Research On Detection Method Of Flame And Smog Based On Fusion Features

In order to improve the accuracy of monitoring system, the paper had researched that the preprocessing of many monitoring images, the selection of features, which involved flame and smog region, and the recognition of flame and smog area.In experimenting sparse dictionary matrixes were structured based on color moment and covariance matrix descriptor and the flame and smoke areas were decomposed by the MP and OMP methods. The paper, sparse representation was compared with AdaBoost and SVM methods, and it was found that the fusion feature could greatly improve the identification accuracy of flame region and smog region, based on sparse representation. In this paper, HOG was used to describe the dynamic and static characteristics of flame area, smog region and object region. In addition, the convex degree, diffusion feature, motion direction, motion intensity and dark channel prior theory were described. The paper, fusion feature was constructed by covariance descriptors, which were expressed in the form of vectors. According to the statistical data, the rationality of the feature selection was analyzed. The sparse expression was solved by matching pursuit and orthogonal matching pursuit.The experiment proved that the sparse expressions of flame and smog region, which were solved by orthogonal matching pursuit, had better accuracy, real-time performance and robustness.