Research On Fog Visibility Detection Algorithms Based On Deep Learning

In recent years,with the development of Chinese economy,a large number of pollutants are discharged.Then it result in frequent haze weather,and the treatment of fog and haze has attracted attention from all walks of society.The occurrence of fog and haze not only endangers people's physical and mental health,but also brings great inconvenience to people's travel because of the reduced visibility caused by fog and haze.Visibility has an important impact on road traffic safety,especially severe haze weather will seriously threaten road traffic safety.Therefore,the accurate detection of visibility can help driving to reduce the potential safety hazards to the greatest extent,but the existing equipment detection methods have the shortcomings of cost-effective and universal applicability.The accuracy and stability of the current methods of visibility detection using image processing need to be improved,so an accurate and stable method of haze visibility detection is needed.As haze visibility detection has important practical value,it has attracted a lot of attention of scholars in the field of image processing and computer vision.With the rise of in-depth learning,in order to overcome the shortcomings of traditional detection methods,this paper proposes a haze visibility detection method based on in-depth learning.The main research contents are as follows:First of all,in view of the shortcomings of traditional methods such as slow detection speed and low accuracy,this paper adopts convolution neural network to detect visibility,uses convolution neural network to learn the features of haze images,uses a lot of manpower to calculate the true value by calculating the mean value according to the images,and uses the true value to optimize by continuously supervised self-learning by reducing the loss function value.The training model obtains the best training model.Two classical convolutional neural network models,deep residual network and VGG network,are used to train the visibility detection model on tensorflow deep learning platform.The model errors are 12% and 10% respectively.Experiments show that the accuracy of visibility detection using convolutional neural network is better than that using traditional algorithm.The validity has been verified.Secondly,in this paper,based on the problem of low stability of detection using convolutional neural networks,an improved smog visibility detection algorithm based on improved convolutional neural network is proposed.Firstly,because the haze images is picked up by video,there is a relationship between the time series of the haze image.According to this relationship,an improved convolution neural network using convolution neural network to extract features and using cyclic neural network to do regression is proposed.Then based on the improved convolutional neural network framework,residual network and VGG are used as features of convolutional neural network,and long short-term memory network is used as regression to build the network.The new convolutional neural network is trained in visibility detection model on tensorflow deep learning platform.Finally,a large number of experiments prove that the effectiveness of the improved convolutional neural network is improved,and the stability of the new network is improved.