End-to-End Signal Classification in Signed Cumulative Distribution Transform Space

Area of Science:

• Signal Processing
• Machine Learning
• Data Science

Background:

• End-to-end signal classification is crucial for various applications.
• Existing methods face challenges in computational cost, data efficiency, and robustness.

Purpose of the Study:

• To introduce a novel end-to-end signal classification method.
• To leverage the signed cumulative distribution transform (SCDT) for simplified classification in the transform domain.

Main Methods:

• Utilizing a transport generative model to frame the classification problem.
• Applying mathematical properties of SCDT to simplify the problem in the transform domain.
• Employing a nearest local subspace (NLS) search algorithm in the SCDT domain for classification.

Main Results:

• The proposed SCDT-based method achieves high accuracy in signal classification.
• Demonstrates computational efficiency and data efficiency compared to existing methods.
• Exhibits robustness to out-of-distribution samples.

Conclusions:

• The SCDT-based method offers a powerful and efficient solution for end-to-end signal classification.
• The approach is suitable for real-world applications due to its accuracy and robustness.
• Implementation is available in Python via the PyTransKit software package.