Research On Haze Visibility Detection Based On Spectrum Analysis And Inproved Inception

In recent years,fog and haze pollution has being becoming a normal trend in China,and all sectors of society have paid more and more attention to the problem of fog and haze pollution.The haze weather can not only damage people's health,but also bring many inconveniences to people's travel.Especially in highway sections,sudden fog weather are prone to occur,which greatly threatens people's driving safety.Therefore,accurate detection of visibility is one of the important steps to solve this problem,but the existing equipment and methods,accuracy and popularity still need to be improved.At present,the visibility detection based on haze video image has attracted wide attention of researchers at home and abroad.It is supposed to be done in this paper towards this problems,which three haze visibility detection algorithm is going to be provided based on the haze visibility video image of Expressway section.The main research contents are as follows:First of all,in view of the Inception network proposed,it has a wide range of applicability in the field of image processing,and can effectively extract image information.For the visibility detection of haze images,Inception V4 network has good performance because of its network structure and low computational complexity,and the training speed of the network is significantly improved compared with the previous network.Therefore,this paper studies the visibility detection of haze based on Inception network.This paper studies and proposes a haze visibility detection algorithm based on Inception V4 network.Inception V4 network is built based on Tensorflow framework,and slightly adjusted in data preprocessing and network structure to adapt to the data set in this paper,so as to build the relationship between haze visibility and haze images.A large number of experimental results verify the effectiveness of the algorithm.Secondly,based on the study of image spectrum theory,a haze visibility detection algorithm is proposed based on spectrum analysis and Inception V4 network.Since the energy of the image captured by the camera gathers to the low frequency part after the haze occurs,this paper uses discrete cosine transform to obtain the image spectrum information,and extracts part of the spectrum low frequency information.As a parameter with higher weight coefficient,it is imported into Inception V4 network to participate in training.Aiming at the relationship between image spectrum information and visibility value constructed by data set using Inception V4 network,the experimental results show that the algorithm has good accuracy.Finally,through the exploration of the above two algorithms,some flawlessness can still be explored.The main reason is that the Inception network structure is very deep.When extracting image features,the low-level details of the feature information loss is large.For haze images,this part of the feature information is helpful for visibility detection.Therefore,based on the analysis of the insufficiency of the relationship between image construction and visibility value in Inception V4 network,this paper proposes an improvement of Inception V4 network in view of the characteristics of haze pictures on Expressway sections,and adds the extraction of detailed features of haze images.Then experiments are carried out for the original image and the image spectrum information.The experimental results show that the improved network has significantly improved the effectiveness of visibility detection.Finally,the algorithms in this paper are compared and analyzed from many aspects.Finally,the algorithm has a good effect on haze visibility of Expressway sections,especially for low visibility,which is worthy of detection and meets the actual needs.