Research Article
An Autonomous Sorting System for Intelligent Warehouse Robots Based on Mask R-CNN
@INPROCEEDINGS{10.4108/eai.18-12-2025.2365267, author={Ying Zhou and Xu Ji and Yifei Guo and Jing Tang and Long jun Wu}, title={An Autonomous Sorting System for Intelligent Warehouse Robots Based on Mask R-CNN}, proceedings={Proceedings of the 13th International Conference on Identification, Information and Knowledge in the Internet of Things, IIKI 2025, 18-21 December 2025, Chengdu, China}, publisher={EAI}, proceedings_a={IIKI}, year={2026}, month={6}, keywords={deep learning intelligent warehouse sorting system Mask R-CNN instance segmentation robotics}, doi={10.4108/eai.18-12-2025.2365267} }- Ying Zhou
Xu Ji
Yifei Guo
Jing Tang
Long jun Wu
Year: 2026
An Autonomous Sorting System for Intelligent Warehouse Robots Based on Mask R-CNN
IIKI
EAI
DOI: 10.4108/eai.18-12-2025.2365267
Abstract
A visual-guidance solution based on Mask R-CNN is proposed for the automated sorting of robots in intelligent warehouses. This approach solves the challenges of identifying and localising multi-category, stacked packages in complex environments. Built a custom dataset with multi-category package labels. Transfer learning was employed to train the Mask R-CNN model, enabling the simultaneous output of the target parcel category, bounding box, and a high-precision pixel-level mask. This mask provides a precise spatial contour and pose for robotic arm grasp planning, effectively preventing misgrabs. The average accuracy (mAP) of the system was 95.7% in the test set and 91.2% in the combination (mIoU). The actual sorting success rate reaches 98.5%, significantly outperforming traditional methods. This validates the solution’s effectiveness and robustness in complex warehouse environments, providing a reliable path to enhance sorting automation.
