Research On Underwater Target Recognition And Location Technology Based On Vision

The underwater target recognition and localization technology is mainly composed of image preprocessing,enhancement and restoration,target recognition and localization and other modules.The image preprocessing operation mainly denoises and grayscale the image,and uses as little atomic information as possible to maintain the structure and texture information of the original image.After the preprocessing is completed,it is necessary to restore the distorted color of the image to enhance the brightness of the image.The thesis first expounds the research status of underwater target recognition and positioning technology,and analyzes and studies underwater image enhancement algorithms,deep learning and target detection algorithms.YOLO-v4 is a typical single-stage target rapid detection algorithm,and it has high detection accuracy and speed.Therefore,the paper improves the input anchor frame,data enhancement method and loss function of the YOLO-v4 algorithm based on the actual application scenarios of underwater targets.By improving the network model The method is evaluated and verified by experiments,and the results show that the accuracy of underwater target recognition can be fully improved.Finally,with the help of the binocular stereo vision positioning system,combined with the camera calibration and stereo matching technology,the underwater target positioning is realized.This thesis analyzes the application and development of underwater image processing technology and studies the key technologies of underwater target identification and positioning.It mainly includes the following contents:(1)In view of the complex and changeable underwater environment,underwater imaging will cause color distortion and low contrast of the observed object,and a multi-scale image enhancement method based on the underwater optical imaging model is adopted.Firstly,with the help of the underwater optical imaging model,the gray-scale world algorithm is applied to the underwater image preprocessing;secondly,the K-SVD algorithm is used to construct a strategy for training atoms for the signal content,and to achieve an over-complete dictionary based on the sparse representation of the signal;Finally,the image dehazing is realized by the blue-green channel enhancement algorithm based on the physical model,and the image enhancement software based on Open CV is designed to carry out comparative experiments to achieve the image enhancement effect.(2)Aiming at the problems of low accuracy of target recognition and lack of robustness of feature extraction in complex underwater environment,an improved algorithm based on YOLO-v4 is designed.Firstly,the original input anchor frame mechanism is analyzed and a new anchor frame calculation method,K-means++ algorithm,is used to complete the accurate anchor frame selection.Meanwhile,an image data enhancement strategy based on geometric and photometric transformations is proposed to optimize the training network and improve the accuracy of the underwater target detection model and the detection rate of the algorithm;finally,by analyzing the performance of the original loss function at the output,the loss function EIOU is used to optimize the model performance and improve the convergence speed of the loss function.In the experimental environment of the thesis,the average accuracy rate of the improved network model reaches 98.37%,and the number of frames per second of image transmission reaches35.69 FPS,which is 4.4% higher than the original algorithm,that is,the improved algorithm detects 1.52 more images per second than the original algorithm picture.(3)In order to obtain the three-dimensional coordinate information of the underwater target and assist the underwater platform to complete the target positioning task,through the analysis and research on the positioning principle of the binocular stereo vision system and the underwater imaging model,combined with the binocular camera calibration and stereo matching technology,a three-dimensional camera is established.Various coordinate systems: the world coordinate system,the camera coordinate system and the image coordinate system,and study their transformation relationships,build a model of the binocular stereo vision system,and design the experimental plan in combination with the experimental scene,which can be obtained through data analysis,within the allowable range of system errors,the positioning accuracy is high.