Learning Sentinel-2 Spectral Dynamics for Long-Run Predictions Using Residual Neural Networks

From LRDE

Abstract

Making the most of multispectral image time-series is a promising but still relatively under-explored research direction because of the complexity of jointly analyzing spatial, spectral and temporal information. Capturing and characterizing temporal dynamics is one of the important and challenging issues. Our new method paves the way to capture real data dynamics and should eventually benefit applications like unmixing or classification. Dealing with time-series dynamics classically requires the knowledge of a dynamical model and an observation model. The former may be incorrect or computationally hard to handle, thus motivating data-driven strategies aiming at learning dynamics directly from data. In this paper, we adapt neural network architectures to learn periodic dynamics of both simulated and real multispectral time-series. We emphasize the necessity of choosing the right state variable to capture periodic dynamics and show that our models can reproduce the average seasonal dynamics of vegetation using only one year of training data.


Bibtex (lrde.bib)

@InProceedings{	  estopinan.21.eusipco,
  author	= {Joaquim Estopinan and Guillaume Tochon and Lucas Drumetz},
  title		= {Learning {Sentinel-2} Spectral Dynamics for Long-Run
		  Predictions Using Residual Neural Networks},
  booktitle	= {Proceedings of the 29th European Signal Processing
		  Conference (EUSIPCO)},
  year		= 2021,
  address	= {Dublin, Ireland},
  month		= aug,
  abstract	= {Making the most of multispectral image time-series is a
		  promising but still relatively under-explored research
		  direction because of the complexity of jointly analyzing
		  spatial, spectral and temporal information. Capturing and
		  characterizing temporal dynamics is one of the important
		  and challenging issues. Our new method paves the way to
		  capture real data dynamics and should eventually benefit
		  applications like unmixing or classification. Dealing with
		  time-series dynamics classically requires the knowledge of
		  a dynamical model and an observation model. The former may
		  be incorrect or computationally hard to handle, thus
		  motivating data-driven strategies aiming at learning
		  dynamics directly from data. In this paper, we adapt neural
		  network architectures to learn periodic dynamics of both
		  simulated and real multispectral time-series. We emphasize
		  the necessity of choosing the right state variable to
		  capture periodic dynamics and show that our models can
		  reproduce the average seasonal dynamics of vegetation using
		  only one year of training data.}
}