Automatic target detection in forward-looking infrared imagery via probabilistic neural networks

Area of Science:

• Computer Vision
• Artificial Intelligence
• Signal Processing

Background:

• Forward-looking infrared (FLIR) imagery presents challenges for automatic target detection due to sensor limitations and environmental factors.
• Existing methods often struggle with distinguishing targets from complex background clutter in FLIR data.
• Accurate target identification is crucial for various applications, including defense and surveillance.

Purpose of the Study:

• To develop and evaluate an automated technique for target detection in FLIR imagery.
• To improve the accuracy and efficiency of target recognition by effectively rejecting background clutter.
• To validate the proposed method using real-world FLIR datasets.

Main Methods:

• Application of mathematical morphology for the initial identification of regions of interest (ROI).
• Development of a clutter rejection module utilizing a probabilistic neural network (PNN).
• Training the PNN with both target and background features, employing ROI shifting for robust classification.

Main Results:

• Demonstrated excellent classification performance in distinguishing targets from clutter.
• Validated the effectiveness of the proposed clutter rejecter module on real-life FLIR imagery.
• Achieved high accuracy in automatic target detection within challenging infrared scenes.

Conclusions:

• The presented technique offers a robust solution for automatic target detection in FLIR imagery.
• The integration of mathematical morphology and a PNN-based clutter rejecter significantly enhances detection performance.
• The method proves effective in real-world scenarios, highlighting its practical applicability.