Super-resolved multi-temporal segmentation with deep permutation-invariant networks

TL;DR

This paper introduces a deep learning model for super-resolved semantic segmentation of satellite images, leveraging permutation-invariant networks and multi-resolution fusion to improve spatial resolution beyond sensor capabilities.

Contribution

It presents a novel permutation-invariant deep learning approach for super-resolved semantic segmentation, outperforming previous methods in multi-temporal satellite image analysis.

Findings

Model won the AI4EO challenge on Enhanced Sentinel 2 Agriculture

Achieved superior segmentation accuracy at higher resolutions

Demonstrated effectiveness of multi-resolution fusion in super-resolution

Abstract

Multi-image super-resolution from multi-temporal satellite acquisitions of a scene has recently enjoyed great success thanks to new deep learning models. In this paper, we go beyond classic image reconstruction at a higher resolution by studying a super-resolved inference problem, namely semantic segmentation at a spatial resolution higher than the one of sensing platform. We expand upon recently proposed models exploiting temporal permutation invariance with a multi-resolution fusion module able to infer the rich semantic information needed by the segmentation task. The model presented in this paper has recently won the AI4EO challenge on Enhanced Sentinel 2 Agriculture.