【摘要】 为满足冷轧带钢表面缺陷快速、准确的在线检测需求，开发了一种高效的新四邻域差分缺陷分割算法。该算法采用图像区块的平均灰度作为局部特征，通过积分图像实现快速计算；同时引入高低两个阈值划分灰度差分值，提高了算法的抗干扰能力。实验结果表明，与传统的SIFT、SURF等分割方法相比，该算法在保持高准确率的同时，显著降低了计算耗时。在多种典型冷轧带钢表面缺陷样本的测试中，该算法展现出优异的分割性能，平均准确率达到95%以上，且处理128×128像素图像的平均时间不超过50 ms。证明该算法能够有效平衡检测精度和速度，为冷轧带钢表面缺陷的在线检测提供了一种实用的解决方案，有望在实际生产中发挥重要作用。更多还原

【Abstract】 In order to meet the requirements of rapid and accurate online detection of surface defects on cold-rolled steel strips, an efficient new four neighborhood difference defect segmentation algorithm has been developed. This algorithm uses the average grayscale of image blocks as local features and achieves fast calculation by integrating the image. It simultaneously introduces high and low threshold values to divide the grayscale difference values to improve the algorithm’s anti-interference ability. The experimental results show that compared with traditional segmentation methods such as SIFT and SURF, this algorithm significantly reduces computation time while maintaining high accuracy. In the testing of various typical cold-rolled strip surface defect samples, the algorithm demonstrated excellent segmentation performance, with an average accuracy of over 95% and an average processing time of no more than 50 ms for 128 × 128 pixel images. It proves that the algorithm can effectively balance detection accuracy and speed, providing a practical solution for online detection of surface defects in cold-rolled steel strips. It is expected to play an important role in practical production.更多还原