Current defect inspection methods for photovoltaic (PV) devices based on electroluminescence (EL) imaging technology lack juggling both labor-saving and in-depth understanding of defects, restricting the progress towards yield improvement and higher efficiency. Herein, we propose an adaptive approach for automatic solar cell defect detection and cl. ••An automatic method is proposed for solar cell defect detection and classification.••An unsupervised algorithm is designed for adaptive defect detection.••A standardized diagnosis scheme is developed for statistical defect classification.••Extensive experimental results verify the effectiveness of the proposed method.Photovoltaic cellAbsolute electroluminescence imagingAutomatic defect detection and classificationReliability diagnosisIn the past few decades, solar power—a recognized alternative to fossil energy—has played an imperative role in the resolution of the global-scale energy crisis due to its safety, reliability, inexhaustibility, and environmental friendliness. Photovoltaic (PV) device that aims to convert solar energy to electricity has achieved record-breaking improvements in conversion efficiencies year by year [,,, ]. However, local defects are ubiquitous in solar cells due to the inherently granular structure and specific procedures employed during their manufacturing, which greatly impair the spatial uniformity and overall conversion efficiency of solar cells [,,, ]. Moreover, exposure under outdoor conditions or even under extremely harsh environments will exacerbate the defects, resulting in the long-term deterioration of cell performance [5,9]. Therefore, in the effort toward higher conversion efficiency, it is imperative to find an effective approach for defect diagnosis to provide conducive and instructive feedback for cell design and fabrication.In practice, conventional characterization techniques such as current-voltage (I–V) characteristics [,, ], capacitance measurements [13,14], and external quantum efficiency (EQE) [,,,, ] can help to monitor the condition of the whole-cell/module. Nevertheless, these global characteriz. 2.1. OverviewThe proposed adaptive automatic solar cell defect detection and classification method mainly consists of the following three steps: solar cell EL image preprocessing, adaptive solar cell defect detection, and solar cell defect classification, as shown in Fig. 1. During the preprocessing step, the effective solar cell regions are firstly detected from the input EL images, then the pixel values (arb. unit) of the effective regions are converted to absolute EL intensities (photons·s−1·cm−2) by using the input image information. The initialization of some parameters is also conducted in this step. In the second step, the proposed adaptive defect detection method takes the preprocessed absolute EL images as input, and outputs solar cell defect positions. During the detection process, the threshold parameters keep updating iteratively until the detection result satisfies the given requirements. Based on the detection results in step two, the injection-current-dependent absolute EL intensity loss rates of the defects are extracted to perform the solar cell defect classification process by matching the numerical simulation results.2.2. Input and preprocessingSolar.
