一种金属带材表面缺陷检测方法

摘要

由于金属板带表面反光强烈，缺陷成因复杂，使得缺陷检测系统性能的进一步提升遇到了瓶颈。为打破瓶颈，提出了一种融合缺陷成因和统计特征的双向协同视觉注意方法。金属表面缺陷的产生与生产设备和工艺流程密切相关，将缺陷成因和统计特征分层量化，通过学习和训练加入先验知识库，在自底向上的视觉注意模型中，分别对初级视觉特征提取，多特征图合并和显著图三个层次施加权值控制，以引导视觉注意过程，从而实现自底向上和自顶向下的双向协同，可有效提升检测效率和检出率。通过实验对比分析，结果表明，该方法在检测性能和效率方面有很大提高。%Because of strong reflection on metal plate surface and complicated defect formations,it is difficult to further improve performance of the detection system.A bidirectional synergy visual attention method,mixing defect formation and statistical characteristics together,is put forward in this paper. Because the generation of metal surface defects is closely related to the production equipment and technological process,the defect formation and statistical characteristics are layered in quantizing and the priori knowledge is joined by learning and training.In order to guide the process of visual attention,in the bottom -up visual attention model,primary visual feature extraction,several characteristic figure merging and salient map are controlled respectively in weight to realize two -way collaboration,and thus testing efficiency and detection rate are promoted.The experiment results show that the method improves detection performance and efficiency obviously.