[1]尤志鹏,黄 硕,李佳坤,等.基于ZYNQ的带钢表面缺陷识别系统设计[J].计算机技术与发展,2025,(10):166-172.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0145]
　YOU Zhi-peng,HUANG Shuo,LI Jia-kun,et al.Design of Strip Surface Defect Identification System Based on ZYNQ[J].,2025,(10):166-172.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0145]
点击复制

基于ZYNQ的带钢表面缺陷识别系统设计

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年10期
页码:
166-172
栏目:
新型计算应用系统
出版日期:
2025-10-10

文章信息/Info

Title:
Design of Strip Surface Defect Identification System Based on ZYNQ
文章编号:
1673-629X(2025)10-0166-07
作者:
尤志鹏1; 黄 硕1; 李佳坤1; 查宇恒1; 孙科学1; 2*
1. 南京邮电大学 电子与光学工程学院、柔性电子(未来技术)学院,江苏南京 210023; 2. 射频集成与微组装技术国家地方联合工程实验室,江苏南京 210023
Author(s):
YOU Zhi-peng1; HUANG Shuo1; LI Jia-kun1; ZHA Yu-heng1; SUN Ke-xue1; 2*
1. School of Electronic and Optical Engineering,School of Flexible Electronics (Future Technology),Nanjing University of Posts and Telecommunications,Nanjing 210023,China; 2. Nation Local Joint Project Engineering Lab of RF Integration &Micropackage,Nanjing 210023,China
关键词:
深度学习; ZYNQ; 带钢缺陷识别; MobileNetV2; 硬件加速
Keywords:
deep learning; ZYNQ; strip defect recognition; MobileNetV2;  hardware acceleration
分类号:
TP274+.2
DOI:
10.20165/j.cnki.ISSN1673-629X.2025.0145
摘要:
针对带钢表面缺陷在尺度多变、背景复杂情况下易出现漏检,误检,准确率低等问题,提出了一种改进的MobileNetV2网络模型以及硬件加速模块。首先,对NEU-CLS数据集进行Cutout增强处理以扩充样本。然后,通过批量归一化(BN)融合与16 bit定点量化网络权重,对MobileNetV2网络的结构参数与ZYNQ平台适配改进。对比分析改进前后的MobileNetV2网络,改进后的网络能够有效提升带钢表面缺陷识别的速度与准确率,降低运算的复杂度。接着,在ZYNQ-7020的PL侧,设计权重输入缓存、数据补零、卷积、平均池化以及全连接的硬件加速模块,实现MobileNetV2网络的识别过程。最后,完成系统的实现与测试,系统的识别速度约为CPU中AMD Ryzen9-5900HX的2.5倍,识别准确率约为86.2%,帧率达到61.2f/s,整体功耗仅2.662W。实验结果表明,设计的带钢表面缺陷识别系统相比目前主流算法和其他改进算法,在保持实时性的同时,可以有效提高检测准确度,降低检测功耗。
Abstract:
To address the problems of missed detection and incorrect detection of surface defects in steel strip with scale variability and complex background,an improved MobileNetV2 network model and hardware acceleration module were proposed. First,the NEU-CLS dataset was enhanced with Cutout to expand the samples. Then,the structure parameters of the MobileNetV2 network were adapted to the ZYNQ platform by fusing batch normalization (BN) and 16-bit fixed-point quantization of network weights. The accuracy and speed of the surface defect recognition were improved effectively by comparing the performance of the improved and unimproved MobileNetV2 networks. Then,a hardware acceleration module was designed on the PL side of the ZYNQ-7020,including weight input cache,data zero filling,convolution,average pooling,and fully connected operations,to implement the recognition process of the MobileNetV2 network.Finally,the system was implemented and tested,with the recognition speed being about 2.5 times that of the CPU's AMD Ryzen9-5900HX and the average recognition accuracy being about 86.2%,the frame rate reaching 61.2 f/s,and the overall power consumption being 2.662W. The experimental results show that the designed steel strip surface defect recognition system can effectively improve detection accuracy and reduce detection power consumption compared with the mainstream algorithms and other improved algorithms while maintaining real-time performance.

