目的 针对包装产品上QR码在采集过程中的运动模糊、失焦模糊，长期磨损形成的自模糊和环境中的噪声等因素，导致QR码无法识别的问题，提出一种基于生成对抗网络的QR码去模糊算法。方法 采用深度学习模型生成对抗网络对模糊核和环境噪声具有的强大拟合和估计能力，提取模糊QR码图像与真实图像的深层特征和差距，并通过生成器与判别器不断迭代对抗，使生成器具有由输入的模糊QR码产生与之对应的去模糊QR码图像的能力。结果 生成器能较好地对模糊核和环境噪声进行估计，而且能够实现对数据集内多种不同模糊程度QR码的去模糊，去模糊QR码图像效果较好，处理时间快，识别率较高。结论 采用基于生成对抗网络的QR码去模糊算法能够广泛应用于包装产品外壳上QR码的预处理过程，泛化能力较好，能有效提高扫描识别率。

Aiming that QR codes on shell packaging products in the process of collecting, motion blur, out-of-focus blur caused by long exposure and wear the fuzzy and environmental factors such as noise, QR code can't identify the problem, the paper aims to put forward a kind of QR codes to fuzzy algorithm based on generative adversarial network. Adversarial network generated by deep learning model had strong fitting and estimation ability on fuzzy core and environmental noise. Deep characteristics and the gap between fuzzy QR code images and real images were extracted. Through constant iterative against generator and discriminator, the generator had the ability of deblurring QR codes of different fuzzy degrees in the dataset. The generator could estimate the fuzzy core and environmental noise well and deblur multiple kinds of QR codes. It was featured with good effect, fast treatment, and high recognition rate on deblurring QR code images. The QR code deblur algorithm based on generative adversarial network can be widely used in the preprocessing of QR codes on the packaging product case and improve the scanning recognition rate.