Scalable Multi-Temporal Remote Sensing Change Data Generation via Simulating Stochastic Change Process

Understanding the temporal dynamics of Earth's surface is a mission ofmulti-temporal remote sensing image analysis, significantly promoted by deepvision models with its fuel -- labeled multi-temporal images. However,collecting, preprocessing, and annotating multi-temporal remote sensing imagesat scale is non-trivial since it is expensive and knowledge-intensive. In thispaper, we present a scalable multi-temporal remote sensing change datagenerator via generative modeling, which is cheap and automatic, alleviatingthese problems. Our main idea is to simulate a stochastic change process overtime. We consider the stochastic change process as a probabilistic semanticstate transition, namely generative probabilistic change model (GPCM), whichdecouples the complex simulation problem into two more trackable sub-problems,\ie, change event simulation and semantic change synthesis. To solve these twoproblems, we present the change generator (Changen), a GAN-based GPCM, enablingcontrollable object change data generation, including customizable objectproperty, and change event. The extensive experiments suggest that our Changenhas superior generation capability, and the change detectors with Changenpre-training exhibit excellent transferability to real-world change datasets.