Artificial Neural Network Data Fusion Method Research And Application

Neural Network Pattern Recognition is a fairly new discipline which has drawn broad attention recently and whose application has become one of the most important areas of the application of Neural Network theory. Data Fusion, as a new front-line discipline combining traditional disciplines with newly developed engineering areas, has been extensively applied in many other areas than military ones to which it has been originally applied. This paper researches the application of combining neural network pattern recognition with data fusion theory. Based on systematic discussion of traditional and singular feature identity recognition(such as speaker recognition and face recognition) methods, it proposes to integrate the feature parameters of face images and voice signals and use neural network to do identity recognition. Simulative experiments are done using the fusion system and traditional ones respectively, and the result demonstrates the better function of the fusion system. The structure of the paper is as followed: The first part reviews traditional identity recognition methods, particularly those for speaker recognition and face recognition. The recognition process and the feature extraction are emphasized, and the achievements and deficiencies of the previous researches home and abroad are examined in kind of details. The second part briefly introduces the theoretical background of the proposed system, including neural network pattern recognition, data fusion, as well as their applications in areas of identity recognition. The third part, on the basis of the previous information in combination, elaborates on the principles, characters, algorithm process and predicted application areas of this neural network fusion recognition system. Extracting voice signals and face images which can reliably denotes individual characters, this system extracts features from these two types of data, and input them into fusion recognition module through integrating data for pattern recognition. At the end of this part, traditional methods and this fusion method are compared by doing real-data simulative experiments. In the concluding part of the paper, researches in the application of neural network data fusion in status recognition are reviewed, possible resolvents for the deficiencies are proposed, and the future of next research is prospected.