基于影像特征随机森林的自然保护区矿区动态监测

Application of random forests on monitoring changes in mining rights in nature reserves

获取矿区地物的变化是监测矿山活动有效手段之一,以贺兰山国家自然保护区为研究区,选用Landsat-8 OLI多光谱遥感影像,通过分析矿区地物光谱信息特点,选取纹理特征、归一化差分植被指数NDVI、改进归一化差异水体指数MNDWI和归一化差值土壤指数NDSI 4种特征信息,采用随机森林分类方法,对研究区地物覆盖物信息进行识别,分类整体精度和Kappa系数分别为92.35%,0.91,分类精度较基于传统随机森林分类和支持向量机分类有明显提高。通过对研究区2016—2019矿业权内进行动态监测。结果表明:①通过图斑提取发现自然保护区内存在越界开采、无序开采及乱采滥挖等情况;②开采场面积增加的区域主要分布在矿业权范围内,减少的开采场面积在矿业权界内外均有分布,减少幅度达56.34%,说明相关矿业部门对矿山管理起到积极作用。基于影像特征的随机森林的分类方法在自然保护区地物覆盖信息提取和动态监测中具有一定可行性。

One of the effective means to monitor mine is to obtain the change of ground feature.This paper selected Helan Mountain National Nature Reserve as the research area,and selected landsat-8 OLI multi spectral remote sensing image,based on the analysis of spectral information characteristics of surface features in mining area,four kinds of feature information were selected,including texture feature,NDVI,MNDWI and NDSI,and the random forest classification method was used to identify the ground cover information in the study area,the overall classification accuracy and Kappa coefficient were 92.35%and 0.91,respectively.The classification accuracy was significantly improved compared with the traditional random forest classification and support vector machine classification.Through the dynamic monitoring of the 2016—2019 mining rights in the study area.The results showed that:1)It was found that there were cross-border mining,disorderly mining and indiscriminate mining in the nature reserve through map patch extraction;2)The areas with increased mining area were mainly distributed within the scope of mining rights,and the reduced mining area was in mining.The rights were distributed in both inside and outside the boundaries,with a decrease of 56.34%,indicating that the relevant mining departments play a positive role in mine management.The classification method of random forest based on image features was feasible in the extraction of land cover information and dynamic monitoring in nature reserves.