湖北汽车工业学院学报2023，Vol.37Issue(4)：48-53,6.DOI:10.3969/j.issn.1008-5483.2023.04.010
基于改进StyleGAN路面缺陷数据增强算法
Improved StyleGAN Algorithm for Road Defect Data Augmentation
摘要
Abstract
In view of the few samples of road defect data and poor image quality in complex scenes,an improved StyleGAN algorithm for road defect data augmentation was proposed.Based on the original StyleGAN,a self-attention mechanism was introduced to enhance the generator's attention to the image texture detail information;an SLE tag encoder was introduced to regulate the texture detail of the gener-ated images;the noise input was increased to enhance the complexity of the training samples and the di-versity of the generated samples.The WGAN-GP loss function was adopted,and the module resolution was adjusted to improve the convergence efficiency of the model.The quality of the generated images by the model was evaluated through ablation experiments,intuitive evaluation methods,and quantitative evaluation methods.The experimental results show that the algorithm generates road defect images with better quality,achieving an IS of 52.1 and an FID of 54.2.After the model is tested with four classical target detection algorithms,an improvement of approximately 30%in the mean average precision com-pared with the original dataset is observed,and the recall rate is enhanced by about 7%.关键词
StyleGAN/数据增强/自注意力机制/路面缺陷检测Key words
StyleGAN/data augmentation/self-attention mechanism/road defect detection分类
信息技术与安全科学引用本文复制引用
刘欢,孙海明,朱焕馨..基于改进StyleGAN路面缺陷数据增强算法[J].湖北汽车工业学院学报,2023,37(4):48-53,6.基金项目
湖北省揭榜制科技项目(2021BEC005) （2021BEC005）
