Computer Vision Method for Extracting an Induced Electroluminescence Signal from Photovoltaic Modules in Daylight Conditions Using Drone-Captured Images

Electroluminescence (EL) imaging is a powerful technique for evaluating the condition of photovoltaic (PV) modules and individual cells. While drones are a cheap and practical imaging medium, they present a challenge in terms of image stabilization. Flying during daylight hours has the advantage of being safer, cheaper, and camera-focus is easier to maintain.

The main drawback is that sunlight introduces sufficient background noise to dominate the EL signal. We present a method for automatically tracking and rectifying a PV module in a stack of EL images captured by drone in daylight, and subsequently rasterizing the EL signal (S/N < 0.1). The method combines feature detection with direct corner alignment, and is applicable to any type of drone-based PV-inspection.

To extract the EL signal, a stabilized image stack is analyzed depth-wise. Background noise is calculated and subtracted, and a Fast Fourier Transform (FFT) analysis is performed to form an EL amplitude map of the module. The analysis is validated by examining adjacent frequencies and by comparison with stationary EL.

Results show promising image quality that fares well compared to stationary EL.