Side Scan Sonar Image Matching Based On Deep Learning And UUV Aided Navigation Application

The rapid development of unmanned underwater vehicles(UUVs)makes the demand for autonomous navigation technology in deep-water scenes more and more urgent.Traditional underwater navigation is mainly based on inertial navigation system.However,due to the inevitable drift error of inertial navigation over time,UUVs usually rely on global positioning system,Doppler velocimeter and other equipment to correct the accumulated error.Because the electromagnetic signal is shielded in deep water,UUVs need to repeatedly float to the surface to receive the signal feedback from the global positioning system,which seriously affects its normal underwater operation and increases the potential risk of the carrier.On the other hand,the complex deep-water environment will lead to large measurement deviation in the use of Doppler velocimeter,which further affects the accuracy and stability of navigation.Therefore,it is necessary to study a new method to assist UUV navigation to improve the accuracy of underwater navigation.With the development of underwater sensors,imaging sonar has become an important equipment for ocean exploration.Side scan sonar is a typical imaging sonar,which has two main advantages: on the one hand,it could obtain high-resolution acoustic images;on the other hand,it is low cost and easy to deploy.The use of side scan sonar images for feature matching to achieve UUV assisted navigation has ushered in an opportunity.This thesis focuses on the feature matching method of side scan sonar images and its application in underwater navigation.In this thesis,a side scan sonar image matching method based on deep learning is proposed to assist UUV navigation.The purpose is to give full play to the feature extraction advantages of machine vision model and improve the accuracy,real-time and robustness of image matching by fully mining the feature information contained in side scan sonar images.Good matching result is the key to ensure the accuracy of navigation parameter calculation.The full research is carried out from four modules:algorithm design,modeling,simulation and experiment.After obtaining rich and stable image matching results of side scan sonar,the vector matrix including position and direction could be obtained through subsequent solution,and then the core parameters of underwater navigation could be finally solved by establishing the through coordinate system between the image,sensor,robot and the world.The innovation of this thesis is to use the sonar images obtained by the side scan sonar sensor which is low-cost and easy to deploy for feature matching,and use the navigation vector data calculated from the matching results to assist in estimating the navigation path of UUV,so as to reduce the floating action of UUV in the underwater detection task,so as to realize the underwater diving of UUV for a long navigation time and avoid navigation drift.