Research On Underwater Target Recognition And Positioning Based On Machine Vision

Underwater detection and underwater operation technologies had been rapidly developed along with the development of marine engineering and equipment technologies.With the development of marine development requirements,underwater robots had been applied more and more widely in such fields as underwater data monitoring,seabed detection,and acquisition and identification of marine targets.The success of these tasks couldn't be separated from the recognition and positioning of targets by underwater robots.Image information was rich in content and was clear in goals,could improve resolution.Therefore it often used for object recognition.This paper applied the machine vision and image processing technology to research the underwater target recognition and positioning technology.Because of the problem of insufficient robustness and low recognition rate of traditional image recognition algorithms in feature extraction under complex background,this paper proposed a combination of traditional recognition algorithm and deep learning algorithm to improve the target recognition rate.As a common model of deep learning,convolutional neural networks had good semantic segmentation capabilities and could effectively improve the image recognition rate.For underwater target location,the binocular stereo vision system was designed to achieve the three-dimensional position information of the target.The target recognition process included image preprocessing,image enhancement and restoration,target extraction,and final recognition.The image preprocessing was mainly used to remove the noise in the image.Due to the large number of underwater suspended matters and the complex imaging environment,the preprocessed image often needed to enhance the contrast of the image and restore the distorted image color.The role of the target extraction was mainly to separate the target from the background,so that the target was so prominent in the background to facilitate subsequent recognition operations.For the research of target recognition technology,this paper first analyzed the traditional BOF algorithm recognition process in detail,then introduced the advantages and characteristics of the convolutional neural network,including its convolution and pooling principle,and finally analyzed network structure of the Alexnet model.In addition to theoretical analysis,this paper also made use of underwater images to verify the design of the identification algorithm,and made its own image database.The experimental results showed that the proposed method could effectively improve the recognition rate of underwater targets.In addition,the principle of binocular vision location algorithm was analyzed in this paper.The hardware system and software system of binocular vision system were designed and the binocular positioning experiment was carried out.