Research On Underwater Image Enhancement And Fish Tracking Technology

Image enhancement and fish tracking in natural waters are of great significance to improving the level of information,automation,and intelligence of fish farming in marine pastures,and have broad application value for fish behavior analysis,species observation,and water quality monitoring.The propagation characteristics of light underwater cause serious degradation of underwater image quality,such as color cast,blur,uneven illumination,and low contrast.At the same time,in the tracking process,fish swimming pattern is changeable,swimming speed is fast,and there is no public dataset.To deal with the above problems,the main work of this thesis are:1.According to the characteristics of underwater images,two underwater image enhancement methods are proposed.1)Underwater image enhancement method based on dark channel prior and Retinex.First of all,the dark channel prior theory is used to dehaze to improve the sharpness of the underwater image.Then the superpixel segmentation algorithm is used to segment the underwater image.The region with high brightness is the foreground,and the region with low brightness is the background.Based on the Retinex theory,the two regions are enhanced separately.Finally,the two regions are combined into a final underwater enhanced image using Poisson fusion.2)Underwater image enhancement method based on.color correction and double exposure fusion.Firstly,color correction based on the gray-world hypothesis theory is performed on the underwater image,and then the problems of uneven illumination and low illumination of the underwater image are solved by double exposure fusion,further the image color and brightness are enhanced.The method has fast processing speed,good real-time performance,and can be well applied to underwater video image sequences.2.A fish tracking algorithm based on improved Kernel Correlation Filters(KCF)is proposed.Based on the KCF tracking algorithm,the HOG(Histogram of Oriented Gradient)feature and CN(Color Name)feature of fish are extracted separately,and the two features are fused to describe the fish,while the fish scale is estimated adaptively through the scale filter,so as to achieved the multi-feature and scale adaptive fish tracking in natural waters.Aiming at the lack of underwater fish tracking dataset,online videos of fish in natural waters are collected and manually labeled.On the self-built dataset,the underwater image enhancement and improved KCF tracking algorithm are verified.Experimental results show that the problems of uneven illumination and blur of the underwater image are effectively solved,and the fish tracking can be achieved in real time and robustly.