Confidence Estimation in Unsupervised Deep Change Vector Analysis

Unsupervised transfer learning-based change detection methods exploit the feature extraction capability of pre-trained networks to distinguish changed pixels from the unchanged ones. However, their performance may vary significantly depending on several geographical and model-related aspects. In many applications, it is of utmost importance to provide trustworthy or confident results, even if over a subset of pixels. The core challenge in this problem is to identify changed pixels and confident pixels in an unsupervised manner. To address this, we propose a two-network model - one tasked with mere change detection and the other with confidence estimation. While the change detection network can be used in conjunction with popular transfer learning-based change detection methods such as Deep Change Vector Analysis, the confidence estimation network operates similarly to a randomized smoothing model. By ingesting ensembles of inputs perturbed by noise, it creates a distribution over the output and assigns confidence to each pixel's outcome. We tested the proposed method on three different Earth observation sensors: optical, Synthetic Aperture Radar, and hyperspectral sensors.