基于多光谱图像的烟雾检测

烟雾检测对于火灾早期防范非常重要，传统的智能视频和图像处理技术易受背景运动信息影响，抗干扰性差，且不容易区分森林水雾和燃烧产生的烟雾，森林防火误报率高。为此提出一种新的多光谱图像检测方法检测烟雾。采用多光谱成像系统，获取400至720 nm波段范围的烟雾、水雾光谱图像序列，对图像进行分层像素整合处理;利用欧氏距离度量不同分块光谱特征差异，获取动态区域光谱特征向量，根据目标与背景间光谱特征向量差异，提取烟雾、水雾区域。室内外试验结果表明：多光谱图像检测方法可用于烟雾检测，能够有效地检测并区分烟雾和水雾，与视频图像方法结合，可有效地用于森林火灾监测，降低森林火灾检测误报率。

Detecting Fire Smoke Based on the Multispectral Image

Smoke detection is very important for preventing forest-fire in the fire early process. Because the traditional technolo-gies based on video and image processing are easily affected by the background dynamic information, three limitations exist in these technologies, i.e. lower anti-interference ability, higher false detection rate and the fire smoke and water fog being not eas-ily distinguished. A novel detection method for detecting smoke based on the multispectral image was proposed in the present pa-per. Using the multispectral digital imaging technique, the multispectral image series of fire smoke and water fog were obtained in the band scope of 400 to 720 nm, and the images were divided into bins. The Euclidian distance among the bins was taken as a measurement for showing the difference of spectrogram. After obtaining the spectral feature vectors of dynamic region, the re-gions of fire smoke and water fog were extracted according to the spectrogram feature difference between target and background.The indoor and outdoor experiments show that the smoke detection method based on muhispectral image can be applied to the smoke detection, which can effectively distinguish the fire smoke and water fog. Combined with video image processing method, the multispectral image detection method can also be applied to the forest fire surveillance, reducing the false alarm rate in forest fire detection.