การเปรียบเทียบกลยุทธ์ ROI สำหรับการนับไก่แบบเรียลไทม์โดยใช้ YOLOv8 และการพัฒนาที่ช่วยเหลือโดย LLM ในโรงฆ่าสัตว์อุตสาหกรรม
Year: 2025
Type: article
Abstract: Accurate chicken counting is essential for operational efficiency and compliance in poultry processing. Manual counting methods are prone to error and unsuitable for high-speed production. This study presents the validation of automated chicken counting in an industrial slaughterhouse using YOLOv8 detection with SORT tracking and ROI-based strategies. While the core pipeline follows established computer vision methods, the novelty lies in systematically benchmarking three ROI strategies under high-speed conveyor conditions where occlusion, motion blur, and unstable lighting are major challenges. Tested on real production line footage, the system was evaluated using precision, recall, and F1-score against ground truth counts. Video-based strategies centred on the conveyor line achieved the highest accuracy, with F1-scores up to 0.998 and a Mean Absolute Error (MAE) of 2.30, a Mean Absolute Percentage Error (MAPE) of 0.74%, and a Root Mean Square Error (RMSE) of 2.70, while image-based approaches undercounted by up to 13%. Confidence variability was markedly lower in video-based methods (CV < 9%), demonstrating robustness under dynamic production conditions. Beyond methodological integration, this work introduces LLM-driven code generation for rapid development of industrial vision systems. The findings provide practical guidance for camera positioning, threshold settings, and deployment in high-speed slaughterhouse environments, establishing a foundation for scalable, high-accuracy poultry processing automation.
Institutions: Naresuan University, Uttaradit Rajabhat University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
Funder: Naresuan University
OpenAlex ID: W7215171941