Flame detection based on GBDT feature for building

摘要

Usually the static or dynamic characteristics of the flame are extracted for flame detection. But the relationship between the various features of flame could not be distinguished by the human eye. the Gradient Boost Decision Tree (GBDT) is thus proposed to combine and optimize the flame shape and texture features, so as to mine the relationship of flame features. Then the more discriminant new flame features is formed and the multicollinearity among the new features could be reduced by information entropy. Based on the new flame features, logistic regression (LR) classifier is used to discriminate whether there is a flame. Experimental results show that the proposed method has the advantages of high detection rate, strong robustness, and a good application prospect.