随着工业自动化检测技术的快速发展，机器视觉逐渐应用在各类工业缺陷检测。目前，市面上超声波轮胎切割表面缺陷检测还处于空白，机器视觉检测通常在PC机上处理，设备体积庞大、成本高昂，难以在资源受限的边缘环境中部署和高效运行，同时，传统嵌入式移动端图像处理存在性能较差，实时性低的缺点。针对上述问题，本文使用多核异构平台，采用FPGA硬件加速和软硬件协同设计方法，设计了一种基于ZYNQ的缺陷检测系统，旨在通过算法优化与硬件架构协同设计，在资源受限条件下实现高效的图像缺陷检测。本文的主要研究内容如下：

首先，针对超声波轮胎切割表面缺陷的检测需求，定制化搭建ZYNQ硬件平台。该平台包含Xilinx公司的ZYNQ7020为主控芯片、DDR3存储器、电源电路、图像采集和显示电路以及众多外设。通过使用Altium Designer软件绘制原理图和高速PCB设计，并完成实物焊接及调试。本设计的系统方案采用异构多核架构和软硬件协同设计方法，在ZYNQ平台上部署了集图像采集、传输、处理和显示一体的SoC系统，设计并测试多个关键模块，包括图像预处理模块、HOG-SVM缺陷检测模块、VDMA数据传输等。

其次，本文在小样本及二分类的现有条件下，采用HOG特征提取和SVM分类算法进行缺陷检测，使用HLS设计工具，采用定点化技术、循环流水线优化、多加速器并行等算法级别优化策略，完成缺陷检测模块的设计。该硬件加速器通过AXI总线受ARM处理器控制和调度，充分利用了FPGA的硬件可编程和ARM的软件可编程优势。在软件方面，一方面完成各IP模块的驱动和配置，实现裸机环境下整体运行，另一方面实现了检测算法动态调度、NMS算法、多分辨率提取、框图绘制等设计。

最后，通过实验对ZYNQ缺陷检测系统的性能、效果、资源和功耗进行全面分析。实验结果表明，图像硬件处理速度较传统软件处理提升32倍，功耗降低至3.599W以下，FPGA资源占用率控制在70%以内。本文在嵌入式缺陷检测应用场景中提供了可复用的硬件实现方案，具备一定的工程应用价值。

With the rapid development of industrial automation detection technology, machine vision is gradually being applied in various types of industrial defect detection. At present, the detection of surface defects in ultrasonic tire cutting on the market is still blank. Machine vision detection is usually processed on PCs, which have large equipment volumes and high costs, making it difficult to deploy and operate efficiently in resource limited edge environments. At the same time, traditional embedded mobile image processing has the disadvantages of poor performance and low real-time performance. In response to the above issues, this article uses a multi-core heterogeneous platform and adopts FPGA hardware acceleration and software hardware collaborative design methods to design a defect detection system based on ZYNQ. The aim is to achieve efficient image defect detection under resource limited conditions through algorithm optimization and hardware architecture collaborative design. The main research content of this article is as follows:

Firstly, a customized ZYNQ control system is established for ultrasonic tire cutting surface defect detection. The system integrates Xilinx’s ZYNQ7020 as the main controller with DDR3 memory, power circuits, image acquisition/display modules, and peripheral interfaces. Schematic design and high-speed PCB layout are implemented using Altium Designer, followed by physical board assembly and debugging. For system implementation, a heterogeneous multi-core architecture with hardware-software co-design is deployed on the ZYNQ platform, creating a complete SoC system integrating image acquisition, transmission, processing, and display. Key modules are designed and validated, including an image preprocessing module with filtering and enhancement functions, the HOG-SVM defect detection module, and VDMA data transmission.

Secondly, under the existing conditions of small sample size and binary classification, this article uses HOG feature extraction and SVM classification algorithms for defect detection. The HLS design tool is used, and algorithm level optimization strategies such as fixed-point technology, loop pipeline optimization, and multi accelerator parallelism are adopted to complete the design of the defect detection module.This module operates under ARM processor control via AXI bus, effectively combining hardware-accelerated parallel processing with software-driven complex algorithm logic.Software development encompasses both driver design/configuration for IP modules and implementation of algorithm scheduling, Non-Maximum Suppression (NMS), multi-resolution feature extraction, and bounding box rendering.

Finally, comprehensive experiments evaluate the system’s performance, effectiveness, resource utilization, and power consumption. Results demonstrate that while maintaining detection accuracy, the system achieves a 32x improvement in frame rate compared to traditional software solutions, reduces power consumption below 5W, and limits FPGA resource usage under 70%. Its low-power operation and high real-time performance meet the stringent cost-efficiency requirements of industrial automation equipment, providing a reusable hardware implementation framework for edge-side machine vision systems with significant engineering application value.

