红外化学遥感系统中分段线性判别算法的研究

A Study on the Algorithm of Piecewise Linear Discrimination for Remote Infrared Chemical Detection

介绍了分段线性分类器的基本原理及其在红外化学遥感监测领域应用的概况,探讨了单边式分段线性分类器应用于红外光谱鉴别的可能性.建立了一个二维数据样本集对这种分类器的判别特性进行了评价,然后用一个甲基膦二甲酯(DMMP)光谱数据样本集对这种分类器的性能进行了检验.二维数据的训练结果表明,单边式分段线性分类器对真实类界面的逼近程度高于传统的树状分段线性分类器.DMMP光谱样本的训练和预测结果表明,一个由4维判别矢量构成的单边式分段线性分类器其性能优于线性分类器和由3步7维判别矢量构成的树状分段线性分类器.

The principle of the piecewise linear classifier and its application to remote infrared chemical detection are briefly introduced. The possibility that the single\|side piecewise linear classifier can be applied to analyze chemical agent infrared spectrum is approached. A 2\|dimension random data set is built up to evaluate the discrimination features of the single\|side classifier. A dimethylmethylphosphonate (DMMP) infrared spectrum data set is used to test the performance of this classifier. It is found from the training of 2\|d data that the class interface produced by the single\|side piecewise linear classifier is approximated to the real interface to greater extent than that of tree\|like piecewise linear classifier. From the training and predicting of DMMP infrared spectrum data, the result is obtained that the performance of the single\|side piecewise linear classifier with 4 discrimination vectors is superior to that of the linear classifier and that of the tree\|like piecewise linear classifier with 3 steps and 7 discrimination vectors.\;