Comparative Study On Models Of Cloud Removal Based On The Features Of Remote Sensing Image

Remote sensing technology has beenused in modernization of national defense, economic construction and other aspects widely whichbenefit by its advantageslike timeliness, objectivity, Its objectivity providing a variety of earth observation data.In recent years, Fast-growing domestic remote sensing satellite industrialization and remote sensing data service capabilities are also becoming an important national frontier, and showing great market prospect in China. Cloudy and rainy mountains and plateau have serious resource and environmental problems. However, except active radar photography can avoid influence of the complex weathereffectively, almost all other remote sensing images obtained by passive optical remote sensing technology could be disturbed by weather such as cloud, mist and haze. This interference makes the remote sensing image blurred, and hazy images can’t meet the actual application requirements. As a result, optical remote sensing image removal research in cloudy and rainy mountains and plateauhas urgent necessity and significance.In this paper, we selected SPOT-5 satellite data of Yongshan area, WorldView-2 satellite data of Wuding area and Mapping Satellite-1 satellite data of Chuxiong area as research objects,which havedifferent typical forms of clouds, such as uniform clouds, inhomogeneous clouds and lumpy clouds. From the statistical characteristics, band features, frequency domain and degradation model of hazy images, this study used methods of GIS and RS to construct the contrast limited adaptive histogram equalization model, the BSHTI(Background Suppressed Haze Thickness Index) combined with VCP(Virtual Cloud Point) model, the homomorphic filtering model as well as the haze removal model were using dark channel prior. Then by applying the above models to experiments, our study evaluated and analyzed the experiment results, advantages and disadvantages of each model and scope. This study explored efficient and convenient methods of haze removal in mountains and plateau.Here are three main research results in this study.(1) The study summarized the imaging process of optical satellite remote sensing systems, the formation of clouds, atmospheric reflectionand scattering mechanism, analyzed the causes and characteristics of hazy image degradation. This paper also analyzed the characteristics of the bands, spatiotemporal distribution, frequency-domain and other aspects of the hazy images;(2) From the statistical characteristics, band features, frequency domain and degradation model of hazy images with different typical forms of clouds, this study constructedthe contrast limited adaptive histogram equalization model, the BSHTI combined with VCP model, the homomorphic filtering model as well as the haze removal model using dark channel prior.(3) The paper compared, evaluated and analyzed the advantages, disadvantages and scope of remote sensing image haze removal models in terms of the nature, the operating environment and the results of the models. The remote sensing image haze removal technologies based on image characteristics are as follow:a. The contrast limited adaptive histogram equalization modelcan achieve the purpose of haze removal by image enchantment effectively. This model achieved good results of the remote sensing images with uniform clouds, so it is suited for the remote sensing images with uniform clouds or high resolution.b. As for the remote sensing images with inhomogeneous clouds, the application of the BSHTI combined with VCP model got a ideal effect of haze removal, so it is suited for the remote sensing images with full optical bands, high inter-bans correlation and uniform transition of cloud.c. There are three kinds of high-pass filter used in the homomorphic filtering model. The result of index high-pass filter or Butterworth high-pass filter is better than the ideal high-pass filter. It is suited for the need of easy and fast haze removal, and commonly used in the remote sensing images for visual interpretation.d. The haze removal model using dark channel prior kept the spectral information of original images, and it also achieved positive effect of lumpy cloud removal.In summary, this paper is a complete study of the feature and imaging mechanism of optical images with three typical types of clouds form such as uniform clouds, inhomogeneous cloud and lumpy cloud, which has made up for the lack of research on mechanism analysis in the past haze removal technology, so the scientific significance is obvious; In addition, it constructs the contrast limited adaptive histogram equalization model, the BSHTIcombined with VCP model, the homomorphic filtering model and the haze removal model using dark channel prior systematically, based on the analysis of hazy images’ features and degradation model mechanism, then by evaluating these models’ demonstration effectqualitatively and quantitatively, the study achieved the optimal construction of haze removal models in mountains and plateau, As a result its technological methods have a great innovation.