Automatización en la detección de potenciales linderos: integración de algoritmos de IA y fusión de imágenes

Outdated cadastral information in Colombia constitutes a structural barrier to the implementation of the Comprehensive Rural Reform (CRR). This research validates a methodology for the automatic extraction of visible boundaries through the fusion of synthetic aperture radar (SAR) and optical imagery, employing artificial intelligence techniques. Machine learning and deep learning approaches were comparatively evaluated, contrasting the foundational Segment Anything Model (SAM) with a retrained edge detector (VGG13_bn). Quantitative results indicate that, while SAM exhibits a higher level of segmentation, the VGG13_bn model achieved an F1-score of 0.405 and an accuracy of 0.888, emerging as the most balanced and operationally viable alternative. This work provides a reproducible methodological workflow that can support cadastral modernization processes in territorially complex contexts.