基于YOLOv5的输送带纵向撕裂检测算法

Longitudinal Tear Detection Algorithm for Conveyor Belt Based on YOLOv5

带式输送机常会发生传送带纵向撕裂的情况,影响工人的安全及生产效率。针对此问题,本文提出一种基于YOLOv5改进的检测方法,使用轻量化网络MobileNetv3代替CSPDarknet作为主干网络进行特征提取,集成轻量级的注意力模型SE,加强了模型对目标权重的学习,提升了主干网络对特征的学习和处理能力,此外,还采用深度可分离卷积模块,有效减少了网络结构的参数量。本研究还采用更高效的边界框回归损失函数SIOU,以解决真实框与预测框之间不匹配的问题,进一步提高了模型训练的速度和推理的准确性。最后使用数据增强的方式优化了数据集的不足和缺陷。实验结果表明,相较于原始模型,改进后的模型在降低参数量的同时还提高了检测的速度和准确率。

Belt conveyors often experience longitudinal tearing of the conveyor belt,which affects the safety and production efficiency of workers.This article proposes an improved detection method based on YOLOv5 to address the issue of longitudinal tearing of conveyor belts in belt conveyors.The lightweight network MobileNetv3 is used instead of CSPDarknet as the backbone network for feature extraction,integrating a lightweight attention model SE to enhance the model's learning of target weights and enhance the backbone network's ability to learn and process features.In addition,a deep separable convolutional module is also used,Effectively reducing the number of parameters in the network structure.This study also addresses the issue of mismatch between the real box and the predicted box,replacing the original bounding box regression loss function GIOU with a more efficient SIOU,further improving the speed of model training and the accuracy of inference.Finally,the shortcomings and deficiencies of the dataset were optimized using data augmentation.The experimental results show that compared to the original model,the improved model not only reduces the number of parameters but also improves the detection speed and accuracy.