|
In the reality of social life and production , the fire is a threat to the safety of life and production , one of the disaster threat to people 's lives and property safety . Community , fire is a common problem of people around the world and the government is troubled not only bring serious loss of property , but also brought about the hazards of life . With the increasing level of social productive forces and material , to prevent the rise of the fire damage , to prevent a fire hazard range expansion , more and more attention of the Governments and people of all countries . This paper first introduces the purpose and meaning of flame recognition , domestic and foreign research status . Followed by discussion of the digital image preprocessing techniques : the graying of the image , image enhancement, image filtering , image edge detection and binarization , image motion detection and object extraction . Flame target in a sequence of frames of the image extracted by the preprocessing algorithm integrates . Digital image pre-processing technology of image recognition and feature extraction to do preparatory work . Then asked the three aspects of the flame image features: color characteristics , static characteristics , dynamic characteristics . First , based on the color features can be the color of the flame information extracted . Second , the static characteristics : characteristics of the image of the moment , curvature characteristics of fractal characteristics . The static characteristic may be flame structure information is extracted . Third, the dynamic characteristics of performance : the overall movement of the flame , the area of ??growth , regional miter, and a high degree of change in shape similar characteristics . Dynamic characteristics can be affected by complex environmental impact flame feature extraction . By color , static , dynamic characteristics can basically identify flame and distinguish between suspected flame disruptors . Finally, the neural network image recognition , introduced the basic idea of the BP neural network algorithm , and propose an improved algorithm for BP algorithm shortcomings , the flame image samples in the interference pattern environment simulation experiments and simulation experiment results and analysis , to verify the superiority of the improved algorithm .
|