A hybrid feature selection and classification strategy was proposed based on the simulated annealing genetic algonthrn and multiple instance learning （MIL）. The band selection method was proposed from subspace decom...A hybrid feature selection and classification strategy was proposed based on the simulated annealing genetic algonthrn and multiple instance learning （MIL）. The band selection method was proposed from subspace decomposition, which combines the simulated annealing algorithm with the genetic algorithm in choosing different cross-over and mutation probabilities, as well as mutation individuals. Then MIL was combined with image segmentation, clustering and support vector machine algorithms to classify hyperspectral image. The experimental results show that this proposed method can get high classification accuracy of 93.13% at small training samples and the weaknesses of the conventional methods are overcome.展开更多
文摘A hybrid feature selection and classification strategy was proposed based on the simulated annealing genetic algonthrn and multiple instance learning （MIL）. The band selection method was proposed from subspace decomposition, which combines the simulated annealing algorithm with the genetic algorithm in choosing different cross-over and mutation probabilities, as well as mutation individuals. Then MIL was combined with image segmentation, clustering and support vector machine algorithms to classify hyperspectral image. The experimental results show that this proposed method can get high classification accuracy of 93.13% at small training samples and the weaknesses of the conventional methods are overcome.
文摘土壤重金属污染高光谱反演的特征波段提取方法和反演模型的选择是影响反演精度的关键;二者如何优化组合,提高反演精度是目前亟需解决的难题。在华南典型铬(Cr)污染区,采集了92组土壤样品,使用电感耦合等离子体质谱(inductively coupled plasma mass spectrometry,ICP-MS)检测Cr含量,并使用ASD Field Spec4地物光谱仪在实验室收集其高光谱信息。光谱信息预处理采用平滑滤波(SG)+标准正态化(SNV)+二阶微分(SD)变换组合,减弱土壤散射和噪声的影响。选择竞争性自适应重加权采样(CARS)、逐步投影算法(SPA)、无信息变量消除(UVE)、遗传算法(GA)四种算法提取特征波段。选择多元线性回归(MLR)、偏最小二乘法(PLSR)、支持向量回归(SVR)和人工神经网络(ANN)四种反演模型建立特征波段与Cr含量之间的关系。通过对比不同特征波段提取方法和反演模型组合对土壤Cr含量反演的结果发现:采用CARS和UVE特征波段提取方法可以显著提高PLSR、MLR和SVR模型的预测效果;SPA方法能够提高ANN模型的预测效果;通过SG+SNV+SD+CARS+PLSR组合方式,提取位于800~1000、1400~1700以及2100~2450 nm之间的98个特征波段,建模后模型验证,决定系数R2为0.97,均方根误差RMSE为5.25 mg·kg^(-1),平均绝对误差MAE为4.35 mg·kg^(-1),相对分析误差RPD为3.94,表明该模型在预测土壤Cr含量具有优异的性能。以土壤Cr污染高光谱反演为例,通过比较不同特征波段提取方法与反演模型组合的反演精度,确定最优模型,为小样本土壤重金属污染反演的建模提供了思路。
