图像数据增强技术原理与发展综述  

Principle and prospect of image data enhancement technology

作　　者：康斓 苏志金 KANG Lan;SU Zhi-jin(PLA Dalian Naval Academy,Dalian 116018,Liaoning Province,China)

机构地区：[1]海军大连舰艇学院,辽宁大连116018

出　　处：《信息技术》2024年第9期176-185,共10页Information Technology

摘　　要：图像数据增强技术在计算机视觉和机器学习领域中扮演着重要的角色。传统的图像数据增强技术包括几何变换、像素级图像变换、图像滤波等方法,但这些方法的效果受到一定的限制。因此,基于深度学习的图像数据增强技术应运而生,涌现出自适应增强、生成对抗式网格(Generative Adversarial Networks, GAN)、弱监督等技术并逐渐成为改进数据集,实现数据的增加和质量的提升,解决深度学习模型过拟合和欠拟合问题的重要手段。文中综述了传统图像数据增强技术和基于深度学习的图像数据增强技术相关原理与应用,并探讨了他们的优缺点以及未来的发展方向。Image data enhancement plays an important role in computer vision and machine learning.The traditional image data enhancement techniques include geometric change,pixel level image change,image filtering and so on,but the effect of these methods is limited.Therefore,image data enhancement technology based on deep learning emerges.The development of adaptive enhancement,GAN,weak supervision and other technologies has gradually become important means to improve the data set,realize the increase of data and the improvement of quality,and solve the overfitting and underfitting problems of deep learning model.The traditional image data enhancement technology and deep learning-based image data enhancement technology are reviewed,and their advantages and disadvantages and future development direction are discussed.

关 键 词：图像数据增强 传统图像处理技术 深度学习 自适应增强 生成对抗式网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]