Yuhan Liu

Yuhan Liu

Associate Professor

MCFS - Coarse-to-Fine Detection

Published:

Method of Infrared Small Moving Target Detection Based on Coarse-to-Fine Structure in Complex Scenes
Y Ma, Y Liu, Z Pan, Y Hu
Published in Remote Sensing, March 2023
DOI | Code

Project Overview

Challenge: Detection in Complex Scenes

Infrared small target detection faces significant challenges in real-world complex scenes:

  • Heavy background clutter (clouds, buildings, terrain)
  • Low signal-to-noise ratio
  • Varying target characteristics (size, intensity, motion)
  • Real-time processing requirements

Solution: Coarse-to-Fine Architecture

MCFS employs a multi-stage detection strategy:

  1. Coarse detection: Quickly identify candidate regions
  2. Fine refinement: Precisely localize true targets
  3. False alarm suppression: Eliminate background clutter

Key Innovations

1. Multi-Stage Pipeline

Coarse Stage:

  • Fast preprocessing with large filters
  • High sensitivity to detect all potential targets
  • High recall but moderate precision

Fine Stage:

  • Detailed analysis of candidate regions
  • Sophisticated feature extraction
  • High precision target confirmation

2. Adaptive Thresholding

Scene-adaptive threshold selection:

  • Automatic background complexity estimation
  • Dynamic threshold adjustment
  • Robust to varying scene conditions

3. Spatial-Temporal Consistency

Multi-frame validation:

  • Temporal continuity check
  • Motion consistency verification
  • Reduce false alarms from noise

Method

System Architecture

Input Frame
    ↓
[Coarse Detection]
├── Background suppression
├── Quick candidate extraction
└── Initial thresholding
    ↓
[Fine Refinement]
├── ROI extraction
├── Detailed feature analysis
└── Precise localization
    ↓
[Post-Processing]
├── Temporal consistency check
├── Shape validation
└── Final output

Results

Performance on benchmark datasets:

  • High detection rate in complex scenes
  • Low false alarm rate
  • Real-time processing capability

Robust to:

  • Various background types
  • Different target sizes
  • Changing illumination conditions

Applications

  • Early warning systems
  • Surveillance and monitoring
  • Search and rescue
  • Autonomous navigation

Citation

@article{ma2023method,
  title={Method of Infrared Small Moving Target Detection Based on Coarse-to-Fine Structure in Complex Scenes},
  author={Ma, Y and Liu, Y and Pan, Z and Hu, Y},
  journal={Remote Sensing},
  volume={15},
  number={6},
  pages={1508},
  year={2023},
  publisher={MDPI}
}