EUVIMEDEUVIMED entdecken
Crossref

IMPROVING OBJECT DETECTION PERFORMANCE USING VGG16 WITH TRANSFER LEARNING

Huynh Tan Loc, Duong Thanh Linh, Nguyen Thi Ngoc Anh

Tạp chí Khoa học Lạc Hồng · 25.06.2026 · 1 (26) · 70-75

DOI: 10.61591/jslhu.26.997 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Tạp chí Khoa học Lạc Hồng, online erschienen am 25.06.2026. Verfasst von Huynh Tan Loc und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Transfer learning has proven to be an effective approach to improve object detection performance, particularly when training data are limited. This study applies the VGG16 convolutional neural network with transfer learning to a custom dataset comprising 16 categories of common objects. Experimental results show that fine-tuning pre-trained ImageNet weights achieves 87.5% accuracy, with mean Average Precision (mAP) values of 74.8% (IoU=0.5) and 59.3% (IoU=0.75), outperforming the baseline model trained from scratch. Although detection of small objects improved (AP_small = 38.2%), challenges remain for very small or occluded targets. These findings highlight the effectiveness of transfer learning in academic and resource-constrained contexts, and suggest future enhancements through multi-scale feature fusion and advanced architectures.

Zitieren

APA

Loc, H. T., Linh, D. T., & Anh, N. T. N. (2026). IMPROVING OBJECT DETECTION PERFORMANCE USING VGG16 WITH TRANSFER LEARNING. Tạp chí Khoa học Lạc Hồng, 1(26), 70-75. https://doi.org/10.61591/jslhu.26.997

Vancouver

Loc HT, Linh DT, Anh NTN. IMPROVING OBJECT DETECTION PERFORMANCE USING VGG16 WITH TRANSFER LEARNING. Tạp chí Khoa học Lạc Hồng. 2026;1(26):70-75. doi:10.61591/jslhu.26.997

BibTeX
@article{Loc2026,
  author = {Loc, Huynh Tan and Linh, Duong Thanh and Anh, Nguyen Thi Ngoc},
  title = {IMPROVING OBJECT DETECTION PERFORMANCE USING VGG16 WITH TRANSFER LEARNING},
  journal = {Tạp chí Khoa học Lạc Hồng},
  year = {2026},
  volume = {1},
  number = {26},
  pages = {70--75},
  doi = {10.61591/jslhu.26.997}
}
Kostenlos starten →