Radon single-pixel flying target classification via texture-fused lightweight differentiable operators.

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Radon single-pixel imaging (SPI) offers rapid sampling and long-range imaging for applications like drone monitoring and airport security.
• Low sampling rates in Radon SPI enhance speed but degrade image quality, challenging object recognition.
• Existing SPI classification methods use shallow networks, limiting their discriminative power for complex tasks.

Purpose of the Study:

• To develop an efficient deep learning model for Radon SPI classification at ultra-low sampling rates.
• To improve the accuracy of identifying fast-moving aerial objects in challenging imaging conditions.
• To leverage Radon SPI characteristics and integrate traditional image processing techniques into a deep learning framework.

Main Methods:

• Utilized state-of-the-art lightweight classification models as a foundation.
• Integrated traditional texture and line-filtering operators into differentiable modules.
• Designed and optimized a novel classification model specifically for Radon SPI data.

Main Results:

• The proposed model achieved the highest Top-1 accuracy on a custom Radon SPI flying target dataset.
• Demonstrated superior performance compared to existing state-of-the-art lightweight classification models.
• Validated the effectiveness of integrating prior knowledge of Radon SPI characteristics into deep learning.

Conclusions:

• Advanced deep learning models, optimized with domain-specific features, can overcome the limitations of low sampling rates in Radon SPI.
• The developed model shows significant promise for enhancing aerial object classification in security and monitoring applications.
• The source code will be publicly released to facilitate further research and development.