A target detection method and system based on a pulse neural network

Abstract

A method for target detection based on pulse neural network, including three steps of image preprocessing, training of multi-scale feature fusion pulse neural network with channel weight adjustment module and input of processed image data into trained weight model for target detection. The multi-scale feature fusion pulse neural network is composed of up-sampling and convolution pulse neural network, receives feature vector sequences of different scales from a backbone network, connects the feature vector sequences processed in different scales through a splicing process, and then uses a convolution pulse neural network to perform further fusion. The channel weight adjustment module is composed of a global average pooling layer and a convolution pulse neural network, and is based on the information correlation between adjacent time steps and the global features of each time step, and adjusts the channel weight of the spliced features by combining the channel attention mechanism. The method can accurately detect targets in images with low parameter quantity by combining the characteristics of the pulse neural network.

Images