimensional Fourier transform is proposed to complete the denoising process in the original weak target data to obtain an accurate detection set,and on the basis of the optimized multi-kernel correlation filtering algorithm,an optimized multi-kernel correlation filter pre-detection tracking algorithm is proposed,combined with the idea of sub-problem division and greedy algorithm,to complete the detection and tracking of weak targets.Simulation experiments show that the proposed Fourier transition threshold detection algorithm has a good detection effect,and the use of lowpass filtering can effectively denoise radar weak targets,and the proposed optimized multi-core correlation filter pre-detection tracking algorithm has high tracking accuracy and false heel number in the radar weak target simulation data with SNR above 1,and has high robustness. |
