中国烟草学报2025，Vol.31Issue(5)：20-30,11.DOI:10.16472/j.chinatobacco.2025.T0041
基于PP-PicoDet的改型轻量级烟丝异物检测算法
A modified lightweight algorithm for foreign object detection in cut tobacco based on PP-PicoDet
摘要
Abstract
[Objective]This study aims to solve the trade-off between speed and accuracy in foreign detection caused by high-speed movement of the vibration transfer groove,and to enhance the detection performance of small and weak contrast targets detection in tobacco filaments.[Methods]An offline background color model for cut tobacco was constructed based on Gaussian Mixture Model(GMM)theory to achieve rapid saliency modeling of abnormal targets.The attention enhancement mechanism of the PP-PicoDet detection network was optimized to strengthen the expression of spatial texture information and improve the perception of the structural features of abnormal targets.A small-sample-friendly loss function was analyzed and constructed,using a Gaussian model to achieve label loss redistribution and enhance the model's incentive for samples with a small proportion.[Results]On a self-built platform,the modified model achieved a 5.8%performance improvement in precision and recall metrics in exchange for a 0.5%additional computational delay.[Conclusion]This technology meets the requirements of timeliness and reliability for foreign object detection tasks in tobacco production,leading to an improvement in cut tobacco production efficiency.关键词
目标检测/PP-PicoDet/GMM建模/显著性检测Key words
object detection/PP-PicoDet/GMM modeling/saliency detection引用本文复制引用
李建军,誉东明,刘会杰,孙熊伟..基于PP-PicoDet的改型轻量级烟丝异物检测算法[J].中国烟草学报,2025,31(5):20-30,11.基金项目
中国科学院安徽工业研究院六安院院企联合基金"LAY202205"(202205) （202205）
