关键词: YOLOv5　, 神经网络　, 目标检测　, 货物卸载

Abstract: To realize automation and intelligence in mineral handling operations:a ton bag recognition and grasping sys tem based on an improved YOLOv5 algorithm has been developed.The system deploys high-definition cameras in the loading/ unloading area to detect mineral ton bags and their handle positions in real time,and drives a robotic arm to perform autono mous and precise grasping and unloading operations,thereby effectively improving handling efficiency and operational safety. Firstly,to enhance the model′s adaptability in real industrial scenes,a custom ton bag image dataset was constructed,covering various lighting conditions and shooting angles,with a particular focus on challenging scenarios such as dim lighting and ob lique viewing angles.Secondly,the GhostNet module is introduced into YOLOv5 to optimize the network architecture and im prove computational efficiency.Test results show that compared with the original YOLOv5 model,the improved YOLOv5 model exhibits advantages in recognition performance;especially after data augmentation,the median and variance of its prediction time are reduced by 2.86% and 19.54%,respectively,demonstrating superior inference speed and stability.Furthermore,the experimental results indicate that the designed mineral ton bag recognition and grasping system based on the improved YOLOv5 can achieve automatic recognition,localization,and grasping of ton bag targets,verifying the feasibility and effectiveness of the system in automatic truck unloading of mineral ton bags.