Research For Fog-Haze Detection Based On Image Processing

With the rapid development of economy and industry in China,the problem of air pollution has received increasing attention.Real time and accurate detection of haze pollution level was realistic and significance for environment and residents'life.There are two main methods to detect haze at present,one is the particle counting method to detect the concentration of various particles in the atmosphere,and the other is a haze detection method based on remote sensing.The development of the above two methods is relatively mature,but they all have the characteristics of high cost and unpopularity.Aiming at the above problems,this paper proposes a detection method based on the basic image processing,aiming to establish a relationship model between the image and the defect,and achieve the purpose of obtaining the environmental defect by analyzing the same target image captured under any one environment.The main contents are as follows:?1?Analyzed the principle of atmospheric visibility decline caused by haze,and analyzed the influence of haze on natural image.Discussed the reason of using PM_2.5concentration to characterize haze,and the whole process principle is summarized.?2?Combine machine vision and image processing technology to preprocess the natural image.Through analysis and comparison,obtain the graying and filtering methods that are most suitable for the study of the image.Because the external environment will cause the deviation in different images,registration is performed according to the image conditions,which provides the basis for the extraction of the region of interest.?3?Select the region of interest according to the characteristics of different features,extract a variety of features in the corresponding region of interest that can reflect the influence of haze on the image.Analyzed Correlations between characteristics and PM_2.5 concentrations,and compared to obtain several features with large correlations with PM_2.5 concentrations.Use brightness contrast,entropy,correlation and peak signal-to-noise ratio as training set for machine learning.?4?A support vector regression model was used to establish the relationship between image features and PM2.5 concentration,and the model's prediction results were analyzed and verified.The haze detection based on image processing proposed in this paper combines image processing technology and machine learning technology.This method is applicable to any scene that contains high-contrast areas.The advantage of this method is that it does not need to set up new equipment.It only needs to photograph the original equipment.Image and environmental data can be modeled to use the existing road conditions camera monitoring system to monitor the monitoring area.The method has a wide coverage,low measurement cost and can realize automatic real-time measurement,and is very suitable for large-scale application promotion.