基于标记分水岭分割算法在矿石异物检测应用研究

Ore Impurities Detection Based on Marker-Watershed Segmentation Algorithm

摘要: 提出一种改进的分水岭分割算法和目标几何特征提取相结合的目标检测方法，应用于传送带矿石中的异物检测。矿石图像噪声及边缘连接严重，分水岭算法产生过分割及欠分割问题。本文通过优化矿石种子区域，改善分割缺陷，并提取分割目标的几何特征，通过特征统计分析，实施矿石异物检测。实验结果表明，该算法可准确获得矿石边界，并能有效的对图像中异物进行标记。

Abstract: This article describes an improved target detection method which combines the watershed seg-mentation algorithm and targeting geometric feature extraction. It is used in the area of detecting impurities, ore image noise and edge connections in conveyor belts. The watershed algorithm causes over-segmentation and under-segmentation in the detection work as the noise area gravely connected to the edge of ore objects. This article concentrates on optimizing the ore seed area, improving segmentation defects to extract the geometric characteristics of the segmentation target and implementation of the ore impurities detection by feature statistical analysis. As shown in the experimental results, such algorithm can obtain the ore boundary accurately, as well as mark the impurities from images effectively.