Two-level classification of target recognition based on neural network

In this paper, we present a classification method, which contains a competitive learning algorithm with a nonlinear map function. Since leakage and error exist in one-level classification, then this method is particularly effective in recognition. A new concept, which is the "two-level classification", is proposed, it and its application to feature extraction and data association are also studied. The aim is to produce a useful track file, which contains groups of information on the moving target. The training of the connection weight in two-level classification is a key problem, and the storage capacity is also an important question. The key to the settlement of the question lies in adjusting adaptive learning rates on parallel distribution. The effectiveness and the correctness of the proposed method are shown in the given results. An input pattern sample is either classified effectively by the former, or classified effectively by the latter.