pith. sign in

arxiv: 2606.00852 · v1 · pith:YYTVWUISnew · submitted 2026-05-30 · 💻 cs.CV · cs.AI· cs.LG

RefDiffNet: Learning to Expose Subtle PCB Defects Before Detection

Vinay Edula , Nilesh Badwe , Priyanka Bagade This is my paper

Pith reviewed 2026-06-28 18:47 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LG
keywords PCB defect detectionreference-based inspectioninput enhancementobject detectionlightweight moduleYOLORT-DETRFaster R-CNN
0
0 comments X

The pith

RefDiffNet compares each PCB image to an aligned defect-free reference and outputs an enhanced version that highlights defects for any downstream detector.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces RefDiffNet as a small plug-and-play block placed before a detector backbone. It takes the inspected PCB image together with its aligned reference image, computes structural differences, and uses a lightweight encoder to produce an output image in which defective regions stand out. This pre-processing step is tested on HRIPCB and DeepPCB with detectors ranging from YOLOv8 through YOLOv26, RT-DETR, and Faster R-CNN. The module adds at most 0.005 million parameters and 0.8 GFLOPs yet yields relative mAP50:95 gains reaching 18 percent.

Core claim

RefDiffNet is a lightweight input-enhancement module that receives a defective PCB image and its aligned defect-free reference, isolates the changes relative to the reference through a learned encoder, and returns the original image with defective areas accentuated so that any standard object detector can locate them more reliably.

What carries the argument

Lightweight encoder that encodes the difference between the defective image and the aligned reference to produce a defect-highlighted output image.

If this is right

  • Any one-stage, two-stage, or transformer detector receives the same relative improvement without retraining its backbone.
  • The added cost stays below 0.25 percent of the host detector's parameter count across the tested families.
  • The same block can be inserted before the backbone of detectors not evaluated in the paper.
  • Detection of small defects that blend with background traces becomes easier because the reference supplies the expected layout.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could transfer to other domains that already collect reference images, such as semiconductor wafer inspection or printed-label verification.
  • If alignment is learned jointly inside the encoder, the requirement for perfect pre-alignment might be relaxed without losing the performance lift.
  • Because the enhancement is detector-agnostic, it might allow smaller or faster base detectors to reach accuracy levels previously requiring larger models.

Load-bearing premise

A perfectly aligned defect-free reference image is available for every inspected PCB.

What would settle it

Running the same detectors with and without RefDiffNet on a test set where reference images are deliberately shifted by a few pixels shows whether the reported gains disappear or reverse.

Figures

Figures reproduced from arXiv: 2606.00852 by Nilesh Badwe, Priyanka Bagade, Vinay Edula.

Figure 1
Figure 1. Figure 1: RefDiffNet Deep learning has significantly improved PCB defect detection by learning data-driven features instead of relying on hand-crafted rules. General object detectors such as Faster R-CNN [30] and modern YOLO families [18, 19, 32, 33, 34, 35] have been widely adapted for PCB defect detection. Recent works improve backbone, neck, attention modules, fea- [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of local contrast normalization under lighting changes. Each 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example PCB image used to illustrate the Haar-band decomposition in Figure [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Haar-band decomposition of the PCB image in Figure [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative visualization of signed residual and morphology-expanded cues. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Encoder and spatial gate blocks used in RefDiffNet. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative visualization of RefDiffNet on representative PCB defect samples. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
read the original abstract

