Cloud Dectection Of Remote Sensing Images Based On Saliency Analysis And Multi-Texture Features

With the vigorous development of Earth Observation Technology,the massive information obtained by remote sensing can be widely used in agricultural investigation,environmental protection,disaster prevention and mitigation,navigation and positioning,geographic mapping,military reconnaissance and other fields,providing decision-making basis and information guarantee for the implementation of important strategies such as geographic national conditions survey and Land and Resources Engineering Construction.However,not all satellite imagery can meet the actual production needs,one of the main reasons is cloud cover.According to the data provided by the International Satellite Cloud Climate Program,cloud cover more than 60% of the earth's surface.In the imaging process,remote sensor will inevitably be disturbed by cloud,resulting in the loss of some information,changing the texture and spectral information of the image.There are many disadvantages in the process of making remote sensing image products,which reduce the utilization rate of images.Therefore,cloud detection of remote sensing image has become the first problem to be solved in the process of remote sensing image processing.In recent years,many scholars at home and abroad have conducted in-depth research on cloud detection of remote sensing images,and proposed various cloud detection algorithms.Most of the algorithms are for hyperspectral images,and some algorithms also need to use DEM data,meteorological data and other auxiliary information.Therefore,it is difficult to obtain ideal detection results for high resolution images with fewer bands and less spectral information and auxiliary data.Based on the above background and the research status at home and abroad,the research content of this paper is determined,that is,to explore a method of cloud detection using less band information.Specifically,a cloud detection method based on saliency analysis and multi-texture features is proposed in RGB three-band space.The main research contents are as follows:(1)Analyzing the causes of cloud formation,the types of cloud,and the radiation,geometric and frequency characteristics of cloud,laying the foundation for the follow-up study.(2)According to the spectral characteristics of clouds,an improved Itti saliency model is proposed to construct the saliency image of cloudy images,and based on this,the coarse detection of clouds is realized.(3)Study various texture features of cloud and underlying surface samples,including fractal dimension,gray level co-occurrence matrix,Tamura texture features and LBP texture.Based on the separability,the features which can distinguish them well are selected.The principal component analysis method is used to compress the features.Finally,train the cloud and underlying samples using Support Vector Machine,and obtain the SVM classifier to extract the cloud.The proposed algorithm takes into account both gray and texture features of cloud layer,and can effectively and quickly detect cloud in remote sensing images by using less band information.Experiments show that the proposed algorithm has high recall and precision.The salient image can highlight the cloud target on the image,and the selected texture features can achieve better cloud-to-ground separation.