基于弹性YOLO模型的移动端珍稀竹种种子缺陷检测系统构建OA

To accurately detect seed defects of Chimonobambusa utilis and Qiongzhuea tumidinoda using portable mobile devices,this paper proposes a lightweight detection method based on an Elastic YOLO model,aiming to address the problem of excessive memory consumption and poor adaptability to mobile devices in existing defect detection approaches.The model introduces elastic convolution layers into the original YOLO model,which can automatically identify filters that have low relevance to seed defect features and be disabled when memory resources are constrained,thereby reducing memory usage while maintaining high detection accuracy.Meanwhile,an integrated defect detection system is built centered on the Elastic YOLO model,which monitors the available memory of mobile devices and the model's memory footprint,formulates a linear programming problem to determine the optimal filter usage strategy,and maximizes detection accuracy under given memory constraints.Based on the dataset containing 2 000 seeds collected in Zhaotong,Yunnan,experiments on three real mobile devices—iPhone 11,Huawei Mate20 X,and NVIDIA Xavier NX(with available memory ranging from 0.5 GB to 3 GB)demonstrate a detection accuracy(mAP@0.5)up to 99.3％,outperforming four advanced models—improved YOLOv5,YOLOv8,YOLOv11,and Mask R-CNN by 14.6％,11.1％,8.78％,and 5.22％,respectively.The Elastic YOLO model can be deployed on a wide range of mobile devices to improve defect detection efficiency.This provides a technical reference for lightweight detection of rare bamboo seeds in mobile devices.