A 50.4 GOPs/W FPGA-Based MobileNetV2 Accelerator using the Double-Layer MAC and DSP Efficiency Enhancement