<p>Express package detection is one of the most fundamental and crucial tasks in the smart logistics warehousing system. Realizing fast and accurate detection for varying scale, cluttered, and rotated objects under limited edge computing capabilities make this task challenging. This study proposes a lightweight rotation detection network to achieve fast and accurate express package detection on CPU, namely, EPDNet. Specifically, the improved CSPDarkNet53 is designed through factorized convolution and a low-cost attention module to maximize detection accuracy and speed. A simple and efficient multi-scale spatial feature extraction network is devised to improve the detection performance for multi-scale objects while maintaining efficiency. An efficient head is devised to alleviate spatial misalignment issues by decoupling the multiple object function. The circular smooth label is adopted to achieve rotated object detection. Meanwhile, multiple data augmentation techniques are implemented to improve the generalization of the network. Our detection network achieves 81.0<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="521_2025_11140_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation> mAP@0.5:95 (90.7<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="521_2025_11140_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation> mAP@0.5) at 46.3 FPS on Intel Core i9-10920X CPU for the express package detection task. This work provides an efficient method for detecting express packages in complex scenarios that can be a reference for implementing rotation detection methods in practical industrial applications.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

EPDNet: a fast and accurate express package detection network on CPU

  • Hao Yin,
  • Gun Li,
  • Guiping Zhang,
  • Bin Tang,
  • Zihao Wang,
  • Wenhua He,
  • Lu Chen

摘要

Express package detection is one of the most fundamental and crucial tasks in the smart logistics warehousing system. Realizing fast and accurate detection for varying scale, cluttered, and rotated objects under limited edge computing capabilities make this task challenging. This study proposes a lightweight rotation detection network to achieve fast and accurate express package detection on CPU, namely, EPDNet. Specifically, the improved CSPDarkNet53 is designed through factorized convolution and a low-cost attention module to maximize detection accuracy and speed. A simple and efficient multi-scale spatial feature extraction network is devised to improve the detection performance for multi-scale objects while maintaining efficiency. An efficient head is devised to alleviate spatial misalignment issues by decoupling the multiple object function. The circular smooth label is adopted to achieve rotated object detection. Meanwhile, multiple data augmentation techniques are implemented to improve the generalization of the network. Our detection network achieves 81.0 \(\%\) % mAP@0.5:95 (90.7 \(\%\) % mAP@0.5) at 46.3 FPS on Intel Core i9-10920X CPU for the express package detection task. This work provides an efficient method for detecting express packages in complex scenarios that can be a reference for implementing rotation detection methods in practical industrial applications.