Related Experiment Video
Updated: Aug 5, 2026
08:08
Evaluating Targeting Accuracy in the Focal Plane for an Ultrasound-guided High-intensity Focused Ultrasound Phased-array System
Published on: March 6, 2019
Cross-Frequency Context-Guided Mamba Network for Infrared Small Target Detection
Hongxin Li1, Nan Li1, Lin Tian1,2
1School of Electronic Engineering, Yili Normal University, Yining 835000, China.
Sensors (Basel, Switzerland)
|July 28, 2026
Summary
CFGMNet enhances infrared small target detection by using a novel Cross-Frequency Context-Guided Mamba Network (CFGMNet). This method effectively suppresses clutter and improves accuracy for detecting small targets in infrared imagery.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Signal Processing
Background:
- Infrared small target detection is challenging due to similar high-frequency responses between targets and background clutter.
- Existing methods struggle with amplifying clutter or diluting target details.
Purpose of the Study:
- To propose CFGMNet, a novel network for infrared small target detection.
- To address limitations of existing local and global methods by reformulating the problem as context-guided verification.
Main Methods:
- CFGMNet employs a Cross-Frequency Context-Guided Mamba module to decompose features into low- and high-frequency components.
- Mamba is used on low-frequency features for long-range dependency modeling, guiding high-frequency response gating.
- Includes local contrast gate, residual attention decoder, and decoupled prediction head for enhanced performance.
Main Results:
- CFGMNet achieves a balance between segmentation accuracy, false alarm suppression, and inference efficiency.
- Demonstrates high performance on NUAA-SIRST (85.27% mIoU, 93.77% F1) and IRSTD-1K (lowest false alarm rate).
- Achieves 141 FPS on an RTX 4090 GPU.
Conclusions:
- CFGMNet effectively suppresses target-like clutter by leveraging low-frequency background context.
- The proposed architecture offers a robust solution for infrared small target detection tasks.
- Achieves state-of-the-art results with high efficiency.
