中文摘要：

针对带钢表面缺陷检测实时性要求高,易受外界环境、缺陷特征弱等视觉检测难题,提出一种基于相位谱和加权马氏距离的带钢表面缺陷显著性检测方法。首先,对缺陷样本图像进行傅里叶变换,归一化幅值信息,获取只保留相位信息的检测模型。并且,证明了模型的有效性,该模型能够保留包含大量相位信息的不均匀缺陷纹理信号。然后,提出了加权马氏距离方法,对显著图像阈值化分割,完成缺陷检测。实验结果表明,该算法检测速度快,单幅图像平均检测耗时仅15.1 ms,能够满足带钢在线实时检测要求。在同一缺陷数据库完成对比实验,对不同带钢缺陷类型,平均检测率达到了94.7%,且漏检率和误检率较低,验证了算法的有效性。

英文摘要：

The defect images in stripe steel are vulnerable to lighting conditions, weak defect characteristics and other factors, which bring many difficult problems to the detection. Thus, a saliency-based defect detection in strip steel by using phase spectrum and weighted Mahalanobis distance was proposed, which can realize the real-time strip surface defect detection. Firstly, the detection model with phase-only Fourier transform by the amplitude information normalized for defect sample image was obtained. Furthermore, the validity of the model was proved by employing some mathematical theory, which help to keep the non-uniformity texture signal from the defect with a large amount of phase information. Then, the weighted Mahalanobis distance method was proposed to significantly enhance the image thresholding, and the defects detection was finished. Some experimental results show that the proposed algorithm is effective. It takes only 15. 1 ms for single image detection, which meets the requirements of the real-time detection. Some comparative experiments in the same defect database illustrate that the average detection rate can reach 94. 7% for different types of defects. In the meantime, missing rate and false positive rate are still low, which validates the effectiveness of the algorithm.