Partitioned Convolutional Spiking Neural Network for Tactile Object Recognition

The human tactile system, which occupies the largest sensory area, has spurred active research on artificial tactile systems for applications in robotics, virtual reality, and e-skin technologies. Applying deep learning models to analyze large-scale, continuous temporal data and detect patterns in such systems is a well-established methodology. However, mimicking biological models for processing large amounts of sequential data is a potentially more economical approach. These biologically inspired models are expected to be more efficient than traditional deep learning models that rely on central processing units (CPUs) or graphics processing units (GPUs). In this study, we implemented an artificial tactile system using a partitioned spiking neural network (SNN). This approach was applied to classify objects detected using an artificial glove equipped with tactile sensors. Our method achieved higher accuracy than the existing SNN architecture. Furthermore, the reduced number of sensor readout cycles and computational requirements demonstrate the potential for faster results than conventional methods.