基于层次化显微特征建模与仿真的缺陷检测方法

doi:10.16182/j.issn1004731x.joss.25-1166

系统仿真学报 ›› 2026, Vol. 38 ›› Issue (1): 1-13.doi: 10.16182/j.issn1004731x.joss.25-1166

• 论文 • 下一篇

基于层次化显微特征建模与仿真的缺陷检测方法

邹静¹, 谭谞¹^,³, 毛俊佶¹, 高海东², 谭建荣³

^1.浙江科技大学计算机学院，浙江杭州 310023
^2.浙江工业大学计算机科学与技术学院，浙江杭州 310058
^3.浙江大学，浙江杭州 310013

收稿日期:2025-11-27 修回日期:2025-12-12 出版日期:2026-01-18 发布日期:2026-01-28
通讯作者: 谭谞
第一作者简介:邹静(2005-)，女，本科生，研究方向为计算机视觉。
基金资助:
国家自然科学基金(62406287)

Defect Detection Method Based on Hierarchical Microscopic Feature Modeling and Simulation

Zou Jing¹, Tan Xu¹^,³, Mao Junji¹, Gao Haidong², Tan Jianrong³

^1.School of Computer Science, Zhejiang University of Science and Technology, Hangzhou 310023, China
^2.School of Computer Science, Zhejiang University of Technology, Hangzhou 310058, China
^3.Zhejiang University, Hangzhou 310013, China

Received:2025-11-27 Revised:2025-12-12 Online:2026-01-18 Published:2026-01-28
Contact: Tan Xu

摘要/Abstract

摘要：

为解决显微图像中小尺寸、低对比度及复杂背景下的缺陷检测难题，提出一种基于层次化显微特征建模与仿真的缺陷检测技术。该方法以RT-DETR(real-time detection transformer)框架为基础，构建HM-RTDETR(hierarchical microscopic RT-DETR)模型，在保持Transformer全局特征建模能力的基础上，引入高密度一对一Mosaic(Dense O2O-Mosaic)仿真增强以提升小样本监督密度，设计深度可分离卷积(depthwise separable convolution，DWConv)局部细节增强模块强化显微纹理特征提取能力，并采用可学习上采样模块(PatchExpand)实现空间语义重建，从而增强模型对微小缺陷的辨识能力。进一步构建DWConv与PatchExpand的协同融合结构，实现多尺度特征自适应整合与轻量化优化。实验结果表明，所提模型在两组离线增强机制下均保持优势：HM-RTDETR的mAP0.5-0.95在设置I与设置II下分别达到57.8%与70.4%，较对应基线分别提升18.9%和21.6%。在精度与实时性之间实现良好平衡，为金相显微组织自动检测提供了一种高效、可推广的解决方案。

关键词: 显微缺陷检测, RT-DETR, Mosaic数据增强, DWConv, PatchExpand, 轻量化网络

Abstract:

To address the challenge of detecting small and low-contrast defects in complex microscopic images, a defect method technology based on hierarchical microscopic feature modeling and simulation is proposed. The method is built on the RT-DETR (real-time detection transformer)framework to construct the HM-RTDETR (hierarchical microscopic RT-DETR) model. It maintains the global feature modeling ability of the Transformer and introduces a Dense O2O-Mosaic,a high-density one-to-one Mosaic augmentation strategy, to increase supervision density for small samples. A depthwise separable convolution (DWConv) module is used to enhance local detail extraction in microscopic textures, and a learnable PatchExpand module is applied for spatial semantic reconstruction, improving the identification of tiny defects. Furthermore, thecooperative fusion of DWConv and PatchExpand achieves adaptive multi-scale feature integration and lightweight optimization.Experimental results show that the proposed model maintains consistent advantages under both offline augmentation settings: HM-RTDETR achieves mAP0.5-0.95 of 57.8% and 70.4% under Setting I and Setting II, respectively, improving over the corresponding baselines by 18.9% and 21.6%.The method achieves a good balance between accuracy and efficiency, providing an effective and extensible solution for automatic microscopic defect detection and simulation analysis.

Key words: microscopic defect detection, RT-DETR(real-time detection transformer), Mosaic data augmentation, DWConv, PatchExpand, lightweight network

中图分类号:

TP 391.9

邹静,谭谞,毛俊佶等 . 基于层次化显微特征建模与仿真的缺陷检测方法[J]. 系统仿真学报, 2026, 38(1): 1-13.

