Research On Underwater Target Recognition And Tracking Technology Based On Deep Learning

The ocean is rich in resources.With the increasing depletion of land resources,human beings are increasingly dependent on marine resources.Therefore,as one of the important tools of marine resources,the development of underwater robot technology becomes more and more important.Compared with the sonar sensor,the optical vision system has a higher resolution in the underwater short-range operation,so that the underwater robot can perceive the external environment more intuitively.Therefore,the underwater machine vision technology has a very broad application prospect and research value.Based on the underwater target recognition and tracking task performed by underwater robots,this paper proposes a new underwater image enhancement method,and improves the underwater target recognition and tracking algorithm based on the deep learning network.The specific research work of my paper is as follows:Firstly,starting from the analysis of the imaging principle of the camera,combined with the characteristics of underwater imaging,the imaging model of the underwater binocular camera is gradually established,and the solution method for the internal and external parameters of the binocular camera is derived.The internal and external parameters of the camera are solved through the calibration experiment of the underwater binocular camera.Aiming at the problems of blurred and blue-green partial underwater optical images caused by uneven lighting in the underwater environment,dispersion and absorption,etc.,an improved MSRCR algorithm is proposed,which incorporates bilateral filtering,channel difference enhancement,gamma correction and other methods to improve The overall performance of the algorithm.Through comparative experiments,the effectiveness of the algorithm is verified by subjective and objective evaluation methods.In view of the difficulty of traditional recognition algorithms to recognize multiple targets in the marine environment,an underwater target recognition algorithm based on improved YOLOv5 is proposed.It is proposed to replace part of conventional convolution with deep separable convolution to increase speed and improve mosaic data enhancement.Algorithm to improve accuracy.By classifying and marking the collected underwater biological pictures,a detailed and sufficient underwater biological target data set has been established.Finally,the improved model is trained with the built data set.The experimental results show that the improved algorithm in this paper improves the speed and accuracy of underwater target detection.Aiming at the difficulty of traditional recognition algorithms to recognize various targets in the marine environment,an underwater target recognition algorithm based on improved YOLOv5 is proposed.It is proposed to replace some conventional convolutions with depthwise separable convolutions to improve speed and improve mosaic data enhancement.algorithm to improve accuracy.Finally,the improved model is trained with the built data set.The experimental results show that the improved algorithm in this paper improves the speed and accuracy of underwater target detection.Finally,since the traditional target tracking algorithm is difficult to apply to the complex underwater environment,this paper adopts the deep learning method for target tracking.The YOLOv5 model is used as the detector and the DeepSort tracking algorithm is used as the tracker,which effectively avoids the problem of target loss during the tracking process and improves the multi-target tracking ability.