Building Damage Mapping with Self-PositiveUnlabeled Learning

TL;DR

This paper demonstrates a self-paced positive-unlabeled learning approach for building damage assessment that performs comparably to supervised methods using limited labeled data and abundant unlabeled data in disaster scenarios.

Contribution

It introduces a novel self-PU learning method for damage mapping that reduces the need for extensive labeled datasets in disaster response.

Findings

Self-PU achieves similar accuracy to supervised learning with fewer labeled samples.

The approach is validated on datasets from multiple major disasters.

Self-PU outperforms traditional PU learning methods.

Abstract

Humanitarian organizations must have fast and reliable data to respond to disasters. Deep learning approaches are difficult to implement in real-world disasters because it might be challenging to collect ground truth data of the damage situation (training data) soon after the event. The implementation of recent self-paced positive-unlabeled learning (PU) is demonstrated in this work by successfully applying to building damage assessment with very limited labeled data and a large amount of unlabeled data. Self-PU learning is compared with the supervised baselines and traditional PU learning using different datasets collected from the 2011 Tohoku earthquake, the 2018 Palu tsunami, and the 2018 Hurricane Michael. By utilizing only a portion of labeled damaged samples, we show how models trained with self-PU techniques may achieve comparable performance as supervised learning.