Segmentation of large-scale remotely sensed images on a Spark platform: A strategy for handling massive image tiles with the MapReduce model

• DOI: 10.1016/j.isprsjprs.2020.02.012
• Publication Date: 2020-02-26T04:38:06Z
• Authors: Ning Wang; Fang Chen; Bo Yu; Yuchu Qin
• Keywords: Segmentation-based object categorization; Merge (version control)

Image segmentation is essential in object-based image analysis. Numerous algorithms have been proposed and widely applied to process remote sensing images, but most of them are designed deal with single scenes. As the volume images grows rapidly, handling machines becoming increasingly difficult, size a composite can be larger than CPU memory computer. To address this problem, distributed strategy paper. The two main steps as follows. First, prepared massive loaded then decomposed into sub-images that across multiple computers; used parallel segment each sub-image large number initial objects. Secondly, object resegmentation method boundary objects order merge these ingested from different computers obtain final image. Two classical employed test eight study areas include urban area, suburban zone agricultural landscape. Both intersection over union F-measure metrics show help solve problem data being too fit on machine, it also performs better comparative strategies. not only has ability very accelerates segmentation-based applications so they match acquisition rate.