Research For Fire Image Recognition Technology In Field

The fire is one kind of common natural disasters. Currently, the field fire alarm has been more mature. The fire is detected by light, smoke, temperature and so on. Then we put out fire, cut off electricity, sprinkle water and alarm directly. But in large space, especially in outdoors, the early fire detection in the world is still a problem. Because there are many factors like special height, thermal barrier, coverage, long-distance signal transmission and so on which affect the fire detection. In these circumstances, traditional detection often has no means. As the image fire detection has non-contact detection features, it becomes an effective mean of fire detection in large factories, warehouses, forests and open space. Through the research and discussion of digital processing technology, we design a variety of detection algorithms and image fire recognition system based on Neural Network.First, the principle of fire occurring, three developing stages and characters of every stage have been discussed systematically in this dissertation. Considering the domestic and international researching about the image recognition technology of the fire detection, the thesis introduces the classification of image fire detection, detection devices, the composition of the fire detection system and basic principle of digital image processing in detail.Then, the algorithm of infrared fire image gain, enhancement, noise filtering, threshold segmentation, region growth and their applications in the fire image processing have been studied in this dissertation. The fire image algorithm which chaging multicolor to gray has been improved and a three-values algorithm which suits the fire image division has been proposed. On the technology of fire detection, the thesis finds out a series of algorithms for the recognition of flame shape features. For improving efficiency, the algorithm of region segmentation is brought forward, which makes the average processing time of the algorithm lessen to 1/4 comparing the classic algorithm.For the fire image recognition, stress has been put on the early fire flame image recognition. The features of fire flame such as area variety, edge variety, shape variety and the whole motion trend have been detailed studied in this dissertation. What's more, the flame flicker frequency is considered as a criterion of fire recognition, which enhances anti-jamming ability.Furthermore, this thesis has discussed the fire recognition scheme based on neural network. The basic concept of nerve net has been introduced at first. And then the topology structure of BP neural network and the design precept of input and output layer have been given. At last, the fire recognition system based on BP neural network has been realized. The experiment results show that the BPNN fire recognition system has good feature of robustness.Finally, we summarize the research work. Then we point out the shortcomings and prospect for the future work.