TCD-Net: A Task-Coordinated Dynamic Network for High-Precision Diamond Clarity Defect Detection
IEEE Access · 2026 · 14 · 104909-104921
DOI: 10.1109/access.2026.3712082 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Ke Rao und 3 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Diamond clarity is a critical indicator of the severity of internal and surface defects and serves as a key criterion for evaluating overall diamond quality. However, diamond clarity defect detection remains challenging owing to significant variations in the target scale, irregular defect distribution, complex backgrounds, and severe sample imbalances. To address these issues, a Task-Collaborative Dynamic Network (TCD-Net) has been proposed based on YOLOv11. Specifically, a C3K2_ConvFormer module was introduced to integrate local feature extraction with global context modeling, thereby enhancing the representation of minute defects. In addition, a Task-Aligned Dynamic Detection Head (TADDH) was designed to improve the detection accuracy in complex scenarios through feature interaction alignment and dynamic convolution. Furthermore, an EMASlideLoss function was incorporated to mitigate the sample imbalance by leveraging sliding weighting and exponential moving averaging mechanisms. Extensive experiments on diamond-DET, NEU-DET, and GC10-DET datasets demonstrated the effectiveness of the proposed method. Compared with YOLOv11, TCD-Net improved mAP@50 from 70.8% to 78.4%, 76.7% to 79.1%, and 71.8% to 73.9%, respectively, whereas mAP@50:95 consistently increased. Meanwhile, the parameter count is reduced from 2.58 M to 2.06 M. These results indicate that TCD-Net achieves a favorable balance between detection accuracy and computational efficiency with a strong generalization capability for complex industrial scenarios, and our self-built Diamond-DET dataset is publicly available at https://ieee-dataport.org/documents/diamond-det
Zitieren
APA
Rao, K., Zhao, F., Wang, Z., & Zhang, K. (2026). TCD-Net: A Task-Coordinated Dynamic Network for High-Precision Diamond Clarity Defect Detection. IEEE Access, 14, 104909-104921. https://doi.org/10.1109/access.2026.3712082
Vancouver
Rao K, Zhao F, Wang Z, Zhang K. TCD-Net: A Task-Coordinated Dynamic Network for High-Precision Diamond Clarity Defect Detection. IEEE Access. 2026;14:104909-104921. doi:10.1109/access.2026.3712082
BibTeX
@article{Rao2026,
author = {Rao, Ke and Zhao, Fengxia and Wang, Zhaoxu and Zhang, Kunpeng},
title = {TCD-Net: A Task-Coordinated Dynamic Network for High-Precision Diamond Clarity Defect Detection},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {104909--104921},
doi = {10.1109/access.2026.3712082}
}
EUVIMED