Mobile Application-Based Detection of Primal Meat Cuts Using a Quantized YOLOv8n Model for Public Market Food Process Monitoring

Public market meat monitoring requires practical computer vision tools that can operate under real-world conditions such as variable lighting, cluttered backgrounds, inconsistent camera angles, and diverse meat presentations. This study developed a mobile application-based detection system for recognizing primal meat cuts using a quantized YOLOv8n model for public market food process monitoring. A total of 3,740 images were captured using Huawei Y7 and iPhone X smartphones under public market conditions and organized into nine meat cut categories: pork head, beef leg, beef ribs, chicken breast, chicken leg, chicken wings, pork belly, pork leg, and pork shoulder. The dataset was divided into 2,992 training images, 374 validation images, and 374 testing images, with image annotations prepared using bounding boxes and class labels in YOLO format. The YOLOv8n model was trained and exported to TensorFlow Lite, and the FP16 model was selected for mobile deployment to prioritize detection accuracy over inference speed. On the test set, the model achieved a precision of 0.994, recall of 0.998, F1-score of 0.996, mAP@50 of 0.993, and mAP@50:95 of 0.817. To provide practical validation beyond the prepared test split, field testing was conducted using a Huawei Y7 smartphone under actual market conditions, achieving 91.56% accuracy across 450 field images. These results indicate that a quantized YOLOv8n model can support accurate and mobile-based detection of primal meat cuts for public market food process monitoring. Further evaluation across additional market locations, lighting conditions, background settings, and smartphone hardware is recommended to assess broader real-world robustness.