Time-Dependent Image Classcifiaction Based on Spiking Neural Network

As one of the important directions in the development of artificial intelligence, spiking neural networks have drawn a lot of interest in the domains of brain-like computing and neuromorphic engineering. To solve the problems of poor generalization and large memory and time consumption of spiking neural networks, This research suggests a spiking neural network-based time-dependent picture categorization technique. Firstly, The introduction of an online training method over time makes up for the kinetic energy lost during gradient descent; secondly, a spatio-temporal synaptic connection algorithm is fused to enhance the network's capacity to handle information effectively; and finally, to improve the network's capacity to catch the significant features in both the channel and spatial dimensions, a convolutional attention mechanism is incorporated. The experimental findings demonstrate that the enhanced method's training memory occupation across the three datasets of CIFAR10, DVS Gesture, and CIFAR10-DVS is reduced by 45.65%, 42.61%, and 10.46%, consequently, and the training speed is increased by 2.72, 1.33, and 2.76 times, in that order, ensuring that the network's performance progressively improves as accuracy increases.