Scalable recurrent neural network for hyperspectral image classification
- PDF / 1,741,447 Bytes
- 17 Pages / 439.37 x 666.142 pts Page_size
- 32 Downloads / 240 Views
Scalable recurrent neural network for hyperspectral image classification Mercedes E. Paoletti1 · Juan M. Haut1 · Javier Plaza1 · Antonio Plaza1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Hyperspectral imaging (HSI) collects hundreds of images over large spatial observation areas on the Earth’s surface, recording scenes at different wavelength channels and providing a vast amount of information. Recurrent neural networks (RNNs) have been widely used for the classification of HSI datasets, understood as a single sequence of pixel vectors with high dimensionality. However, the RNN model scales poorly when dealing with HSI scenes with large dimensionality. In order to mitigate this problem, this paper presents a new RNN classifier based on simple recurrent units that performs HSI classification in a highly scalable and efficient way. Our experimental results (conducted on four real HSI datasets) reveal very good performance, not only in terms of classification accuracy (in line with existing methods), but also in terms of computational performance when dealing with large datasets. Keywords Hyperspectral image · Recurrent neural networks · CUDA
1 Introduction The significant advances in computing technology achieved in the last decade, coupled with the newest developments in imaging spectroscopy [18], have allowed the development of new Earth observation (EO) missions with powerful airborne and satellite hyperspectral imaging (HSI) sensors, which can capture high-quality Source codes: https://github.com/mhaut/scalable_RNN_HSI. * Juan M. Haut [email protected] Mercedes E. Paoletti [email protected]
Javier Plaza [email protected]; [email protected]
1
Department of Technology of Computers and Communications, University of Extremadura, Escuela Politecnica, Avda. de la Universidad s/n, Cáceres, Spain
13
Vol.:(0123456789)
M. E. Paoletti et al.
images composed by hundreds of measurements (at different wavelength channels) over extensive spatial areas, acquiring information in hundreds of continuous and narrow bands, ranging from the visible to the near-infrared (NIR) and shortwaveinfrared (SWIR) [8] parts of the electromagnetic spectrum. As a result, current spectrometers are able to produce very large HSI data cubes, where each pixel contains the spectral signature of the observed materials. These spectral signatures collect the physical–chemical behavior of materials in the presence of solar light, being unique for each kind of terrestrial object, and allowing to describe and identify each element of the scene, not only at an object level, but also at pixel (and even sub-pixel) level of detail [19], providing abundant information for the characterization of the surface of the Earth. Such information can be used in a wide range of human activities, such as hydrology [12], forestry [20], geology [22] and mineralogy [5]), as well as precision agriculture [7], urban planning [27], and prevention and management of disasters. In this context, the analysis and processing of
Data Loading...
