基于改进YOLOv8的输送带异物检测研究

Research on Foreign Body Detection in Conveyor Belt Based on Improved YOLOv8

针对基于深度学习的输送带异物检测模型参数量大、检测速度慢等问题,提出一种基于改进YOLOv8的输送带异物检测方法。利用Ghost模块设计主干网络CSPGhostNet,减少模型参数量和计算量;采用Alpha-IoU损失函数作为目标框回归损失函数,提高网络模型的收敛速度和检测精度;利用网络剪枝对检测模型剔除冗余参数。在嵌入式平台NVIDIA Jetson Xavier NX上进行实验,结果表明,相较于原YOLOv8l模型,改进模型mAP为90.5%,帧率提高375%,运算速度达到30帧/s,实现了有限计算资源下输送带异物的实时检测。

Aiming to the problems such as the large number of parameters and slow detection speed in the existing conveyor belt foreign body detection model based on deep learning,a conveyor belt foreign body detection method based on improved YOLOv8 was proposed.To reduce model parameters and calculation amount,a Ghost module was utilized to design the CSPGhostNet backbone network.The Alpha-IOU loss function was employed as the target frame regression loss function to enhance both convergence speed and detection accuracy of the network model.Network pruning were applied to eliminate redundant parameters in the detection model.The experiments were conducted on the embedded platform NVIDIA Jetson Xavier NX.The results show that in comparison to the original YOLOv8l model,the improved model achieves mAP of 90.5%,increases the frame rate by 375%,and attains an operational speed of 30 frames per second.Real-time detection of foreign body on conveyor belt with limited computational resources is successfully achieved.