The Application Of Fuzzy Neural Network In Target Recognition

Underwater target recognition is the key technique in accomplishing intelligent of underwater acoustic equipment and weapon, which is an unsolved difficult subject for researchers by many years. Echo recognition is the main method to identify underwater targets in long distance. The subject of this dissertation is to distinguish between true and false target by echo recognition. Using the existing acoustic echo feature extraction, the dissertation adopts Fuzzy Neural Network (FNN) technique, designs two classifiers that have the ability of adaptiveness and generalness. Referring to actual target and environment where it will be used in, the dissertation has designed two classifiers, tests and analyzes these classifiers with the data gained in experimentation.Fuzzy Neural Network combines the advantages of neural network and fuzzy system. It is good at expressing knowledge with fuzzy logical and the capacity of adaptation and learning. It can be used for recognizing complex and fuzzy characters of underwater targets, and can self-extract of fuzzy rules and adaptive surroundings.In the dissertation the first classifier is designed according to the Fuzzy Cluster Neural Networks, which is a kind of Fuzzy Kohonen Cluster Networks, its substance is the calculate way that uses the Kohonen algorithms to carry out fuzzy c-average clustering; It can quickly obtain the central point of data which would be assembled; It combines together with the classic Self-Organizing Feature Map, this way would pick up learning speed of Self-Organizing Feature Map. This classifier uses clustering algorithms for obtaining number of fuzzy rules and form of member functions, its algorithm is to direct demarcate the whole input space, which uses multiple dimensions member functions, and makes member functions to overlay the whole input space as more as possible It reduced regular number greatly.In the dissertation the second classifier is designed according to the Adaptive Neuro-Fuzzy Inference Network (ANFIS), which produces fuzzy rules from the input- output data, adjusting through the BP network study, then constituting the produced rule and experience rule to fuzzy logical system at last. The second classifier is considered to the identification of the data rule.The author adopts VC++ to write the data manifestation and characteristic software, and uses Matlab to accomplishing the design of two classifiers. The dissertation detects the identifiable ability of classifiers with the test data. In the paper, I analyze and deal with the primal data of mine echo to get the ratio of recognition and make a conclusion that the fuzzy neural network arithmetic is efficient.