Abstract
Change detection typically involves identifying regions with changes between bitemporal images taken at the same location. While significant changes between images have traditionally been the focus, gradual or subtle changes are also critically important in real-world monitoring scenarios. For instance, weak changes often serve as precursors to major hazards in scenarios like slopes, embankments, and tailings ponds. Therefore, a change detection network capable of simultaneously detecting slow and fast changes is highly needed in practical applications. To address this issue, we propose a novel method named Flow-CDNet, a dual-branch neural network composed of an optical flow branch and a binary change detection branch. The optical flow branch employs a multi-scale pyramid structure to extract displacement changes at various spatial resolutions. The binary change detection branch integrates features from a ResNet-based convolutional neural network with outputs from the optical flow branch, facilitating the detection of rapid changes. Subsequently, to supervise and evaluate this new change detection framework, a self-built change detection dataset Flow-Change, a loss function combining binary Tversky loss and L2 norm loss, along with a new evaluation metric called FEPE are designed. Quantitative experiments conducted on Flow-Change dataset demonstrated that our approach outperforms the existing methods. Ablation studies further confirm that the two branches reinforce each other, improving the overall detection accuracy. Additional evaluation using real-world monitoring images from embankments illustrates the practical applicability and effectiveness of the proposed method, reinforcing its suitability for real-life hazard monitoring tasks associated with both slow deformation processes and abrupt collapses.