Printed circuit board (PCB) defect detection is challenging because many defects are small and difficult to distinguish from complex background patterns. Most deep learning-based PCB inspection methods rely only on the inspected PCB image for defect detection, ignoring the defect-free reference image that encodes the expected layout of traces, pads, and other PCB structures. In this work, we propose RefDiffNet, a lightweight plug-and-play input enhancement block placed before the detector backbone to enhance the image before defect detection. RefDiffNet brings one proven idea from classical inspection into the deep learning era, using a defect-free reference image to reveal defects. RefDiffNet compares the defective image with the aligned reference, captures structural changes relative to the reference, and uses a lightweight encoder to output the original image with defective regions highlighted, thereby making the downstream detector's task easier. Results on HRIPCB and DeepPCB show that RefDiffNet consistently improves performance across detector families, including one-stage detectors from YOLOv8 to YOLOv26, the transformer-based RT-DETR, and the two-stage Faster R-CNN. It achieves up to 18% relative mAP50:95 gain with negligible overhead, introducing only 0.004 - 0.005M additional parameters and 0.7 - 0.8 GFLOPs, amounting to at most 0.25% of the parameter count of any evaluated detector. Results establish RefDiffNet as a lightweight, plug-and-play, detector-agnostic input enhancement module that substantially improves PCB defect detection with minimal computational cost.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes RefDiffNet, a lightweight plug-and-play input enhancement module placed before a detector backbone. It takes a test PCB image and a defect-free reference image, computes their difference to capture structural changes, encodes the result with a small network, and outputs an enhanced image with defects highlighted. Experiments on HRIPCB and DeepPCB report consistent mAP50:95 gains (up to 18% relative) across YOLO variants, RT-DETR, and Faster R-CNN, with added cost of only 0.004-0.005M parameters and 0.7-0.8 GFLOPs.

Significance. If the reported gains hold under realistic conditions, the work demonstrates a practical, detector-agnostic way to incorporate classical reference-based inspection into modern deep detectors for subtle PCB defects. The negligible overhead and plug-and-play design are clear strengths that could be adopted broadly if robustness is established.

major comments (2)
  1. [§3] §3 (method description): RefDiffNet is constructed around the assumption of pixel-perfect alignment between the test image and reference; the difference map is fed to the encoder to isolate defects. No experiments, ablations, or stress tests evaluate robustness to realistic misalignment (translation, rotation, or scale offsets typical in production lines), so it is unknown whether the claimed 18% mAP gains survive the registration errors that occur at inference time.
  2. [§4] §4 (experiments): The evaluation reports gains on HRIPCB and DeepPCB but supplies no details on how reference images were obtained, registered, or whether any misalignment was present or simulated. This leaves the central empirical claim dependent on an untested precondition.
minor comments (2)
  1. [Abstract, §1] The abstract and §1 could more explicitly state the alignment requirement as a precondition for the method.
  2. [Figure 2] Figure 2 (architecture diagram) would benefit from explicit notation on the alignment step before subtraction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. The two major comments correctly identify an important practical consideration for reference-based inspection. We address both below and will incorporate the requested clarifications and experiments in the revised manuscript.

read point-by-point responses

  1. Referee: [§3] §3 (method description): RefDiffNet is constructed around the assumption of pixel-perfect alignment between the test image and reference; the difference map is fed to the encoder to isolate defects. No experiments, ablations, or stress tests evaluate robustness to realistic misalignment (translation, rotation, or scale offsets typical in production lines), so it is unknown whether the claimed 18% mAP gains survive the registration errors that occur at inference time.

    Authors: We agree that the current method description in §3 presupposes aligned inputs, consistent with the classical reference-based inspection paradigm and with the wording “aligned reference” used throughout the manuscript. Because no misalignment experiments appear in the original submission, the robustness claim cannot be substantiated from the existing results. In the revision we will add a dedicated ablation subsection that introduces controlled misalignment (translations of 0–8 pixels, rotations of 0–3°, and isotropic scale factors of 0.98–1.02) on both HRIPCB and DeepPCB, reporting mAP50:95 for each detector with and without RefDiffNet. We will also discuss whether a lightweight pre-registration step (e.g., phase-correlation) can be inserted before RefDiffNet without materially increasing the reported overhead. revision: yes

  2. Referee: [§4] §4 (experiments): The evaluation reports gains on HRIPCB and DeepPCB but supplies no details on how reference images were obtained, registered, or whether any misalignment was present or simulated. This leaves the central empirical claim dependent on an untested precondition.

