A Memory Reduction with Neuron Pruning for a Binarized Deep Convolutional Neural Network: Its FPGA Realization