Buoy Detection under Extreme Low-light Illumination for Intelligent Mussel Farming.

Mussel farming in New Zealand (NZ) is a thriving and crucial industry that supports local communities and underscores aquaculture’s significance. A key task in mussel farming involves buoy inspection to monitor mussel growth for efficient harvesting. Currently, this process heavily relies on manual labor, resulting in labor and time-intensive spot operations for farmers. To address this, AI methods have recently been studied for fully automated buoy detection from captured images, showing promising results. However, these methods have mainly focused on images with good illumination, limiting their applicability in low-light conditions, such as dawn and dusk. This paper presents the first exploration of AI-based automated buoy detection in low-light environments, taking steps toward full-day buoy inspection. We investigate the impact of low-light conditions on deep learning-based buoy detection performance and explore techniques like intrinsic decomposition and low-light image enhancement to mitigate these effects. Incorporating these analyzed and enhanced results into a YOLO-based deep learning model results in significant improvements in low-light detections.