|本期目录/Table of Contents|

[1]罗芷萱,田 斌*,黎 曦*.基于视觉变换器动态优化的瓷砖瑕疵检测模型[J].武汉工程大学学报,2025,47(05):539-547.[doi:10.19843/j.cnki.CN42-1779/TQ.202409022]
　LUO Zhixuan,TIAN Bin*,LI Xi*.A ceramic tile defect detection model based on dynamic optimization of vision transformer[J].Journal of Wuhan Institute of Technology,2025,47(05):539-547.[doi:10.19843/j.cnki.CN42-1779/TQ.202409022]
点击复制

基于视觉变换器动态优化的瓷砖瑕疵检测模型
(/HTML)
分享到：

《武汉工程大学学报》[ISSN:1674-2869/CN:42-1779/TQ]

卷:
47
期数:
2025年05期
页码:
539-547
栏目:
智能制造
出版日期:
2025-10-31

文章信息/Info

Title:
A ceramic tile defect detection model based on dynamic optimization of vision transformer

文章编号:
1674 - 2869（2025）05 - 0539 - 09
作者:
武汉工程大学电气信息学院，湖北 武汉 430205
Author(s):
School of Electrical and Information Engineering，Wuhan Institute of Technology，Wuhan 430205，China
关键词:
Keywords:
分类号:
TP391.4
DOI:
10.19843/j.cnki.CN42-1779/TQ.202409022
文献标志码:
A
摘要:
瓷砖表面瑕疵呈现出面积较小、形状不规则、尺度多变、难与复杂花纹背景区分等特点，这使得高精度瑕疵检测的实现异常困难。为解决此问题，提出了一种高性能瓷砖表面瑕疵检测模型，该模型结合了视觉变换器（ViT）网络与YOLOv8n高效架构，在精度和计算量之间取得了良好平衡。首先，使用高效ViT网络作为模型的主干网络，增强模型对图像全局语义和局部特征的联合建模能力，同时保持计算效率。其次，使用空间到深度卷积替换普通下采样卷积，以保留更多细粒度信息。此外，在颈部网络末端添加空间通道重组卷积，使模型更好地聚焦关键特征信息。最后，引入动态头机制，通过动态调整注意力权重，提升模型对复杂瑕疵特征的检测能力。在阿里云天池官方提供的瓷砖瑕疵检测数据集上的实验结果表明，改进后的模型平均检测精度达到71.5%，较原模型YOLOv8n提升了11.6%，且在瓷砖表面微小瑕疵和密集瑕疵检测等复杂场景中表现出色，并兼具较低的部署成本，能够满足实际生产中实时质量检测的需求。
Abstract:
Ceramic tile surface defects are characterized by their small size, irregular shapes, varying scales, and the challenge of differentiating them from complex patterned backgrounds, rendering high-precision defect detection extremely difficult. To address this issue, we proposed a high-performance ceramic tile surface defect detection model that seamlessly integrates a vision transformer (ViT) network with the efficient YOLOv8n architecture, achieving a favorable balance between accuracy and computational efficiency. First, an efficient ViT network was adopted as the backbone to enhance the model’s capability for joint modeling of global image context and local features while maintaining computational efficiency. Then, space-to-depth convolution replaced standard downsampling convolution to preserve more fine-grained information. And a spatial-channel reorganization convolution module was incorporated at the end of the neck network to focus on critical feature information. Finally, a dynamic head mechanism was introduced to dynamically adjust attention weights, significantly improving the model’s ability to detect complex defect features. Experiments on the Alibaba Cloud Tianchi ceramic tile defect detection dataset demonstrated that this optimized model achieved a mean average precision (mAP) of 71.5%, outperforming the original YOLOv8n by 11.6%. The proposed model excels in detecting tiny and densely distributed defects on tile surfaces while maintaining low deployment costs, fulfilling real-time quality inspection requirements in industrial production.

参考文献/References:

［1］ WANG K, LI Z R, WANG X. Concatenated network fusion algorithm (CNFA) based on deep learning: improving the detection accuracy of surface defects for ceramic tile[J]. Applied Sciences, 2022, 12(3): 1249.
［2］ CAO T L, SONG K C, XU L K, et al. Balanced multi-scale target score network for ceramic tile surface defect detection[J]. Measurement, 2024, 224: 113914.
［3］ KUMAR A, ZHANG Z J, LYU H B. Object detection in real time based on improved single shot multi-box detector algorithm[J]. EURASIP Journal on Wireless Communications and Networking, 2020, 2020: 204.
［4］ REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2016: 779-788.
［5］ REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
［6］ CAI Z W, VASCONCELOS N. Cascade R-CNN: high quality object detection and instance segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(5): 1483-1498.
［7］ ZOU Z X, CHEN K Y, SHI Z W, et al. Object detection in 20 years: a survey[J]. Proceedings of the IEEE, 2023, 111(3): 257-276.
［8］ 苏佳，贾泽，秦一畅，等. 面向工业表面缺陷检测的改进YOLOv8算法［J］. 计算机工程与应用，2024，60（14）：187-196．
［9］ 赵楚，段先华，苏俊楷. 改进Faster RCNN的瓷砖表面瑕疵检测研究［J］. 计算机工程与应用，2023，59（14）：201-208．
［10］ YU X L, YU Q C, MU Q Y, et al. MCAW-YOLO: an efficient detection model for ceramic tile surface defects[J]. Applied Sciences, 2023, 13(21): 12057.
［11］ 余松森，薛国鹏，何皇，等．改进YOLOv8的轻量级瓷砖表面缺陷检测［J］．计算机工程与应用，2024，60（18）：88-102．
［12］ LIU X Y, PENG H W, ZHENG N X, et al. EfficientViT: memory efficient vision transformer with cascaded group attention[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2023: 14420-14430.
［13］ HUI Y, YOU S J, HU X H, et al. SEB-YOLO: an improved YOLOv5 model for remote sensing small target detection[J]. Sensors, 2024, 24(7):2193.
［14］ LI J F, WEN Y, HE L H. SCConv: spatial and channel reconstruction convolution for feature redundancy[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2023: 6153-6162.
［15］ WU M J, YUN L J, WANG Y B, et al. Detection algorithm for dense small objects in high altitude image[J]. Digital Signal Processing, 2024, 146: 104390.
［16］ GUPTA C, GILL N S, GULIA P, et al. A novel finetuned YOLOv6 transfer learning model for real-time object detection[J]. Journal of Real-Time Image Processing, 2023, 20(3): 42.
［17］ WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2023: 7464-7475.
［18］ ZHAO Y, Lü W Y, XU S L, et al. DETRs beat YOLOs on real-time object detection[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2024: 16965-16974.

相似文献/References:

备注/Memo

备注/Memo:
收稿日期：2024-09-28
基金项目：国家自然科学基金（62367006）
作者简介：罗芷萱，硕士研究生。Email：sql9617@163.com
*通信作者：田 斌，博士，教授。 Email:121494671@qq.com
黎 曦，博士，副教授。Email:305642677@qq.com
引文格式：罗芷萱，田斌，黎曦. 基于视觉变换器动态优化的瓷砖瑕疵检测模型［J］. 武汉工程大学学报，2025，47（5）：539-547，555.

更新日期/Last Update: 2025-11-03