Image data set for AI-assisted reliability assessment for gravure offset printing system

In total, 299 images of printed lines were collected using the in-house roll-based gravure offset printing system. Then they were labeled for overall printing quality classification and local printing defect detection tasks. For overall printing quality classification, these images were divided into two classes: 225 with satisfactory quality and 74 with defects. For local printing defect detection, all 74 images with distinctive local defects were chosen and labeled following the YOLO object detection format. During the training process, these images were split for training and validation as follows: 75/25 % for the overall printing quality classification model and 80/20 % for the local printing defect detection model respectively.本研究采用自研卷筒式凹版胶印系统，共采集299幅印刷线条图像。随后针对整体印刷质量分级与局部印刷缺陷检测两项任务分别完成标注。在整体印刷质量分级任务中，全部299幅图像被划分为两类：225幅质量合格样本，74幅存在印刷缺陷的样本。针对局部印刷缺陷检测任务，研究团队选取全部74幅带有典型局部缺陷的图像，按照YOLO目标检测格式进行标注。训练阶段中，两类任务的数据集划分比例如下：整体印刷质量分级模型采用75%训练集、25%验证集；局部印刷缺陷检测模型则分别采用80%训练集、20%验证集。