|
Enhancing drone-based solar panel inspection with thermal image super resolution |
|---|---|
| รหัสดีโอไอ | |
| Title | Enhancing drone-based solar panel inspection with thermal image super resolution |
| Creator | Niwat Jamrunsin |
| Contributor | Sasiporn Usanavasin, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | Solar PV farm, Inspection, Drone, Thermal image, Super-resolution, Deep learning, Anomaly detection, LCOE, Diagnostic, O&M, RECALL, F1 score, Accuracy, Precision, Diffusion, Bicubic interpolation |
| Abstract | This paper focuses on an operational issue for large-scale photovoltaic (PV) solar farms: how to perform speedy, cost-efficient inspections without compromising the image quality required to enable fault detection with great precision. Drone-based thermography is now the de facto standard for solar O&M, enabling rapid coverage of extensive sites and revealing non-visible defects such as hotspots and emerging cell failures. However, for today’s practice, there is a basic altitude–resolution trade-off involved: low-altitude flights generate high-resolution thermal images that would be able to detect subtle anomalies but are slow, battery-intensive, and expensive, while high-altitude flights work more effectively but lead to low-resolution images that hide important defects and may result in expensive ground re-inspections. To overcome this limitation, the thesis proposes a deep learning-based super resolution framework tailored for drone-captured thermal images of solar panels. The core idea lies in computationally enhancing low-resolution images obtained from efficient high-altitude flights to reconstruct finer spatial detail which approximates low altitude inspections. The study involves the curation of a paired dataset where lower resolution thermal images can match high-resolution ones, designing an appropriate super-resolution model and evaluating quantitative performance by standard image quality metrics. In addition to visual improvement, the framework is evaluated based on the impact on downstream automated anomaly detection algorithms and economic benefits for solar O&M processes, from how it would reduce re-inspection work and increase efficiency in flight-planning process optimization. Through the convergence of computer vision, deep learning, and renewable energy asset management, this work seeks to facilitate implementation of inspection protocols that will offer coverage and cost benefits of high-altitude flight and also the diagnostic precision of high-resolution thermography. The desired results are a more robust, scalable, and economical mechanism for PV farm maintenance, resulting in increased system efficiency, lower LCOE, and longer-term sustainable solar energy. |
