Research On Cloud Detection Algorithm For MODIS Data

When we use satellite remote sensing data for resources exploration,environment and disaster monitoring and research, the cloud is a huge barrier, theexistence of the cloud seriously affects the signal transmission and affects thequality of the remote sensing information acquisition, which reduces the datautilization and sometimes even the data cannot be used. Therefore, to use of satelliteremote sensing observation data cloud detection, we should first detect cloud. EOS(Earth Observation System) carries satellite equipment with MODIS and use of theinstrument people can detect land use and land cover research, seasonal andinterannual climate research, natural disaster monitoring and analysis, long-termchanges in the rate of climate change and the change of atmospheric ozone research,etc. So MODIS data for a variety of monitoring and analysis has been widely used.There we deciede to study the cloud detection algorithm based on MODIS data inthis paper.First we analyse the purpose and signficance of cloud detection,then summarizethe various algorithms and basic principles for different sensors domestic andforeign in-depth. And learn about satellite platforms and sensors for MODIS data.Then we begin to learn about the MODIS data storage format, and stored datainformation, and on the basis of data, we carry out a series of data pre-processing,and obtained the data of the reflectivity of cloud detection and bright temperaturethat can be used for spectral analysis, and diacuss the various features and spectralcharacteristics of clouds. So we lay the root for the subsequent three cloud detectionalgorithms.Then we carry out the traditional cloud algorithm based on MODIS dataaccording to reflectance or temperature and using absolute threshold to identifyclouds. This method is very effective for the typical cloud and we can obtainsatisfactory results. But because the threshold can affect the results, the algorithmusing fixed threshold to detect cloud is no longer useful, it can lead to Virtualscreening or leak detection. To solve this problem we deceide to use Otsu to detectcloud. First set a low initial threshold for the test, and then secondary detectionusing Otsu algorithm, the same time solving the virtual inspection problem. To someextent, this method improved the detection effect of the cloud, but the initialthreshold setting will affect the test results, there are still leak problem.Finally, after theoretical analysis, we decide to use the K nearest neighboralgorithm for cloud detection because K neighbor algorithm is widely used in dataclassification, using this method do not need the threshold, every sentence is according to the characteristic value with multiple band, so for low clouds, cloudspectral characteristics is easy to be disturbed by earth surface especially iceinterference image, setting the threshold detection effect is not ideal, while choosingthe typical samples by using K neighbor algorithm can solve the problem ofthreshold value well, then this method can more accurate separate the cloud andground objects, solve virtual inspection and leak problem better.