High-frame-rate target detection method based on deep pulse neural network

The invention relates to a high-frame-rate target detection method based on a deep spiking neural network. A YOLOv3-Tiny neural network model is converted to obtain a spiking neural network; on the channel dimension of each layer of the spiking neural network, performing normalization processing on the weight and the bias through the maximum activation value; setting positive and negative activation thresholds of neurons by using a signed neuron method with unbalanced thresholds, and selecting neurons needing to be activated in the spiking neural network; based on an STDP unsupervised learning training method, training the optimized pulse neural network to obtain a high-frame-rate target detection model; tensor decomposition and quantitative compression are carried out on the high-frame-rate target detection model to realize lightweight processing; and inputting a to-be-detected image into the lightweight high-frame-rate target detection model, and outputting a target category and a target boundary position c