Abstract: To improve the angle detection accuracy of blurred workpiece image in the complex industrial production environment for performing the effective deblurring operations on the workpiece images,a deblurring method based on Generative Adversarial Networks was proposed,which minimized the distance between deblurred image and sharp image by zero-sum game training between generative network and discriminative network.To avoid problems such as line error detection or breakage,an improved line segment detector algorithm based on line segment detector was proposed.Comparing with the multi-scale convolutional neural network deblurring method,it could be found that the proposed method had improved the detection accuracy by about 13% through experiments and data analysis.

Key words: machine vision, workpiece image, angle detection, image deblurring, generating adversarial networks, line detection

摘要： 为提升复杂的工业生产环境中模糊工件图像的角度检测精度,对工件图像进行有效的去模糊操作,提出基于生成对抗网络的去模糊方法,该方法通过生成网络与判别网络间的对抗性训练,最小化去模糊图像与清晰图像间的距离。为避免直线错检、断线等问题,基于直线检测算法提出改进的直线检测算法。通过对比实验与数据分析发现,所提方法比多尺度卷积神经网络去模糊方法提升了约13%的检测精度。

关键词: 机器视觉, 工件图像, 角度检测, 图像去模糊, 生成对抗网络, 直线检测