Spark-based adaptive Mapreduce data processing method for remote sensing imagery

摘要

Existing Hadoop-based remote sensing data processing approaches are insufficient for efficiently meeting the requirements of applications, especially when large remote sensing datasets are involved. This paper proposes an adaptive Spark-based remote sensing data processing method on the cloud that achieves improved efficiency and stability. The method includes a remote sensing data storage scheme on the cloud that employs the Hadoop Distributed File System (HDFS) and adaptive MapReduce mechanisms for use with remote sensing data; specifically, a mapping strategy for use with image tiles, a reducing strategy for use with adjacent tiles, and a mechanism for merging the results are proposed. An image classification experiment is conducted using Land Remote-Sensing Satellite System (Landsat) Thematic Mapper (TM) data, and the proposed method displays improved performance, stability and scalability compared to the existing Hadoop-based method. Hence, the proposed method is more suitable for processing large volumes of remote sensing data.