机载中波红外影像分类性能分析

Classification Performance of Airborne Mid-wave Infrared Imagery

由于中波红外谱段复杂的辐射特性以及红外探测技术的限制,目前学界对中波红外的遥感分类应用探索较少。该文是在国内首幅可见光-中波红外高分辨率(中波红外0.6m分辨率)多光谱影像的基础上,探索地物的中波红外辐射特性,挖掘中波红外谱段的潜在价值,进而融合地物的中波红外与可见光的特征,分析中波红外影像的地物分类性能,提高遥感地物分类的精度。中波红外谱段的光谱辐射特性不同于可见光与热红外谱段,既包含地面反射辐射,也包含地面物体的发射辐射能量。研究中基于多尺度分割算法与随机森林分类器分别对可见光影像和中红外+可见光四波段融合影像进行面向对象分类。该方法融合了地物的可见光与中波红外特征,并且评估了光谱、形状、纹理等特征在分类中的重要程度,定量分析了融合中波红外波段后的特征空间。研究结果表明:针对中红外特征,最有效特征为中红外与可见光其中两波段组合HIS空间各分量特征,其次为灰度共生矩阵纹理信息;中波红外波段的引入可以稳定地提高地物分类的总精度;中波红外波段对于人工地物的分类效果优于非人工地物类型,其中建筑物的分类精度提升最为显著。

The main objective of this study is to explore the utility of aerial mid-wave infrared image for the land cover classification.The proposed classification procedure was implemented in a selected study area that is located in Zhanghe airport,Hubei province,China.During the classification,in addition to mid-wave infrared band,the visible bands were incorporated.In this study,random forest(RF)classification was used,which is a promising machine learning algorithm.The obtained accuracy assessment results indicate the success of RF classifier in land-cover classification from the visible and mid-wave infrared merged images with overall accuracy rising to 89.63%.The result demonstrates following conclusions:referring to mid-wave infrared band,the most effective feature is its HSI color space combined with two of the visible bands,and the second important feature is the GLCM texture information;using mid-wave infrared band in land cover classification can improve the overall accuracy stably;the mid-wave infrared band can work effectively on the classification of the man-made objects with the most remarkable increase in the accuracy of the buildings.