漆膜缺陷检测训练结果数据

本数据集面向轨道交通领域漆膜缺陷自动检测研究与机器视觉算法验证需求构建，旨在为高铁车体漆膜缺陷检测算法训练、性能评估及模型优化提供标准化、可复用的数据支撑。数据依托国家重点研发计划相关课题，由武汉理工大学汽车工程学院硕博研究生依据国家标准 GB/T 40659-2021《智能制造 机器视觉在线检测系统 通用要求》与GB/T 42755-2023《人工智能 面向机器学习的数据标注规程》在武汉理工大学汽车工程学院完成，数据采集与整理工作于2023年6月18日至2025年7月1日期间实施，总规模为738 MB。数据集基于高铁车体漆膜缺陷原始图像构建，初始采集1348张原始图像，随后通过旋转、平移、错切、裁剪、引入噪声及Cutout等方式对图像进行增强，扩充至5704张图像。增强后的数据按照训练集和测试集进行划分，其中训练集5181张、测试集523张，并使用LabelImg软件对所有图像逐一标注缺陷位置。数据集同时记录了训练过程的参数及结果，包括训练批次、精确率、召回率、mAP@0.5及mAP@0.5:0.95等关键指标，完整呈现每轮训练性能变化情况。模型训练在搭载Intel i7-12700K处理器、NVIDIA RTX 3080Ti（12 GB）显卡及64 GB内存的工作站上进行，使用Python语言和PyTorch深度学习框架构建YOLO网络，并在PyCharm 2020.1集成开发环境下完成训练。训练初始学习率设为0.001，权重衰减系数0.0005，输入图像分辨率为 640×640，批大小为16，总训练周期为300轮，并记录每轮训练数据，最终生成训练效果最佳的权重文件。数据集内容包括原始图像、增强后的训练集和测试集图像、对应的标注标签、训练结果数据表以及训练后的权重文件，系统呈现从数据采集、标注、增强到模型训练和权重生成的完整流程，为漆膜缺陷检测模型开发、算法性能验证及工业应用提供可靠的数据基础。同时，配套的训练结果数据和权重文件可直接用于模型复现和算法优化研究。

This dataset is constructed for the needs of automatic detection research of paint film defects in rail transit and machine vision algorithm verification, aiming to provide standardized and reusable data support for the training, performance evaluation and model optimization of paint film defect detection algorithms for high-speed train car bodies. The data was developed under relevant projects of the National Key R&D Program of China, and completed by master and doctoral students from the School of Automotive Engineering, Wuhan University of Technology in accordance with the national standards GB/T 40659-2021 *General Requirements for Machine Vision Online Inspection Systems in Intelligent Manufacturing* and GB/T 42755-2023 *Artificial Intelligence Data Labeling Procedures for Machine Learning* at the School of Automotive Engineering, Wuhan University of Technology. The data collection and organization work was carried out from June 18, 2023 to July 1, 2025, with a total size of 738 MB. This dataset is built based on original images of paint film defects on high-speed train car bodies. Initially, 1348 original images were collected, then augmented via methods such as rotation, translation, shearing, cropping, noise injection and Cutout, and expanded to 5704 images. The augmented data is divided into training set and test set, with 5181 images in the training set and 523 images in the test set. All images were manually annotated for defect locations using LabelImg software. The dataset also records the parameters and results of the training process, including key indicators such as training batches, precision, recall, mAP@0.5 and mAP@0.5:0.95, fully presenting the performance changes of each training epoch. The model training was conducted on a workstation equipped with an Intel i7-12700K processor, an NVIDIA RTX 3080Ti (12 GB) graphics card and 64 GB of memory. The YOLO network was built using Python language and PyTorch deep learning framework, and the training was completed in the PyCharm 2020.1 integrated development environment. The initial learning rate of training was set to 0.001, the weight decay coefficient was 0.0005, the input image resolution was 640×640, the batch size was 16, and the total training epochs were 300. The training data of each epoch was recorded, and finally the weight file with the best training effect was generated. The dataset includes original images, augmented training set and test set images, corresponding annotation labels, training result data tables and trained weight files. It systematically presents the complete workflow from data collection, annotation, augmentation to model training and weight generation, providing a reliable data foundation for paint film defect detection model development, algorithm performance verification and industrial applications. In addition, the supporting training result data and weight files can be directly used for model reproduction and algorithm optimization research.

搜集汇总

背景与挑战

背景概述

本数据集面向高铁车体漆膜缺陷检测算法训练与验证需求构建，基于原始图像通过增强处理生成5704张图像，并划分为训练集和测试集。数据集包含图像、标注标签、训练结果数据及权重文件，为深度学习模型开发与性能优化提供标准化数据支撑。