Automatic Recognition Of Forest Fire

Forest Fire is increasingly emphasized by the government as an important natural disaster in the world. Forest Fire has the characteristics of rapid transmission and the difficulty of fire extinguishing and rescuing, therefore, the problem of avoiding forest fire effectively and detecting forest fire as soon as impossible becomes more and more urgent. As a new-style and effective measure for the detection of early fire, much attention has been focused on image fire detection technology.At first, the main physical phenomenon during forest fire happens, has been introduced systematically in this dissertation. Based on this, we discuss the early forest fire smoke character and flame character. The law and characteristic property have been discovered and provide important basis for forest fire detection.Moreover, according to the analysis of the threshold method for image segmentation based on the traditional two-dimensional gray level histogram which partitions the areas with some wrong pixels and processes slowly, the two-dimensional gray level histogram is improved and a novel threshold method based on this two-dimensional histogram gray level histogram is brought forward, and the particle swarm optimization algorithm is realized successfully in the process of searching the optimal solution. Fire image segmentation is done by using the optimal solution.At last, early forest fire image recognition has been discussed in the paper, and the recognition was made by using RBF neural network in combination with the characteristic of the flame. According to the analysis of the features of fire flame such as area growth, edge variety, shape variety and the whole motion trend in detail, and put forward the corresponding image recognition algorithm for each characteristic. We can recognize forest flame image by using these algorithms. The experiment results show that the RBFNN fire detection system has good anti-interference capability.