“…Qu et al [16] proposed an improved SSD network and a pyramid convolution-based cigarette appearance defect detection model with an mAP of 90.26%. Zhao [17] achieved an mAP of 97.64% by improving NMS on YOLOv5 with CIoU loss instead of IoU loss. Lu et al focused on the defect detection of the pulling thread on the transparent plastic outer packaging of tobacco rods.…”