Research On Target Detection And Tracking Algorithm In Forward-Looking Sonar Image Based On Deep Learning

With the national strategy shift,more and more attention has been paid for the defense of marine security and the exploitation of marine resources.Forward-looking sonar(FLS)plays a great role with its excellent ocean exploration performance.Therefore,the research on the application of target detection and tracking technology in FLS image is particularly important.There are some problems with FLS image,such as low resolution,few effective features,nonrigid deformation of target and interference of highlighted background.These problems bring challenges to the use of target detection and tracking in the field.Compared with traditional detection and tracking algorithms,deep learning algorithm has better feature extraction ability,which provides a new idea to improve the performance of target detection and tracking algorithm in FLS image.Therefore,this paper studies the FLS image augmentation algorithm and target detection and tracking algorithm based on deep learning.According to the problems that FLS images affect the detection and tracking performance,some solutions are proposed.Firstly,the augmentation technology and its influence on the target detection performance of FLS image are studied.Two FLS image augmentation methods called pole rotation and random filling is proposed,and a FLS image augmentation technology scheme based on generative adversarial network is applied,which enhances the generalization ability of FLS image target detection and tracking algorithm.Secondly,the general framework of target detection algorithm based on deep learning is studied.A backbone network based on dilated convolution is designed.The pool mode of neck network in detection algorithm and the overall architecture of classification regression network is improved.Experiments show that the algorithm solves the problem of detection failure due to low resolution of FLS image and non-rigid deformation of target.Then,the basic framework of target tracking algorithm based on deep learning is studied.The adaptive strategy of attention mechanism is improved and applied to the siamese networks.The attention mechanism is also used to improve the cross-correlation network,and the classification and central quality branches are combined to improve the performance of the network.Experiments show that the algorithm can track the target more accurately.Finally,the detection and tracking system is studied and built,which can judge whether the system needs to be corrected with tracking quality score.Experiments show that the system can ensure the stability of target tracking.