用改进后的BP网络识别地图符号

Recognizing Map Symbol Using An Improved BP Neural Network

讨论基于BP神经网络的点状地图符号识别,在分析介绍BP网络模型及其学习算法之后,根据BP模型本身所存在的一些不足之处,提出了相应的改进措施。改进后的BP网络学习速度明显提高,性能得到增强,因而可以更有效地识别点状地图符号。另外,还对网络的输入模式加以讨论,给出了在BP网络学习训练过程中的一些体会,并提出了三种减小网络规模的具体方法。

Automatic recognition of map elements in scanning maps is an advanced technique in the field of cartography nowadays. It is one of the key techniques to realize the automation of map-making. The point-shaped element is the core and base in the course of automatic recognition of map element. In this paper, we discuss the recognition of point-shaped map symbol based on BP neural networks. Due to the deficiency of BP model, we propose a relevant imporved measure after analysing and introducing the BP networks model and learning algorithm. And its learning speed is obviously raised and its capacity is enhanced. In the meantime, we also discuss the input pattren of networks, give some of experiences from the course of BP network learning and training, put forward three methods to decrease networks scale.