De-DDPM:可控、可迁移的缺陷图像生成方法
De-DDPM:A Controllable and Transferable Defect Image Generation Method
岳忠牧 1张喆 1吕武 2赵瑞祥 2马杰1
作者信息
- 1. 华中科技大学人工智能与自动化学院 武汉 430074
- 2. 中国船舶集团有限公司航海科技有限责任公司 北京 100070
- 折叠
摘要
基于深度学习的表面缺陷检测技术是工业上的一项重要应用,而缺陷图像数据集质量对缺陷检测性能有重要影响.为解决实际工业生产过程中缺陷样本获取成本高、缺陷数据量少的痛点,提出了一种基于去噪扩散概率模型(Denoising diffusion probabilistic model,DDPM)的缺陷图像生成方法.该方法在训练过程中加强了模型对缺陷部位和无缺陷背景的差异化学习.在生成过程中通过缺陷控制模块对生成缺陷的类别、形态、显著性等特征进行精准控制,通过背景融合模块,能将缺陷在不同的无缺陷背景上进行迁移,大大降低新背景上缺陷样本的获取难度.实验验证了该模型的缺陷控制和缺陷迁移能力,其生成结果能有效扩充训练数据集,提升下游缺陷检测任务的准确率.
Abstract
Surface defect detection technology based on deep learning is an important application in industry and the quality of defect image dataset has a significant impact on defect detection performance.A defect image genera-tion method based on denoising diffusion probabilistic model(DDPM)is designed to address the pain points of high cost of obtaining defect samples and low amount of defect data in actual industrial production processes.This meth-od enhances the model's differential learning of defect locations and defect free backgrounds during the training process.Through the defect control module during the generation process,this method accurately controls the cat-egory,morphology,saliency and other features of generated defects.Through the background fusion module,de-fects can be migrated on different defect free backgrounds,which greatly reducing the difficulty of obtaining defect samples on new backgrounds.The experiment has verified the defect control and defect migration capabilities of the model,and its generated results can effectively expand the training dataset and improve the accuracy of down-stream defect detection tasks.
关键词
数据增强/数据集扩充/缺陷图像生成/深度学习Key words
Data augmentation/dataset expansion/defect image generation/deep learning引用本文复制引用
基金项目
国家自然科学基金(U1913602)
国家自然科学基金(61991412)
装备预先研究基金(50911020603)
出版年
2024
国家科技期刊平台
NSTL
万方数据