Dice-Tversky-Based Instance Segmentation of Small Defects in Automotive Sealing Strips

Ye Zhu

doi:10.6919/ICJE.202609_12(9).0030

Authors

Ye Zhu ShanDong University of Science and Technology, Qingdao 266590, China

DOI:

https://doi.org/10.6919/ICJE.202609_12(9).0030

Keywords:

Automotive Sealing Strip; Small-defect Segmentation; Instance Segmentation; YOLOv8s-seg; Dice-Tversky Loss.

Abstract

Small surface defects in automotive sealing strips, including cracks, indentations and local surface anomalies, usually occupy only a small foreground area and have fragmented boundaries. Conventional mask losses can therefore be dominated by background pixels, leading to missed small regions or insufficient boundary fitting. To address this issue, this study improves YOLOv8s-seg by introducing Dice loss and Tversky loss into the original BCE mask loss, forming a Dice-Tversky mask overlap constraint. The Dice term strengthens region-level overlap supervision, whereas the Tversky term adjusts the relative costs of false positives and false negatives. Experiments were conducted on the automotive sealing strip surface defect dataset ASSD using four settings: baseline, Dice-only, Tversky-only and Dice-Tversky. Compared with the baseline, Dice-Tversky improved Mask P, Mask R, Mask50 and Mask50-95 by 2.1, 1.3, 2.8 and 10.1 percentage points, respectively. The largest gain appeared in Mask50-95, indicating that the proposed constraint mainly improves high-IoU mask fitting for slender and small-area defects.

Downloads

Download data is not yet available.

References

[1] Yang, H. Y. (2025). Design of a tire X-ray image defect detection system based on YOLOv8 and PyQt5. Computer and Network, 51(3), 265–274.

[2] Zheng, M. J., Pan, Q., & Xu, H. Y. (2025). Design of chip packaging defect detection system based on deep learning. Computer Measurement & Control, 33(8), 45–53.

[3] Zhao, Y. H., Li, Z., Jin, S., et al. (2026). Rapid detection of photovoltaic defects based on improved YOLOv8. Acta Energiae Solaris Sinica, 47(3), 584–593.

[4] Ren, S. Q., He, K. M., Girshick, R., et al. (2017). Faster R-CNN: towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6), 1137–1149. https://doi.org/10.1109/TPAMI.2016.2577031

[5] Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 779–788). IEEE.

[6] He, K. M., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask R-CNN. In Proceedings of the IEEE International Conference on Computer Vision (pp. 2961–2969). IEEE.

[7] Božič, J., Tabernik, D., & Skočaj, D. (2021). Mixed supervision for surface-defect detection: from weakly to fully supervised learning. Computers in Industry, 129, 103459. https://doi.org/10.1016/j.compind.2021.103459

[8] Milletari, F., Navab, N., & Ahmadi, S. A. (2016). V-Net: fully convolutional neural networks for volumetric medical image segmentation. In 2016 Fourth International Conference on 3D Vision (pp. 565–571). IEEE.

[9] Zhang, H. C., Zheng, S., Zhu, Y. H., et al. (2023). Surface defect detection of wooden boards based on multi-modal fusion. IEEE.

[10] Lin, T. Y., Maire, M., Belongie, S., et al. (2014). Microsoft COCO: common objects in context. In European Conference on Computer Vision (pp. 740–755). Springer.

[11] Huang, J. J., Jiang, X. J., & Liu, Y. (2024). CAGK-YOLO: an efficient detection model for micro-scale and irregular surface defects based on deep learning. IEEE.

[12] Sun, H. Y., An, J. X., Zhang, Z. S., et al. (2025). Based on YOLOv11 steel surface defect detection. IEEE.

[13] Liu, K. H., Chen, S. J., & Liu, T. J. (2020). Unsupervised UNet for fabric defect detection. IEEE.