The Analysis And Research Of Fire Flame Based On Computer Vision And Neural Network Technique

In the real social production,fire is one of disasters which threaten to security of our production,life,property.In the morden society,How to avoid fire is a issue that puzzled people and government all over the world,fire cause serious damage not only to the properties,but aslo to people's lifies.with the development of the social productivity and material level,issues about reducing damage and extent of the fire attract more and more attention from people and governments all around the world.Firstly,this paper introduce the purpose and significance of studying Flame Recognition, both in and out of china. Secondly,the digital Image Preprocessing technique,Image Grayness,Image Enhancement,Image Filtering,Image Edge Detection,Binaryzation and Object Extraction have been discussed.The flame object can be obtained in image sequence by integrating preprocessing algorithm.The Digital Image Preprocessing Technique do a good job in preparing for image detection.Then the Digital Image Preprocessing Technique can show three characteristics of flame image:colour feature, static feature and dynamic feature.First,the colour information can be obtained by the colour feature.Second,the static feature characterized by the image zernike feature,curvature feature and fractal feature.The static feature can shows the structure information of the flame.Three,the dynamic feature characterized by flame bodily movement,area growth,area sharp corner,altitude variation and similar shape feature.The dynamic feature can shows the flame colour feature be conditioned by complicated environment.It can identify flame and discriminate some suspected flame chaff interferent. At last the application of neural networks for image detection be disscussed and the BP algorithm basic thought of neural networks can be introduced,then put forward a improved algorithm to improve the defect of the BP algorithm,in order to make simulation experiment in the interference pattern for image flame sample and give the result and analysis to prove the superiority of the improved algorithm.