Zou Jing,Tan Xu,Mao Junji,et al . Defect Detection Method Based on Hierarchical Microscopic Feature Modeling and Simulation[J]. Journal of System Simulation, 2026, 38(1): 1-13.

图/表 11

图 1

图2

图3

表1

表2

图4

图5

表3

表4

图6

图7

参考文献 26

[1]	Sun L G, Wu G, Wang Q, et al. Nanostructural Metallic Materials: Structures and Mechanical Properties[J]. Materials Today, 2020, 38: 114-135.
[2]	白雄飞, 龚水成, 李雪松, 等. 基于泊松融合数据增强的焊缝金相组织缺陷分类研究[J]. 上海交通大学学报, 2023, 57(10): 1316-1328.
	Bai Xiongfei, Gong Shuicheng, Li Xuesong, et al. Defect Classification of Weld Metallographic Structure Based on Data Augmentation of Poisson Fusion[J]. Journal of Shanghai Jiao Tong University, 2023, 57(10): 1316-1328.
[3]	鲁斌, 杨烜, 杨振宇, 等. 自适应采样与重影多尺度特征融合的轻量化焊缝缺陷检测[J]. 系统仿真学报, 2025, 37(8): 1978-1990.
	Lu Bin, Yang Xuan, Yang Zhenyu, et al. Adaptive Sampling and Ghost Multi-scale Fusion for Lightweight Weld Defect Detection[J]. Journal of System Simulation, 2025, 37(8): 1978-1990.
[4]	赵朗月, 吴一全. 基于机器视觉的表面缺陷检测方法研究进展[J]. 仪器仪表学报, 2022, 43(1): 198-219.
	Zhao Langyue, Wu Yiquan. Research Progress on Surface Defect Detection Methods Based on Machine Vision[J]. Chinese Journal of Scientific Instrument, 2022, 43(1): 198-219.
[5]	Zhao Yaling, Wang Hai, Xie Xiaoming, et al. An Enhanced YOLOv5-based Algorithm for Metal Surface Defect Detection[J]. Applied Sciences, 2023, 13(20): 11473.
[6]	胡依伦, 杨俊, 许聪源, 等. PIC2f-YOLO: 金属表面缺陷检测轻量化方法[J]. 光电工程, 2025, 52(1): 240250.
	Hu Yilun, Yang Jun, Xu Congyuan, et al. PIC2f-YOLO: A Lightweight Method for Metal Surface Defect Detection[J]. Opto-Electronic Engineering, 2025, 52(1): 240250.
[7]	张寅, 朱桂熠, 施天俊, 等. 基于特征融合与注意力的遥感图像小目标检测[J]. 光学学报, 2022, 42(24): 2415001.
	Zhang Yin, Zhu Guiyi, Shi Tianjun, et al. Small Object Detection in Remote Sensing Images Based on Feature Fusion and Attention[J]. Acta Optica Sinica, 2022, 42(24): 2415001.
[8]	Müller Martin, Stiefel Marie, Bachmann Björn-Ivo, et al. Overview: Machine Learning for Segmentation and Classification of Complex Steel Microstructures[J]. Metals, 2024, 14(5): 553.
[9]	Hütten Nils, Meyes Richard, Meisen Tobias. Vision Transformer in Industrial Visual Inspection[J]. Applied Sciences, 2022, 12(23): 11981.
[10]	王改华, 李柯鸿, 龙潜, 等. 基于知识蒸馏的轻量化Transformer目标检测[J]. 系统仿真学报, 2024, 36(11): 2517-2527.
	Wang Gaihua, Li Kehong, Long Qian, et al. Object Detection of Lightweight Transformer Based on Knowledge Distillation[J]. Journal of System Simulation, 2024, 36(11): 2517-2527.
[11]	Zhao Yian, Wenyu Lü, Xu Shangliang, et al. DETRs Beat YOLOs on Real-time Object Detection[EB/OL]. (2024-04-03) [2025-11-21]. .
[12]	Wenyu Lü, Zhao Yian, Chang Qinyao, et al. RT-DETRv2: Improved Baseline with Bag-of-freebies for Real-time Detection Transformer[EB/OL]. (2024-07-24) [2025-11-21]. .
[13]	Wang Shuo, Xia Chunlong, Feng Lü, et al. RT-DETRv3: Real-time End-to-end Object Detection with Hierarchical Dense Positive Supervision[C]//2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). Piscataway: IEEE, 2025: 1628-1636.
[14]	Sauer A, Boesel F, Dockhorn T, et al. Fast High-resolution Image Synthesis with Latent Adversarial Diffusion Distillation[EB/OL]. (2024-03-18) [2025-11-21]. .
[15]	Howard A G, Zhu Menglong, Chen Bo, et al. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications[EB/OL]. (2017-04-17) [2025-09-19]. .
[16]	Carion N, Massa F, Synnaeve G, et al. End-to-end Object Detection with Transformers[C]//Computer Vision – ECCV 2020. Cham: Springer International Publishing, 2020: 213-229.
[17]	Zhu Xizhou, Su Weijie, Lu Lewei, et al. Deformable DETR: Deformable Transformers for End-to-end Object Detection[C]//ICLR 2021 Conference. New York: ICLR, 2021: 1-16.
[18]	Meng Depu, Chen Xiaokang, Fan Zejia, et al. Conditional DETR for Fast Training Convergence[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2021: 3631-3640.
[19]	Zhang Hao, Li Feng, Liu Shilong, et al. DINO: DETR with Improved DeNoising Anchor Boxes for End-to-end Object Detection[C]//ICLR 2023 Conference. New York: ICLR, 2023: 1-19.
[20]	Li Feng, Zhang Hao, Liu Shilong, et al. DN-DETR: Accelerate DETR Training by Introducing Query DeNoising[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 13609-13617.
[21]	Sun Peize, Zhang Rufeng, Jiang Yi, et al. Sparse R-CNN: End-to-end Object Detection with Learnable Proposals[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 14449-14458.
[22]	He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Deep Residual Learning for Image Recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 770-778.
[23]	Sandler M, Howard A, Zhu Menglong, et al. MobileNetV2: Inverted Residuals and Linear Bottlenecks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 4510-4520.
[24]	Cao H, Wang Y, Chen J, et al. Swin-unet: UNet-like Pure Transformer for Medical Image Segmentation[J]. IEEE Transactions on Medical Imaging, 2022, 41(2): 302-312.
[25]	Ultralytics. YOLO(Version 8.0)[EB/OL]. (2023-01-10) [2025-11-21]. .
[26]	Khanam R, Hussain M. YOLOv11: An Overview of the Key Architectural Enhancements[EB/OL]. (2024-10-23) [2025-11-21]. .

