Related Experiment Video
Updated: May 6, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.0K
HWANet: A Haar Wavelet-based Attention Network for remote sensing object detection
Baohua Jin1, Fukang Yin1, Wenpeng Cai1
1School of Computer Science and Technology, Zhengzhou University of Light Industry, Zhengzhou, China.
Plos One
|September 4, 2025
Summary
This study introduces HWANet, a novel Haar wavelet-based attention network for remote sensing object detection. It effectively handles scale variations, achieving high accuracy with fewer parameters.
Area of Science:
- Computer Vision
- Machine Learning
- Remote Sensing
Background:
- Remote sensing object detection (RSOD) faces challenges with scale variations.
- Current deep learning methods lose information during downsampling and lack context awareness.
Purpose of the Study:
- To propose a novel network, HWANet, for improved RSOD.
- To address information loss and enhance context modeling for multi-scale objects.
Main Methods:
- Developed a Haar wavelet-based Attention Network (HWANet).
- Introduced Low-frequency Enhanced Downsampling Module (LEM) to preserve object information.
- Integrated Haar Frequency Domain Self-attention Module (HFDSA) and Spatial Information Interaction Module (SIIM) for context-aware multi-level feature integration.
Main Results:
- HWANet achieved 93.1% mAP50 on NWPU VHR-10 and 99.1% mAP50 on SAR-Airport-1.0.
- The model demonstrates superior performance with only 2.75M parameters.
- Outperformed existing state-of-the-art methods in RSOD.
Conclusions:
- HWANet effectively mitigates information loss during downsampling using Haar wavelets.
- The network enhances context-aware modeling for superior multi-scale object detection.
- HWANet offers a parameter-efficient and high-performing solution for RSOD.
