Research On Underwater Image Segmentation And Target Feature Extraction And Recognition Technology

With the rapid development of China's industry and the enhancement of environmental awareness,the pressure to treat industrial wastewater has increased.The recovery and cleaning of sundries in waste water tank is an important link in industrial waste water treatment.As the carrier of underwater environment,the superb visual processing technology is helpful to improve the safety and autonomous ability of underwater intelligent robot.The underwater intelligent robot based on light vision is widely used in underwater operation due to its good environmental perception ability.Therefore,it is of great significance to study the visual processing technology of underwater intelligent robot.This paper mainly studies from four aspects: image enhancement,image segmentation,feature extraction and target recognition.The specific research contents are as follows:Firstly,an underwater gray image enhancement method is studied to solve problems of uneven illumination,low contrast and large noise.This method fuses three classical image enhancement methods: homomorphic filtering,histogram stretching and wavelet threshold denoising to improve the uneven illumination of the image and enlarge the gray difference between the target and the background.Secondly,two underwater image segmentation methods are researched for the segmentation of underwater targets with special underwater imaging environment and different gray levels: the level set segmentation method for specifies grayscale target and the level set segmentation method of multiple grayscale targets.On the basis of the c-v model,the level set segmentation method for the specified grayscale target adds a small range of distance constraint terms to make it local and can be used to segment any one of the multiple grayscale targets.On the basis of c-v segmentation model,the level set segmentation method for multi-gray targets adds the internal energy function and edge gradient function of li chunming's segmentation model,which has good segmentation results and good noise immunity.Thirdly,an underwater target feature extraction method based on shape feature is researched because the color distortion and the weak texture feature of underwater images.The combined invariant moments based on Hu invariant moment,NMI,wavelet moment and affine invariant moment are constructed,and used PCA to reduce dimension and optimize the combined invariant moments.Experimental results show that the feature extraction method improves the speed and efficiency of underwater target recognition.Finally,a BP neural network algorithm based on improved PSO(CGPS-BP)was studied to overcome the disadvantages of slow convergence and local minimum in BP neural network learning.This method combines chaos strategy and genetic algorithm with PSO,and the improved PSO algorithm is used to optimize the BP neural network.Experimental results show that this method can accelerate the convergence speed of BP neural network and prevent it from falling into local minima.at the same time,the performance of BP neural network in underwater target identification is improved.