本研究旨在透過利用YOLOv8演算法，提升布料異常檢測的效能，特別是對中型缺陷的檢測能力。布料缺陷的檢測，如孔洞、撕裂及不規則線條，是紡織行業中至關重要的品質保證任務。如果未能及時發現缺陷布料，將導致供應鏈中出現巨大的經濟損失和效率低下。早期且準確的缺陷檢測能防止缺陷材料進入後續生產流程，從而減少浪費、最小化停工時間並保持產品品質。

YOLOv8作為YOLO（You Only Look Once）家族演算法的最新版本，相較於早期版本如YOLOv5、YOLOv7及YOLOv9，在檢測準確性、計算效率及即時性能方面有顯著的進步。本研究基於這些優勢，通過結構優化、超參數調整以及先進數據預處理和增強技術的應用，進一步改進模型性能。

關鍵創新包括整合了如**通道壓縮與激勵（Channel Squeeze and Excitation, CSE）和空洞空間金字塔池化（Atrous Spatial Pyramid Pooling, ASPP）**等模組，通過增強空間及通道注意力來提升特徵提取能力。這些改進輔以無錨檢測機制，使模型能更有效地處理不同規模和質地的缺陷。

實驗結果證明，優化後的YOLOv8模型在多種配置和預處理技術中於Precision（準確率）、Recall（召回率）及F1-Score（F1分數）方面均有顯著提升。特別是，結合**自適應均衡（Adaptive

Equalization）**預處理與CSE_ASPP模組的配置，對於難以識別的中型異常檢測達到了最高準確率。此外，優化後的模型保持了較低的計算開銷，適用於即時工業應用場景。
關鍵字：布料異常檢測，YOLOv8，紡織行業，缺陷布料

This study aims to enhance the performance of fabric anomaly detection with an emphasis on medium-sized defects, leveraging the YOLOv8 algorithm. The detection of fabric defects, such as holes, tears, and irregular lines, is a critical quality assurance task in the textile industry. Defective fabrics, if undetected, can lead to substantial economic losses and inefficiencies in the supply chain. Early and accurate defect detection prevents defective materials from progressing through production, reducing waste, minimizing downtime, and maintaining product quality. YOLOv8, the latest iteration in the YOLO (You Only Look Once) family of algorithms, offers significant advancements over earlier versions like YOLOv5, YOLOv7, and YOLOv9 in terms of detection accuracy, computational efficiency, and real-time performance. This study builds upon these strengths, further refining the model through architectural enhancements, hyperparameter tuning, and the application of advanced data preprocessing and augmentation techniques. Key innovations include the integration of modules such as Channel Squeeze and Excitation (CSE) and Atrous Spatial Pyramid Pooling (ASPP), which improve feature extraction by enhancing spatial and channel-wise attention. These modifications are complemented by anchor-free detection mechanisms, enabling the model to handle varying defect scales and textures more effectively. The experimental results demonstrate the superiority of the optimized YOLOv8 model, which achieved significant improvements in Precision, Recall, and F1-Score across multiple configurations and preprocessing techniques. Notably, the combination of Adaptive Equalization preprocessing and the CSE_ASPP module delivered the highest detection accuracy, particularly for medium-sized anomalies that are challenging to identify. Additionally, the optimized model maintained low computational overhead, making it suitable for real-time industrial applications.

List of Contents

1.	Introduction	1
2.	Relative Work	3
2.1 Image Watermarking	3
2.2 Diffusion Models	4
2.3 Watermark in Diffusion Models	5
3.	Methodology	8
3.1 Problem Statement and Objectives	8
3.2 Algorithm Framework	8
3.3	Proposed Method	12
3.4 Preliminaries of Diffusion Models	15
Forward Diffusion Process	15
Reverse Denoising Process	16
Pre-training Watermark Decoders	16
Optimization Objective	17
Robustness Against Transformations	17
4.	Experiment Results	18
4.1 Evaluation Metrics	18
Frechet Inception Distance (FID)	18
Structural Similarity Index (SSIM)	18
Peak Signal-to-Noise Ratio (PSNR)	19
Trace metrics	20
4.2 Experiment Setting	22
4.3 Implementation Details	22
4.4 Result	23
5.	Conclusion and Future Work	37
5.1 Conclusion	37
5.2 Future Work	37
References	38

 
List of Illustration
Figure 1: A conventional Image Watermarking Model [13].	3
Figure 2: Diffusion Models Architecture [4]	4
Figure 3: Fingerprinting in Diffusion Models Architecture.	5
Figure 4: WaDiff Diffusion model Architecture [16]	6
Figure 5:  Recipe for Watermarking DMs in Different Generation Paradigms. [10]	7
Figure 6: Illustration of Watermark Bits	9
Figure 7: Illustration of Concatenation.	9
Figure 8: Our’s Propose Architecture.	12
Figure 9: Our’s Propose Workflow.	13
Figure 10: Performance of WaDiff, across different stages of training.	24
Figure 11: Performance of Ours-hybrid, across different stages of training.	25

