基于无人机多光谱影像和机器学习的铁尾矿库恢复植被精细识别  

Fine Recognition of Vegetation Restoration in Iron Tailings Reservoir Based on UAV Multi-spectral Image and Machine Learning

作　　者：王紫威 孙冉冉 李小光 尹轩 张晓彤 满卫东 WANG Ziwei;SUN Ranran;Li Xiaoguang;YIN Xuan;ZHANG Xiaotong;MAN Weidong(College of Mining Engineering,North China University of Science and Technology,Tangshan Hebei 063210,China;Tangshan Key Laboratory of Resources and Environmental Remote Sensing,Tangshan Hebei 063210,China;Hebei Industrial Technology Institute of Mine Ecological Remediation,Tangshan Hebei 063210,China;Collaborative Innovation Center of Green Development and Ecological Restoration of Mineral Resources,Tangshan Hebei 063210,China)

机构地区：[1]华北理工大学矿业工程学院,河北唐山063210 [2]唐山市资源与环境遥感重点实验室,河北唐山063210 [3]河北省矿区生态修复产业技术研究院,河北唐山063210 [4]矿产资源绿色开发与生态修复协同创新中心,河北唐山063210

出　　处：《华北理工大学学报(自然科学版)》2025年第2期85-93,共9页Journal of North China University of Science and Technology:Natural Science Edition

基　　金：国家自然科学基金项目(52274166):基于空-天-地多源数据的铁尾矿库恢复植被与土壤养分耦合关系研究;中央引导地方科技发展资金项目(246Z4201G):矿区生态修复效果动态监测关键技术及综合评价平台研发。

摘　　要：无人机搭载多光谱相机能够更便捷、精确的获取矿区恢复植被类别,对于促进生态修复效果评价和保障矿区可持续发展具有重要意义。本研究选取了唐山市迁安大石河铁尾矿库典型样地,开展铁尾矿恢复植被的精细化分类研究。基于多光谱影像数据,构建包含光谱特征和植被指数的多维特征集,并通过ReliefF特征优选方法确定了最优特征,采用面向对象与随机森林算法,对铁尾矿恢复植被进行了分类。研究结果表明:在铁尾矿恢复植被的分类中,归一化差值植被指数(NDVI)和比值植被指数(RVI)显示出较高的贡献度,其重要性分别为0.1583和0.1530。此外,光谱特征中的红边波段相较于其它波段具有更高的分类重要性,达到0.1164;仅利用光谱特征进行的随机森林分类,总体分类精度为82.60%。引入植被指数后,分类精度显著提升至87.24%;当采用经过特征优选的变量进行分类时,分类精度进一步提升至91.64%,相较于仅使用光谱特征和光谱特征结合植被指数的方法,分别提高了9.04%和4.40%。其中槐树分类精度提升最大,用户精度为77.41%。这表明,优选后的特征集不仅减少了数据集的冗余性,也显著提高了分类效率和精度。本研究结果可为利用低成本无人机开展植被精细化分类提供方法借鉴。Equipped with multi-spectral cameras,Unmanned Aerial Vehicles(UAVs)provide an efficient and accurate method for assessing vegetation types in mining area restoration.This is essential for evaluating ecological recovery and ensuring the sustainable development of mining regions.This study focuses on a representative site within the Dashihe iron tailings reservoir in Qian’an,Tangshan City,to conduct an in-depth classification of vegetation restoration in iron tailings.Using multi-spectral image data,a multi-dimensional feature set,including spectral features and vegetation indices,was developed.Optimal features were identified through the Relief F feature selection method.Vegetation restoration classification in iron tailings was performed using object-oriented and random forest algorithms.The results show that the Normalized Difference Vegetation Index(NDVI)and the Ratio Vegetation Index(RVI)made significant contributions to the classification,with importance values of 0.1583 and 0.1530,respectively.Additionally,the red edge band demonstrated greater classification importance compared to other spectral bands,with an importance value of 0.1164.The overall classification accuracy for random forest classification based on spectral features alone was 82.60%.When vegetation indices were included,the accuracy improved significantly to 87.24%.With the optimized feature set,classification accuracy increased further to 91.64%,a 9.04%and 4.40%improvement over the spectral-only and combined spectral features with vegetation indices approaches,respectively.Among the classified vegetation types,Sophora japonica exhibited the highest user accuracy,reaching 77.41%.These results indicate that the optimized feature set not only reduces dataset redundancy but also significantly improves classification efficiency and accuracy.The findings of this study offer valuable insights for using low-cost UAVs in detailed vegetation classification.

关 键 词：铁尾矿库 特征优选 ESP2 面向对象 随机森林 

分 类 号：X87[环境科学与工程—环境工程]