Research On Atmospheric Visibility Estimation Algorithm Based On Dark Channel Prior And Deep Learning

Atmospheric visibility is an important indicator of meteorological observation.It restricts road traffic,aviation,navigation and military activities.It is very important for highway management departments,airlines and environmental monitoring to timely and accurately estimate the current visibility and improve the accuracy of atmospheric visibility identification.However,the accuracy and stability of the existing visibility detection methods using image processing still need to be improved.Therefore,an accurate,stable and efficient haze visibility detection method is urgently needed.Firstly,the relationship between visibility and surface meteorological observation factors is explored and analyzed by using linear regression model and integrated learning model,the relationship expression based on elasticnet model is constructed,and the main meteorological factors affecting the change of visibility index are explored.The establishment of relational model can assist the estimation of visibility to a certain extent,but because the detection of meteorological factors depends on specific instruments,it is not suitable for monitoring anytime and anywhere.Therefore,based on the visibility detection technology and deep learning,this paper makes a theoretical research on the existing algorithms,and proposes two new visibility estimation methods.In this paper,a visibility detection model based on dark channel a priori algorithm and monodepth2 depth algorithm is proposed.This method indirectly calculates the visibility value by measuring the change of image features.Firstly,the atmospheric transmittance is calculated by the dark channel a priori algorithm,and then the scene depth is obtained based on the depth of field estimation model monodepth2.Finally,the atmospheric transmittance and depth of field are substituted into the fog imaging model to obtain the atmospheric extinction coefficient.Finally,the atmospheric extinction coefficient is used to calculate the estimated value of visibility.Aiming at the problem that the physical model has no obvious effect on fog prediction and weak robustness,this paper studies the haze visibility detection method based on depth learning algorithm,and proposes an atmospheric visibility detection algorithm based on 3D convolution depth residual network.This method replaces the 2D convolution in the depth residual network with 3D convolution,and integrates the attention mechanism module of the convolution module to help the model search the farthest visibility distance.The experimental results show that compared with the physical model,the model has better accuracy,high visibility scene recognition accuracy and good robustness.