Research On Underwater Image Segmentation And Pattern Recognition Based On Particle Swarm Optimization

In recent years, with the development of ocean exploitation and the need of national defenses, Autonomous Underwater Vehicle (AUV) has been widely applied as an important component of ocean high-tech. Complexities and uncertainties of working environments make the vision system of AUV's stand out especially. This paper carries out with a certain study-in-advance project named "information processing and understanding technology of underwater optical vision" in intelligent underwater robot technology.Image segmentation and pattern recognition technologies for underwater object are two important links in underwater optical vision technology. Image segmentation is a classical problem in image processing especially applied in the underwater image, while pattern recognition is the critical step of the target identification work. At present, there are various methods being used in pattern recognition, among which pattern recognition with neural networks is one of the most extensive and effective methods. The purpose of this thesis is to investigate a set of image segmentation and target recognition algorithms with good real-time performance and robustness through subject interrelated research, and then to construct an optical vision based underwater target recognition system.This thesis first briefly reviews the subject of optimization. This is followed by an elaborate discussion of swarm intelligent based optimization technique, i.e. particle swarm optimization (PSO) algorithm. Several aspects of PSO such as basic structure, convergent conditions and parameters selection etc are then systematically discussed, which serves as as basis of following research in our work.Subsequently, considering how to make full use of original data in the following information processing to supply accurate information, this thesis begins with reliability of information handling, and investigates entropy theory in the application of image segmentation in depth. Then two entropic thresholding methods are put forward through the analysis of underwater imaging.In the phase of feature extraction, this thesis makes a detailed discussion on moments and invariant moments, and constructs the affine invariants based on region moments. Then a new method of recognition based on invariant moments and neural network is proposed, in which extracting the invariant features, regularizing the feature vectors and designing the classifying cone are combined.In order to improve the efficiency of information handling, this thesis also investigates the application of PSO to the field of image segmentation and pattern recognition. In the former, PSO is used to search the optimal threshold, which can find better solutions with much little complexity. Aiming at the drawbacks of traditional BP neural network, such as converging slowly and tending to get into the local minimizer, PSO is introduced into the training of neural network.Finally, this thesis summarizes the context and constructs an underwater optical vision based target recognition system using definite hardware and software. As the experiment results show, these methods proposed here are feasible and effective, and significant in the field of underwater optical vision.