A Design Methodology for Efficient Implementation of Deconvolutional Neural Networks on an FPGA