植被种类的准确分类是实现煤矸石山植被修复效果评价的基础。利用无人机遥感技术获取不同季节的煤矸石山可见光影像，通过色彩空间转换和纹理滤波充分挖掘可见光影像中丰富的色彩、结构及纹理等特征；然后对传统人工特征选取方法做出改进，该方法可快速、简单、高效地筛选特征信息以获取最优分类特征，并将优选结果与RGB影像融合以获得多特征融合影像；最后，利用3种监督分类模型分别对两期RGB影像及多特征融合影像进行分类处理并对结果进行精度评价及植被动态变化分析。结果表明：基于改进的人工特征选取方法可筛选出不同季节煤矸石山影像的最优分类特征，所选特征不仅能有效反映各类地物的差异性，同时可降低特征信息冗余以提高影像分类精度及效率。支持向量机（Support Vector Machine Classification, SVM）分类方法结合多特征融合影像的分类精度最高，总体分类精度最高可达90.60%，相应Kappa系数为0.8780，较同期RGB影像分别提高了9.74%和0.1265；而最大似然（Maximum Likelihood Classification, MLC）和神经网络（Neural Network, NNC）分类方法精度提高较少，总体分类精度较同期RGB影像最多可分别提高6.95%和3.93%，相应Kappa系数分别提高0.0845和0.0541。同时，基于最优分类结果从植被覆盖度和植被配置模式2个角度对常村煤矸石山植被修复效果进行评价，结果表明：该煤矸石山采用了多种不同的植被配置模式，且秋夏两季的植被覆盖度均高于75%，植被修复的整体效果较好。研究可为基于无人机可见光影像的煤矸石山植被信息识别分类提供参考，同时为煤矸石山植被修复的后期管理、维护等提供意见或建议。

The accurate classification of vegetation species is the basis for the evaluation of vegetation restoration effect of coal gangue pile. In this paper, the visible image of coal gangue pile in different seasons was obtained by UAV remote sensing technology. The color space conversion and texture filtering were used to adequately explore the rich features of color, structure and texture in the visible image. Then, the traditional artificial feature selection method was improved, which could quickly, simply and efficiently screen features information to obtain the optimal classification features, and the optimized results were fused with RGB images to obtain multi-feature fusion images. Finally, based on two stages of RGB images and multi-feature fusion images, the vegetation of coal gangue pile was classified by three supervised classification methods, including support vector machine (SVM), maximum likelihood (ML) and neural network (NN). Meanwhile, the accuracy of classification results was evaluated by confusion matrix and the dynamic changes of vegetation were analyzed. The results showed that the improved artificial feature selection method could screen out the optimal classification features of coal gangue pile vegetation in different seasons. The selected classification features can not only effectively reflect the differences of various ground features, but also reduce the redundancy of feature information to improve the accuracy and efficiency of image classification. The classification result based on Support Vector Machine Classification (SVM) combined with multi-feature fusion image had highest classification accuracy, and the overall classification accuracy could reach 90.60%, and the corresponding Kappa coefficient is 0.8780, which was 9.74% and 0.1265 higher than that of RGB image of the same period, respectively. And, the accuracy of MLC and NNC classification methods was less improved. Compared with the RGB images of the same period, the overall classification accuracy could be improved by 6.95% and 3.93%, respectively, and the corresponding Kappa coefficient could be improved by 0.0845 and 0.0541, respectively. At the same time, based on the result of optimal classification, this paper evaluated the vegetation restoration effect of coal gangue pile in Changcun from the perspectives of vegetation coverage and vegetation allocation pattern. The results showed that a variety of different vegetation allocation patterns were adopted by the coal gangue pile, and the vegetation coverage in autumn and summer is higher than 75%. The overall effect of vegetation restoration was better. This study could provide reference for the identification and classification of coal gangue piles vegetation information based on UAV visible light image, and meanwhile provide opinions or suggestions for the later management and maintenance of coal gangue piles vegetation restoration.