基于多特征的高分辨率影像松材线虫病树识别

为了提高松材线虫病树的监测效率,减少其对林业生产造成的损失,利用在高分辨率遥感影像上提取松材线虫病树的光谱特征、空间特征等多特征,然后进行Relief特征选择算法,提取的合适特征为归一化植被指数NDVI(Normalized Vegetation Index)、差值植被指数DVI(Difference Vegetation Index)、OHTA颜色模型作为病树与非病树的光谱特征,对目标影像进行自动筛选,得到疑似病树像元。运用DBscan空间聚类算法对疑似病树像元进行聚类,并以周围一定范围内有一定数量的健康树像元为空间分布参考,对拍摄地点30°1′N/111°43′E附近、分辨率为0.1 m的3幅高分辨率遥感影像筛选病树。自动筛选耗时分别是人工筛选的43.99%、51.08%和46.62%,相对于人工筛选的数量准确度分别为79.37%、77.85%和82.56%。结果表明:采用光谱特征与空间特征相结合的方法在高分辨率遥感影像上识别松材线虫病树识别效率更高。

High-Resolution Image Pine Nematode Tree Identification Based on Multi-feature

In order to improve the monitoring efficiency of pine nematode tree and reduce its loss to forestry production,the spectral features and spatial features of pine nematode tree were extracted from high-resolution remote sensing images;relief feature selection algorithm was carried out,and the appropriate features were derived as normalized vegetation index NDVI(Normalized Vegetation Index),differential vegetation index VI(Difference Vegetation Index),OHTA color model as a spectral feature of the tree and non-disease tree.DBscan spatial clustering algorithm is used to cluster suspected disease tree cells.With a certain number of healthy tree cells in a certain range around as spatial distribution reference,disease trees are screened in the shooting location of 30 degrees 1′N/111′43′E near the resolution of 0.1 meters of 3 high-resolution remote sensing image.The time-consuming of automatic screening was 43.99%,51.08% and 46.62%of manual screening,respectively,and the accuracy of the relative manual screening was 79.37%,77.85%and 82.56%,respectively.The results show that it is more efficient to identify pine nematode trees on high-resolution remote sensing images by combining spectral features with spatial features.