摘要
本文利用HJ-1A星HSI2级产品数据对大兴安岭塔河地区的森林优势树种组进行识别研究。通过对影像数据进行一阶微分变换、对数变换、对数变换后的导数变换、二阶微分变换、三阶微分变换以后,分别对原始数据和5种变换后数据进行MNF（最小噪声分离变换）变换。用SVM（支持向量机）分类器分别对6种数据监督分类后,进行精度的验证及评价。结果表明：影像数据经过5种变换后,分类总精度均高于未经变换的原始数据,精度提高幅度为1.5%~4.8%,二阶微分变换分类精度最高（精度为89.5%,Kappa系数为0.802）;二阶微分变换下的4个优势树种组各自的制图精度和用户精度均高于其他变换方法,平均精度分别为90.4%和90.7%,总平均精度为90.5%。
We classified the dominate forest tree species group by one image data of the HJ-1A satellite Level2 HSI product in the Tahe area. We analyzed the image data by first derivative, log, first derivative of log, the second derivative and the third derivative methods, and deposed the original data and five kinds transformed data by MNF （ Minimum Noise Frac- tion）, then verified and appraised them with the SVM （Support Vector Machine） after combining with the ground data computing confusion matrix for validation. The classification accllracy with five kinds of transformation was higher than that without any transformation, and the accuracy increased by from 1.5% to 4.8%. The accuracy of second derivative was 89.5%, which was the best with Kappa of 0. 802. The mapping accuracy and user accuracy of four dominate forest tree species groups deposed by the second derivative was better than others, and the average accuracies were 94.7 % and 91.2%, re- spectively, with the total average accuracy of 90.5%.
出处
《东北林业大学学报》
CAS
CSCD
北大核心
2015年第5期134-137,共4页
Journal of Northeast Forestry University
基金
国家高技术研究发展计划(2012AA102001)
关键词
HJ-1A卫星
数据变换
SVM
树种识别
HJ-1A satellite
Data transformation
Support Vector Machine （SVM）
Tree species recognition
