Infrared Small Target Detection Based on Multiscale Kurtosis Map Fusion and Optical Flow Method.

Area of Science:

• Computer Vision
• Infrared Imaging Technology
• Signal Processing

Background:

• Small infrared target detection is challenged by variable target sizes and complex backgrounds, leading to poor performance.
• Existing methods struggle to effectively enhance targets while suppressing background noise and clutter.

Purpose of the Study:

• To develop a robust and accurate algorithm for small infrared target detection in complex natural scenes.
• To improve target representation by integrating shape, size, and motion information.
• To enhance target detection performance beyond current state-of-the-art methods.

Main Methods:

• A novel structure combining multiscale kurtosis maps and optical flow fields for infrared small target detection.
• A multi-scale kurtosis map fusion strategy with a novel weighting mechanism to enhance targets of varying sizes and suppress noise.
• An improved optical flow method utilizing scale confidence parameters for optimal neighborhood selection and clutter suppression.

Main Results:

• The proposed method effectively enhances small targets across different scales while suppressing noise and background edges.
• The integrated optical flow method further refines detection by suppressing residual clutter and improving target integrity.
• Experimental results demonstrate superior performance compared to seven state-of-the-art methods on complex infrared scenes.

Conclusions:

• The combined multiscale kurtosis map fusion and optical flow method offers a significant advancement in small infrared target detection.
• The algorithm's ability to handle varying target sizes and complex backgrounds leads to enhanced detection accuracy and robustness.
• The method provides superior subjective visual effects and objective evaluation metrics (BSF, SCRG, ROC).