Fast recognition method of surface defects on micro-steel balls based on improved AlexNet-SVM convolutional neural network
-
摘要: 微型钢球由于高反射、球体需要全覆盖的特点,其表面缺陷的质量控制尤为困难。针对人工检测方法效率低且准确度不足的问题,文章提出一种改进的AlexNet的卷积神经网络和SVM模型的钢球表面缺陷快速识别方法。该模型删减了后3个卷积层,保留全连接层FC7提取的特征,采用SVM代替原始Softmax分类器以防止过拟合,提高模型泛化能力。此外,研究了基于K-CV的改进网络搜索算法确定分类器最佳参数。实验采用混淆矩阵对提出模型的识别结果进行性能评估,结果表明,该方法平均准确率达到99.43%,运算时间为17.2 ms。对比原模型及其他网络模型,具有较高的准确度和推理速度,能够满足工业现场检测的需求。Abstract: The quality control of surface defects on micro-steel balls is particularly difficult because of high reflectivity and the need for full coverage on the spherical surface. In response to the problems of low efficiency and inadequate accuracy of manual inspection methods, an improved method is proposed for fast identification of surface defects on steel balls with a combination of an improved AlexNet convolutional neural network and an SVM model. The last three convolutional layers are removed and the features extracted by the fully connected layer FC7 are reserved in our model. The original Softmax classifier is replaced by SVM to prevent overfitting and improve the model's generalization ability. In addition, an improved network search algorithm based on K-CV is used to determine optimal parameters for the classifier. Experimental evaluation of the proposed model's recognition results is performed with a confusion matrix. The results show that this method achieves an average accuracy rate of 99.43% with an operating time of 17.2 ms. Compared to the original model and other network models, it has higher accuracy and inference speed, meeting the requirements of industrial field inspection.
-
Key words:
- micro-steel balls /
- CNN /
- AlexNet /
- SVM /
- surface defect inspection
-
表 1 经典AlexNet模型的混淆矩阵
目标类型 输出类型 样品 孔伤 凹坑 开裂 锈蚀 磨损 孔伤 171 0 5 0 0 凹坑 0 173 3 0 0 开裂 0 3 173 0 0 锈蚀 0 0 0 170 6 磨损 0 0 2 4 170 表 2 本文AlexNet-SVM模型实验混淆矩阵
目标类型
输出类型样品 孔伤 凹坑 开裂 锈蚀 磨损 孔伤 176 0 0 0 0 凹坑 0 176 0 0 0 开裂 0 1 175 0 0 锈蚀 0 0 0 174 2 磨损 0 0 0 2 174 表 3 提出AlexNet-SVM模型性能对比
指标 精度/(%) 召回率/(%) F1分数 提出模型 改进前 提出模型 改进前 提出模型 改进前 孔伤 100 97.16 100 97.16 1 0.971 6 凹坑 100 96.64 99.43 98.30 0.997 0.974 6 开裂 99.43 93.01 100 98.30 0.997 0.955 8 锈蚀 100 96.60 98.86 96.60 0.9886 0.966 0 磨损 98.86 93.41 98.86 96.60 0.9886 0.938 6 表 4 网络模型对比
模型 特征提取模型 分类器 mAP/(%) 时间/ms CNN CNN CNN 99.05 13.2 DST_GLCM +SVM DST_GLCM SVM 96 24.8 BGP +SVM BGP SVM 99.11 25.23 AlexNet CNN CNN 98.4 21.31 提出的AlexNet+SVM CNN SVM 99.43 17.2 -
[1] 孙虹,王仲,付鲁华,等. 一种新型钢球表面检测方法的分析与系统实现[J]. 机械科学与技术,2016,35(1):118-121. [2] 栗琳,王仲,裴芳莹. 双图像传感器的球表面展开方法[J]. 仪器仪表学报,2012,33(7):1641-1646. doi: 10.3969/j.issn.0254-3087.2012.07.029 [3] 宋宇航,王仲,付鲁华,等. 钢球表面质量视觉检测仪照明系统优化[J]. 光电工程,2016,43(11):13-18. [4] 张艳萍. 钢球表面缺陷涡流探伤仪分析[J]. 哈尔滨轴承,2007,28(2):32-33. doi: 10.3969/j.issn.1672-4852.2007.02.002 [5] 徐淑琼,蒋沂萍. 轴承球缺陷的超声波检测方法[J]. 机械制造与自动化,2005,34(5):5-44. [6] 周杭超,董晨晨,陈锋,等. 钢球表面缺陷检测方法综述[J]. 装备制造技术,2018(10):8-11. [7] Jiang M H,Fu L H,Wang Z,et a1. A new method for specular curved surface defect inspection based on reflected pattern integrity[J]. Journal of Measurement Science and Instrumentation,2016,7(3):221-228. [8] 付鲁华,庞家明,孙长库,等. 基于轻量化卷积神经网络的陶瓷球表面缺陷快速识别方法[J]. 传感技术学报,2023,36(7):1041-1027. doi: 10.3969/j.issn.1004-1699.2023.07.006 [9] 李宗祐,高春艳,吕晓玲,等. 基于深度学习的金属材料表面缺陷检测综述[J]. 制造技术与机床,2023(6):61-67. [10] Yi L,Li G Y,Jiang M M. An end-to-end steel strip surface defects recognition system based on convolutional neural networks[J]. Steel Research,2017,88(2):176-187. [11] Ashour M W,Khalid F,AbdulHalin A,et al. Surface defects classification of hot-rolled steel strips using multi-directional shearlet features[J]. Arabian Journal for Science and Engineering,2019(44):2925-2932. [12] 布申申,田怀文. 基于卷积神经网络的带钢表面缺陷检测算法[J]. 机械设计与制造,2022(7):29-33. doi: 10.3969/j.issn.1001-3997.2022.07.007 [13] Krizhevsky A,Sutskever I,Hinton G. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM,2017,60(6):84-90. doi: 10.1145/3065386 [14] Liang H B,Cheng G,Zhang Z D,et al. Research on ultrasonic defect identification method of well control manifold pipeline based on IAFSA-SVM[J]. Measurement,2022,194:110854. doi: 10.1016/j.measurement.2022.110854 [15] Chen C,Zhang X Y. Research on laser ultrasonic surface defect identification based on a support vector machine[J]. Science Progress,2021,104(4):1-17. [16] 廖益龙. 钢球表面缺陷检测仪控制技术及图像处理算法研究[D]. 哈尔滨:哈尔滨理工大学,2015. [17] 吴秀东,李东兴,常晓刚,等. 基于Halcon 的钢球表面缺陷检测系统设计[J]. 轴承,2018(10):53-58.