Data Fusion, Neural Computing Methods

Data fusion technology is used for how to utilize multi-source information to overcome the inaccuracy while thoroughly describing objects concerned for correct judgment. This technology has been widely applied to pattern recognition, target tracing and automation fields. Although it is rapidly progressing, it is far less mature realm that no integral theory emerged and many problems pending. Neural computing facilities the construction of knowledge library during data fusion process and fulfills automatic knowledge acquire and associate reasoning to express complex environment. Neural network has the ability of large scale parallel processing, error tolerance and little dependency on transcendent knowledge. The main advantages of neural networks based data fusion over traditional methods are that real-time processing can be achieved and data fusion is done adaptively. But traditional neural network for multi-sensor data fusion lacks uniform theory and people select some parameters and neural network empirically, and most of its algorithms have deficiencies. These facts hinder its applications and development greatly. In particular, the following aspects of research work have been done in this paper:(1) Analyze and induce the theories and methods of data fusion based on artificial neural network.(2) Introduce the basic concepts of neural network and analyze its internal mechanism. Analyze the relationship between neural network and pattern recognition and prove the neural network fusion theory qualitatively.(3) The integration of modular neural network and fuzzy theory resolves the problems brought by each of the two methods.(4) The application of fuzzy neural network in fusion area is discussed. Counter-propagation network (CPN) based data fusion at feature level for target classification and the modified algorithm for training the CPN are brought about.(5) Based on modular neural network and fuzzy theory, a framework for human consciousness is brought about.The research performed in this paper illustrates that the applications of neural networks in data fusion have significance and advantages but many problems need to be settled. Although neural network model is only a partly simulation to human brain, the life science based methodology will play a more and more important role in information processing field in the future.