摘要
在对数据进行归一化处理的基础上,将概率神经网络用于遥感影像分类,并探讨样本区的选择和高斯基函数标准差对分类精度的影响。用西藏波密地区1999年的TM遥感影像进行分类试验,并将分类结果和经典的最大似然法进行比较。结果表明:概率神经网络的总体分类精度和Kappa系数分别为94.5%和0.934,取得了较为理想的识别和分类效果。
This paper has analyzed the remote sensing classification using the probabilistic neural network（PNN） on the basis of data normalization,for the best classification accuracy,the picking of sample area and the standard deviation of the basis Gauss function has been discussed.PNN classification model was applied to classify the TM image in Tibet.Based on error matrix,the classification result of the maximum likelihood was contrasted with that of PNN model.The results show that the overall accuracy and Kappa coefficient of PNN model reach 94.5% and 0.934,respectively,is superior to that of the traditional Maximum likelihood method.
出处
《高原山地气象研究》
2011年第3期26-29,共4页
Plateau and Mountain Meteorology Research
基金
国家自然科学基金(50779042)
国家发改委重点项目"西藏自治区生态环境遥感监测与服务系统"
关键词
概率神经网络
遥感影像
最大似然法
误差矩阵
分类
probabilistic neural network
remote sensing image
maximum likelihood
error matrix
classification
