基于图像型的煤矿早期外因火灾预测及识别方法研究

针对传统煤矿井下火灾预测易受环境影响的不足,文中提出了一种基于图像型的煤矿外因火灾预报方法。文中基于火灾图像特征根据运动检测提取出疑似火灾区域,再根据颜色决策排除掉不具有火焰颜色的物体,最后再利用离散分形布朗增量场进行识别。实验结果表明,该算法对于小样本、非线性的分类问题效果显著,该法能较好地将煤矿井下的火源与干扰源区别开来,有较高的识别率和较强的鲁棒性,可降低误报、漏报。

Image-based prediction and recognition method for exogenous-fire in coal mine

Considering the impacts of environmental factors to the traditional fire detection in coal mine, a new methodology based on image for exogenous fire is proposed. Firstly, the image features are ana lyzed, the candidate fire region is extracted by motion detecting, then the objects without fire color are eliminated, finally, the images are classified by Discrete Fractal Brownian Incremental Random field. It is demonstrated by experiments that the approach is significant to small samples and nonlinear problems. It can classify the potential fire source and interference sources with high detection rate and strong ro bust. The false positive and leakage forecast can be reduced