A Spark-Based Remote Sensing Time-Series Processing System

In recent years,with the improvement of remote sensing technology,China's high-resolution remote sensing satellite technology has developed at a rapid pace.The data level of high resolution remote sensing images is becoming more and more massive.Large amounts of remote sensing image data bring more information.However,it also brings great challenges.Due to the characteristics of the rotation of the satellite around the earth,the same area will be taken many times at different time by satellite.By detecting the changes of the image in the same area at different time,it is helpful to find the change of land cover in this area.The change detection algorithm can be divided into pixel level,feature level and object level according to the level of image analysis.According to the mechanism of data analysis,the change detection algorithm can be divided into supervised and unsupervised.Traditional change detection method can process small remote sensing image data efficiently.However,facing high-resolution remote sensing images,it processes slowly and even sometimes can't handle.We designed a remote sensing time-series change detection system based Spark,a computational engine based on memory calculation,to speed up the detection speed.We detect the change area by supervised pixel-level change detection method.In order to process high-resolution remote sensing efficiently,the image data is reduced dimension firstly.Then the region samples are extracted and are labelled artificially.According to the pixel information and the spatial information in the samples,the multi-classification model is trained.The original remote sensing image is divided into different regions by the model.By comparing the changes in different regions of different time,the change analysis report can be made.