期刊文献+

基于YOLO的铝型材料表面小缺陷检测 被引量:11

Detection of small defects on aluminum profile surface based on YOLO
下载PDF
导出
摘要 铝型材料是世界上使用量较多的合金材料之一,精确检测铝型材料表面缺陷的类型和位置对提高铝型材料的利用率有着重要作用。现有物体表面缺陷检测的方法在准确度、灵活性以及检测尺度上存在不足,为弥补这些不足,提出了一种基于YOLO-v5的多尺度物体表面缺陷检测模型。该模型在保证检测性能的情况下扩张了检测尺度,改进了锚框的计算,攻克了作为表面缺陷检测主要难点之一的小缺陷检测。在训练过程中引入了迁移训练的方法进行多次迁移训练,提升了短期训练的模型检测精度。实验结果表明:在铝型材料表面缺陷的小缺陷检测任务中,该模型的平均正确率(AP)较改进前有了明显的提升,其余缺陷检测任务的AP也不低于改进前,准确率均值(mAP)也有了明显的提升,并且检测速度(FPS)没有明显下降。 Aluminum profile material is one of the most widely used alloy materials in the world.Accurate detection of the type and location of surface defects of aluminum profile material plays an important role in improving the utilization efficiency of aluminum profile material.Existing methods for surface defect detection of objects have shortcomings in accuracy,flexibility and detection scale.In order to make up for these deficiencies,a multi-scale object surface defect detection model based on YOLO-v5 is proposed.The model expands the detection scale and improves the calculation of anchor frame while ensuring the detection performance.It overcomes the detection of small defects,which is one of the main difficulties in surface defect detection.In the training process,the migration training method is introduced to carry out multiple migration training,which improves the model detection accuracy of short-term training.The experimental results show that in the small defect detection task of aluminum surface defects,the average precision(AP)value of the model is significantly improved than that before the improvement,the AP values of other defect detection tasks are not lower than those before the improvement.The mean average precision(mAP)has also been significantly improved,and the frame per second(FPS)has not decreased significantly.
作者 沈希忠 吴迪 SHEN Xizhong;WU Di(School of Electrical and Electronic Engineering,Shanghai Institute of Technology,Shanghai 201418,China)
出处 《浙江工业大学学报》 CAS 北大核心 2022年第4期372-380,共9页 Journal of Zhejiang University of Technology
关键词 铝型材料 小缺陷 YOLO-v5 尺度扩张 迁移训练 aluminum profile material small defect YOLO-v5 scale expansion migration training
  • 相关文献

参考文献5

二级参考文献24

共引文献40

同被引文献136

引证文献11

二级引证文献18

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部