Abstract
Machine learning techniques are widely used for pixel-wise classification of hyperspectral images. These methods can achieve high accuracy, but most of them are computationally intensive models. This poses a problem for their implementation in low-power and embedded systems intended for on-board processing, in which energy consumption and model size are as important as accuracy. With a focus on embedded and on-board systems (in which only the inference step is performed after an off-line training process), in this paper we provide a comprehensive overview of the inference properties of the most relevant techniques for hyperspectral image classification. For this purpose, we compare the size of the trained models and the operations required during the inference step (which are directly related to the hardware and energy requirements). Our goal is to search for appropriate trade-offs between on-board implementation (such as model size and energy consumption) and classification accuracy.
Highlights
Fostered by significant advances in computer technology that have taken place from the end of the last century to the Earth observation (EO) field has greatly evolved over the last 20 years [1]
It is worth mentioning that, for a tree based model, gradient boosting decision Trees (GBDT) achieves great accuracy values which are very close to those obtained by neural networks and the support vector machine (SVM), which always provide higher values than the random forests (RFs), which is a tree based model
The distribution of the different models keeps the same behavior described for the rest of the data sets, with the particularity that the multinomial logistic regression (MLR) model outperforms the GBDT in this case
Summary
Fostered by significant advances in computer technology that have taken place from the end of the last century to the Earth observation (EO) field has greatly evolved over the last 20 years [1]. Focusing on hyperspectral imaging (HSI) [19], this kind of data comprises abundant spectral–spatial information for large coverage, obtained by capturing the solar radiation that is absorbed and reflected by ground targets at different wavelengths, usually ranging from the visible, to the near (NIR) and short wavelength infrared (SWIR) [20]. In this sense, HSI data obtained by airborne and satellite platforms consist of huge data cubes, where each pixel represents the spectral signature of the captured object. As it was proposed as the benchmark data set for the 2013 IEEE Geoscience and Remote Sensing Society data fusion contest [148], it is already divided into training and testing sets, with 2832 and 12,197 pixels, respectively
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
