基于电子鼻的SVM模式识别算法的研究与实现
传统的模式识别算法需要有足够多的样本进行训练，而实际情况下，样本数往往难以达到要求，而在样本数有限的时候又很难达到预期的学习效果。支持向量机（SVM）是1995年提出的新模式识别算法，其能解决小样本情况下机器学习困难的问题，同时克服了神经网络过学习的难题，所以本文选择使用支持向量机作为电子鼻上的模式识别算法。
本文研究了机器嗅觉中的数据预处理方法，应用min-max线性转换进行数据归一化，并使用PCA进行降文。本文还具体研究了SVM分类方法的基本原理及相关的实现方法。在参考了国内外学者进行的研究并结合相关的理论研究和实验方案后，提出了三种针对SVM参数寻优的方案——遗传算法、粒子群算法、网格搜索，并采用交叉验证的方法自主设置训练和测试的比例来验证模型的准确率。
最后算法验证中，本文对比了三种参数优化方法在6种气体数据集上的识别结果。首先，从是否进行降文数据预处理进行比对，分析了进行降文处理的必要性，因为采用的数据集文数较高（128文），发现进行降文后参数寻优和训练的时间效率均提升了50%以上，正确率各提高了1%~2%不等，说明在数据集文数较高时进行降文是正确而且是很必要的。然后，从时间效率上和准确率上进行了比对，分析了三种参数优化方法的优缺点：当数据集包含的样本数目比较小时，利用网格搜索算法优化支持向量机的参数比较理想；当数据集包含的样本数目较大时，可以选择遗传算法或粒子群算法优化支持向量机参数，遗传算法优化支持向量机参数虽然在运行时间上不如粒子群算法，但是在分类正确率上占优势。
毕业论文关键词： 电子鼻；支持向量机（SVM）；特征提取；模式识别
Based on the electronic nose SVM pattern recognition research and implementation of the algorithm
Abstract:The electronic nose has raised along with the development of machine learning and artificial intelligence.It is the main approach to solve the difficulty about identification of odor by machines. In the recent two decades, for the both domestic and overseas study of the electronic nose mainly consists two aspects: １: hardware electronic nose system research and the composition of sensor array, 2: the improvement of the electronic nose pattern recognition algorithm.Pattern recognition is to collect the sensor array of data sets by data pre processing and then to study the qualitative or class model, and finally to identify the unknown smelling. Therefore,the research of electronic nose pattern recognition algorithm has relatively high theoretical and practical significance.
The traditional pattern recognition algorithm based on statistics requires having enough samples,however,in the actual situation,it is often difficult to meet the requirements, and limited number of samples increased the difficulty to achieve the desired learning effect. Support vector machine (SVM) was put forward in 1995 as a new pattern recognition algorithm, it can solve the problem of small sample situations, and overcome the difficulty of neural network through learning at the same time, so this article choose support vector machine (SVM) as the pattern recognition algorithm on the electronic nose.