多特征融合与相关向量机的火灾烟雾识别方法

A Method of Identification for the Smog in the Fire Based on Multi-feature Fusion and Relevance Vector Machine

针对当前单一特征以及简单组合特征描述火灾烟雾状态的不足,以提高火灾烟雾识别准确性为目标,本文提出了一种多特征融合和相关向量机的火灾烟雾识别方法(MF-RVM)。首先获取火灾烟雾的可疑区域,并提取火灾烟雾可疑区域的静态和动态特征,然后利用主成分析法对静态和动态特征进行融合,消除特征之间冗余,最后利用相关向量机对融合特征进行训练,建立火灾烟雾识别模型。采用多个火灾烟雾视频图像在Matlab2012平台上对MF-RVM的识别性能进行仿真测试。结果表明,MF-RVM能够有效地对火灾烟雾进行识别,平均识别率达到了95%以上,并且提高火灾烟雾识别效率,以满足火灾烟雾识别的实时性要求。

Aiming at a defect in identification for the fire smog described by a single feature and a simple combination in order to improve the accuracy of identification for the fire smog, this paper put tbrward a new method which fire smog was identified by the Multi-feature thsion and Relevance Vector Machine（MF-RVM）. Firstly, the static and dynamic features were obtained in a suspected fire smog area and then combined them with principal component analysis to eliminate the redundancy message betweeii foatures. Lastly, relevance vector machine was used to train the fusion tbatures and established an identification model for a fire smog to carried out the simulation test on the Matlab 2012 platform. The results showed that the proposed method could effectively, identify a fire smog to be more than 95% an average recognition accuracy and increase the efficiency of identification so as to satisfy the real time requirements of identification for fire smog.