CNN-Based System for Photovoltaic Potential Estimation on Rooftops in San Pedro Sula, Honduras

Honduras depends mainly on fossil fuels for electricity generation, with the need to diversify its energy matrix. This study presents a CNN-based system for estimating rooftop photovoltaic potential in San Pedro Sula, Honduras. A convolutional neural network was trained on aerial imagery from nearby cities to detect and segment rooftops, achieving 94.7% mAP, 91.2% precision, and 88.7% recall. The CNN was then applied to a representative sample of 383 images from 27 neighborhoods in San Pedro Sula. A Python algorithm calculated usable roof area, number of potential solar modules, nominal capacity in DC, energy generation, and avoided greenhouse gas emissions based on the CNN detections. The analysis revealed significant photovoltaic potential, estimating a total nominal capacity of 750.73 MWp, annual energy generation of 1,041,634.34 MWh, and potential avoided GHG emissions of 636,855.23 tCO2 across the sampled neighborhoods. These results provide valuable insights for energy planning and highlight the opportunity for decentralized photovoltaic systems to address electricity distribution challenges in Honduras. The methodology developed offers an adaptable tool for estimating urban solar potential, for more comprehensive future studies.