[1]张云电, 高衡, 吴子侠, 等. 橡胶超声切割装置的研制[C]. 中国机械工程学会特种加工分会.第16届全国特种加工学术会议论文集（下）.杭州电子科技大学机械工程学院; 2015: 441-444.

[2]胡月, 隋宏艳, 李大为, 等.超声辅助切割橡胶试验研究[J]. 工具技术, 2011, 45(10): 47-50.

[3]朱严己. 基于机器视觉的转向节缺陷检测及边缘打磨轨迹生成系统[D]. 河北: 河北科技师范学院, 2022.

[4]孟凡超. 基于FPGA的工件表面在线图像采集与缺陷检测系统研究[D]. 山东: 齐鲁工业大学, 2023.

[5]钱成铎. 基于ZYNQ的图像采集与表面缺陷检测系统[D]. 安徽: 淮北师范大学, 2022.

[6]Schmidberger E, Ahlers R. Quality control with a robot-guided electro-optical sensor[C]. Proceedings of the 4th International Conference on Robot Vision and Sensory Controls, 9-11 October 1984, London, UK. IFS,4: 27.

[7]Yang J X, Zhang F Y, Song X X, et al. New Method of Spherical Surface Defect Detection Based on Machine Vision[J]. Advanced Materials Research, 2011, 295-297: 1274-1278.

[8]Jian C, Gao J, Ao Y. Automatic Surface Defect Detection for Mobile Phone Screen Glass Based on Machine Vision[J].Applied Soft Computing, 2016. 236-241.

[9]Aminzadeh M, Kurfess T. Automatic thresholding for Defect Detection by Background Histogram Mode Extents[J]. Journal of Manufacturing Systems, 2015, 37: 83-92.

[10]Li C, Lan H Q, Sun Y N, et al. Detection Algorithm of Defects on Polyethylene Gas Pipe Using Image Recognition[J]. International Journal of Pressure Vessels and Piping, 2021, 191: 1-27.

[11]Hocenski Z, Aleksi I, Mijakovic R. Ceramic tiles failure detection based on FPGA image processing[C]. 2009 IEEE International Symposium on Industrial Electronics, 2009: 2169-2174.

[12]Choi J, Kim C. Unsupervised detection of surface defects: A two-step approach[C]. 2012 19th IEEE International Conference on Image Processing, 2012: 1037-1040.

[13]Iyshwerya K, Janani B, Krithika S, et al. Defect detection algorithm for high speed inspection in machine vision[C]. 2013: 103-107.

[14]Praveenkumar B ,Eswaran P . Hardware acceleration for object detection using YOLOv4 algorithm on Xilinx Zynq platform [J]. Journal of Real-Time Image Processing, 2022, 19 (5): 931-940.

[15]Liangwei C ,Ceng W ,Yuan X . A Real-Time FPGA Accelerator Based on Winograd Algorithm for Underwater Object Detection [J]. Electronics, 2021, 10 (23): 2889-2889.

[16]Qingbo J ,Chong D ,Changbo H , et al. Real-time embedded object detection and tracking system in Zynq SoC [J]. EURASIP Journal on Image and Video Processing, 2021, 2021 (1).

[17]V.R.S. M ,A. S ,N. A . Performance comparison of CNN, QNN and BNN deep neural networks for real-time object detection using ZYNQ FPGA node [J]. Microelectronics Journal, 2022, 119.

[18]鲁湛, 代作晓. 基于异构多核处理器的实时红外人体目标检测技术[J]. 半导体光电, 2018, 39(05): 753-757.

[19]李林, 张盛兵, 吴鹃. 基于深度学习的实时图像目标检测系统设计[J]. 计算机测量与控制, 2019, 27(07): 15-19.

[20]李向阳. 基于ZYNQ的车载目标检测系统设计与实现[J]. 机械设计, 2020, 37: 35-38.

[21]张宝朋, 康谦泽, 李佳萌, 等. 轻量化的YOLOv4目标检测算法[J]. 计算机工程, 2022, 48(08): 206-214.

[22]戴伟杰, 王衍学, 李昕鸣, 等.面向FPGA部署的改进YOLO铝片表面缺陷检测系统[J]. 电子测量与仪器学报, 2023, 37(09): 160-167.

[23]赵冬冬, 谢墩翰, 陈朋, 等.基于ZYNQ的轻量化YOLOv5声呐图像目标检测算法及实现[J]. 光电工程, 2024, 51(01): 60-72.

[24]黄宇新, 裴成梅, 吴春儒, 等. 基于无线数据传输的智能电机监测系统设计[J]. 物理通报, 2021, (08): 138-139.

