Related Experiment Video
Updated: Aug 9, 2026
07:13
Early Detection of Cyanobacterial Blooms and Associated Cyanotoxins using Fast Detection Strategy
Published on: February 25, 2021
SMN: Signal Modulation Network for Tiny Object Detection in Remote Sensing Imagery
Summary
This study introduces a novel Signal Modulation Network (SMN) to improve tiny object detection (TOD) in remote sensing. The SMN effectively addresses foreground-background signal imbalance, enhancing accuracy for small objects.
Area of Science:
- Computer Science
- Remote Sensing
- Artificial Intelligence
Background:
- Tiny object detection (TOD) in remote sensing is challenging due to weak foreground signals and background interference.
- Existing methods struggle with foreground-background signal modulation imbalance (FBSMI).
Purpose of the Study:
- To propose a novel Signal Modulation Network (SMN) to mitigate FBSMI for improved remote-sensing TOD.
- To enhance the detection accuracy of very tiny and tiny objects.
Main Methods:
- The proposed Signal Modulation Network (SMN) incorporates an adaptive Wiener filter modulator (AWFM) to suppress noise and preserve target signals.
- A denoising diffusion transformer (DDT) operates in feature space, expanding local representations for weak object evidence.
Main Results:
- SMN effectively mitigates the foreground-background signal modulation imbalance (FBSMI).
- Experiments on benchmark datasets (AI-TOD, SODA-A, DOTAv2.0, DIOR-R) show improved detection accuracy, especially for tiny objects.
- SMN outperforms state-of-the-art methods in tiny object detection.
Conclusions:
- The Signal Modulation Network (SMN) offers a robust solution for tiny object detection in remote sensing.
- The combination of AWFM and DDT effectively addresses signal imbalance and enhances detection performance.
