摘要
在利用BP网络识别点状地图符号的过程中,提出了三项改进措施,简化了数据获取,提高了网络收敛速度。通过对训练样本图像的格式分析,实现了输入神经元的简单获取;对神经元、权值数组采取堆内存动态分配的方式,减少了内存占用,极大地提高了权收敛速度;通过对网络拓扑的分析,调整了节点的输出方式,加快了权值改正。
In the course of mark map symbol recognition with BP(Back Propagation) neural network, some innovations are put forward and passed the verification by experiment. Through the analysis of the format of the training sample images, a simple new method of getting data for in-nerve cell is found; The stack memory way is adopted in managing nerve cell and weight, then the memory occupation is reduced and the convergence speed is increased hugely; Through the analysis of the network topology, the output mode of node is adjusted, so that the weight correction is quickened.
出处
《测绘与空间地理信息》
2004年第5期18-20,共3页
Geomatics & Spatial Information Technology
基金
辽宁省自然科学基金资助项目(002099)
