论文部分内容阅读
化学模式分类问题通常是非线性的,而且比较复杂,难以用经典统计方法建立分类判别模型。以支持向量机(SVM)构建的分类器具有更好的分类性能。对于非线性分类,SVM通过核函数将其映射到高维特征空间中,然后再进行线性分类。因此,核函数往往是决定SVM非线性分类性能的关键。实际应用时,一般通过选择几种核函数,并对其参数进行优化,然后根据分类器的预测性能来决定,训练过程非常耗时,而且结果难以保证最优。为此,采用一种通用性的核函数,即PersonⅦ核函数(PUKF),它可取代目前常用的几种核函数,可避免SVM非线性分类器训练过程的核函数选择问题。本研究将基于PUKF的SVM分类器应用于两个化学模式分类问题,均取得了较好的结果。对于多类分类,设计了一种子分类器的构造方法,它在分类性能保持较好的情况下,简化了多类分类器结构,大大降低了计算量。
Chemical model classification is usually non-linear, and more complex, it is difficult to use classical statistical methods to establish classification discriminant model. Classifiers built with support vector machines (SVMs) have better classification performance. For non-linear classification, SVM maps it into high-dimensional feature space by kernel function and then linearly classifies it. Therefore, kernel function is often the key to determine the performance of SVM nonlinear classification. In practice, several kernel functions are generally selected and their parameters are optimized, and then the performance of the classifier is determined according to the predictive performance of the classifier. The training process is very time-consuming and the results are difficult to guarantee optimal. For this reason, we adopt a universal kernel function, namely, Person VII kernel function (PUKF), which can replace several commonly used kernel functions to avoid the selection of kernel function in SVM nonlinear classifier training process. In this study, the PUKF-based SVM classifier was applied to the two chemical classification problems and achieved good results. For the multi-class classification, a sub-classifier construction method is designed. It simplifies the structure of multi-class classifier while keeping the classification performance well, which greatly reduces the computational cost.