【摘要】 铸坯表面缺陷直接影响后续钢材质量和性能，对高温连铸坯进行表面在线检测对于控制铸坯表面质量和提高钢材质量非常重要。本文介绍了开发的连铸坯表面缺陷检测系统，采用光学成像和图像识别方法，可以在线检测高温异形连铸坯表面缺陷。系统采用短波长的蓝色激光照明技术，并通过精密的窄带滤波方法，采集高温异形坯表面高清图像。由于异形坯端面复杂、规格多样，系统采用两台高分辨率线阵CCD摄像机分别采集异形坯左右部分，并开发了一种适用于异形坯表面的图像拼接方法，对不同相机拍摄的图像进行拼接，形成异形坯表面完整图像。开发了基于YOLOv5的目标检测算法，融入注意力机制，增强模型的鲁棒性，提高了异形连铸坯表面缺陷检测的准确率。检测模型对于常见的裂纹缺陷的mAP0.5指标达到95.8%,对于其他不常见缺陷的mAP0.5指标均达到80%以上。更多还原

【Abstract】 The surface defects of the billets directly affect the quality and performance of the steels.On-line surface detection of high-temperature continuous casting billets is very important to surface quality control of billets and steels.In this paper, an developed on-line surface defect detection method of high-temperature special-shaped continuous casting billets is introduced.Optical imaging and image recognition methods are used to detect surface defects of high-temperature special-shaped continuous casting billets online.The method utilizes short-wavelength blue laser illumination combined with precision narrow-band filtering to capture high-definition images of the high-temperature special-shaped continuous casting billets.Because of the complex surface and various specifications of the special-shaped billets,two high-resolution linear CCD cameras are used to separately capture the left and right parts of the special-shaped billets. An image stitching method suitable for the surfaces of special-shaped billets was developed to merge images captured by different cameras,forming a complete image of the special-shaped continuous casting billet. Additionally,a YOLOv5-based object detection algorithm was developed,incorporating an attention mechanism to enhance model robustness and improve the accuracy of detecting surface defects in special-shaped billets. The mAP0. 5 value of crack detection is 95. 8%,while the mAP0. 5 values of uncommon defect detection are more than 80%.更多还原