Research On Fog Detection And Prediction Method

In recent years,fog and haze weather occur frequently,and visibility reduction brings many inconveniences to people's daily life.It is urgent to detect and forecast the visibility of fog days.Nowadays,visibility detection instruments are expensive and cannot be widely used.Visibility detection methods based on image processing are becoming more and more mature.At the same time,a large number of cameras installed on highways can be used to study visibility detection methods.At present,most of the visibility prediction methods such as multiple regression model and synoptic chart analysis require the conditions that the observation time is long enough and the data is complete,and depend on the experience and model obtained from previous experiments.Aiming at the above problems,the background of the collected highway images is extracted,the atmospheric transmittance of the images is obtained based on the dark channel prior theory and refined by the steering filter method.After edge detection using Canny algorithm,the lane is detected quickly by the telescopic rectangular window,and the transmittance at both ends of the lane is calculated.The extinction coefficient of the image is obtained by the distance information of the lane,and the visibility distance of the current detection image is deduced.The distributed real-time detection of the visibility is realized by using the lane of the expressway.The detection relative error is less than 18%,which meets the requirements of the national standard,has good robustness and high accuracy.This paper designs a visibility prediction method based on artificial neural network,establishes BP neural network visibility prediction model,optimizes and adjusts parameters of network layer based on genetic algorithm,trains GA-BP neural network visibility prediction model by using meteorological data provided by National Meteorological Information Center,and trains GA-BP neural network and BP neural network visibility prediction model.The predicted data of the neural network are compared with the observed data of visibility,and the correlation coefficient,root mean square error and mean absolute error are evaluated.The predicted accuracy of visibility of GA-BP neural network model is higher,the predicted stability is better,the predicted performance is better and provide reliable visibility forecast information for people to travel.