Classification of big satellite images using hadoop clusters for land cover recognition

The amount of remote sensing data including satellite imagery is increasing due to the growth in the number and the better accuracy of the imaging devices. Seeing that these images must be classified in a parallel and distributed storage-and-processing framework, in this study, a new framework is proposed based on the Hadoop Distributed File System and the Spark data processing engine for classifying big satellite imagery data in order to land cover recognition using clustered common-hardware. The linear Support Vector Machine and multi-layer Perceptron algorithms are compared in this framework. Results show the Spark and Hadoop can perform better in utilizing the Multi-Layer Perceptron Classification, offering a shorter execution time. Better accuracy compared with common methods such as the linear Support Vector Machine is obtained.