Automating the Detection of Unplanned Urban Constructions through AI-Powered Super-Resolution and Multi-modal Data Fusion

Unauthorized urban development significantly threatens sustainable development objectives (SDG 11), requiring robust and scalable surveillance systems. While Algeria's ALSAT-2 satellite offers consistent territorial coverage, its inherent 10m spatial resolution (2.5m post-pansharpening) proves inadequate for detecting small-scale illegal structures. This study presents a novel dual-stage artificial intelligence pipeline designed to augment existing national satellite data for urban sprawl monitoring. Initially, a Generative Adversarial Network was developed using Sentinel-2 training datasets to execute spectral enhancement, generating higher-resolution imagery with supplementary spectral channels from original ALSAT-2 data. Subsequently, a U-Net architecture processed this refined imagery to automatically identify and delineate unauthorized construction areas. To address local material discrimination challenges, Land Surface Temperature information was incorporated to distinguish concrete surfaces from prevalent alternative building materials in the study region. Performance analysis revealed that spectral enhancement significantly boosted the classification accuracy relative to standard pansharpened baselines, with thermal data integration providing further improvements in the precision and sensitivity of detection. The methodology successfully mapped illegal construction zones, achieving an IoU of 0.7712, Dice coefficient of 0.8350, Precision of 0.8210, Recall of 0.9285, and Overall Accuracy of 0.9516, which was verified through field validation in Arzew, Oran, Algeria, during the years 2022 and 2025. This research demonstrates artificial intelligence's potential to transcend the hardware limitations of satellite sensors, providing an economical and adaptable solution for urban governance, planning compliance, and sustainable territorial management.