Research On UUV Underwater Target Recognition Method Based On Deep Learning

With the continuous development of target recognition technology in the field of computer vision and the rise of deep convolution neural network,the application of deep learning in target recognition gradually exceeds the traditional target recognition methods.Some of these new algorithms based on neural network have high accuracy,some have fast recognition speed,and some are compatible.How to better apply these methods to practical needs,we need to continue to study and explore,which has become a research hotspot in recent years.In this paper,the underwater unmanned underwater vehicle is used as the carrier to identify the underwater biological targets,and different model algorithms are used for comparative study.In order to recognize underwater biological targets accurately,MUTI-YOLOV3 and UD-YOLOV3 algorithms are proposed.On this basis,in order to meet the requirements of real-time detection of UUV,the lightweight of MUTI-YOLOV3 is studied.The specific research contents of this paper are as follows:Firstly,this paper summarizes the development of target algorithm based on deep learning in recent years,then compares the program platform,selects the appropriate platform,introduces and compares the performance of different network models,and selects YOLOV3 as the basic model of this paper.According to different optimization methods in neural network,convolution neural network is established for experimental comparison,and the optimization method used in this paper is determined.Secondly,in the case of complex water environment and excellent UUV training model equipment,the senet network structure and new feature extraction layer which can make full use of channel network information are added to the selected YOLOV3 algorithm,and the up sampling method is optimized,and the MUTI-YOLOV3 algorithm is proposed.In the case of complex water environment and common UUV training model equipment,the UD network structure and the optimized SPPNET are proposed to expand the receptive field of the network layer based on the selected YOLOV3 algorithm.The loss function is optimized,and the UD-YOLOV3 algorithm is proposed.The simulation results show that the above two models prediction ability.Finally,on the basis of the MUTI-YOLOV3 model,the lightweight is improved,which can meet the task of some UUV equipment is more common,the water area is more clear.This paper proposes four lightweight target recognition algorithms,compares their performance through experiments,selects two algorithms with better comprehensive performance,and compares the detection effect with the original author's YOLOV2-tiny and YOLOV3-tiny,which verifies the feasibility of the lightweight target recognition algorithm designed in this paper.