To achieve manufacturing and logistics informatization management for steelworks, it is of crucial importance to automatically recognize the slab management numbers (SMNs) sprayed on the steel slabs. However, due to the poor quality of spraying and various interferences, SMN detection is a major challenge for subsequent recognition in the steel-slab product line. This paper proposes a corner-clustering method, which can extract the SMN from a changeable background precisely and promptly. In our method, the FAST algorithm is modified to extract the image corners by adaptively adjusting the local threshold of corner detecting with the change of image contrast. Then, the DBSCAN algorithm is implemented to group the corners into several clusters, which includes the SMN regions and interference regions. Finally, a classifier based on HOG features and SVM is applied to discriminate SMN and non-SMN regions. For experimental validation, the proposed method was implemented to a substantial amount of acquired images. A good performance has been achieved as the detection accuracy can reach as high as 98.96% for SMN on the steel slabs.

中文翻译：

---

一种用于检测喷洒在钢板上的板坯管理数的角聚类方法

为了实现钢厂的制造和物流信息化管理，自动识别喷涂在钢板上的板坯管理编号（SMN）至关重要。然而，由于喷涂质量差和各种干扰，SMN检测是板坯生产线后续识别的一大挑战。本文提出了一种角点聚类方法，可以准确、快速地从多变的背景中提取SMN。在我们的方法中，FAST算法被修改为通过随着图像对比度的变化自适应地调整角点检测的局部阈值来提取图像角点。然后，实施 DBSCAN 算法将角点分成几个簇，其中包括 SMN 区域和干扰区域。最后，应用基于 HOG 特征和 SVM 的分类器来区分 SMN 和非 SMN 区域。对于实验验证，所提出的方法已应用于大量获取的图像。取得了良好的性能，对钢板上的SMN检测准确率高达98.96%。