Dual-Region Search Detection

Published:

An Infrared Small Moving Target Detection Method Based on Dual-Region Search in Complex Scenes
H Cao, Y Hu, Z Wang, J Yang, G Zhou, W Wang, Y Liu
Published in Remote Sensing, January 2025
DOI


Project Overview

Challenge: Target Diversity

Complex scenes contain diverse target types:

  • Fast moving jets
  • Slow hovering drones
  • Boats with wake patterns
  • Ground vehicles with different signatures

Single strategy cannot handle all types effectively.

Solution: Dual-Region Strategy

Adaptive search based on target characteristics:

  1. Region A: Fast targets, simple backgrounds
  2. Region B: Complex backgrounds, dim targets
  3. Dynamic allocation: Automatic region selection

Key Innovations

1. Automatic Region Classification

Scene analysis:

  • Background complexity estimation
  • Target motion prediction
  • Automatic region assignment

2. Specialized Detectors

Region A detector:

  • Optimized for speed
  • Motion-based detection
  • Low computational cost

Region B detector:

  • Deep analysis capability
  • Background suppression
  • High sensitivity

3. Intelligent Fusion

Decision-level fusion:

  • Combine results from both regions
  • Resolve conflicts intelligently
  • Optimize overall performance

Method

System Architecture

Input Image
    ↓
[Scene Analysis Module]
├── Background complexity
├── Target characteristics
└── Region assignment
    ↓
[Parallel Detection]
├── Region A Detector (fast)
└── Region B Detector (deep)
    ↓
[Result Fusion]
├── Conflict resolution
├── Confidence weighting
└── Final output

Results

Comprehensive evaluation on diverse datasets:

  • Handles 5+ target categories
  • Adapts to 10+ scene types
  • Superior to single-strategy methods

Applications

  • Multi-threat detection: Various aerial targets
  • Maritime surveillance: Ships, boats, swimmers
  • Traffic monitoring: Mixed vehicle types
  • Urban security: Complex city environments

Citation

@article{cao2025dual,
  title={An Infrared Small Moving Target Detection Method in Complex Scenes Based on Dual-Region Search},
  author={Cao, H and Hu, Y and Wang, Z and Yang, J and Zhou, G and Wang, W and Liu, Y},
  journal={Remote Sensing},
  volume={17},
  number={2},
  pages={323},
  year={2025},
  publisher={MDPI}
}