Research On Human Target Detection And Tracking Algorithm On The Sea Surface Based On Deep Learnin

There is a vast ocean with rich resources in China.With the in-depth exploration of the ocean and the continuous development of marine resources,the technique of target detection and tracking on the sea surface is widely used in military and civilian fields.Frequent maritime accidents are caused due to the increase of maritime activities so that it is particularly important to search and rescue human.Therefore,it is of great significance to study the technique of human target detection and tracking on the sea surface.In this thesis,the radar echo image is processed based on deep learning algorithm,and a series of studies on the technique of human target detection and multi-target tracking is briefly summarized as follows:1.Aiming at the characteristics of complex sea background,strong sea clutter,low signalto-noise ratio(SNR)and small human target,the sea echo model is established.Considering the dynamic change of sea background and the distribution of sea clutter,the distribution of sea clutter is simulated.And it is transformed into images that the radar echo data with the echo of human targets in different SNR and signal-to-clutter ratio(SCR).According to the universality and randomness of data,the dataset of sea human target detection in complex background is produced.At the same time,the motion model of human target is established.Combined with the characteristics of sea clutter and different motion states,multiple radar echo images are generated and the dataset of sea human target tracking is produced.2.In this thesis,a human target detection method based on deep learning in low SNR and strong sea clutter is proposed.The Faster R-CNN is improved.Due to the characteristics of small targets and clutter occlusion,multi-scale feature fusion and context information fusion are carried out.Then,the dataset is trained and tested.The performance verification and analysis are carried out,and there is a comparison with the Constant False-Alarm Rate(CFAR)algorithm based on statistical characteristics,and the Support Vector Machine(SVM)algorithm based on machine learning.The results show that the target detection accuracy of the proposed algorithm is greatly improved,and it has great advantages for the detection of human targets in the complex background of the sea surface.3.On the basis of the human target detection algorithm,the tracking of human targets is realized.Firstly,the Meanshift algorithm is used to track the single human target.Then,a multitarget tracking method based on the detection is proposed to solve the problems of low SNR,clutter interference and the less difference of human target features in radar echo images.The proposed method combines the improved Faster R-CNN algorithm with data association and adds the Kalman filter and Hungary algorithm.It is shown that the accuracy of target association between frames is improved and the tracking accuracy is greatly increased.Compared with the original algorithm,the results show that the proposed algorithm can perform high-precision multi-target tracking and maintain high tracking stability.