基于无人机影像特征优选的山地地区植被分类

Vegetation Classification in Mountainous Areas Based on UAV Image Feature Selection

摘要: 针对山地丘陵地区无人机影像分类尺度难以确定，特征数量维数过高，分类精度较低的问题。研究首先确定最优分割尺度，结合Relief F算法和CFS算法分别对先验特征数据集进行优选，最后利用随机森林算法(Random Forest, RF)完成分类。以湖南山地丘陵地区为研究区，在同质性与Moran’s I联合评价的最优分割尺度160基础上，采用优选的特征子集，构造出3种面向对象分类方案。结果表明，经过最优尺度计算、CFS特征优选和机器学习方法的分类结果精度最高，总体精度达到90.3%，Kappa系数达到0.873。证明了该方法适用于山地丘陵地区土地覆盖分类。

Abstract: In view of the problems that the classification scale of UAV images in mountainous and hilly areas is difficult to determine, the feature quantity dimension is too high, and the classification accuracy is low. The research first determines the optimal segmentation scale, and combines the Relief F algo-rithm and the CFS algorithm to optimize the prior feature data set respectively, and finally uses the Random Forest (RF) algorithm to complete the classification. Taking the mountainous and hilly area of Hunan as the study area, based on the optimal segmentation scale 160 jointly evaluated by ho-mogeneity and Moran’s I, three object-oriented classification schemes were constructed using the optimal feature subset. The results show that the classification results obtained by optimal scale calculation, CFS feature optimization and machine learning method have the highest accuracy, with an overall accuracy of 90.3% and a Kappa coefficient of 0.873. It is proved that the method is suita-ble for land cover classification in mountainous and hilly areas.