基于SVM的大屯矿区遥感影像变化检测  

Change Detection of Remote Sensing Images in Datun MiningArea Based on Support Vector Machine

作　　者：龙亦凡 乔雯钰 孙静[1] LONG Yifan;QIAO Wenyu;SUN Jing(Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]山东科技大学,山东青岛266590

出　　处：《测绘与空间地理信息》2020年第12期107-110,115,共5页Geomatics & Spatial Information Technology

摘　　要：矿区土地覆盖变化信息的高精度提取,是区域资源环境保护和地质灾害防治领域的关键问题。以SPOT-5高分辨率卫星影像为数据源,以分类后比较法为变化信息提取方法,利用分类性能良好的支持向量机(SVM)为分类算法,分别设置不同SVM核函数及参数对实验样区进行分类,并评价分析不同核函数及参数对分类效果的影响。实验发现在4种核函数中高斯核函数分类精度最高,为87.1%;Sigmoid和多项式核函数分类精度适中,线性核函数分类精度最低,为78.7%。因此,利用高斯核函数分别对3个时相的大屯矿区影像进行分类,提取出变化信息空间分布并定量统计分析其变化特征。结果表明:矿区植被覆盖面积在近十年内持续减少18.539 km 2,塌陷面积共扩大4.379 km 2,而矿区厂矿增加、煤矿开采规模扩大、采煤塌陷区面积扩大以及耕地占用面积增大等人为因素,是威胁大屯矿区生态环境的主要因素。The high-precision extraction of land cover change information in mining areas is a key issue in the field of regional resource and environmental protection and geological disaster prevention and control.SPOT-5 high-resolution satellite images are used as data sources,and the post-classification comparison method is used as the change information extraction method.Using the support vector machine(SVM)with good classification performance as the classification algorithm,different SVM kernels and parameters are set to classify the sample areas,analysis and evaluation the influence of different kernel and parameters on the classification effect.The experimental results show that among the four kernel functions,gaussian kernel function has the highest classification accuracy(87.1%),Sigmoid and polynomial kernel function have the moderate classification accuracy,and linear kernel function has the lowest classification accuracy(78.7%).Therefore,the gaussian kernel is used to classify the images of three time phases in Datun mining area,extract the spatial distribution of change information,and analyze the change characteristics quantitatively.The results show that the vegetation cover area of mining area has been continuously reduced by 18.539 km 2 and the collapse area has been expanded by 4.379 km 2 in the past decade.The main factors threatening the ecological environment of Datun mining area are the increase of factories and mines,the expansion of coal mining scale,the expansion of mining subsidence area and the increase of cultivated land occupation area.

关 键 词：SPOT-5数据 变化检测 分类后比较 支持向量机 矿区 

分 类 号：P237[天文地球—摄影测量与遥感] TP79[天文地球—测绘科学与技术]