    Authors: The original §4 indeed omits explicit description of reference-image provenance and registration. Both datasets supply paired defect-free references that were already aligned by the dataset creators; however, the manuscript does not state this or quantify residual registration error. In the revision we will expand the dataset subsection of §4 with: (i) the exact source of each reference image, (ii) the registration procedure reported by the dataset authors, and (iii) a short statement that all reported numbers assume the provided alignment. The new misalignment experiments described in the response to comment [§3] will further quantify sensitivity to deviations from that alignment. revision: yes

Circularity Check

0 steps flagged

No circularity; claims rest on empirical results with no self-referential derivations

full rationale

The paper presents RefDiffNet as an empirical input-enhancement module whose value is demonstrated solely through reported mAP gains on HRIPCB and DeepPCB across multiple detectors. No equations, first-principles derivations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided text. The central claim (performance improvement with negligible overhead) is therefore independent of any circular reduction and is evaluated against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the domain assumption that aligned reference images exist and on the unstated modeling choice that a lightweight encoder can isolate defect signals without harming downstream detection.

axioms (1)
  • domain assumption Aligned defect-free reference images are available for every test PCB
    The method description explicitly requires comparison with a reference image.

pith-pipeline@v0.9.1-grok · 5824 in / 1213 out tokens · 24746 ms · 2026-06-28T18:47:32.319573+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 29 canonical work pages · 5 internal anchors

  1. [1]

    A comprehensive review of deep learning-based pcb defect detection.IEEE Access, 11:139017–139038, 2023

    Xing Chen, Yonglei Wu, Xingyou He, and Wuyi Ming. A comprehensive review of deep learning-based pcb defect detection.IEEE Access, 11:139017–139038, 2023. doi: 10.1109/ACCESS.2023.3339561

    work page doi:10.1109/access.2023.3339561 2023

  2. [2]

    A high-accuracy pcb defect detection algorithm based on improved yolov12.Symmetry, 17(7), 2025

    Zhi Chen and Bingxiang Liu. A high-accuracy pcb defect detection algorithm based on improved yolov12.Symmetry, 17(7), 2025. ISSN 2073-8994. doi: 10.3390/ sym17070978. URLhttps://www.mdpi.com/2073-8994/17/7/978

    2025

  3. [3]

    Dvqi: A multi-task, hardware-integrated artificial intelligence system for automated visual inspection in electronics manufacturing, 2023

    Audrey Chung, Francis Li, Jeremy Ward, Andrew Hryniowski, and Alexander Wong. Dvqi: A multi-task, hardware-integrated artificial intelligence system for automated visual inspection in electronics manufacturing, 2023. URLhttps://arxiv.org/ abs/2312.09232

    work page arXiv 2023

  4. [4]

    Yu-Shan Deng, An-Chun Luo, and M. J. Dai. Building an automatic defect verification system using deep neural network for pcb defect classification.2018 4th International Conference on Frontiers of Signal Processing (ICFSP), pages 145–149, 2018. URL https://api.semanticscholar.org/CorpusID:54215616

    2018

  5. [6]

    URLhttps://ietresearch.onlinelibrary.wiley.com/doi/ abs/10.1049/joe.2019.1188

    work page doi:10.1049/joe.2019.1188 2019

  6. [7]

    Evangelidis and Emmanouil Z

    Georgios D. Evangelidis and Emmanouil Z. Psarakis. Parametric image alignment using enhanced correlation coefficient maximization.IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(10):1858–1865, 2008. doi: 10.1109/TPAMI. 2008.113

    work page doi:10.1109/tpami 2008

  7. [8]

    Pcb defect detection via local detail and global dependency information.Sensors, 23(18):7755, 2023

    Bo Feng and Jian Cai. Pcb defect detection via local detail and global dependency information.Sensors, 23(18):7755, 2023. doi: 10.3390/s23187755

    work page doi:10.3390/s23187755 2023

  8. [9]

    Changechip: A reference- based unsupervised change detection for pcb defect detection, 2021

