Intelligent Conveyor System for Small-Scale Logistics Using Deep Learning

초록

This paper introduces an intelligent conveyor system to automate critical processes in small logistics centers, which often face difficulties in adopting AI due to cost constraints and limited infrastructure. By leveraging deep learning, the proposed system integrates three main stages to comprehensively acquire cargo information, including classification, volume analysis, damage inspection, and barcode localization. A light detection and ranging sensor and high-speed camera capture real-time data for subsequent processing to extract key attributes such as cargo volume, classification, damage status, and barcode location. Following a comparison of 29 YOLO-based models, we identified YOLOv11n as the most suitable for real-time cargo analysis on resource-constrained devices. Trained on a custom dataset, this integrated model demonstrated high validation performance, achieving an overall mean average precision (mAP) of 99.5% for cargo classification and an average precision (AP) of 99.1% specifically for damage inspection. Scenario tests simulating real-world logistics environments further confirmed the system's efficacy, with accuracies exceeding 95% for volume measurement, classification, and damage inspection. Through its modular design and efficient information exchange, the proposed intelligent conveyor system reduces computational load while supporting step-by-step AI adoption, thus paving the way for scalable, cost-effective automation in small logistics centers.

키워드