Unsupervised Change Detection

Unsupervised change detection aims to identify differences between images or point clouds acquired at different times without relying on labeled training data, a significant advantage in scenarios with limited resources or inaccessible ground truth. Current research focuses on leveraging foundation models like Segment Anything Model (SAM) and contrastive learning techniques, along with algorithms such as optimal transport and generative adversarial networks (GANs), to achieve robust and efficient change detection across various data modalities (e.g., optical imagery, LiDAR point clouds, SAR time series). This field is crucial for applications like disaster monitoring, environmental assessment, and autonomous robotic navigation, offering faster and more cost-effective solutions compared to supervised methods.