Enhanced multi-view object detection via complex group equivariant convolutional neural networks with Student’s-t loss and diffeomorphic classification
The Computer Journal · 29.06.2026
DOI: 10.1093/comjnl/bxag068 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Computer Journal, online erschienen am 29.06.2026. Verfasst von Xiaomin Liu und 4 weiteren Autor:innen. Die Arbeit nennt 76 Literaturangaben. Gefördert von: National Natural Science Foundation of China, the Science Foundation of Heilongjiang, National Natural Science Foundation Training Project of Jiamusi University.
Abstract (Originalsprache)
Object detection is a pivotal technology in intelligent perception systems, with significant applications in autonomous driving, marine monitoring, and disaster rescue. However, the geometric variations in object images captured from different angles pose challenges for local feature extraction. This paper proposes an enhanced multi-view object detection method built upon complex group equivariant convolutional neural networks, which guarantee rotation, scaling, and gray invariance in feature extraction. To align the regression loss with the object detection evaluation function, we employ the Student’s-t distribution for multi-attribute fusion-based regression loss computation. Furthermore, diffeomorphic similarity calculation is used to derive the classification loss, enhancing detection accuracy. Experimental results on the DOTA.v1.0 and HRSC2016 datasets demonstrate strong performance, with mean average precision reaching 81.37% and 90.6%, respectively.
Förderung
- National Natural Science Foundation of China
- the Science Foundation of Heilongjiang
- National Natural Science Foundation Training Project of Jiamusi University
- Doctoral Project of Jiamusi University
- Jiamusi University Horizontal Program
- Space-land Collaborative Smart Algriculture Innovation Team Project
- Jiamusi University “East Pole” Academic Team
Zitieren
APA
Liu, X., Li, S., Zhang, J., Li, J., & Zhao, H. (2026). Enhanced multi-view object detection via complex group equivariant convolutional neural networks with Student’s-t loss and diffeomorphic classification. The Computer Journal. https://doi.org/10.1093/comjnl/bxag068
Vancouver
Liu X, Li S, Zhang J, Li J, Zhao H. Enhanced multi-view object detection via complex group equivariant convolutional neural networks with Student’s-t loss and diffeomorphic classification. The Computer Journal. 2026. doi:10.1093/comjnl/bxag068
BibTeX
@article{Liu2026,
author = {Liu, Xiaomin and Li, Shangren and Zhang, Jiaqing and Li, Jun-Bao and Zhao, Huaqi},
title = {Enhanced multi-view object detection via complex group equivariant convolutional neural networks with Student’s-t loss and diffeomorphic classification},
journal = {The Computer Journal},
year = {2026},
doi = {10.1093/comjnl/bxag068}
}
EUVIMED