模型	P/%	R/%	F₁/%	mAP0.5-0.95/%	FPS
RTDETR-ResNet101	67.5	60.1	63.6	38.9	108.2
RTDETR-ResNet50	73.6	55.9	63.6	38.7	173.5
YOLOv8	95.0	59.3	73.0	53.4	238.3
YOLOv11	81.7	52.6	63.8	48.2	104.7
Deformable DETR	79.2	89.1	83.8	53.8	58.5
HM-RTDETR	85.9	70.4	77.4	57.8	106.1

模型	P/%	R/%	F₁/%	mAP0.5-0.95/%	FPS
RTDETR-ResNet101	84.5	67.1	74.8	48.8	108.2
RTDETR-ResNet50	68.9	68.8	68.8	47.3	173.5
YOLOv8	88.6	71.8	84.5	61.9	238.3
YOLOv11	87.0	63.7	73.5	56.8	104.7
Deformable DETR	79.7	88.3	83.8	68.7	58.5
HM-RTDETR	88.4	87.1	87.7	70.4	106.1

方法	P/%	R/%	F₁/%	mAP0.5-0.95/%	FPS
baseline	67.5	60.1	63.6	38.9	108.2
+Dense O2O Mosaic	62.7	58.0	60.2	40.1	106.1
+Dense O2O Mosaic+DWConv	71.6	58.1	64.1	42.3	104.0
+Dense O2O Mosaic+PatchExpand	85.2	69.1	76.3	51.4	105.8
+Dense O2O Mosaic+DWConv+PatchExpand	85.9	70.4	77.4	57.8	106.1

方法	P/%	R/%	F₁/%	mAP0.5-0.95/%	FPS
baseline	84.5	67.1	74.8	48.8	108.2
+Dense O2O Mosaic	79.9	73.6	76.6	54.0	106.1
+Dense O2O Mosaic+DWConv	88.5	74.6	80.9	62.1	104.0
+Dense O2O Mosaic+PatchExpand	86.0	78.1	81.9	61.1	105.8
+Dense O2O Mosaic+DWConv+PatchExpand	88.4	87.1	87.7	70.4	106.1

基于层次化显微特征建模与仿真的缺陷检测方法

Defect Detection Method Based on Hierarchical Microscopic Feature Modeling and Simulation

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 26

相关文章 1

编辑推荐

Metrics

本文评价