基于Gabor小波的火灾烟气识别新方法.pdfVIP

学兔兔 第31卷 第1期 仪 器 仪 表 学 报 、，0l-31 NO．1 2010年 1月 Chinese Journal of Scientific Instrument Jan．2010 基于Gabor小波的火灾烟气识别新方法术 吴章宪 ，杨国田 ，刘向杰 ，杨鹏远 ，刘思飞 (1 华北电力大学自动化系 北京 102206；2 中国科技大学电子科学与技术系 安徽 230027) 摘 要：研究提出一种基于Gabor小波的火灾烟气模型，以提高基于视频图像的火灾烟气检测的准确性。本文采用混合高斯 模型与色度方差相结合的方法定位火灾烟气目标区域，并基于Gabor小波对烟气纹理及边缘特征进行描述，从而提出烟气变 化能量模型与方向角分布模型来分析纹理变化的动态性质。最后采用 Gentle Boosting设计分类器对模型特征进行分类与匹 配，并实现在训练的同时进行最优特征选择。与传统方法相比，基于本文方法的火灾烟气榆测在降低漏报率与误报率方面拥 有更好的性能。 关键词：烟气建模；Gabor小波；混合高斯模型；Gentle Boosting 中图分类号：TP391 文献标识码：A 国家标准学科分类代码：520．2040 Novel method for fire smoke recognition based on Gabor wavelet wu Zhangxian ，Yang Guofian ，Liu Xiangjie ，Yang Pengyuan ，Liu Sifei (1 Department ofAutomation．North China Electric Power University,Beijing 102206,China；2 Department ofElectronic Science and Technology,Chinese University ofScience and Technology,Anhui 23002Z China) Abstract：In 0rder to improve the performance of fire smoke detection based on video sequences。a novel model for fire smoke detection is proposed in this paper．An algorithm combining Gaussian mixture model and color smoothness function is presented to locate the suspicious smoke regions．Gabor wavelet is utilized to describe the textures and edges of the fire smoke．Meanwhile，to analyze the dynamic characteristics of the fire smoke textures， energy vafi~ion model and model of texture varying orientation are proposed．Gentle Boosting is employed to design the classifier and match the features．Optimal features are selected during the training process．Experimen- tal results prove the superior perform ance of the proposed method in reducing the rates of both misrecognition and false recognition compared with previous methods． Key words：model of fire smoke；Gabor wave