模糊控制BP网络的遥感图象分类方法研究

针对遥感图象分类中经常采用的BP算法存在训练时间长、不易收敛缺点，提出了一种改进方法，即采用模糊规则有效控制BP网络学习率的方法，该方法使网络具有自适应能力，从而不易陷入局部最小，导致收敛速度大大加快，训练时间大大缩短。最后以徐州地区TM图象土地利用分类为例，将模糊控制BP网络模型同BP算法及学习率自调整算法进行了比较。结果表明新方法确实大大加快了网络收敛速度，一定程度上提高了图象分类精度，是一种有效的图象分类方法。

Research on Classification Method of Remote Sensing Image Using Fuzzy Controlled BP Network

In order to solve the problem of long learning time and difficult converging in remote sensing image classification using BP network, a new method of using fuzzy rule to control efficiently the learning rate was put forward. The method makes the neural network be self adaptive and difficult to become the local minimum, so that the convergence rate can be greatly speeded and the learning time shortened. Taking the land classification of Xuzhou TM image as a sample, the authors compared the new method with the old ones. The results suggest that the new method can greatly quicken the convergence rate and efficiently improve the image classification.