基于深度学习的皮带机运行状态多任务集成视觉检测方法研究

Research on a Multi-task Integrated Visual Detection Method for Belt Conveyor Operating Status Based on Deep Learning

摘要:

在煤矿、港口这类工业场景中，皮带输送机能否稳定运行至关重要。当前，基于机器视觉的检测技术虽已在输送带跑偏、撕裂、异物识别等单一任务上应用较多，但难以实现同时检测多种状态，并且多网络并行部署会导致系统复杂、硬件成本高昂。由此，提出一种新的基于深度学习的集成检测方法。该方法首次将语义分割网络接到目标检测网络YOlOv5上，构建了一个统一的双头网络架构，能同时处理输送带模糊负载测量、跑偏状态检测、大异物识别及皮带损伤诊断等多种任务。通过自主构建的多状态数据集，对网络进行训练与优化。实验结果表明，所提方法在保持高精度的同时，还能满足实时性要求，平均检测精度(mAP@0.5)超过97%，负载分割精度达99%，检测速度最高可达91 FPS。该研究为输送机系统智能化运维提供了一种高效、集成的解决方案，在技术和工程上都具备显著的应用潜力。

Abstract:

In industrial environments such as coal mines and ports, ensuring the stable operation of belt conveyors is critically important. In recent years, the integration of the Internet of Things (IoT) and artificial intelligence has enabled real-time and intelligent monitoring of conveyor operational status, which has become a key approach to enhancing both safety and efficiency. Although machine vision-based methods are widely used for single tasks—such as detecting belt misalignment, tears, or foreign objects—they struggle to perform integrated multi-status detection. Moreover, deploying multiple networks in parallel increases system complexity and hardware costs. To address these challenges, this paper proposes a novel integrated detection method based on deep learning. For the first time, a semantic segmentation network is combined with the object detection network YOLOv5, forming a unified double-head architecture. This design enables simultaneous and integrated processing of multiple tasks, including fuzzy load measurement, misalignment detection, large foreign object identification, and belt damage diagnosis. Experimental results demonstrate that the proposed method maintains high accuracy while meeting real-time requirements. The average detection precision (mAP@0.5) exceeds 97%, the load segmentation accuracy (MIoU) reaches 99%, and the inference speed achieves up to 91 FPS. This research provides an efficient and integrated solution for the intelligent operation and maintenance of conveyor systems, offering significant technical advancement and strong potential for engineering applications.