Research On Technology Of Forest Fire Monitoring Based On Image Processing

The forest fire monitoring technology based on image processing involves the integrated knowledges of computer vision technology, digital image processing, pattern recognition, artificial intelligence, and so on. The research combines with these subjects’knowledge, by application of analysis and comparison of the forest fire image and due to pattern classification technology, the suspected flame region of forest fire image is oritented and verified, and then a series of features information are extracted by analysis of the suspected flame features, which are used to recognize the suspected flame. Finally, an image forest fire monitoring system has been designed and implemented.The forest fire monitoring system based on image processing includes four parts, which includes image acquisition, image preprocessing and segmentation of interested region, extraction of features for interested region and classification of the suspected flame. Firstly there was a brief introduction to the key technologies of Delphi7.0software development platforms used in the paper and the hardware configuration in image acquisition, meanwhile, each function module in the system is introduced briefly, and then a detailed introduction of the related technologies and implementation involved in the four parts of the system is presented.In the part of forest fire image preprocessing, we introduced some common used image enhancement technologies including gray stretching, noise filtering of the forest fire image and so on,as for segmentation of the suspected flame region, which is implemented by analyzing the color feature distribution in five color spaces(RGB,HSV,HLS,HIS and CMYK) of the suspected flame region, comparing the segmentation results of four segmentation methods(improved segmentation method based on one-dimensional Otsu, improved segmentation method based on two-dimensional Otsu, segmentation method based on two peaks in CMKY color space, segmentation method of two peaks based on improved meanshift algorithm) and then getting the optimal segmentation algorithm for forest fire image. In the part of feature extraction of the suspected flame region, several algorithms have been analyzed which mainly deals with the first-order, second-order and third-order color moment, circular degree and rectangular degree, and which are also compared by affections on the classification results of forest fire image. Based on these features, an idea of fusion feature is inferred and presented. As for recognition technologies, two common fire classification algorithms have been analyzed:template matching and Bayes, a fast fire image classification method based on AdaBoost algorithm has been posed and implemented, in which the affections of weak classifiers’number on the algorithm’s efficiency has been researched carefully and then the optimal parameter of weak classifiers’ number has been gained,meanwhile, the generic algorithm has been used to preliminary orientation of the suspected flame region, and then the information entropy theory is used for further accurate verification, combining different features of the suspected flame region, the four kinds of classification algorithms have been coded for a comparison in order to improve recognition rate and reduce time consuming.The paper integrates all discussed optimal algorithms gained from each step to recognize forest fire image, from which the experimental results show that the forest fire monitoring system researched and developed in this paper appears higher accuracy, quicker recognition speed, lower rate of false alarm and stronger anti-jamming ability, which has certain practical significance to the forest fire recognition.