 
List of Tables
Table 1: Encoder Fingerprint Upsample Shape.	10
Table 2: Ground truth contain binary arrays representing pre-generated watermark patterns.	11
Table 3: Comparison of Three Models Across Various Aspects.	14
Table 4: SSIM Table of Two Model Comparison Across Different Training Steps.	26
Table 5: PNSR of Two Model Comparison Across Different Training Steps.	27
Table 6:  FID of Two Model Comparison Across Different Training Steps.	28
Table 7: Time Training of Two Model Comparison Across Different Training Steps.	29
Table 8: WaDiff and Our Time Generation.	30
Table 9:  Watermark identified the true fingerprint of two model.	31
Table 10: Noise attack identified the true fingerprint of two model.	32
Table 11: Adjust brightness attack identified the true fingerprint of two model.	32
Table 12: Random mask attack identified the true fingerprint of two model.	33
Table 13: Overall performance under different attack conditions.	34
Table 14: Performance Summary Table Across Two Models.	35

References

[1] 	J. Redmon, S. Divvala, R. Girshick and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," Computer Vision and Pattern Recognition, pp. 779-788, 2016. 
[2] 	J. Torres, "YOLOv8 Architecture: A Deep Dive into its Architecture," YOLOV8, 15 01 2024. [Online]. Available: https://yolov8.org/yolov8-architecture/#google_vignette.
[3] 	H. Jie, S. Li, A. Samuel, S. Gang and W. Enhua, "Squeeze-and-Excitation Networks," Computer Vision and Pattern Recognition, 2018. 
[4] 	L. C. Chen, P. George, K. Lasonas, M. Kevin and Y. L. Alan, "Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs," Computer Vision and Pattern RecognitionDeepLab: Semantic, 2017. 
[5] 	G. Jocher, "YOLOv5 Documentation Ultralytics," Ultralytics, 2020. [Online]. Available: https://docs.ultralytics.com/models/yolov5/.
[6] 	L. P. Paul, "Medium," Squeeze-and-Excitation Networks, 18 10 2017. [Online]. Available: https://towardsdatascience.com/squeeze-and-excitation-networks-9ef5e71eacd7.
[7] 	GeeksforGeeks, "Understanding the Confusion Matrix in Machine Learning," GeeksforGeeks, 8 July 2024. [Online]. Available: https://www.geeksforgeeks.org/confusion-matrix-machine-learning/.
[8] 	Y. Guo, X. Kang and J. Li, "Automatic Fabric Defect Detection Method Using AC-YOLOv5," Advances in Computer Vision and Deep Learning and Its Applications, vol. 12, no. 13, 2023. 
[9] 	Z. Huanhuan, M. Jinxiu, J. Junfeng and L. Pengfei, "Fabric Defect Detection Using L0 Gradient," Fabrication, 2019. 
[10] 	T. Mahmud, S. Juel, C. J. Rana and F. Jannat, "Fabric Defect Detection System," Advances in Intelligent Systems and Computing, vol. 1324, 2021. 
[11] 	K. Rahimunnisa, "Textile Fabric Defect Detection," Journal of Innovative Image Processing, vol. 4, no. 3, pp. 165-172, 2022. 
[12] 	C. Y. Wang, H. Y. Bochkovskiy and M. Liao, "YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors," Computer Vision and Pattern Recognition, 2022. 
[13] 	A. Mehra, "Evolution of YOLO Object Detection Model From V5 to V8," LABELLERR, 06 06 2024. [Online]. Available: https://www.labellerr.com/blog/evolution-of-yolo-object-detection-model-from-v5-to-v8/.
[14] 	J. Pedro, "Detailed Explanation of YOLOv8 Architecture," Medium, 4 12 2023. [Online]. Available: https://medium.com/@juanpedro.bc22/detailed-explanation-of-yolov8-architecture-part-1-6da9296b954e.
[15] 	VK, "YoloV8 Architecture & Cow Counter With Region Based Dragging Using YoloV8," Medium, 22 11 2023. [Online]. Available: https://medium.com/@VK_Venkatkumar/yolov8-architecture-cow-counter-with-region-based-dragging-using-yolov8-e75b3ac71ed8.
[16] 	E. A. Team, "How to interpret a confusion matrix for a machine learning model," evidentlyai, 1 October 2024. [Online]. Available: https://www.evidentlyai.com/classification-metrics/confusion-matrix.
[17] 	J. Nathapat, R. Nantachaporn and C. C. Chien, "Automatic Anomaly Mark Detection on Fabric Production Video by Artificial Intelligence Techniques," International Conference on Knowledge Innovation and Invention, 2022 . 
[18] 	C. C. Chien, H. W. Chia and S. L. Cheng, "Fast Detection of Fabric Defects based on Neural Networks," 2023 Sixth International Symposium on Computer, Consumer and Control (IS3C), pp. 322-325, 2023.