Underwater Acoustic Image Processing And Pattern Recognition

Algorithms in the field of underwater acoustic image processing and pattern recognition are studied and developed in the thesis. Basic image processing algorithms, image processing algorithms based on Pulse Charged Neural Network (PCNN), morphologic image processing algorithms and target recognition algorithms are involved.The image rotation algorithm and non-integer times scale algorithm which are included in the basic image processing are use to implement rotation invariability and scale invariability. Gray-level histogram equilibrium is always used to resolve the problem on the low intensity pixels in the underwater acoustic images. A novel bio-value segmentation algorithm is proposed and a novel multi-threshold segmentation algorithm is formed by deduction. Besides, a self-adapt bio-value segmentation algorithm called Otsu algorithm is thoroughly studied and widely used in other parts of the thesis.PCNN algorithm is a novel one and a pop problem in the field of image processing at present. Edge extraction, thinning and noise reduction based on PCNN in bio-value image are given emphasis in the thesis. And the bio-value image noise reduction algorithm origins from the gray-level image noise reduction algorithm proposed by craft brother by modified the essential parameter, and good results are produced.Morphological image processing algorithm is widely and deeply studied. Middle results with high quality are produced for the following target recognition task by bio-value morphological opening and closing. Morphological smooth algorithm and morphological gradient algorithm are compared with the routine algorithms which have the same functions, and all their advantages, shortages and other characters are clarified. Top-hat is used in the sidescan sonar image processing to obtain the shade and it's detail which can be used in target recognition. Finally, on the basis of the all the above algorithms, sidescan sonar targets auto detection algorithm, highly symmetric area and other areas auto recognition algorithm and rectangular area and ellipsoid/circular area auto classification algorithm are firstly proposed by calculating the areas' eigenvector. Meantime, a novel algorithm is proposed which can work out the perimeter length of an area without a hole, this algorithm conquer the trap point which is a difficult problem.To sum up, underwater acoustic image preprocessing, feature extraction and pattern recognition are widely and deeply studied, some pop problems are resolved and some original productions are achieved.