SSVEP Harmonic Fusion for Improved Visual Field Reconstruction with CNN

Steady-state visually evoked potentials (SSVEPs) occur due to a repetitive visual stimulus, which results in periodic responses from the visual cortex at the stimulus frequency and its harmonics. Prior studies show that the fundamental SSVEP frequency response can be used to produce a visual reconstruction of what is shown to the human eye. However, due to interference coming from the source and the sensing device, the resulting captured image contains salt-and-pepper noise and random value noise. This study investigates whether information present in the SSVEP harmonics is useful in denoising and enhancing the captured visual reconstruction. The proposed convolutional neural network architecture methods are compared against the SSVEP fundamental and naive additive reconstructions. The results show that combining harmonics and reconstructions from different signal processing methods into the neural network architecture enhances the resulting image.