Abstract:

To address the challenges of slow response, high false alarm rates, and poor environmental adaptability in traditional coal slurry overflow detection methods, a novel approach based on ground markers was proposed. By deploying ground markers and utilizing the YOLOv8 deep learning model, the visibility of markers was monitored in real time, with an automatic overflow alarm triggered when markers were covered by coal slurry. A dataset of 60 images from a coal preparation plant was collected, including 40 simulated images generated by pouring coal slurry over markers. Data augmentation techniques, such as brightness adjustment, contrast enhancement, and noise addition, were applied to simulate environmental variations, expanding the dataset to 100 images for YOLOv8 model training. To validate the effectiveness of the proposed method, a comparative study with an image classification approach was conducted. The results demonstrate that this method significantly reduces the dependency on large-scale labeled data and effectively addresses the challenge of limited data availability, achieving a detection accuracy of 95%. Moreover, with substantially enhanced robustness against environmental variations such as illumination and weather changes, it is particularly well-suited for the complex and varying industrial environments of coal preparation plants.

Key words: YOLOv8, coal slurry overflow, coal preparation plant, object detection, marker