遥感影像的神经网络分类及遗传算法优化  

Neural Network Classification with Optimization by Genetic Algorithms for Remote Sensing Imagery

作　　者：童小华[1] 张学[1] 刘妙龙[1] 

机构地区：[1]同济大学测量与国土信息工程系

出　　处：《同济大学学报（自然科学版）》2008年第7期985-989,共5页Journal of Tongji University:Natural Science

基　　金：国家自然科学基金资助项目(40771174);教育部新世纪优秀人才支持计划基金资助项目(NCET-06-0381);上海市“曙光学者”计划资助项目(07SG24);高等学校博士学科点专项科研基金资助项目(20070247046)

摘　　要：针对传统遥感影像分类方法难以辨识波谱特性相似的地物,而标准反向传播学习(back propagation,BP)神经网络分类方法存在网络训练速度慢、局部极值等收敛性问题,探讨了采用遗传算法(genetic algorithms,GA)优化BP网络结构方法进行遥感影像分类.在BP网络分类的基础上,着重阐述了遗传算法实现BP网络隐含层神经元数、阈值和连接权值的优化方法,提出了遗传算法的变长实数编码方式,改进了遗传进化方式使BP网络进化达到最优.最后,以淀山湖区域的陆地卫星专题制图仪(Landsat thematic mapper,TM)影像分类为例,应用本文改进算法与其他分类方法进行了分析比较,得到了较高的分类精度,验证了采用遗传算法优化神经网络的可行性和有效性.This paper presents a new method based on genetic algorithms （GA） optimized back propagation neural network classification for remote sensing imagery, which overcomes the defects that traditional approaches based on statistical principle have difficulties in distinguishing the objects with similar spectral characteristics, and the back propagation （BP） neural network method has difficulties in sufficiency and convergence. On the basis of the typical back propagation neural network classification, the optimization for the structure of neural network by genetic algorithms is presented, including the thresholds and connection weights of neural network and the number of neural nodes in the hidden layer of neural network. A coding method on float coding with variable length for genetic algorithms is then introduced, and the evolution method is improved to obtain an optimal back propagation neural network. In the end, an experimental test on remote sensing classification by using Landsat thematic mapper （TM） imagery in Dianshan Lake is conducted. The classification with the proposed approach is the most accurate, which proves its feasibility and validity.

关 键 词：遥感分类 人工神经网络 反向传播学习 遗传算法 精度评价 

分 类 号：P237[天文地球—摄影测量与遥感]