PWIoU-YOLO: A Position-Wise IoU-Optimized Detector With Multiscale Context Fusion for Aerial Oriented Objects

View at publisher Summarise with VIRAT
Authors: Shihao Chen, Junwei Yu, Hao Yang, Xinmo Zhao, Chunhua Zhu
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Accurate oriented object detection in aerial imagery is crucial for remote sensing applications, yet remains challenging due to arbitrary orientations, extreme aspect ratios, and dense distributions of objects. Existing rotated detectors often suffer from unstable training gradients and inadequate geometric representation in Intersection over Union (IoU) computation. This study proposes PWIoU-YOLO, a novel detection framework that introduces position-wise IoU (PWIoU) as the core optimization strategy. Specifically, we develop: (1) a contextual spatial pyramid pooling fast (CSPPF) module that enhances multi-scale feature representation while suppressing background interference; (2) a multi-scale channel attention mechanism (MSCAM) that adaptively fuses features across scales; and (3) the PWIoU loss, which combines the smallest oriented enclosing rectangle, probabilistic IoU, and dynamic gradient modulation to stabilize training and improve regression accuracy. Extensive experiments on DOTA-v1.0 demonstrate that our method achieves 81.2% mAP, outperforming state-of-the-art approaches by 2.5% while maintaining computational efficiency.

Fields of Study & Subjects

Advanced Neural Network Applications Advanced Image and Video Retrieval Techniques Visual Attention and Saliency Detection

Publication details

Journal / source
IEEE Geoscience and Remote Sensing Letters
Volume
23
Pages
1–5
Publication date
1 January 2026
Language
EN