YOLO-Ee：一种用于烟草粉螟检测的改进型自标检测方法

打开文本图片集

中图分类号：S379.5 文献标志码：A 文章编号：2096-9902（2025）23-0032-06

Abstract：TomitigatequalitylossescausedbyEphestiaelutella（tobaccomoth）duringstorageandtoovercomethe ineficiencyandsubjectivityofmanualmonitoring，thisstudyaimstodevelopahigh-precision，deployableautomateddetection model.noveldetector，YOLO-Ee，wasbuiltontheYOLOv12backbonebyincorporating windmillshapedconvolutions（SConv） andanenhancedPSC3k2module.Atrap-boardimagedatasetwascompiledandenlarged throughdata-augmentation techniques. The modelwas evaluatedagainst mainstream detectors-FasterR-CNN，YOLOv5，andYOLOv8-using mean AveragePrecisionat 0.5IoU（mAP @0.5 5）and related metrics.YOLO-Ee achieved a mAP @0.5 of 94.7% ，outperforming Faster R-CNN， YOLOv5，and YOLOv8by6.34.1，and.8percentagepoints，respectivelywithsuperiorperformanceonsmallanddenselyclusteredtargets. TheimprovedYOLO-Eemodeldeliverseficientandaccurateidenticationoftobaccomothsontrapboards，oferingareliable solution for warehouse pest monitoring and paving the way for future edge-computing applications.

Keywords： Ephestiaelutella; object detection; YOLO-Ee; PSConv; pest monitoring; deep learning

烟草粉螟（Ephestiaelutella）是一种世界性分布的多食性仓储害虫，主要侵害烟草、小麦、可可及干制坚果等农产品，对烟草行业危害尤甚[-2]。（剩余9665字）