Thermal and visible images acquired from Unmanned Aerial Vehicles (UAV) have proven to be indispensable for detecting faults in photovoltaic panels. Thermal orthomosaics enable rapid identification of defective modules, while visible imagery assists in interpreting the nature of these defects. However, accurate orthomosaic generation remains challenging due to the low spatial resolution of thermal images and the repetitive patterns of photovoltaic installations. This paper aims to optimize orthomosaic generation through the fusion of UAV thermal and visible images. An extensive literature review was conducted, followed by an analytical comparison of nine image fusion methods: ADF, CBF, FPDE, GLPF, GTF, LatLRR, MGFF, MSVD, and VGG. The methods were evaluated using qualitative visual assessment and quantitative metrics. Results indicate that MGFF outperforms the other approaches, achieving an entropy value of 1.0356 and significantly enhancing image detail, which improves tie-point detection and image alignment. Accordingly, a dedicated processing pipeline combining MGFF with a customized registration workflow was developed and validated on a challenging dataset of 500 thermal and visible image pairs, demonstrating substantial improvements in orthomosaic quality.
Bamessaoud et al. (2026) studied this question.