Research On Underwater Sea Cucumber Image Recognition Technology Based On Machine Vision

With the improvement of people’s living quality,the consumption market of sea cucumber is expanding year by year,and the sea cucumber breeding industry is growing rapidly.However,the sea cucumber detection and recognition methods are still in the initial stage of research,which cannot meet the needs of sea cucumber identification for the tasks of aquaculture monitoring and underwater robot fishing.Therefore,in this paper,the existing underwater image processing algorithm and underwater target recognition technology are studied.In view of the characteristics of underwater image imaging,an underwater sea cucumber image target recognition system based on machine vision is designed.The system is divided into underwater image enhancement and target recognition.In the process of underwater image enhancement,an underwater image enhancement method based on image fusion is proposed.The homomorphic filtering,MSRCR and the dark channel prior enhancement algorithm based on guided filtering are comprehensively used to process underwater images.Then,the multi-channel linear fusion method based on the sharpness weight of the image is used to fuse the result,and then the USM is used to sharpen it,and finally the enhanced result image is obtained.Then the Brenner gradient,Tenengrad gradient,Laplacian gradient,SMD2,energy gradient,MSE,PSNR and SSIM of the processed image are calculated to verify the enhancement effect of the algorithm.Finally,the SIFT feature matching experiment shows that the algorithm can effectively improve the image feature information density and establish a good condition for the target recognition.In the process of target recognition,400 enhanced sea cucumber images are selected to make image data sets,and a neural network model of sea cucumber target recognition based on YOLOv3 is built and trained.After 500 times of training,the loss and val loss of the model decreases to 12.479 and 14.954 respectively,and then gradually become flat.The neural network model for underwater sea cucumber image recognition is obtained.Finally,through numbers of underwater sea cucumber image target recognition simulation experiments,the recognition accuracy of the system is analyzed,and its multi-class average accuracy(mAP)can reach 98.87%.When the IOU threshold is set to 0.5 and the confidence threshold is set to 0.6,the recognition accuracy of sea cucumber target can reach 97.49%.The experimental results show that the method designed in this paper is in line with the design expectation,can meet the task requirements of underwater holothurian image target recognition well,and has high recognition accuracy and model accuracy.