    Yehonatan Fridman, Matan Rusanovsky, and Gal Oren. Changechip: A reference- based unsupervised change detection for pcb defect detection, 2021. URLhttps: //arxiv.org/abs/2109.05746

    work page arXiv 2021

  9. [10]

    A comprehensive review of research on surface defect detection of pcbs based on machine vision.Results in Engi- neering, 27:106437, 07 2025

    Zihan He, Yudong Lian, Yulei Wang, and Zhiwei Lu. A comprehensive review of research on surface defect detection of pcbs based on machine vision.Results in Engi- neering, 27:106437, 07 2025. doi: 10.1016/j.rineng.2025.106437

    work page doi:10.1016/j.rineng.2025.106437 2025

  10. [11]

    Coordinate attention for efficient mobile network design

    Qibin Hou, Daquan Zhou, and Jiashi Feng. Coordinate attention for efficient mobile network design. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 13713–13722, 2021

    2021

  11. [12]

    A lightweight and high-accuracy framework for printed circuit board defect detection.Engineering Applications of Artificial Intelli- gence, 148:110375, 2025

    Yingqiang Hou and Xindong Zhang. A lightweight and high-accuracy framework for printed circuit board defect detection.Engineering Applications of Artificial Intelli- gence, 148:110375, 2025. doi: 10.1016/j.engappai.2025.110375. 16REFDIFFNET: LEARNING TO EXPOSE SUBTLE PCB DEFECTS BEFORE DETECTION

    work page doi:10.1016/j.engappai.2025.110375 2025

  12. [13]

    MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

    Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient con- volutional neural networks for mobile vision applications, 2017. URLhttps: //arxiv.org/abs/1704.04861

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  13. [14]

    Squeeze-and-excitation networks

    Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 7132–7141, 2018

    2018

  14. [15]

    Hripcb: a chal- lenging dataset for pcb defects detection and classification.The Journal of Engineering, 2020(13):303–309, 2020

    Weibo Huang, Peng Wei, Manhua Zhang, and Hong Liu. Hripcb: a chal- lenging dataset for pcb defects detection and classification.The Journal of Engineering, 2020(13):303–309, 2020. doi: https://doi.org/10.1049/joe.2019

    work page doi:10.1049/joe.2019 2020

  15. [16]

    URLhttps://ietresearch.onlinelibrary.wiley.com/doi/ abs/10.1049/joe.2019.1183

    work page doi:10.1049/joe.2019.1183 2019

  16. [17]

    Wavelet-based printed circuit board inspection algorithm.Integrated Computer-Aided Engineering, 12:201–213, 04 2005

    Zuwairie Ibrahim and Syed Al-Attas. Wavelet-based printed circuit board inspection algorithm.Integrated Computer-Aided Engineering, 12:201–213, 04 2005. doi: 10. 3233/ICA-2005-12206

    2005

  17. [18]

    S.H Indera Putera and Z. Ibrahim. Printed circuit board defect detection using math- ematical morphology and matlab image processing tools. In2010 2nd International Conference on Education Technology and Computer, volume 5, pages V5–359–V5– 363, 2010. doi: 10.1109/ICETC.2010.5530052

    work page doi:10.1109/icetc.2010.5530052 2010

  18. [19]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Kevin Jarrett, Koray Kavukcuoglu, Marc’Aurelio Ranzato, and Yann LeCun. What is the best multi-stage architecture for object recognition? In2009 IEEE 12th Interna- tional Conference on Computer Vision, pages 2146–2153, 2009. doi: 10.1109/ICCV . 2009.5459469

    work page doi:10.1109/iccv 2009

  19. [20]

    Ultralytics yolo26, 2026

    Glenn Jocher and Jing Qiu. Ultralytics yolo26, 2026. URLhttps://github. com/ultralytics/ultralytics

    2026

  20. [21]

    YOLOv11: An Overview of the Key Architectural Enhancements

    Rahima Khanam and Muhammad Hussain. Yolov11: An overview of the key architec- tural enhancements, 2024. URLhttps://arxiv.org/abs/2410.17725

