Visibility Detection Algorithm Based On Deep Learning

In recent years,with the development of industrialization in the world,the problem of haze caused by air pollution is becoming more and more serious,and haze weather has become a major problem of environmental governance.At the same time,the harm caused by haze weather is gradually expanding,which not only threatens human health,but also reduces the visibility of the airport,causing great interference to the flight time.Therefore,the detection of visibility is becoming more and more important,and has attracted the attention of experts at home and abroad.For the traditional visibility detection methods,there are some shortcomings,such as low efficiency,long timeconsuming,inaccurate measurement results and so on.In order to improve the detection results,domestic and foreign experts have achieved good results through virtual target extraction,contrast detection,dark primary color prior channel extraction transmittance and other methods.In recent years,with the continuous development of deep learning,visibility detection has also attracted deep learning They proposed spectrum analysis,improved Inception network and improved CNN-LSTM network for visibility detection,and achieved good experimental results.However,due to the influence of network input image noise,the trained network effect is greatly reduced.In order to solve this problem,this paper carries out a series of research on the network preprocessing problem and the sample size problem.Finally,through the traditional image processing method combined with deep learning,the visibility level detection is carried out,and good results are obtained.The research content of this paper is mainly divided into four parts.Firstly,this paper briefly introduces the research status of visibility detection at home and abroad,and then determines the research scope and structure of the article.The second part mainly studies the existing visibility detection algorithms.Here we introduce the visibility detection algorithm based on image processing and the visibility detection algorithm based on deep learning.The visibility detection algorithm based on image processing mainly includes daytime visibility detection,night visibility detection and camera calibration technology.For the visibility detection based on deep learning,the visibility detection algorithm based on image processing mainly includes daytime visibility detection,night visibility detection and camera calibration technology In terms of algorithm,the principle of convolutional neural network and the application of deep learning in visibility detection are introduced.In the third part,the principle of InceptionV3 neural network is introduced firstly.Through the establishment of visibility classification model by using the network directly,the over fitting phenomenon is found,so the application of transfer learning is proposed.The preprocessed data is imported into the InceptionV3 network based on transfer learning for training,and then predicted.The experimental results show that this method is feasible.In the fourth part,in order to reduce the data redundancy caused by image noise,this paper proposes a depth model visibility detection algorithm based on ROI by combining with traditional image processing algorithm.The algorithm is based on deep learning,and extracts the region of interest before the image is imported,and uses the transfer learning method to model the image with only the region of interest,this not only reduces the impact of image noise,but also greatly reduces the demand for tags.Finally,experiments show that the algorithm is better than the use of a single deep learning model,and has a greater effect on the accuracy of the classification model.