Trusting small training dataset for supervised change detection

TL;DR

This paper investigates how supervised deep learning models for change detection can perform well with small, geographically diverse training datasets, introducing a confidence indicator to assess model trustworthiness.

Contribution

It demonstrates that geographically diverse small datasets improve supervised change detection and proposes a confidence indicator to evaluate model reliability.

Findings

Diverse training data enhances supervised model performance.

A confidence indicator helps verify model trustworthiness.

Unsupervised methods outperform supervised ones on less confident test cases.

Abstract

Deep learning (DL) based supervised change detection (CD) models require large labeled training data. Due to the difficulty of collecting labeled multi-temporal data, unsupervised methods are preferred in the CD literature. However, unsupervised methods cannot fully exploit the potentials of data-driven deep learning and thus they are not absolute alternative to the supervised methods. This motivates us to look deeper into the supervised DL methods and investigate how they can be adopted intelligently for CD by minimizing the requirement of labeled training data. Towards this, in this work we show that geographically diverse training dataset can yield significant improvement over less diverse training datasets of the same size. We propose a simple confidence indicator for verifying the trustworthiness/confidence of supervised models trained with small labeled dataset. Moreover, we show that for the test cases where supervised CD model is found to be less confident/trustworthy, unsupervised methods often produce better result than the supervised ones.