Change Detection in Heterogeneous Remote Sensing Images Based on Copula-Supervised Neural Networks

Change detection (CD) in heterogeneous remote sensing images has been extensively applied in response to emergencies. There have been substantial improvements during the past decades in addressing heterogeneous CD challenges, driven by the rapid advancements in deep neural networks (DNNs). However, most DNN-based heterogeneous CD methods often exhibit black-box behavior, limiting their interpretability and controllability in many practical CD applications. To overcome these issues, we propose a neural copula change detection (NCCD) framework, an approach for the heterogeneous CD problem that leverages an interpretable neural network supervised by copula theory. In NCCD, copula’s mathematical characteristics are incorporated as loss functions to supervise a neural network in modeling dependence between bi-temporal heterogeneous remote sensing images. The trained neural network works as a copula function to estimate the degrees of dependence that serve as the indicator for identifying changes. Experimental results on two real heterogeneous CD datasets substantiate the effectiveness of our NCCD.