GLOW-DL: Generalized Light-Optimized Workflow with Deep Learning for Contamination Detection Using Fluorescence Imaging in Variable Conditions

Abstract Purpose This study aims to improve fluorescence imaging techniques for detecting surface contamination under various ambient light conditions. The major challenge addressed is the interference from ambient light, which diminishes fluorescence contrast and hinders accurate contaminant detection. Methods and Results We optimized key imaging parameters, including exposure time, synchronization of pulsed LEDs with camera exposure, and background subtraction. A noise-aware training approach was also applied using the YOLOv8 deep learning model to increase the model’s robustness to real-world noise. Results demonstrated that LED pulse synchronization enhanced image quality by reducing the impact of ambient light and increasing the signal-to-noise ratio by 25%. Extending exposure times from 3 to 21 ms increased fluorescence intensity by 35%, although it introduced a risk of motion blur. A refined background subtraction method significantly improved contrast, with up to a 30% enhancement, particularly under high ambient light levels, while maintaining controlled noise levels that were consistently lower in higher light conditions. Including Gaussian, Poisson, and stripe noise in training datasets substantially increased detection precision from 62.2 to 71.8% in low-noise environments and maintained precision at 60.8% in high-noise conditions. Conclusion The study confirms that optimized exposure settings, synchronized pulsed illumination, and noise-aware training substantially enhance the accuracy and reliability of fluorescence imaging for contamination detection. These strategies collectively offer a robust solution for improving contamination monitoring in environments with variable and challenging lighting, broadening the practical applications of fluorescence imaging.