The Method Of Image De-noising By Fusing Multi-temporal Remote Sensing Images

The emergence of remote sensing technology makes us receive the basic information of the observed object through the sensing device without contact with the object directly.This avoids that the information cannot be received in the area which is remote or steep.So it becomes the only way to observe data dynamically on a global scale so far and is widely used in a number of areas and plays a significant role in economic growth and social development.However,due to the influence of weather,remote sensing equipment and transmission medium,remote sensing images are often polluted by noise in the process of imaging and transmission.The most common noise is Gaussian noise,cloud noise and fog noise.This will directly affect the further processing,analysis and application of remote sensing images.The goal of remote sensing image de-noising is to remove noise to the utmost extent under the premise of maximizing protecting the details to improve the readability and effectiveness of the data.At present,the de-nosing methods for Gaussian noise such as thermal noise and shot noise remove noise mainly according the spatial or frequency domain features through a single remote sensing image.However,there is a problem in this kind of de-noising method,which is the contradiction between retaining the image edge and removing the noise resulting in edge blur or the removal of noise unideal.For thin cloud,because it not only contains the information related to the cloud,but also the effective information such as terrain information,it attracts a lot of people to research.The common approach is to weaken the cloud information while enhancing the terrain information.And for the thick cloud,as the terrain information is completely covered and almost have no useful information,removing thick clouds by a single remote sensing image would cause information empty.This shows that the amount of information in a single remote sensing image is insufficient,and it is necessary to combine the multi-temporal remote sensing data with complementary information according to certain methods to obtain a more remote sensing image with more information.According to the above analysis,this paper studies the method of multi-temporal remote sensing images' fusion de-noising based on DS evidence theory,which mainly starts from the following three aspects:(1)It analyzes the features of many types of noise in remote sensing images and researches of status quo,and analyzes the feasibility of using DS evidence theory in multi-temporal remote sensing images' fusion de-noising.DS evidence theory is emerged in the 1970's as a reasoning theory and belongs to the scope of artificial intelligence.It can integrate multiple evidences and make decisions.And then make a reasonable explanation of reasoning.It can effectively solve the problem of uncertainty caused by the inaccurate cognition or cognitive loss of the research object.In the remote sensing image,the noise is random and uncertain.DS evidence theory can take into account uncertain information from multiple sources and is suitable for multi-temporal remote sensing images' fusion de-noising.(2)It puts forward a method of removing Gaussian noise using multi-temporal remote sensing images based on the DS evidence theory.It designs four noise detection models that are the two-state Gaussian mixture model,the mean value detection model,the median value detection model and edge analysis model to obtain basic probability assignment for evidence.They are used to analyze whether each gray value is related to noise or to the terrain.Then,according to the rule of DS evidence,the method fuses the four evidences of each remote sensing image to a whole and get the probability of each pixel related to noise or related to the terrain.Then,the method uses DS evidence theory again to synthesize multiple evidences of multi-temporal remote sensing images and get the final conclusion.Finally,according to the conclusions and decision rules,remove noise for remote sensing image by fusing multiple information.Experimental results show that,the proposed algorithm is superior to traditional ones in Gaussian noise removal and edges preserving according visual,variance and signal-noise ratio.(3)It puts forward a method of removing cloud using multi-temporal remote sensing images based on the DS evidence theory.It designs two cloud detection models according gray-scale statistical changes and frequency domain statistical changes to obtain basic probability assignment for evidence.Firstly,the method divides the multi-temporal remote sensing images into several small areas according to the same standard.Each of the small regions is judged by the above two models to determine whether each region is related to cloud or terrain.Secondly,according to the rule of DS evidence,the method synthesizes the two evidences of each remote sensing image to a whole and gets the probability of each region related to cloud or related to terrain.Thirdly the method uses the DS evidence theory again to synthesize multiple evidences of multi-temporal remote sensing images and get the final conclusion.Finally,according to the conclusions and decision rules,removing cloud for remote sensing image.Experimental results show that,the proposed algorithm obtains richer images by using the effective complementary information.