Very Efficient Convolutional Neural Network Based on the Discrete Hirschman Transform

Convolutional Neural Networks (CNNs) play a crucial role in computer vision and machine learning applications, but they are often associated with high computational demands. To tackle this challenge, researchers have turned to the Fast Fourier Transform (FFT) for spectral convolution to help reduce complexity. However, the Discrete Hirschman Transform (DHT) has emerged as a more efficient alternative for performing linear convolutions. In this study, we introduce a novel CNN methodology based on the principles of the DHT. Our experimental results highlight the impressive efficiency of this approach, significantly lowering both computational complexity and processing time. Additionally, we implement the DHT-based method in hardware to validate its performance in real-world applications, demon-strating its effectiveness in practical scenarios.