多重编码的三维多目标光学模式识别研究

STUDY OF OPTICAL PATTERN RECOGNITION OF 3 D MULTIPLE TARGETS BASED ON MULTI ENCODING METHOD

利用Ｋｏｈｏｎｅｎ侧抑制自组织神经网络算法和改进的增量算法，对三维多目标识别中的同一目标不同训练样本由单一编码改为多重编码，减轻了多层级联神经网络中第一级网络权重的学习负担，在不降低网络识别目标正确率的条件下。

The Kohonen self organized neural network and the improved increment algorithm were used in multi encoding to replace single encoding for different training sets of the same targets in 3 D multi target pattern recognition. It reduces the first layers learn intensity and learn time for multi layer cascaded network.The result of computer simulation indicated that the weights convergence speed can be improved in the condition that the recognition ratio of this neural network is not reduced.