Recognition And Forecast Of The Fire Based On The Video Image

Fire is one of the most common serious disasters in human society, especially largespace building fires, mine fires, forest fires, tunnel fires and other accidents to human life andserious property damage. Therefore, in order to reduce the maximum degree of harm causedby fire, to strengthen the real-time monitoring and forecasting of fire is particularly important.Traditional temperature, smoke detection method style often results in a complexenvironment is not ideal, and image-based fire detection has a non-contact detection offeatures, so the technology is effectively used in the buildings, forests, tunnels and other largespace and the wild field for fire detection. In this paper, based on the characteristics of the fire,the fire will fuse a variety of feature information, in order to construct a neural network-basedfire image recognition system.. Through the early identification of smoke and fire through theearly identification for early warning of fire.This paper systematically discusses the physical characteristics of fire and some of thetraditional lack of fire detection technology, and describes the current new fire detectiontechnology, image analysis, based on this type of fire detection technology advantages andcharacteristics.The paper also studied the difference method of image to detect moving objects, imageenhancement, image segmentation and edge detection algorithms used in the fire imageprocessing. Based on the traditional Otsu thresholding algorithm,ant colony algorithm andmathematical morphology is proposed for improving Otsu thresholding algorithm.Ant colonyalgorithm is a random search algorithm whichi inspired by true cooperative behavior ofgroups, considerably reduced the times of traditional Otsu threshold algorithm, and achievingthe binary thresholding image filtering. through mathematical morphological operators. Thealgorithm is effective to improve the efficiency of image segmentation, improvesegmentation results, especially for such large-scale image-based fire detection imageprocessing system, will greatly save system cost.The paper make a detailed analysis of the fire flame and other sources of interferenceincluding color features, morphological characteristics, the changes in size and masscharacteristics and texture features, on this basis, then measure the GLCM of image to extractenergy, entropy, moment of inertia and local smoothness feature vectors as identified andmake an analysis of these characteristic of the law of development, the amount of some of thecharacteristics are selected as neural network input vector. Based on the neural network, the combination of genetic algorithm and BP neuralnetwork is proposed for fire image recognition. Finally, the extracted image features fireinformation as neural network input for training neural network, and use a series of firesimage and some sources of interference images samples for verification of the neuralnetwork.Experimental results show that using the neural network can identify the image offire quickly and effectively and it owns a strong capability of anti-jamming.Comprehensive analysis of early fire smoke characteristics。 Through watershedsegmentation algorithm baseed on morphological reconstruction marks to achieve imagesegmentation of smoke, then make a analysis of similarity characteristics of the earlysequence images of fire smoke and other sources of interference such as cloud. Experimentsshow that: the algorithm to distinguish the early fire smoke and other sources of interferenceeffectively, and early fire smoke image recognition can play a good role in fire prediction.