OS-YOLO: A robust object detection model for storage pest monitoring in American ginseng

Storage pest poses a serious threat to the quality and safety of Chinese medicinal materials such as American ginseng ( Panax quinquefolium ). However, intelligent detection of pests in ginseng storage remains challenging due to the lack of targeted datasets and models. Specifically, the detection task is hindered by the small size of pest targets, complex storage backgrounds, and poor robustness and generalization of existing models.This study propose OS-YOLO, an improved lightweight object detection model based on the YOLOv11n architecture, specifically optimized for pest monitoring in American ginseng storage. A specialized image dataset, Aginseng-Pest, was constructed, containing 10,207 annotated images across five key pest species under diverse lighting and background conditions. OS-YOLO introduces a unified design principle of hierarchical multi-granularity perception and adaptive feature coordination. By integrating BiFPN, ODConv, TODL, DCO-CSPB, and DB-SEv2, it achieves collaborative enhancement of multi-scale feature fusion, small-object recognition, and channel attention optimization, forming a structurally coherent and functionally complementary system-level detection framework.. Experimental results show that OS-YOLO achieved a precision of 96.1%, recall of 91.6%, F1-score of 94.0%, [email protected] of 96.2%, and [email protected]:0.95 of 76.3%, outperforming several state-of-the-art detection models. The proposed model significantly improves pest detection accuracy and robustness under complex storage scenarios, particularly for small and occluded insects. This study provides a key technical reference for intelligent pest prevention and control in the storage of Chinese medicinal materials. This study proposes OS-YOLO, an improved lightweight detection model optimized for monitoring storage pests in American ginseng. By integrating TODL, BiFPN, ODConv, DCO-CSPB, and DB-SEv2 modules, the model enhances tiny target detection and feature extraction in complex storage backgrounds. Experimental results on the newly constructed Aginseng-Pest dataset (10,207 images) demonstrate superior performance, achieving an F1 score of 94.0% and [email protected] of 96.2%. This method provides an effective solution for intelligent pest detection in the storage of functional foods. • OS-YOLO is proposed for pest monitoring in American ginseng storage as a robust detection model. • The Aginseng-Pest dataset has 10,207 images of five pests under varied conditions. • OS-YOLO reaches 96.2% [email protected] and 94.0% F1-score, outperforming eight detectors.