Synthetic Dataset Generation for Change Detection in Mining Areas Using Image Inpainting

Change Detection is a critical task in Remote Sensing, aimed at identifying and quantifying alterations in a specific area over time by analyzing satellite or aerial imagery. While numerous datasets and models have been developed for this purpose, most existing benchmarks focus primarily on urban or agricultural environments, neglecting the particular challenges of mining areas. Moreover, ground truth data in current datasets tend to be coarse and insensitive to small-scale modifications, reducing their suitability for precise applications such as environmental monitoring or land-use management in mining zones. To address this limitation, this work presents the creation of a custom synthetic dataset for Change Detection tailored to mining areas, generated using inpainting techniques to simulate realistic temporal variations in satellite imagery. This paper details the dataset creation pipeline, discusses challenges encountered during the inpainting process, and analyzes the suitability of synthetic data generation. Experimental results demonstrate that models trained with the proposed synthetic data achieve improved sensitivity to small-scale changes compared to baseline datasets. Overall, the main contribution of this work is the proposal and evaluation of a controlled synthetic data generation strategy for Change Detection in mining areas.