Target recognition-oriented method and device based on residual pulse neural network

The invention relates to a target recognition-oriented method and device based on a residual pulse neural network, and the method comprises the steps: inputting an obtained input image into a designed lightweight target recognition network, carrying out the pulse coding of the input image through a pulse coding layer, and obtaining a pulse feature map; secondly, extracting trunk feature maps with different sizes and hierarchical structures from the pulse feature map through a trunk network, finally, selecting different trunk feature maps from the trunk network through a detection head with two groups of different independent prediction branches, fusing the different trunk feature maps with a middle feature map of the detection head, and performing prediction output by using the fused feature maps. And the prediction result is screened and filtered by using a non-maximum suppression method to obtain a target recognition result of the input image, so that the target recognition performance is greatly improved.