结合参数寻优LIBSVM模型的采矿用地提取研究

Research on Open-pit Mining Land Extraction Combined with Parameter Optimization LIBSVM Model

采用国产GF-1卫星数据,分析该数据中的光谱、纹理、植被指数、地形等特征,构建对露天采矿用地提取具有一定应用价值的多特征集合,最后对比应用价值较高的BP神经网络和改进的参数寻优LIBSVM模型对露天采矿用地的提取效果。结果表明:1)在2种算法的应用中,可以获取较高精度的特征集合都为B(光谱)+G(纹理)+V(植被)+D(地形);2)结合参数寻优的LIBSVM模型与B+G+V+D特征集对露天采矿用地提取可以获取较高精度,总体精度为91.76%,F1函数为94.55%,Kappa系数为0.8972。基于B+G+V+D的多特征集合,结合参数寻优的LIBSVM模型在露天采矿用地提取中具有较高的应用价值,为自然资源管理部门提供了一种较好的管理方法。

This paper uses the domestic GF-1 satellite data to analyze the spectrum,texture,vegetation index,terrain and other features on the data,and constructs a multi-feature set with certain application value for the extraction of open-pit mining land,and finally compares the BP neural network with higher application value and the extraction effect of the improved parameter optimization LIBSVM model for the open-pit mining land.The results show that:(1)In the application of the two algorithms,the feature sets that can be obtained with higher precision are B(spectrum)+G(texture)+V(vegetation)+D(terrain);(2)Combined with parameter optimization LIBSVM model and B+G+V+D feature set,high precision can be obtained for open-pit mining land extraction,with the overall accuracy of 91.76%,F1 function of 94.55%,and kappa coefficient of 0.8972.Based on the multi-feature set of B+G+V+D,the LIBSVM model combined with parameter optimization has good application value in the extraction of open-pit mining land,and provides a better management method for natural resource management departments.