Cloud Implementation of Multinomial Logistic Regression for UAV Hyperspectral Images

Hyperspectral imaging (HSI) is a hot topic within the remote sensing field due to the great amount of spectral information acquired over a wide range of the electromagnetic spectrum. Traditionally, aerial and satellite platforms have been used to transport the spectrometers. However, nowadays there is a very clear trend in the use of small satellites (smallSats) and unmanned aerial vehicles (UAVs). In this sense, the miniaturized sensors are currently providing a massive amount of data, characterized by its variety, velocity, and volume. The proliferation of this kind of data, coupled with its characteristics, has resulted in the concept of big remote sensing data. The management of the massive amounts of HSI data is a highly demanded task. However, HSI processing is a challenging task due the huge amount of information comprised in these images, which contain hundreds of contiguous and narrow spectral bands. High performance computing (HPC) is an efficient way to address the enormous computation requirements introduced by HSI processing, particularly cloud computing, which offers a natural solution for the management of large and complex datasets. In this article, a new cloud computing approach has been developed for efficient HSI processing. Based on the Apache Spark distributed platform, we specifically provide a cloud implementation of the multinomial logistic regression (MLR) probabilistic classifier. Conducted experiments over two real HSI scenes reveal that cloud computing architectures provide an efficient processing of large HSI datasets while maintaining the accuracy performance.