Study On Identifying Method For Forest Fire Based On Support Vector Machine(SVM)

The forest fire is one kind of common natural disasters and cause serious ecological destruction and economic loss. Currently, the field fire alarm has been more mature. The fire is detected by light, smoke, temperature sensor and so on. Then we put out fire, cut off electricity, sprinkle water and alarm directly. But the forest fire is take place in large space, so the traditional detectors can’t play a role. At present, to prevent forest fires, forest fires intelligent recognition is an urgent need to solve the problem.Applicable to forest fire detection, fire detection method based on support vector machine is a new and effective fire detection technology. It can effectively overcome the defects of conventional fire detector. This paper which include image enhancement, image segmentation, morphological image processing, feature extraction, support vector machines to in-depth analysis of forest fire detection, forest fire detection method based on support vector machine on this basis.First of all, The system using CCD camera acquisition forest fire video and convert video into BMP image, and then collect RGB color images and change it into the gray image, specific to the complex forest fire image used median filtering pre-processing filter technology to be enhance.Secondly, the system using HIS color segmentation, regional growing, morphological technology to eliminate noise and accurately segment the flame area from the images. And put forward method on image segmentation of fire based on HIS and region growing.And then analyses the color characteristics, shape characteristic and dynamic characteristics of the suspected flame area, put forward by using inter-frame similarity, red saturation, value of area changing, eccentricity five characteristics and get the flame and suspected the characteristics of flame area data.Finally, model of the separability of the establishment of fire support vector machine to identify suspected fire characteristics data is extracted as the input to identify images of forest fires. The results show that:using a suspected fire area features support vector machine identification method can effectively identify forest fires, the model has good resistance to the recognition accuracy rate of97%.