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  21. [22]

    Selective kernel networks

    Xiang Li, Wenhai Wang, Xiaolin Hu, and Jian Yang. Selective kernel networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 510–519, 2019

    2019

  22. [23]

    A deep context learn- ing based pcb defect detection model with anomalous trend alarming system.Results in Engineering, 17:100968, 2023

    JiaYou Lim, JunYi Lim, Vishnu Monn Baskaran, and Xin Wang. A deep context learn- ing based pcb defect detection model with anomalous trend alarming system.Results in Engineering, 17:100968, 2023. doi: 10.1016/j.rineng.2023.100968

    work page doi:10.1016/j.rineng.2023.100968 2023

  23. [24]

    Printed circuit board defect detection methods based on image processing, machine learning and deep learning: A survey.IEEE Access, 11: 15921–15944, 2023

    Qin Ling and Nor Ashidi Mat Isa. Printed circuit board defect detection methods based on image processing, machine learning and deep learning: A survey.IEEE Access, 11: 15921–15944, 2023. doi: 10.1109/ACCESS.2023.3245093

    work page doi:10.1109/access.2023.3245093 2023

  24. [25]

    Deep siamese se- mantic segmentation network for pcb welding defect detection.IEEE Transactions on Instrumentation and Measurement, 71:1–11, 2022

    Zhigang Ling, Aoran Zhang, Dexin Ma, Yuxin Shi, and He Wen. Deep siamese se- mantic segmentation network for pcb welding defect detection.IEEE Transactions on Instrumentation and Measurement, 71:1–11, 2022. doi: 10.1109/TIM.2022.3154814. REFDIFFNET: LEARNING TO EXPOSE SUBTLE PCB DEFECTS BEFORE DETECTION17

    work page doi:10.1109/tim.2022.3154814 2022

  25. [26]

    Yolo-bfrv: An efficient model for detecting printed circuit board defects.Sensors, 24:6055, 09 2024

    Jiaxin Liu, Bingyu Kang, Chao Liu, Xunhui Peng, and Yan Bai. Yolo-bfrv: An efficient model for detecting printed circuit board defects.Sensors, 24:6055, 09 2024. doi: 10.3390/s24186055

    work page doi:10.3390/s24186055 2024

  26. [27]

    Tao Luo, Yongbing Zhou, Donglin Shi, Qinglin Yun, Shuying Wang, Jian Zhang, and Guofu Ding. A lightweight defect detection transformer for printed circuit boards combining image feature augmentation and refined cross-scale feature fu- sion.Engineering Applications of Artificial Intelligence, page 111128, 2025. doi: 10.1016/j.engappai.2025.111128

    work page doi:10.1016/j.engappai.2025.111128 2025

  27. [28]

    Detrs beat yolos on real-time object detection, 2023

    Wenyu Lv, Shangliang Xu, Yian Zhao, Guanzhong Wang, Jinman Wei, Cheng Cui, Yuning Du, Qingqing Dang, and Yi Liu. Detrs beat yolos on real-time object detection, 2023

    2023

  28. [29]

    Torchvision: Pytorch’s computer vision library.https://github.com/pytorch/vision, 2016

    TorchVision maintainers and contributors. Torchvision: Pytorch’s computer vision library.https://github.com/pytorch/vision, 2016

    2016

  29. [30]

    Cost-sensitive siamese net- work for pcb defect classification.Computational Intelligence and Neuroscience, 2021(1):7550670, 2021

    Yilin Miao, Zhewei Liu, Xiangning Wu, and Jie Gao. Cost-sensitive siamese net- work for pcb defect classification.Computational Intelligence and Neuroscience, 2021(1):7550670, 2021. doi: https://doi.org/10.1155/2021/7550670. URLhttps: //onlinelibrary.wiley.com/doi/abs/10.1155/2021/7550670

    work page doi:10.1155/2021/7550670 2021

  30. [31]

    Detection of bare pcb defects by image subtraction method using machine vision.Proceedings of the World Congress on Engineering 2011, WCE 2011, 2, 07 2011

