支持向量机在危险化学品被动红外遥测光谱鉴别中的应用

Applications of support vector machine in identification of passive IR remote sensing spectrum of dangerous chemicals

为提高危险化学品被动红外遥测光谱鉴别正确率,提出应用支持向量机建立鉴别模型。利用野外实测氨气被动红外遥测光谱样本集,变换惩罚因子C对比高斯核函数与多项式核函数的效能,结合网格遍历法搜寻最佳模型参数,建立了基于支持向量机的鉴别模型。基于40个训练样本得到的模型,对包含267个样本的测试样本集的鉴别正确率可达93.6%,明显优于3层网络结构的BP神经网络鉴别模型。实验结果表明,支持向量机鉴别模型是一种有效的危险化学品红外遥测光谱鉴别方法。

To improve the identification effect, a sample set of passive IR remote sensing spectrum of NH3 is obtained in the field, which is used to build an identification model based on support vector machine. Penalty factor is transformed to compare the effects of Gaussian kernel and polynomial kernel, and the grid traversal method is used to search for the best model parameters. The discriminant rate of test sample set cantaining 267 samples is 96.3% based on a support vector machine model buit with only 40 samples, which is better than the BP Neural Network model obviously. It suggests that the support vector machine model is appropriate to identify passive IR remote sensing spectrum of dangerous chemicals.