MHF-YOLO: An Improved YOLOv8n-Based Approach Developed for Commercial-Grade Honeysuckle Recognition

Honeysuckle (Lonicera japonica Thunb.) is a medicinal and edible plant whose flower buds are widely used in functional foods, herbal beverages, and natural antioxidant products. Accurate identification of harvestable buds is important because flower developmental stage affects the accumulation of bioactive constituents and the commercial value of raw materials. In this study, an improved YOLOv8n-based detection model, termed MHF-YOLO, was developed to identify immature, harvestable, and over-mature honeysuckle buds under field conditions. A custom dataset containing 400 original images was constructed and annotated into three categories. Large Separable Kernel Attention (LSKA) was introduced into the backbone to enhance feature representation under complex backgrounds, and the original CIoU loss was replaced with WIoU for bounding-box regression. Ablation experiments showed that the combined modifications improved mAP50 from 76.4% to 85.6% and mAP50:95 from 46.5% to 50.8%, while increasing the parameter count by only 3.4%. Compared with representative detection models, MHF-YOLO achieved a precision of 84.3%, a recall of 74.8%, an mAP50 of 85.6%, and an inference time of 7.09 ms. Compared with the recent lightweight YOLO11n baseline, MHF-YOLO achieved a slightly lower overall mAP50 (85.6% vs. 86.1%) and lower inference efficiency, but obtained a higher precision (84.3% vs. 83.7%) and a slightly higher mAP50:95 (50.8% vs. 50.1%) under the present experimental setting. Category-wise evaluation showed that MHF-YOLO achieved the best class-wise performance for harvestable buds, with a precision of 88.2%, a recall of 76.1%, an mAP50 of 93.7%, and an mAP50:95 of 63.2%. A PyQt5-based recognition system was further developed to support single-image, batch, and camera-based continuous recognition. These results indicate that the proposed MHF-YOLO-based platform is a feasible visual-recognition and decision-support prototype for commercial-grade honeysuckle assessment under the present field conditions.