基于视觉显著性的印刷辊筒表面缺陷分类方法研究

Study on Classifying Surface Defects of Printed Roller Based on Visual Saliency

针对印刷辊筒表面缺陷分类精度不高、效率低等问题,本研究提出基于视觉显著性的印刷辊筒表面缺陷分类方法。通过视觉显著性算法与深层信息融合算法抑制背景纹理中的高频分量,采用多组卷积并联结构充分提取图像特征信息,以加强网络的多尺度表达能力并提升分类性能,通过激活函数PReLU保留推理过程中的负值信息,提升网络的非线性表达能力。实验结果表明,该方法可有效区分印刷辊筒表面缺陷,准确率可达98.50%,基本满足印刷工业的生产要求。

For the problems of low accuracy and efficiency of the classification of surface defects of printed rollers,a classifying surface defects of printed roller based on visual saliency was proposed in this study.The visual saliency algorithm and deep information fusion algorithm were used to suppress the high-frequency components in the background texture,and the multi-group convolutional parallel structure was used to fully extract the image feature information in order to enhance the multi-scale expression capability of the network and improve the classification performance,and the activation function PReLU was used to retain the negative information in the inference process to improve the nonlinear expression capability of the network.The results showed that the method can effectively distinguish the surface defects of printed roller,the accuracy can reach 98.50%,which basically meets the production requirements of the printing industry.