    Ajay Pal, Singh Chauhan, and Sharat Bhardwaj. Detection of bare pcb defects by image subtraction method using machine vision.Proceedings of the World Congress on Engineering 2011, WCE 2011, 2, 07 2011

    2011

  31. [32]

    Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real- time object detection with region proposal networks, 2016. URLhttps://arxiv. org/abs/1506.01497

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  32. [33]

    Online PCB Defect Detector On A New PCB Defect Dataset

    Sanli Tang, Fan He, Xiaolin Huang, and Jie Yang. Online pcb defect detector on a new pcb defect dataset, 2019. URLhttps://arxiv.org/abs/1902.06197

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  33. [34]

    YOLOv12: Attention-Centric Real-Time Object Detectors

    Yunjie Tian, Qixiang Ye, and David Doermann. Yolov12: Attention-centric real-time object detectors, 2025. URLhttps://arxiv.org/abs/2502.12524

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  34. [35]

    Yolov8: A novel object detection algorithm with enhanced performance and robustness

    Rejin Varghese and Sambath M. Yolov8: A novel object detection algorithm with enhanced performance and robustness. In2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS), pages 1–6, 2024. doi: 10.1109/ADICS58448.2024.10533619

    work page doi:10.1109/adics58448.2024.10533619 2024

  35. [36]

    Yolov10: Real-time end-to-end object detection,

    Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jungong Han, and Guiguang Ding. Yolov10: Real-time end-to-end object detection, 2024. URLhttps:// arxiv.org/abs/2405.14458

    work page arXiv 2024

  36. [37]

    Y ., & Liao, H

    Chien-Yao Wang, I-Hau Yeh, and Hong-Yuan Mark Liao. Yolov9: Learning what you want to learn using programmable gradient information, 2024. URLhttps: //arxiv.org/abs/2402.13616

    work page arXiv 2024

  37. [38]

    Eca-net: Efficient channel attention for deep convolutional neural networks

    Qilong Wang, Banggu Wu, Pengfei Zhu, Peihua Li, Wangmeng Zuo, and Qinghua Hu. Eca-net: Efficient channel attention for deep convolutional neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11534–11542, 2020. 18REFDIFFNET: LEARNING TO EXPOSE SUBTLE PCB DEFECTS BEFORE DETECTION

    2020

  38. [39]

    Cbam: Con- volutional block attention module

    Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon. Cbam: Con- volutional block attention module. InProceedings of the European Conference on Computer Vision (ECCV), pages 3–19, 2018

    2018

  39. [40]

    Pcb defect detection algorithm based on cdi-yolo.Scientific Reports, 14, 03 2024

    Gaoshang Xiao, Shuling Hou, and Huiying Zhou. Pcb defect detection algorithm based on cdi-yolo.Scientific Reports, 14, 03 2024. doi: 10.1038/s41598-024-57491-3

    work page doi:10.1038/s41598-024-57491-3 2024

  40. [41]

    Yolov8-dee: A high-precision model for printed circuit board defect detection.PeerJ Computer Science, 10:e2548, 2024

    Feifan Yi, Ahmad Sufril Azlan Mohamed, Mohd Halim Mohd Noor, Fakhrozi Che Ani, and Zol Effendi Zolkefli. Yolov8-dee: A high-precision model for printed circuit board defect detection.PeerJ Computer Science, 10:e2548, 2024. doi: 10.7717/peerj-cs. 2548

    work page doi:10.7717/peerj-cs 2024

  41. [42]

    Msa-detr: Multi-scale alignment detection transformer for pcb defect detection with enhanced cross-scale feature fusion

    Renjie Zhang, Yanjue Gong, Fu Zhao, and Jinkai Fan. Msa-detr: Multi-scale alignment detection transformer for pcb defect detection with enhanced cross-scale feature fusion. Measurement Science and Technology, 36:125401, 2025. doi: 10.1088/1361-6501/ ae2152

    work page doi:10.1088/1361-6501/ 2025