Change Detection Based on Ordinal Entropy of Self-Supervised Segmentation

Change detection in multi-pass imagery may be subject to unknowable and/or unstructured variations. This challenges supervised approaches, since annotations and examples may not exist to explain what is observed. This paper proposes two components that jointly characterize this class of change. The first component is a self-supervised clustering network that is data-driven and can be applied at single or multiple-pass datacubes at once. The second component builds on several recently introduced measures of information entropy in images that are applied to the clusters detected in the first stage. This framework is versatile and can be used in an end-to-end pipeline. It consumes raw data and computes qualitative and quantitative metrics on the data without using priors.

Paper

Full text

PDF

Change Detection Based on Ordinal Entropy of Self-Supervised Segmentation

Semantic Scholar · Computer Science · 2023

Abstract

Change detection in multi-pass imagery may be subject to unknowable and/or unstructured variations. This challenges supervised approaches, since annotations and examples may not exist to explain what is observed. This paper proposes two components that jointly characterize this class of change. The first component is a self-supervised clustering network that is data-driven and can be applied at single or multiple-pass datacubes at once. The second component builds on several recently introduced measures of information entropy in images that are applied to the clusters detected in the first stage. This framework is versatile and can be used in an end-to-end pipeline. It consumes raw data and computes qualitative and quantitative metrics on the data without using priors.

Similar papers

© 2026 NYSGPT2525 LLC