Research On Cloud Detection And Thin Cloud Removal Technology Of High Resolution Satellite Remote Sensing Imagery

With the development of space exploration technology,the time resolution,space resolution and spectral resolution of satellite imagery have improved continuously.High resolution satellite images play an important role in many applications,such as surveying and mapping,resource exploration,disaster monitoring,military reconnaissance and so on.Because of the climate,cloud contamination is a common interference factor which appears frequently in imagery.It decreases the quality of remote sensing data,thereby bringing a great challenge to data analysis and transmission.Consequently,reducing the interference of cloud noise and increasing the utilization of satellite imagery make a lot of sense.In this paper,we focus on this problem and do research on the relative topics: cloud detection and thin cloud removal.The main work of our paper is as follows:(1)We analyze the feature extraction technology of clouds.First,the physical properties and image properties of clouds are introduced in detail,and based on the attribute difference between cloud and underlying surface,three classes of feature are selected for feature description.Then,we analyze the typical methods of feature extraction from the corresponding features.Finally,the classification ability of features in three categories is analyzed according to the experiments.(2)We study the cloud detection technology which aims at the calculation of cloud cover.Based on the characters of high resolution panchromatic images,a cloud detection method based on SVM is proposed.Experiments demonstrate the effectiveness of the proposed method in visual effect,detection accuracy and the computation efficiency.Furthermore,we propose a method of detecting the cloud covering regionally,which is beneficial to evaluate the image quality.(3)We study the thin cloud removal technology which aims to improve the definition of the imagery.We propose a thin cloud removal method based on the character of the thin cloud distribution.The experiments show that the proposed method can remove thin cloud and meanwhile keep the spectral characters of the cloudless regions free of change.