Research On Safety Early Warning Technology Of Fire Smoke Detection Based On Image Processing

Fire,as a kind of uncontrolled combustion,is a factor threatening the safety of human production and life.Hence,fire early warning technology is increasingly concerned.With the rapid development of computer vision,image processing and pattern recognition technology,a new idea for fire detection is available.The generation and diffusion of smoke in a certain space is one of the remarkable characteristics in the early stage of fire,so video smoke detection is of great significance to fire early warning.Based on digital image processing technology,this paper probes into moving object detection of smoke images,the feature selection and classification algorithms of smoke regions.The main research contents and achievements of this paper are as follows:(1)In the suspected smoke object detection part,an ameliorative background update arithmetic was proposed for background modelling,and the suspected smoke area was delineated by the method of background subtraction to improve the foreground segmentation efficiency.(2)In the suspected smoke segmentation part,the RGB color judgment criteria for white smoke and black smoke were constructed,which are 80 ≤ ≤ ≤ ≤ 240 and 10 ≤ ≤ ≤ ≤ 240,respectively.A genetic algorithm optimized KSW and rough set and region growing method were proposed to segment white smoke and black smoke,respectively.The results show that the proposed method is better than other methods,such as Ostu algorithm,and the segmentation foreground objective is complete and accurate,effectively avoiding over-segmentation or under-segmentation.Meanwhile,the necessity and operability of different segmentation methods proposed in this paper are verified.(3)The smoke feature description scheme from three aspects of spatial,texture and energy was designed.A Gabor filter was used to extract the spatial information of the smoke related to the flow direction,the local binary pattern was applied to extract the texture information of the smoke,and a two-dimensional discrete wavelet transform was employed to extract the energy information of the smoke.And a 127-dimensional feature vector that effectively characterizes the smoke was constructed.(4)The early warning framework of smoke detection was summarized,and the early warning scheme after identifying smoke was described.The Light GBM algorithm was proposed to be applied to smoke discrimination,and the Bayesian optimization method and grid search method were used to train the algorithm and find the best hyperparameters.The results show that the Bayesian optimized Light GBM algorithm has the best performance on ROC curve,with an AUC value of 0.993,which is better than other algorithms.The proposed method in this paper has high accuracy and robustness in different scenes,both indoors and outdoors under different weather conditions such as wind and clouds.