Fine-Tuning with Attention Mechanisms for LULC Classification: A Case Study of Tianjin, China

The repository contains:
Training and testing datasets: Processed Sentinel-2 image tiles (10-channel) and the corresponding 7-class LULC labels used for model training and testing.
Evaluation data: Numerical values required to reproduce the reported metrics, including per-class precision, recall, F1-score, overall accuracy (OA), and Kappa coefficient. Raw confusion matrices are also included.
Figure and table data: The underlying tabular values used to generate all figures and tables in the manuscript, including classification accuracy plots, comparative charts, and performance summaries.
Original Sentinel-2 Level-2A imagery is publicly available from the European Space Agency (ESA) Copernicus Open Access Hub (https://scihub.copernicus.eu/) and is therefore not redistributed here. The shared repository provides all processed and labelled data required to replicate the findings of this study.