Multi-stage neural operator learning with application for convolutions