EUVIMEDEUVIMED entdecken
Crossref

A lightweight physical attention YOLO based inventory counting model for warehouse

Jun Chen, Jiansha Lu, Yiping Shao, Zhilong Xu, Zhi Liu, Jun Yuan, Jianping Ma

Engineering Research Express · 01.06.2026 · 8 (11) · 115402

DOI: 10.1088/2631-8695/ae72ff ↗

Kurz gefasst

Fachartikel in der Zeitschrift Engineering Research Express, online erschienen am 03.06.2026. Verfasst von Jun Chen und 6 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Gefördert von: Zhejiang Science and Technology Plan Project, National Natural Science Foundation of China, Zhejiang Provincial Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate and reliable inventory counting is fundamental to maintaining continuous production and improving supply chain efficiency. Conventional approaches generally require substantial manual labor and considerable time expenditure, while the growing availability of smart glasses offers a practical alternative by enabling hands-free image capture on the shop floor. Building on this opportunity, a lightweight physical attention YOLO (LPA-YOLO) based on YOLOv8 is proposed for warehouse inventory counting. To address the coupled requirements of boundary discrimination, lightweight computation, multi-scale perception, and count-level reliability in warehouse inventory counting, LPA-YOLO is designed as a task-oriented lightweight detection framework based on YOLOv8. A C2f-Ghost module is used to reduce redundant computation, while a physical intrinsic edge module is introduced at the input stage to fuse explicit Sobel-based edge priors with learnable semantic texture features. Coordinate attention, convolutional block attention module, and dynamic head are further integrated into different stages of the network to enhance feature representation and improve object perception for warehouse objects with varied appearances and spatial arrangements. A novel evaluation metric termed count deviation rate (CDR) is further proposed to directly quantify the discrepancy between predicted and ground-truth object counts, and mean absolute error and exact count accuracy are also adopted to provide complementary image-level counting evaluation. These metrics provide a measurement-level assessment of counting accuracy that conventional detection metrics cannot offer. Experiments show that LPA-YOLO outperforms YOLOv8 in detection and counting reliability. It achieves an mAP@0.5 of 0.87 and a CDR of 0.13, corresponding to an increase of 0.05 in mAP@0.5 and a reduction of 0.05 in CDR. In addition, LPA-YOLO reaches 143.3 FPS with only 2.23 million parameters, representing a 25.9% parameter reduction relative to YOLOv8. These results indicate that LPA-YOLO provides a favorable balance among detection accuracy, counting reliability, and computational efficiency for warehouse inventory counting.

Förderung

  • Zhejiang Science and Technology Plan Project
  • National Natural Science Foundation of China
  • Zhejiang Provincial Natural Science Foundation of China

Zitieren

APA

Chen, J., Lu, J., Shao, Y., Xu, Z., Liu, Z., Yuan, J., & Ma, J. (2026). A lightweight physical attention YOLO based inventory counting model for warehouse. Engineering Research Express, 8(11), 115402. https://doi.org/10.1088/2631-8695/ae72ff

Vancouver

Chen J, Lu J, Shao Y, Xu Z, Liu Z, Yuan J, et al. A lightweight physical attention YOLO based inventory counting model for warehouse. Engineering Research Express. 2026;8(11):115402. doi:10.1088/2631-8695/ae72ff

BibTeX
@article{Chen2026,
  author = {Chen, Jun and Lu, Jiansha and Shao, Yiping and Xu, Zhilong and Liu, Zhi and Yuan, Jun and Ma, Jianping},
  title = {A lightweight physical attention YOLO based inventory counting model for warehouse},
  journal = {Engineering Research Express},
  year = {2026},
  volume = {8},
  number = {11},
  pages = {115402},
  doi = {10.1088/2631-8695/ae72ff}
}
Kostenlos starten →