Cloud Detection Of Remote Sensing Satellite Imagery Based On Multi-feature Fusion And Graph-cut Model

With the rapid development of earth observation technology,the resolution of remote sensing satellite sensor is higher and higher.Those high resolution imagery products are widely used in environmental monitoring,disaster relief,agroforestry construction,intelligence traffic engineering and other aspects.However,clouds are above the earth all year round.And it is inevitable that these clouds will be captured by satellite sensors when they shot the surface of the earth.The existence of the clouds on the satellite images,on the one hand,has blocked the terrain,causing the surface information loss,bringing adverse effect for target recognition,changing detection,object classification and so on.On the other hand,in the process of 4D products,images which have lots of clouds will be abandoned.This causes a large waste of image data.Therefore,the detection of cloud on remote sensing images is of great significance for the follow-up processing like cloud area repairing,image analysis,image matching,target detection and extraction.Aiming at the characteristics of the high resolution remote sensing images,such as the less band number,narrow spectral range and rich details of the features,this paper put forward a method for cloud detection of remote sensing satellite imagery based on multi-feature fusion and graph-cut model.Firstly,images are transferred to HSL color space for generating the spectral features basement images and then super pixel segmentation are used to segment the images into pixel blocks.The brightness threshold method are carried out after this for crude extracting cloud area.Secondly,according to the feature that cloud should not contain linear structure,the Gabor filter is used to detect the linear structure for excluding the non-cloud area.Thirdly,expansion and contraction operations are conducted to observe cloud seeds,non-cloud seeds and unclassified pixels for graph-cut model.When using graph-cut model optimization method,this paper analyzes the disadvantages of classical graph-cut model,proposing an improved graph-cut model which is more suitable for cloud detection of high resolution remote sensing images.Finally,several representative high resolution remote sensing images were selected to carry on the experiment.Based on qualitative analysis and quantitative evaluation,the results of each step were compared and analyzed,and each step was verified to optimize the results.In addition,this paper compares the improved graph-cut model with the classical graph-cut model,evaluating the precision,recall,error rate and the operation efficiency of each algorithm.The experimental results show that our method which this paper mentioned has higher precision,higher recall,lower error rate and is faster than classical graph-cut model.Through the results,the feasibility and practicability of cloud detection algorithm that this paper put forward have been proved.