Research On The Method Of Forest Fire Recognition Based On Digital Image Processing

Forest fire is one of the most devastating disasters for the social,environment and economic development.In order to reduce the loss of forest fire,every country attaches great importance to the monitoring of forest fire around the world.Traditional methods for fire detection are smoke detection, temperature detection,light & infrared detection,and satellite monitoring.For such a large space of forest fire monitoring,a fire detection method based on the visual characters of image is proposed.Based on the analysis of development of monitor to forest fire home and abroad and advanced technologies,this paper is trying to combine digital image processing technology with BP neural network pattern recognition theory,and applying them to the fire detection process.(1) Based on probability sampling theory and signal theory,propose a method to sample from original image.It can reduce the amount of data,without losing the information of image.(2) Based on fully analyzing the common method and theory of image segmentation,propose a suitable method to fire detection,which is based on RGB and HIS.(3) The fire images were processed by the following steps:threshold of segmentation,median filter and gray level transformation,which rely on the image processing technology.Seek the best method through theoretical analysis and experimental contrast.(4) This paper described the principle of fire feature extraction and analyzed the characteristic quantity which is common used to be extracted and the factors of fire formation,thus shape feature, color feature and dynamic characterstics were chosen as fire pattern characteristics to be the multidimensional input of the BP neural network.(5) Based on the algorithms of fire feature extraction,the software of fire feature extraction system is developed.For ensuring the expansibility of the source code,each algorithm was designed by modular programming.(6) According to numerous specific fire images which were collected and analyzed a fire BP neural network pattern recognition system was established by carrying out Matlab engine calling the functions provided by Matlab neural network toolbox in VC++6.0 development environment.The feasibility of the system could be proved from three areas which included the recognition accuracy and minimum resolution.