基于BP神经网络的针阔混交林TM遥感图像自动分类技术研究

在对标准BP神经网络试验分析的基础上，通过输入矢量归一化处理、主成分分析、增加验证集、改进训练学习算法、扩大隐层和输出层规模等措施，对BP神经网络自动分类系统进行改进；利用改进后的BP系统对吉林省汪清林业局的典型针阔混交林TM遥感图像进行辩识、分类试验研究。结果表明：改进后的BP网络分类系统自动分类精度提高了19．14％，比传统无监督自动分类精度提高8．55％，达到了区分森林类型的分类要求。研究还显示了该改进系统应用于针阔混交林TM遥感图像自动分类识别的精度随网络规模增大而提高。

Automatically Classifying and Identifying the TM Remote Sensing Images of Forest Mixed with Conifer and Broadleaves Using Improved BP ANN

The automatically classifying and identifying the TM remote sensing images of forest plays an important role in the monitoring and management of forest resources. In order to improve the performance of BP artificial neural network(BP ANN), many measures, such as standardizing input vectors, increasing verifying set volume, promoting training study algorithm, expanding layers of input_output and main factor analysis, were applied in the TM image data processing. Taking Wangqing Forestry Bureau of Jilin Province as the example study area, the authors studied the automatically classifying and identifying the TM remote sensing images of forest mixed with conifer and broadleaves using the improved BP ANN. The results show that accuracy of automatically classification and identification has been increased significantly, 19.14% higher than that of the traditional ANN method and 8.55% higher than that of traditional unsupervised classifying method respectively. The research also indicates that the classification and identification accuracy rate can be increased further with expanding the BP ANN network volume.{