Image recognition based on sparse spike neural network

Spiking neural networks (SNNs) has made great achievements in pattern recognition. Among the existing SNNs, unsupervised SNNs, especially those using spike-time-dependent plasticity (STDP), have a biological mechanism more in line with human brain cognition and are considered to have great potential in simulating the learning process of the biological brain. However, most of the existing SNNs based on STDP will be directly or indirectly used in the full connection layer, which leads to large computational overhead and over-training of the network. On the basis of predecessors, this paper uses the concurrent sparse spike neural network and achieves very good recognition accuracy when the number of synapses is reduced by nearly 1/10.