Exploration and Performance Improvement of YOLOv8 Model for Object Detection in Low-Light Conditions: A Case Study of Supermarket Retail Products

YOLOv8, a computer vision algorithm known for its exceptional object detection capabilities, has become an essential tool in various automation technologies. With its integration of new convolutional layers and the introduction of mosaic augmentation, YOLOv8 was engineered to deliver precise object predictions and superior feature extraction, making it a powerful solution for diverse object detection tasks. However, despite its advancements, challenges persist in specific scenarios, particularly under low-light conditions. These conditions often lead to detection inaccuracies, thereby increasing the likelihood of inference errors and potential financial or operational losses when applied in sectors such as retail and other industries. This paper presents the exploration of YOLOv8’s performance in such scenarios and proposes improvements through architectural modifications with the introduction of GhostConv. Additionally, the use of EnlightenGAN is explored to enhance detection capabilities under low-light conditions. Results demonstrate that these approaches effectively address YOLOv8’s limitations in low-light conditions, showcasing significant performance improvements and better reliability overall. As a result, this research contributes to advancing YOLOv8’s reliability in challenging scenarios, paving the way for broader and more accurate applications in real-world automation tasks. This work emphasizes the need for adaptive, innovative, and efficient solutions to enhance YOLOv8’s effectiveness under various operational constraints.