文档介绍：摘要森林是陆地生态的主体,具有很高的生态效益和经济效益。鉴于目前我国森林防火的严峻形势,必须开发有效技术解决森林火灾的监测问题。传统火灾探测器多采用单一时刻的火灾参量作为判断标准,在外界干扰下易引起频繁误报或漏报。近年来提出的基于机器视觉的火灾报警系统,利用数字图像处理技术来实现火灾自动报警。基于火焰和烟雾的图像特征,本文研究了一种识别自然环境下火灾的机器视觉方法。火灾发生过程中,主要的图像信息是燃烧时产生的烟雾和火焰,通过对烟雾和火焰的图像信息研究发现,烟雾和火焰本身具有一定的规律性,以此为依据设计有针对性的算法,从图像中识别出烟雾和火焰,判断火灾是否发生。首先,论文阐述了森林防火技术及图像型火灾检测技术的发展和现状,并对图像分割和滤波方法中的关键技术进行了详细介绍,在此基础上,本文分别讨论了火焰和烟雾的分割与识别。然后,对于火焰分割,针对不同情况下的火焰研究了三种不同的分割技术,实现了火焰区域的准确分割。对于火焰的特征检测,主要进行颜色和动态特征的分析,通过建立火焰颜色模型进行颜色识别,再进一步进行火焰的四个动态特征的识别。对于烟雾分割,由于烟雾颜色的复杂性,采取颜色提取法进行分割,并运用视觉一致性的聚类算法对其进行了改进。对于烟雾的特征检测,主要进行小波特征及动态特征的分析,通过对比烟雾图像与背景图像小波系数进行小波特征识别,再对识别结果进一步进行动态特征识别,包括烟雾的不规则性和扩散性,最终确定视频中是否存在烟雾。最后,综合以上分析,给出了森林防火系统中火灾的识别的整体流程以及火焰和烟雾分别的识别流程。实验证明,综合火焰和烟雾的静态特征及动态特征的火灾识别方法,识别率高。在火灾检测技术中,具有较好的发展前景。关键词火焰识别;烟雾识别;图像分割;动态特征 Abstract Forest is the main terrestrial ecosystem, with high ecological benefits and economic benefits. In view of the current forest fire in China's serious situation, effective technology must be develop to solve the problem of forest fire monitoring so that people's lives and property safety can be effectively protected. More traditional fire detectors use a single moment of the parameters asa standard, in the under- interference caused frequent false positives or omissions. In recent years the fire alarm system based on the machine vision uses digital image processing techniques to achieve automatic fire alarm. Based on the flame and smoke image characteristics, a machine vision method of identifying the natural fire is proposed in this thesis. In the course of the fire, the main image information is bustion of smoke and flame. Through the study of smoke and flames image information, smoke and flame phenomenon itself has certain regularity. So targeted algorithm can be designed , identify the smoke and flame from image and judge whether the fire occurred based on this kind of found. First of all the thesis explains the techniques status and development of forest fire prevention and fire-detection using digital image processing techni