基于多维特征优选支持向量机算法的城市土地利用变化遥感监测  

Remote Sensing Monitoring of Urban Land Use Change Based on Multi-Feature Optimal Support Vector Machine Algorithm

作　　者：周柱灿 郑云云 刘亚群 ZHOU Zhucan;ZHENG Yunyun;LIU Yaqun(Chongqing Institute of Surveying and Monitoring for Planning and Natural Resources,Chongqing 401120,China;Institute of Geographic Sciences and Natural Resources Research,Chinese Academy of Sciences,Beijing 100101,China)

机构地区：[1]重庆市规划和自然资源调查监测院,重庆401120 [2]中国科学院地理科学与资源研究所,北京100101

出　　处：《三峡生态环境监测》2022年第1期73-87,共15页Ecology and Environmental Monitoring of Three Gorges

基　　金：重庆市规划和自然资源局2020年科技项目(KJ-2020017)。

摘　　要：随着遥感影像时、空、谱、辐分辨率和数据处理能力的提升,综合多维影像特征已成为提高土地利用分类精度的关键。目前并非所有特征均有助于分类,且传统分类仍拘泥于单一特征,因此,急需有效的特征优化选择方法。基于光谱指数、穗帽变换、最小噪声分离、高斯滤波、灰度共生矩阵等变换提取了Landsat TM/ETM+/OLI影像的31维特征,提出了特征优化选择指标和模型,基于最优特征组合开展了支持向量机算法(support vectormachine,SVM)分类,识别了1996—2014年渝北区城市土地利用分布,并揭示了土地利用时空变化特征。结果表明:(1)不同特征在分类时各有优劣,多维特征组合的可分性明显高于单一特征,但掺杂无效特征会降低可分性。(2)基于全局Jeffries-Matusita(J-M)距离的特征优化选择模型(feature optimalselectionmodel,FOSM)可有效遴选最优特征组合,有利于提升分类精度。(3)1996—2014年渝北区城市用地扩张了518.11%,主要占用林地和耕地,不利于生态系统可持续发展,退耕或撂荒还林还草和耕地开荒并存降低了生态保护成效。With the improvement of temporal,spatial,spectral,radiation resolution and data processing ability of remote sensing imag⁃es,comprehensive multidimensional image features have become the key to improve the accuracy of land use classification.At pres⁃ent,not all features are helpful for classification,and the traditional classification is still constrained by a single feature.Therefore,ef⁃fective feature optimization methods are urgently needed.In this study,31-dimensional features of Landsat TM/ETM+/OLI images were extracted based on spectral index,ear cap transformation,minimum noise separation,Gaussian filtering,gray level co-occur⁃rence matrix and other transformations,and the index and model of feature optimization selection were proposed.Based on the opti⁃mal feature combination,the support vector machine(SVM)classification was carried out,and the urban land use distribution in Yu⁃bei District from 1996 to 2014 was identified,and the spatio-temporal change characteristics of land use were revealed.The results show that:(1)the OFC had a higher separability than any single feature,but introducing useless one would compromise accuracy;(2)the global Jeffries-Matusita distance(GJM)based feature optimal selection model(FOSM)is more efficient and accurate for land use classification;(3)from 1996 to 2014,the urban land area of Yubei expanded by 518.11%.This urban sprawl tended to consume cropland and forest around the city,and was detrimental to the sustainable development of ecosystem.The widespread cropland recla⁃mation along with the grain-for-green and cropland abandonment has undermined the effectiveness of ecological protection measures.

关 键 词：城市土地利用 干旱调节归一化植被-水指数(dNDVWI) 全局J-M距离(GJM) 特征优化选择模型(FOSM) 可持续城市发展 渝北区 

分 类 号：P237[天文地球—摄影测量与遥感]