Cloud And Fog Removal Method Of Optical Remote Sensing Image Based On Dark Channel Prior Model

In the imaging process of optical remote sensing image,it is often interfered by weather and other factors,especially cloud and fog,which will affect the imaging effect of the image,resulting in the reduction of clarity,which is detrimental to subsequent visual interpretation.Although active radars can effectively penetrate clouds and fog to avoid the influence of weather and can remove clouds from optical remote sensing images by means of multi-source compensation.Most of them work in single-polarization mode,resulting in the reduction of image resolution.In addition,the revisit period of radar is long and it is in different orbit from the imaging satellite of optical remote sensing image,so the same ground object may not be captured at the same time.Besides,image registration remains a difficult problem in multi-source compensation,and the location of ground objects cannot be accurately determined.It can be seen that cloud removal of single source image is also of great significance in real life.Therefore,this paper takes single-source optical remote sensing images as the research object to study cloud removal,which can be summarized as follows:Firstly,the degradation model of cloud image is established,the corresponding principle of dark channel is analyzed in detail,and the spectrum,histogram and frequency characteristics of cloud image are discussed.On the one hand,atmospheric scattering and reflection are the main reasons that clouds affect image imaging,and the dark channel prior is further derived from the atmospheric scattering model.On the other hand,the image histogram interfered by clouds is too concentrated and the cloud information is mainly distributed in the low-frequency components.These analyses provide theoretical support for further research.Second,the single-temporal image is studied,in view of the cloud and fog removal algorithm is easy to cause color distortion.Based on dark channel prior model,this paper put forward the method called as multi-scale correction dark channel.This method is based on frequency characteristics of the image and combine the multi-scale transform and dark channel prior model.At the same time,the atmospheric light in the dark channel prior model is corrected.It can effectively remove the fog and cloud in the degraded image,and minimize the color distortion in the process of restoration,and approximate estimate the clear image accurately.Finally,the multi-temporal cloud image is studied.Aiming at the problems of serious color distortion and incomplete cloud removal,an association and collaboration method based on multi-scale fusion of dark channels is proposed to make use of images of different phases for information compensation.In this method,the sparse representation is combined with the dark channel prior model from the perspective of multi-scale transformation.Besides,artificial control parameters are introduced to optimize the Laplace sharpening for the problem of image edge blurring after fusion.The algorithm can remove most of the cloud and reduce the spectral distortion.