基于决策树判别的高温目标遥感识别方法

High Temperature Targets Remote Sensing Recognition Based on Decision Tree Discrimination

通常,高温目标与常温地物间光谱特征差异显著;但研究发现,Landsat8 OLI遥感影像中彩钢屋顶像元却与高温目标(林火)的光谱特征颇为相似,使用以往高温目标识别方法识别效果不佳。为实现高温目标的精确识别,引入决策树判别法;根据不同地物类型的相似程度构建决策树模型,针对各分支结点的相似地物类型,按定量指标分别进行特征波段筛选,确定反映地物间本质区别的判别函数,并经分类统计确定判别阈值。研究表明,所构建的决策树能够准确划分地物类型,在实现同一般常温地物有效区分的同时,能有针对性地区分高温目标与彩钢屋顶建筑,高温目标识别精度为97. 67%。

Generally, spectral differences between high temperature targets and normal temperature objects are significant. However, it is found that pixels of color steel roof in the Landsat8 OLI remote sensing image are quite similar to the high temperature targets (forest fires), and using previous high temperature target recognition method to identify not effectively.In order to realize accurate identification of high temperature targets, decision tree discriminant method is introduced. The decision tree model is constructed according to the similarity degree of different feature types. For the similar feature types under each branch node, sensitive bands are screened according to quantitative indicators, discriminant functions reflecting the essential difference between objects are determined, and discriminant thresholds are determined by classification statistics. The research shows that the constructed decision tree can accurately classify feature types, and classify high temperature targets and color steel roof buildings in a targeted way while achieving effective separation from normal temperature objects. The high temperature targets recognition accuracy is 97.67%.