针对目前铝型材表面缺陷检测存在的准确率、检测效率较低等问题,提出了一种基于改进的YOLOv3铝型材表面缺陷检测方法。首先通过k-均值聚类算法对采集到的数据集进行聚类分析,选取尺寸最优的目标候选框;考虑到铝型材表面缺陷较大,对YOLOv3的网络层级结构进行调整,并将目标检测层之前的6个CBL单元改成4个,再补充2个残差单元,以提高特征的复用。将提出方法用于铝型材表面缺陷检测,并与经典的卷积网络Faster-RCNN和SSD方法进行比较,实验结果表明,采用提出的算法准确率达到97%,检测速度达到47帧/s,明显优于经典的卷积网络Faster-RCNN和SSD,适于在有高精度快速性要求的铝型材表面缺陷检测中推广应用。
Abstract
Aiming at the problems of low accuracy and low efficiency in the detection of surface defects of aluminum profile, an improved method based on YOLOv3 is proposed. Firstly, the k-means clustering algorithm is used to cluster the collected data sets, and the target candidate box with the optimal size is selected. Then, considering the large surface defects of aluminum profile, the network hierarchical structure of YOLOv3 is adjusted. Six CBL units before the target detection layer are changed into four CBL units, and two residual units are added to improve the reuse of features. Compared with the classical convolution network Faster-RCNN and SSD, a large number of experimental results show that the accuracy of the proposed algorithm can reach 97%, and the detection speed can reach 47 frame/s. The proposed method is obviously better than Faster-RCNN and SSD, which is suitable for the aluminum profile surface defect detection with high accuracy and rapidity.
关键词
计量学 /
表面缺陷检测 /
铝型材 /
深度学习 /
YOLOv3方法 /
k-均值聚类
Key words
metrology /
surface defect detection /
aluminum profile /
deep learning /
YOLOv3 method /
k-means clustering
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
［1］邵伟, 彭鹏, 周阿维, 等. 工件表面低对比度缺陷快速准确识别方法［J］. 计量学报, 2019, 40(5): 793-797.
Shao W, Peng P, Zhou A W, et al. A fast and accurate identification method for low contrast defects on workpiece surface［J］. Acta Metrologica Sinica, 2019, 40(5): 793-797.
［2］王宪保, 李洁, 姚明海, 等. 基于深度学习的太阳能电池片表面缺陷检测方法［J］. 模式识别与人工智能, 2014, 27(6): 517-523.
Wang X B, Li J, Yao M H, et al. Surface defect detection method of solar cell based on deep learning［J］. Pattern Recognition and Artificial Intelligence 2014, 27(6): 517-523.
［3］郭保苏, 吴文文, 付强. 基于支持向量机分类策略的多晶硅电池片色差检测［J］. 计量学报, 2019, 40(6): 1013-1019.
Guo B S, Wu W W, Fu Q. Chromatic Aberration Detection of Polysilicon Cells Based on Support Vector Machine Classification Strategy［J］. Acta Metrologica Sinica, 2019, 40(6): 1013-1019.
［4］徐义鎏, 贺鹏, 任东, 等. 基于改进faster RCNN的木材运输车辆检测［J］. 计算机应用, 2020, 40(S1): 209-214.
Xu Y L, He P, Ren D, et al. Detection of timber transport vehicle based on improved Faster RCNN［J］. Computer Application, 2020, 40(S1): 209-214.
［5］李兰, 奚舒舒, 张才宝, 等. 基于改进SSD模型的工件表面缺陷识别算法［J］. 计算机工程与科学, 2020, 42(9): 1608-1615.
Li L, Xi S S, Zhang C B, et al. Surface defect recognition algorithm based on improved SSD model［J］. Computer Engineering and Science, 2020, 42(9): 1608-1615.
［6］代小红, 陈华江, 朱超平. 一种基于改进Faster RCNN的金属材料工件表面缺陷检测与实现研究［J］. 表面技术, 2020, 49(10): 362-371.
Dai X H, Cheng H J, Zhu C P. A study on Surface Defect Detection and Realization of Metal Workpiece Based on Improved Faster RCNN［J］. Surface technology, 2020, 49(10): 362-371.
［7］Wei R B, Song Y H, Zhang Y L. Enhanced Faster Region Convolutional Neural Networks for Steel Surface Defect Detection［J］. ISIJ International, 2020, 60(3): 539-545.
［8］沈远. 基于深度学习的改进热镀锌板表面缺陷检测算法［D］. 杭州: 浙江大学, 2020.
［9］蔡彪, 沈宽, 付金磊, 等. 基于Mask R-CNN的铸件X射线DR图像缺陷检测研究［J］. 仪器仪表学报, 2020, 41(3): 61-69.
Cai B, Shen K, Fu J L, et al. Defect Detection of Casting X-ray DR image based on Mask R-CNN［J］. Journal of Instrumentation, 2020, 41(3): 61-69.
［10］胡珊珊, 肖勇, 王保帅, 等. 基于深度学习的PCB缺陷检测研究［J］. 电测与仪表, 2021, 58(3): 139-145.
H S S, Xiao Y, Wang B S, et al. Research on PCB Defect Detection Based on Deep Learning［J］. Electrical measurement and Instrumentation, 2021, 58(3): 139-145.
［11］Redmon J, Farhadi A. Yolov3: An incremental improvement［EB］. https://ui.adsabs.harvard.edu/abs/2018arXiv180402767R, 2018.
［12］Lin C. Detection Method of Offshore Droppers Based on UAV and YOLOv3［J］. World Scientific Research Journal, 2019, 5(9): 288-295.
［13］Luo Z, Yu H L, Zhang Y Z. Pine Cone Detection Using Boundary Equilibrium Generative Adversarial Networks and Improved YOLOv3 Model［J］. Sensors, 2020, 20(16): 4430: 1-18
［14］谭芳, 穆平安, 马忠雪. 基于YOLOv3检测和特征点匹配的多目标跟踪算法［J］. 计量学报, 2021, 42(2): 157-162.
Tan F, Mu P A, Ma Z X. Multi-target Tracking Algorithm Based on YOLOv3 Detection and Feature Point Matching［J］. Acta Metrologica Sinica, 2021, 42(2): 157-162.
［15］Zhao L Q, Li S Y. Object Detection Algorithm Based on Improved YOLOv3［J］. Electronics, 2020, 9(3): 537: 1-11.
［16］Yang F, Yang D M, He Z M, et al. Automobile Fine-Grained Detection Algorithm Based on Multi-Improved YOLOv3 in Smart Streetlights［J］. Algorithms, 2020, 13(5): 114: 1-22.
基金
国家自然科学基金(51675259);江苏省智能感知技术与装备工程研究中心开放基金(ITS202103);南京工程学院研究生创新(TB20211604)
PDF(67379 KB)
