【摘要】 针对当前钢材表面缺陷种类多、形态复杂等原因导致的检测精度低的问题，本文提出了一种基于改进YOLOv5目标检测网络的缺陷检测方法.首先，对于在检测中小目标缺陷易被漏检、错检的问题，增加了小目标检测层；其次，对于缺陷图像表现的背景复杂，且部分缺陷交叉、重叠等问题，引入了Transformer encoder block模块和Convolutional block attention model(CBAM)模块，使网络能更加有效地对抗复杂背景信息，专注于目标对象的检测；最后，使用NEU-DET数据集对该改进模型进行了实验.结果表明，相较于原YOLOv5模型，该方法在缺陷检测方面的精度提升了6.5%;相较于Faster-RCNN模型，其精度提高了约10%.因此，该方法在钢材表面缺陷检测上，具有较好的检测精度和效率.更多还原

【Abstract】 In order to solve the problem of low detection accuracy caused by the variety and complexity of the surface defects of steel, we proposed a defect detection method based on improved YOLOV5 object detection network.Firstly, a small target detection layer is added to the problem that small defects are often to be missed and misdetected.Then, the Transformer encoder block module and the Convolutional block attention model(CBAM) are used to deal with the complex background of the defect image, which makes the network more effective against complex background information and more focus on object detection.Finally, the improved model is tested on NEU-DET datasets.The results show that the precision of the proposed method is 6.5% higher than that of the original YOLOv5 model and about 10% higher than that of the Faster-RCNN model.The experimental results show that the method has good accuracy and efficiency in steel surface defect detection.更多还原