[25]王肃国, 李龙华. 基于Zernike正交矩的图像亚像素边缘检测算法改进[J]. 信息与电脑, 2021, 33(02): 71-74.

[26]李佩斌. 基于高速串行总线的DSP+FPGA架构图像处理系统设计[J]. 集成电路与嵌入式系统, 2024, 24(12): 45-51.

[27]杨振雷, 刘承敏, 青先国, 等. 基于Zynq-7000的千兆以太网传输系统设计与实现[J]. 核技术, 2021, 44(02): 37-44.

[28]钟鑫. 基于眼动数据分析的认知负荷评估研究[D]. 北京: 北京邮电大学, 2023.

[29]宁云. 基于深度学习的冠状动脉造影图像分割算法[D]. 浙江: 浙江理工大学, 2022.

[30]刘文汇, 巢渊, 唐寒冰, 等. 移动机器人视觉目标检测与跟踪方法研究综述[J]. 电光与控制, 2022,29(04): 59-67+88.

[31]郭双权, 郝欢, 王阳. 基于HOG特征的隔离开关红外图像故障识别方法[J]. 红外技术, 2025, 47(02): 243-249.

[32]牛朝旭, 孙海江. 基于FPGA的Winograd算法卷积神经网络加速器设计与实现[J]. 液晶与显示, 2023, 38(11): 1521-1530.

[33]李宁, 肖昊. 基于FPGA的稀疏卷积神经网络加速器设计[J]. 电子测量技术, 2024, 47(05): 1-8.

[34]江瑜, 朱铁柱, 蒋青松, 等. 基于FPGA的卷积神经网络硬件加速器设计[J]. 电子器件, 2023, 46(04): 973-977.

[35]陈逸, 刘博生, 徐永祺, 等. 混合精度频域卷积神经网络FPGA加速器设计[J]. 计算机工程, 2023, 49(12): 1-9.

[36]熊阿龙. 夜间图像增强算法研究及其基于HLS仿真[D]. 哈尔滨: 哈尔滨工业大学, 2020.

[37]王云峰, 范正吉, 何鑫, 等. 基于Vivado HLS的内窥镜实时暗部增强算法设计 [J]. 电子测量技术, 2022, 45 (23): 31-37.

[38]张艳, 赵平, 余建英, 等. 基于GF-1 PMS/GF-6 PMS的水稻种植区识别方法研究[J]. 测绘与空间地理信息, 2024, 47(07): 18-21.

[39]袁儒明, 陈迎春, 汪杨, 等. 基于UVM的总线仲裁模块的验证研究[J]. 电子制作, 2021, (13): 90-92.

[40]马飞, 刘琦, 包斌. 基于FPGA的AXI4总线时序设计与实现[J]. 电子技术应用, 2015, 41(06): 13-15+19.

[41]邹敏, 鲁澳宇, 邹望辉, 等. 一种带Cache加速的HyperRAM控制器设计与验证[J]. 现代电子技术, 2024, 47(06): 91-96.

[42]史雁辉. 基于ZYNQ的CANFD-FlexRay汽车网关设计[D]. 哈尔滨: 哈尔滨理工大学, 2023.

[43]王咏星, 张国兵, 宋俊霞. Zynq-7000“FPGA+ARM”架构的数字图像处理系统设计[J]. 单片机与嵌入式系统应用, 2021, 21(07): 43-45+50.

[44]武世雄. 基于FPGA的卷积神经网络加速器的研究与实现[D]. 沈阳: 沈阳化工大学, 2022.

[45]唐寒冰, 巢渊, 刘文汇, 等. 基于机器视觉的大尺寸零件测量方法研究综述[J]. 电子测量技术, 2021, 44(17): 33-40.

[46]吴婧仪, 周哲海, 赵爽, 等. 基于双光栅和柱透镜的高分辨率近红外微型光谱仪[J]. 光谱学与光谱分析, 2024, 44(04): 1144-1150.

[47]郭思超, 刘婷婷, 陈奎. 基于ANSYS的高速PCB谐振仿真及优化[J]. 电脑编程技巧与维护, 2025, (03): 136-139.

[48]刘晛, 吴瑞琦, 高尚尚, 等. 基于ZYNQ的通用型卷积神经网络设计与实现[J]. 电子器件, 2023, 46(01): 121-125.

[49]宋杨, 殷焕松. 基于FPGA的实时图像预处理系统研究[J]. 安徽建筑大学学报, 2024, 32(02): 49-54+66.

[50]胡晓炜, 朱庆生, 沙孝鸣, 等. 应用于人卫激光测距光尖图像处理方法研究[J]. 激光与红外, 2020, 50(03): 380-384.