Channel-Based Attention for LCC Using Sentinel-2 Time Series

TL;DR

This paper introduces a channel-based attention mechanism within a neural network architecture for land cover classification using Sentinel-2 satellite time series, enhancing interpretability and efficiency.

Contribution

It proposes a novel DNN architecture with channel attention and shared kernels to improve interpretability and reduce complexity in satellite image time series classification.

Findings

Promising classification accuracy on Sentinel-2 data

Effective channel importance weighting demonstrated

Reduced model complexity through shared kernels

Abstract

Deep Neural Networks (DNNs) are getting increasing attention to deal with Land Cover Classification (LCC) relying on Satellite Image Time Series (SITS). Though high performances can be achieved, the rationale of a prediction yielded by a DNN often remains unclear. An architecture expressing predictions with respect to input channels is thus proposed in this paper. It relies on convolutional layers and an attention mechanism weighting the importance of each channel in the final classification decision. The correlation between channels is taken into account to set up shared kernels and lower model complexity. Experiments based on a Sentinel-2 SITS show promising results.