基于图像分割的森林火灾早期烟检测算法研究

针对森林这样的大空间、复杂场景下的火灾检测，提出一种在单帧视频序列图像中的烟检测方法，并研究一种新的超像素合并算法，改进现有的天地线检测算法。该方法对图像进行SLIC（Simple Linear Iterative Clustering）超像素分割，并用一种新的超像素合并算法解决过分割问题；通过改进的天地线分割算法，排除天空中云对于烟检测的干扰；根据光谱特征，运用支持向量机（SVM）对超像素块进行分类。实验结果表明，超像素合并算法高效简洁，易于编程实现，基于图像分割的烟检测技术能排除云雾等噪声对烟雾检测的干扰，在森林场景下的烟雾检测正确率为77%，可以作为人工森林火灾监测的辅助手段。

Early forest fire smoke detection algorithm based on image segmentation research

Focused on fire detection in large and complicated space such as forest region, a method is presented for forest fire smoke detection in a single frame, and a new superpixels merging algorithm is studied, and then an existed horizon detection algorithm is improved. Firstly, Simple Linear Iterative Clustering（SLIC） is performed to compute superpixels and a new superpixels merging  algorithm is used. Secondly, a robust horizon detection algorithm is proposed to reduce false alarm rata by cloud interference. Finally, these superpixel areas are characterized using spectrum information, and support vector machine classification is applied to distinguish smoke and background. Experimental results show that this supepixels merging algorithm is efficient and concise, which is easy to be programmed. The smoke detection algorithm is based on image segmentation method, which can reduce noise interference, and smoke detection rate is 77% under forest scene.