Automated Change Detection in Land Cover Using Deep Learning

Land cover changes, such as deforestation, plays a crucial role in understanding environmental transformations and supporting sustainable land management practices. This study explores the application of deep learning models to automate change detection using temporal remote sensing imagery. Specifically, U-Net and Vision Transformer (ViT) architectures are utilized and compared for their effectiveness in identifying and quantifying changes across time-separated image pairs. The dataset comprises high-resolution remote sensing images organized (`time1` and `time2`) with corresponding labeled changes for training and evaluation. Different performance metrics are employed to assess both models. Preliminary findings reveal that U-Net delivers strong performance in preserving intricate details, while ViT excels in capturing complex spatial relationships. This comparative analysis highlights the advantages and limitations of each approach, contributing valuable insights to the field of automated land cover change detection.