Research Of Fire Feature Classification And Detection Algorithm Based On Concept And Rough-svm

Fire detection technology based on video image can effectively solve the problem of fire detection in large space, and has important application value. The technology includes video capture, image preprocessing, segmentation, feature extraction and flame identification several parts. In order to improve the accuracy and robustness of detection algorithms, aiming at the inadequacy of existing algorithms, this paper researches on related algorithms of flame segmentation, feature selection and identification. Mainly in the following aspects:(1) Research on flame segmentation algorithm. Target segmentation is the foundation and key of feature extraction and object recognition. It has great significance in reducing data amount and improving algorithm performance. According to the analysis and comparison of existing algorithms, this paper proposes a flame segmentation method based on CIE Lab color model and fuzzy C-means. Firstly, the flame color model is built in CIE Lab space and divided by color components, then the discrete targets in the same area are combined with fuzzy C-means algorithm, so the accuracy of the flame suspected segmentation is improved.(2) Research on flame feature selection and recognition algorithms. The principles and characteristics of fire detection is outlined and a new flame image feature selection and detection algorithm based on concept lattice rough set and support vector machine is proposed, for the current problem that the flame image features can not be selected adaptively with the monitor scene. Firstly the flame characteristics is analyzed and calculated, then the attribute reduction methods of each concept lattice and rough set is summarized, and their characteristics and advantages of the reduction are analyzed, and further the concept lattice and rough set are combined to apply in the characteristics reduction of fire flame. Finally the minimalist feature set is input to support vector machine for testing. Experimental results show that the recognition accuracy of this method is significantly higher than that only use rough set feature for feature selection and artificial selection, it achieves the purpose of improving efficiency and reducing false positives.