Fractional Fourier transform pre-processing for neural networks and its application to object recognition

Area of Science:

• Signal Processing
• Artificial Intelligence
• Machine Learning

Background:

• Neural networks are powerful tools for signal analysis.
• Traditional Fourier transforms are widely used for signal pre-processing.
• Object recognition and position estimation from sonar data present challenges.

Purpose of the Study:

• To investigate the use of fractional Fourier transform (FFT) for pre-processing input signals to neural networks.
• To demonstrate the effectiveness of FFT in enhancing neural network performance for sonar data analysis.
• To explore the impact of the FFT order parameter on recognition and estimation accuracy.

Main Methods:

• Applying the fractional Fourier transform with a variable order parameter to raw sonar return signals (amplitude and time-of-flight).
• Utilizing neural networks for object recognition and position estimation tasks.
• Comparing the performance of neural networks with and without FFT pre-processing.

Main Results:

• Fractional Fourier transform pre-processing led to improved neural network performance.
• Reduced errors were observed in both object recognition and position estimation.
• The choice of the fractional Fourier transform order parameter significantly influenced the results.

Conclusions:

• Fractional Fourier transform is a valuable technique for pre-processing signals for neural networks.
• This method offers a promising approach for enhancing sonar-based object recognition and position estimation.
• Further research into optimizing the FFT order parameter can yield even greater performance gains.