船舶钢板表面焊接质量检测AI模型训练数据

随着工业4.0时代的发展,现代制造业对质量检测技术提出了更高要求。传统手工检测方法效率低下,难以满足大规模制造流水生产线的需求。与此同时,基于深度学习的计算机视觉技术在许多检测任务中均取得优异效果,为制造业数据驱动的智能化转型带来新机会。
在制造业领域,由于生产过程中的注重保密性,公开的检测数据集面临供给不足的困境,这也成为研究和产业应用难点之一。尤其是在船舶等规模化领域,公开相关产品质量检测数据集能够极大促进技术研发与产业升级。
数据集于2023年构建完成,数据源于某船厂实际生产线采集。数据集包含298张高质量的船舶钢板表面照片,井然有序的进行了分类和标注工作。标注采用COCO格式,对图片中的4类检测目标(焊接工件、缺陷、焊接线和工作件)进行了物体框定和分类标注。
数据集特点：
真实场景采集,近乎于工业检测任务需求;
丰富的样本数量,适用于深度学习模型训练;
细致的分类和标注工作,有助于学习特征提取;

With the advancement of the Industry 4.0 era, modern manufacturing has put forward higher requirements for quality inspection technologies. Traditional manual inspection methods suffer from low efficiency and fail to meet the demands of large-scale manufacturing assembly lines. Meanwhile, deep learning-based computer vision technology has achieved excellent results in numerous inspection tasks, bringing new opportunities for the data-driven intelligent transformation of the manufacturing industry.
In the manufacturing sector, due to the emphasis on confidentiality during the production process, publicly available inspection datasets face a shortage of supply, which has become one of the difficulties in research and industrial applications. Especially in large-scale fields such as shipbuilding, releasing public product quality inspection datasets can greatly promote technological research and development and industrial upgrading.
This dataset was completed in 2023, with data collected from the actual production line of a shipyard. It contains 298 high-quality photographs of ship steel plate surfaces, and has undergone systematic classification and annotation. The annotations adopt the COCO format, performing object bounding box annotation and classification labeling for the four types of inspection targets (welded workpieces, defects, weld lines and workpieces) in the images.
Dataset Features:
1. Collected from real-world scenarios, which closely aligns with the requirements of industrial inspection tasks;
2. Abundant sample quantity, suitable for training deep learning models;
3. Detailed classification and annotation work, which facilitates feature extraction learning;