Remote Sensing Digital Image Processing Methods Based On Hadoop

Currently remote sensing digital images are widely used in the fields of military,agriculture, environmental science, geoscience and mapping. With the continuousupdate of space remote sensing technology, the total amounts of the data generated byremote sensing satellites of multi-temporal, multi-resolution, multi-sensor andmulti-band remote sensing satellites grow rapidly. The storage and parallel processingof these vast amounts of remote sensing data have become a hot research, cloudcomputing provides us with the direction of the study.Characteristics of the cloud computing are ultra-large-scale virtualization, highscalability, high reliability, on-demand service, versatile, extremely cheap. Hadoop,an open source cloud computing system, is mainly composed by HDFS(HadoopDistributed File System) and the parallel computing framework MapReduce. Thispaper is based on Hadoop and mainly introduces the enhancement processing ofremote sensing digital images and clustering after enhanced with MapReduce, thengives a comparative study with the PC serial processing.For the low brightness and the poor visual effect of remote sensing digital image,and the traditional methods of image enhancement is not conducive to visualinterpretation and subsequent processing, this paper improves the losslessenhancement method with all pixels in the effective area, the experiments show thatthe method achieves a better visual effect.This paper proposes a parallel processing Waterfall Model for the large amount ofremote sensing digital images, and achieves this model on the Hadoop environmentwith MapReduce. It not only solves the input problem of various image formatshandled by MapReduce directly, but also speeds up the processing speed of theremote sensing digital images. Finally, clustering of the remote sensing digital imagesenhanced by the proposed method achieves a good result with experimentalverification. At the same time the K-Means algorithm for clustering is implementedby MapReduce with a little improvement that it increased the screening of the pixelsand the segment selection of the initial cluster centers.