基于LSA和SVM的火灾烟雾检测算法

为了解决视频烟雾检测中特征提取难度较大、复杂度较高的问题,提出一种基于潜在语义(Latent Semantic Analysis,LSA)特征和支持向量机(Support Vector Machine,SVM)的烟雾检测算法。该算法首先将烟雾图像库中的每幅图像进行有重叠分块,提取每个分块的小波纹理与HSV颜色特征;再对所有分块特征进行聚类、量化成"视觉字",并且根据每个"视觉字"在每幅烟雾图像中出现的频率,建立"词-文档"矩阵;然后采用LSA方法获得每幅烟雾图像的潜在语义特征;最后结合SVM,实现视频烟雾检测。对比实验表明,该算法特征提取简便,可以更快检测烟雾的发生,提高了烟雾检测效率。

A fire smoke detection algorithm based on SVM and LSA features

Feature extraction algorithms in video smoke detection are complex with high computational load which make it difficult to realize real-time smoke detection.To solve this problem,a smoke detection algorithm based on LSA feature and SVM is proposed in this paper.In this algorithm,firstly the images in the library are divided into overlapping blocks and both wavelet texture and HSV color feature of each block are extracted.Secondly,the feature of each block is clustered into different class and every block is quantified into a＂visual word＂.According to the frequency of each＂visual word＂appearing in each smoke image,a＂word-document＂matrix is established.Thirdly,latent semantic features of each image are obtained by using latent semantic analysis（LSA）method.Lastly video smoke detection is achieved by the LSA-SVM algorithm combining support vector machine（SVM）.Comparative experiments show that this algorithm is simple and convenient in feature detection,can detect smoke more quickly,and therefore can improve the efficiency of smoke detection.