Study On Cloud Detection Method For High Resolution Satellite Remote Sensing Image

In order to improve remote sensing satellite data transmission efficiency, forremote sensing image information loss caused by cloud mask, a cloud detection methodbased on high resolution optical satellite images has been proposed in this dissertation.Simulation results have shown that the proposed method’s cloud detection probability is98.3%and its detection error is2.6%, has higher performance than other methods. Themain research works and results are as follows:(1) According to the characteristics of high-resolution panchromatic remotesensing images, a remote sensing image cloud detection overall program was proposed,overcoming the deficiencies of traditional methods based on multispectral threshold;(2) Based on the differences of clouds and underlying surface properties inhigh-resolution remote sensing images, from the radiation characteristic, geometriccharacteristics and texture characteristics, the classification feature parameters wereextracted and the feature space was established;(3) According to the information redundancy caused by high classification featurespace dimension, a feature selection method based on separable degree threshold wasproposed at first. Then an algorithm for feature space linear dimension compressionbased on maximum likelihood estimation was proposed on this basis, which caneffectively remove the classification features redundancy, greatly improve theseparation effect of two samples.(4) In compressed subspace, the improved FCM and SVM clustering algorithmwere used for clustering training and the cloud detection classifier design optimizationwas completed, the performance of the classifiers were verified by simulationexperiments. Based on this, a cloud detection software system based on GUI wasdeveloped.The cloud detection method for high resolution satellite remote sensing imagestudied in this dissertation will provide a theoretical basis for clouds removal in orbitand image data compression.