用模糊ARTMAP算法对CBERS-2数据进行分类

Classification of CBERS-2 Imagery with Fuzzy ARTMAP Classifier

摘要: 用模糊ARTMAP(fuzzy adaptive resonance theory map)神经网络算法对CBERS-2数据进行了分类实践。首先介绍了模糊ARTMAP神经网络的算法原理和具体训练分类过程;然后用2004年9月新疆石河子地区的影像数据进行土地利用分类试验,并将分类结果与基于统计的最大似然法(MLC)、反向传播神经网络(BP)的分类结果作比较,总分类精度比MLC和BP算法分别提高9.9%和4.6%。结果表明,模糊ART-MAP对试验区CBERS-2影像上的裸地识别能力很强,对高分辨率的CBERS-2影像可获得很好的分类结果。

Abstract: This paper adopts fuzzy ARTMAP classifier to do the classification CBERS-2 imagery.The fundament theory and material process about the algorithm were firstly introduced,followed with a landuse classification experiment in Shihezi Municipality on CBERS-2 high resolution imagery.Three classifiers were compared: maximum likelihood classifier(MLC),error back propagation classifier(BP),fuzzy ARTMAP classifier.The assessment shows that fuzzy ARTMAP classifier has a comparably better result,with overall classification accuracy higher 9.9% and 4.6% than MLC and BP.The result also shows that fuzzy ARTMAP classifier has a better discernment to identify the bare soil on CBERS-2 imagery.Finally a superiority of the fuzzy ARTMAP classifier on CBERS-2 high resolution imagery is concluded.