ChangeBind: A Hybrid Change Encoder for Remote Sensing Change Detection

arXiv:2404.17565v1 Announce Type: new
Abstract: Change detection (CD) is a fundamental task in remote sensing (RS) which aims to detect the semantic changes between the same geographical regions at different time stamps. Existing convolutional neural networks (CNNs) based approaches often struggle to capture long-range dependencies. Whereas recent transformer-based methods are prone to the dominant global representation and may limit their capabilities to capture the subtle change regions due to the complexity of the objects in the scene. To address these …

abstract arxiv change cnns convolutional convolutional neural networks cs.cv dependencies detection encoder fundamental hybrid networks neural networks semantic sensing struggle transformer type