基于YOLO-CIRCLE算法的圆形钢卷检测

Detection of Round Steel Rolls Based on YOLO-CIRCLE Algorithm

YOLOV3是计算机视觉领域一种优秀的模型,在目标检测领域有着极为广泛的应用,但在圆形目标检测上有较大的检测冗余。工业生产环境中圆形目标极为常见,为高效准确地确定圆形钢卷目标位置及轮廓,本文针对针对圆形目标检测提出YOLO-CIRCLE模型,使用圆形先验框机制进行回归,并针对圆形目标的IOU计算提出CIRCLE-IOU和更适合圆形目标的损失计算函数CIRCLE-LOSS。并采集2000张工业生产环境中的钢卷图片进行训练。实验结果表明:YOLO-CIRCLE模型能够有效检测出圆形钢卷目标,平均精度为99.92%,平均有效面积超过95%,虽然检测速度低于原YOLOV3模型,但平均有效面积提升了20%,平均有效精度提升了0.08%,降低了空间复杂度;相比霍夫圆检测算法,速度提升56%,并且减少了参数调整的过程。本论文提供的方法可以减少大量圆形目标的识别冗余,并具有较高的鲁棒性,有助于工业产品质量检测自动化系统的建立。

As one of the excellent models in the field of computer vision, YOLOv3 has been widely used in the field of object detection. However, it has a large detection redundancy in circular object detection.Circular targets are very common in industrial production environment. In order to efficiently and accurately determine the target position and contour of circular steel rolls, this paper improves YOLO-CIRCLE model for advanced detection of circular targets and makes regression using circular prior box mechanism.It improvesYOLOv3, the advanced target detection model. It proposes CIRCLE-IOU for IOU calculation of round targets and proposes CIRCLE-LOSS, the loss calculation function, which is more suitable for circular targets. After collecting 2000 steel coil images in the industrial production environment for training, the results show that the YOLO-CIRCLE model can effectively detect the circular steel coil target, with the average accuracy of 99.92% and average effective area exceeding 95%.Although the detection speed is lower than the original YOLOv3 model, the average effective area increases by 20%,and precision increases by 0.08%, which reduces the spatial complexity. Compared with the Hove circle detection algorithm, the speed increases by 56%, reducing the process of parameter adjustment.The methods can reduce the identification redundancy of large circular targets and have high robustness, contributing to the establishment of automated systems for quality detection of industrial products.