Research on Underwater Image Segmentation and Pattern Recognition Based on Particle Swarm Optimization
|School||Harbin Engineering University|
|Course||Design and manufacture of ships and marine structures|
|Keywords||Underwater light vision Image Segmentation Entropy Particle Swarm Optimization Neural Networks|
In recent years, with the deepening of the marine research and development as well as the needs of national defense, smart as an important part of the marine high-tech underwater robot has been widely used. Underwater robots often need to operate in the harsh and complex environment, so that the visual system is particularly important to highlight. This study is combined the underwater light visual information processing and understanding technology \Underwater image segmentation and pattern recognition techniques are two important aspects of the underwater light vision technology. Image segmentation is a classic problem in image processing has been applied to the underwater images is more difficult at; while to achieve the target of pattern recognition is a key step of the entire target recognition system, pattern recognition many ways, the most widely used , the effect is one of the best method of neural network identification. The purpose of this paper is to through issues related to the theory of learning, study a set of real-time processing ability and robustness of image segmentation and object recognition, and as a basis to build a light vision underwater target recognition system . The paper first reviews the optimization of some basic concepts and definitions, and then study a particle swarm optimization (PSO) algorithm optimization technique that is based on swarm intelligence, and, respectively, from the basic structure of the algorithm, the convergence conditions, parameter selection, etc. carried out a detailed analysis of the particle swarm optimization algorithm. These contents will be behind the theory of algorithms and improved to provide the essential basis. Shortly thereafter, the paper starting from the reliability of information processing in the visual system, that is to consider how to take full advantage of the system to obtain the original data (for example, optical image acquisition) to provide more effective input information for subsequent processing sectors, in-depth study of the entropy theory and entropy-based image processing methods. And the characteristics of the underwater images, design two entropy-based thresholding methods, to ease the impact of underwater imaging result in loss of information. Feature extraction stage of moments and invariant moments theory, the paper made a more detailed description, and then constructs a Moment Invariant Feature and combined with neural network theory, based on the the area moments affine transform invariant; neural network identification method, designed to identify the model includes the same moment feature extraction the feature vector standardization as well as the design of neural networks and discriminant mechanism design. The papers also consider the efficiency of information processing in the visual system, to get useful information related to the task at the same time, try to shorten the processing time for all aspects of system processes, papers respectively using particle swarm optimization algorithm to search for the optimal segmentation threshold and Training neural network, in order to achieve the optimization of the process of image segmentation processing and pattern recognition. Finally, use of the hardware platform and software system to build an underwater light visual target identification system. Experimental results show that the task of image segmentation and recognition for underwater targets, this study method is feasible and effective underwater light vision technology has important practical significance.