人工神经网络在遥感图像森林植被分类中的应用

应用人工神经网络模型对陆地卫星TM多光谱图像进行了森林植被分类的研究 ,共选取了 8种主要植被类型 ,重点是研究在不同背景条件下存在同谱异物现象的云杉、油松和落叶松等针叶林树种的分类方法 .所采用的网络模型为 3层误差后向传播神经网络模型 ,鉴于贺兰山自然植被垂直带谱明显 ,利用误差后向传播网络模型的并行分布式结构 ,研究中引入高程数据作为一个独立波段与 3个多光谱波段一起直接进行分类 ,取得了很好效果 .该方法与常规的最大似然法相比 ,存在同谱异物现象的云杉、油松和落叶松的分类精度平均提高了 2 7 5个百分点 .对存在同物异谱现象的阔叶林的分类精度也有一定程度的提高 .

Forest vegetation classification of TM images using artificial neural network

The authors studied the forest vegetation classification of TM images with the error back propagation(BP) model. Eight main vegetation types were classified, the classification of various conifer tree species with similar spectral characteristics was the key part of the research. Because of the evident vertical distribution of the vegetation in Helan Mountain, northwest China, the elevation data were imported to the error back propagation model as a separated channel with the parallel distributed structure of the model. The elevation channel, combining with the three multi spectral channel, could improve the classification accuracy. Compared with the accuracy of the maximum likelihood classifier, the classification accuracy of spruce, pine and larch with error back propagation model was increased by 27 5%. The BP model also improved the accuracy of the same deciduous tree species with different spectral values.