BCNN: Binary complex neural network

Binarized neural networks, or BNNs, show great promise in edge-side applications with resource limited hardware, but raise the concerns of reduced accuracy. Motivated by complex this paper we introduce representation into BNNs and propose Binary network – a novel design that processes binary inputs weights through convolution, still can harvest extraordinary computation efficiency BNNs. To ensure fast convergence rate, BCNN based batch normalization weight initialization strategies. Experimental results on image radio signal classifications achieve better accuracy compared to original BNN models. improves strengthening its learning capability extending applicability complex-valued input data. Our code is available at https://github.com/flying-Yan/BCNN.