相似文献/References:

[1]陈强锐,谢世朋.基于深度学习的肺部肿瘤检测方法[J].计算机技术与发展,2018,28(04):201.[doi:10.3969/ j. issn.1673-629X.2018.04.043]
　CHEN Qiang-rui,XIE Shi-peng.Lung Cancer Detection Method Based on Deep Learning[J].,2018,28(10):201.[doi:10.3969/ j. issn.1673-629X.2018.04.043]
[2]施泽浩,赵启军.基于全卷积网络的目标检测算法[J].计算机技术与发展,2018,28(05):55.[doi:10.3969/j.issn.1673－629X.2018.05.013]
　SHI Ze-hao,ZHAO Qi-jun.Object Detection Algorithm Based on Fully Convolutional Neural Network[J].,2018,28(10):55.[doi:10.3969/j.issn.1673－629X.2018.05.013]
[3]黄法秀,张世杰,吴志红,等.数据增广下的人脸识别研究[J].计算机技术与发展,2020,30(03):67.[doi:10. 3969 / j. issn. 1673-629X. 2020. 03. 013]
　HUANG Fa-xiu,ZHANG Shi-jie,WU Zhi-hong,et al.Research on Face Recognition Based on Data Augmentation[J].,2020,30(10):67.[doi:10. 3969 / j. issn. 1673-629X. 2020. 03. 013]
[4]陈浩翔,蔡建明,刘铿然,等. 手写数字深度特征学习与识别[J].计算机技术与发展,2016,26(07):19.
　CHEN Hao-xiang,CAI Jian-ming,LIU Keng-ran,et al. Deep Learning and Recognition of Handwritten Numeral Features[J].,2016,26(10):19.
[5]高翔,陈志,岳文静,等.基于视频场景深度学习的人物语义识别模型[J].计算机技术与发展,2018,28(06):53.[doi:10.3969/ j. issn.1673-629X.2018.06.012]
　GAO Xiang,CHEN Zhi,YUE Wen-jing,et al.Human Semantic Recognition Model Based on Video Scene Deep Learning[J].,2018,28(10):53.[doi:10.3969/ j. issn.1673-629X.2018.06.012]
[6]贺飞翔,赵启军. 基于深度学习的头部姿态估计[J].计算机技术与发展,2016,26(11):1.
　HE Fei-xiang,ZHAO Qi-jun. Head Pose Estimation Based on Deep Learning[J].,2016,26(10):1.
[7]徐 融,邱晓晖.一种改进的 YOLO V3 目标检测方法[J].计算机技术与发展,2020,30(07):30.[doi:10. 3969 / j. issn. 1673-629X. 2020. 07. 007]
　XU Rong,QIU Xiao-hui.An Improved YOLO V3 Object Detection[J].,2020,30(10):30.[doi:10. 3969 / j. issn. 1673-629X. 2020. 07. 007]
[8]曾志平[] [],萧海东[],张新鹏[]. 基于DBN的金融时序数据建模与决策[J].计算机技术与发展,2017,27(04):1.
　ZENG Zhi-ping[] [],XIAO Hai-dong[],ZHANG Xin-peng[]. Modeling and Decision-making of Financial Time Series Data with DBN[J].,2017,27(10):1.
[9]李全兵,文 钊*,田艳梅*,等.基于 WGAN 的音频关键词识别研究[J].计算机技术与发展,2021,31(08):26.[doi:10. 3969 / j. issn. 1673-629X. 2021. 08. 005]
　LI Quan-bing,WEN Zhao *,TIAN Yan-mei *,et al.Research on Audio Keywords Recognition Based on WassersteinGenerative Adversarial Network[J].,2021,31(10):26.[doi:10. 3969 / j. issn. 1673-629X. 2021. 08. 005]
[10]李宏林. 分析式纹理合成技术及其在深度学习的应用[J].计算机技术与发展,2017,27(11):7.
　LI Hong-lin. Analyzed Texture-synthesis Techniques and Their Applications in Deep Learning[J].,2017,27(10):7.

更新日期/Last Update: 2025-10-10