A modeling and validation of semantic relationships between land object changes and scene changes based on Bayesian networks

With the improvement of remote sensing image resolution, the landscape and texture information represented in the images is becoming increasingly rich. Encoding the landscape distribution and structural patterns in high-resolution remote sensing images and subsequently assigning semantic labels to image scenes can realize scene-level image analysis. Scene-level image analysis includes scene classification, scene segmentation, and scene change detection. Utilizing multi-temporal remote sensing images to identify land use changes at the scene level can provide a new perspective for analyzing and monitoring urban development. This paper proposes a scene-level change detection method based on a Bayesian network model to model the relationship between low-level object changes and scene change semantics and to verify its effectiveness based on a self-built dataset.