基于BP神经网络的点状地图符号识别

点状地图符号的识别是实现扫描方式地图自动数字化的重要研究内容之一，本文首先分析介绍ＢＰ网络模型及其学习算法，然后根据ＢＰ模型本身所存在的一些不足之处，提出相应的改进措施，还以实际扫描地图图像进行相应的实验验证，最后，对网络的输入模式加以讨论，提出了三种减小网络规模的具体方法。

BP NEURAL NETWORK BASED RECOGNITION OF MAP SYMBOL

The recognition of point shaped map symbol is one of important research contents to realize map automatic digitizing based on scanned images. In this paper, the BP networks model and learning algorithm is analysed and introduced first. Then a relevant imporve measure is proposed according to the deficiency of Bp model. In the meantime, an experiment is done by use actual scanned map image. The input pattern of networks is discussed, some of experiences from the course of BP networks learning and training is given, and three methods to decrease networks scale is put forward at last.