中文摘要：

针对目前火焰图像特征不能随监控场景自适应选择的问题，提出基于概念格粗糙集一支持向量机的启发式火焰图像特征选择与探测新算法。通过对火焰特征数据离散化，建立概念格的形式背景，计算形式背景的区别矩阵，再利用属性重要性指标对不同属性的重要性进行计算，最后将得到的最简特征分类集输入支持向量机中进行测试。实验证明，该方法的识别准确率明显高于单独应用粗糙集进行特征选择和人工进行特征选择时的识别准确率，达到了提高效率，减少误报等的目的。

英文摘要：

Aiming at the problem that currently the flame image features cannot be adaptively selected along with the monitoring scenes, this paper put forward a new algorithm of heuristic flame image features selection and detection,which is based on concept lattice and rough set-support vector machine (SVM).Through the discretisation of fire flame feature data,we built the formal background of concept lattice, calculated its difference matrix,and then used attribute importance indexes to calculate the importance of different attributes,and finally we put the derived simplest feature classification set into SVM for test.Experiments proved that the recognition rate of this method was obviously higher than that of feature selection using sole rough set and manual work,and reached the goal of efficiency and false alarm reduction.