近年来，多传感器数据融合技术和支持向量机算法越来越受到重视并得到广泛应用。为了解决多类分类问题，论文将支持向量机算法引入到数据融合中，把多传感器数据融合技术与支持向量机算法结合，保持各自的优点，弥补不足之处，并以车内有害气体多传感器观测数据作为实验对象，对支持向量机在多传感器数据融合中的应用做了分析研究。 论文在介绍数据融合原理的基础上，分析了多传感器数据融合结构、层次以及几种常见的融合方法，包括自适应加权融合方法、贝叶斯融合方法以及卡尔曼滤波融合方法，对这几种融合方法加以比较并总结了各自的优缺点。从统计学习理论相关原理出发，讨论了支持向量机分类原理，分析比较了几种常见的多类支持向量机算法，包括一次性求解算法、一对多算法、一对一算法、有向无环图SVM算法，决策树支持向量机算法等，总结了它们的性能及优缺点。论文在决策树支持向量机理论基础上给出了基于改进决策树支持向量机和卡尔曼滤波的数据融合算法，建立了混合式融合算法的模型；采用两层融合方式，在数据层中利用卡尔曼滤波融合去除数据样本中大量的噪声污染，并将改进决策树的支持向量机融入到特征层中，用来解决多类分类问题。 最后，将改进决策树支持向量机和卡尔曼滤波的多传感器数据融合算法应用于车内有害气体识别中，通过交叉验证法选择令分类识别错误率较低的核参数，并以此得到支持向量及多类分类识别决策树模型，实现了对有害气体的有效识别。仿真结果表明该算法能够有较快的运算速度，同时可以得到较为准确的融合分类结果。

In recent years, multi-sensor data fusion technology and support vector machines attract more and more attention and are widely used. In order to solve the problem of multi-class classification, the support vector machine algorithm for data fusion is introduced into this thesis, and the multi-sensor data fusion technology is combined with support vector machines to maintain their advantages and to make up for the deficiencies. The application of support vector machine to multi-sensor data fusion is analyzed; meanwhile, the multi-sensor observation data of the harmful gases inside a car are used as the experiment object. Based on the introduction of data fusion theory, the analysis of multi-sensor data fusion structure and hierarchy and several common methods are conducted. The common methods include the adaptive weighted fusion method, Bayesian fusion method and Kalman filtering fusion method. These fusion methods are compared and summarized to get their advantages and disadvantages. From the relative principle of the statistical learning theory, this thesis discusses the support vector machine classification principle; and analyzes and compares of several common multi-class support vector machine algorithms which include one versus one algorithm, one versus rest algorithm and the acyclic graph algorithm, decision tree support vector machine. The performances and advantages and disadvantages of these algorisms are summarized. Then the improved decision tree support vector machine data fusion algorithm is given and a hybrid model of fusion is constructed. There are two layers in the model. Kalman filtering is used to removal the noise pollution in the data layer. The skewness support vector machine decision tree solves the multiclass classification problem in the characteristics of layer. Finally, the improved decision tree support vector machine data fusion algorithm is applied to the recognition of harmful gases inside the vehicle. The cross-validation method is used to choose the applicable kernel classification parameters. Then recognition model with the applicable multi-class support vector classifier is realized to classify the harmful gases. Simulation results show that the algorithm can effectively improve the operation speed of data fusion, and get a more accurate classification results.