Towards Tactile Texture Classification Using Low Spectral Resolution Analysis with Multi-Probes

This study explores two methods for classifying a tactile texture dataset: a deep learning approach using convolutional neural networks (CNNs) and a Fast Fourier Transform (FFT)-based method. The CNN model performed well with longer signals, while the FFT-based method proved effective when combined with careful bin selection to reduce dimensionality. Furthermore, multiple probes improved the FFT approach, whereas the CNN maintained high performance even with a single probe. These findings offer practical insights for developing efficient, real-time IoT solutions and assistive tools.