Research Of Fire Image Recognition Technology In Field Based On Dsp

Fire is one of the most common serious natural disasters. Currently, the fire in indoors is detected by light, smoke, temperature and so on. Then we take measures to put out fire, cut off electricity, sprinkle water and alarm directly. But in large space, especially in outdoors, traditional detection usually has no means because there are many factors that affect the fire detection, such as special height, thermal barrier, coverage, air speed. In these circumstances, we can only use image-based fire detection technology. The fire detection system based on PC is bulky, inconvenient to install, and lack of flexibility in use. The signal transmission is complex and the burden on the computer is large. It can not be applied in some places of computer inconvenient placement. Thus, image-based fire detection technology based on DSP can overcome these shortcomings. It can reduce the burden on the computer to raise the level of the system's intelligent.This thesis analyzes the domestic and international researching about the image recognition technology of the fire detection. Considering that, the thesis introduces the application of digital image processing and DSP in image recognition technology of the fire detection. It introduces the components of image-based fire detection system based on DSP, including color cameras with infrared filters, fire image capture and processing platform.This thesis focuses the recognition technologyon of the early fire flame image based on DSP embedded platform. It discusses image pre-processing, the characteristics of the flame criterion and flame recognition. In the DSP platform, it uses median filter, OTSU method for image binarization, regional growth and removal of interfering area for image pre-processing. We propose five flame criterions. Eigenvalues of the flame criterions are extracted in the DSP platform. In the PC, the weights and thresholds calculating by BP neural network training with the input signal of each value of criterions enhance the system's ability to identify the flame. Then, we use the weights and thresholds in the DSP to identify the images and do a large number of fire recognition experiments. The results of the experiments show that the flame recognition's ability and accuracy of the system has been improved.Finally, we summarize the research work and point out the shortcomings and prospect for the future work.