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

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...

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

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.