Spiking Neural Network based on YOLO-C3 for Object Detection

This research proposes a spiking neural network for object detection based on YOLOv5, in which the activation function in $\mathbf{C} 3$ module is replaced by a spiking neuron, and moved before the original initial module to better conform to the working principle of spiking neural network. Three different depths of spiking neural networks are studied, and their power consumption is theoretically analyzed. The proposed spiking neural network is tested by foreign objects detection on overhead power transmission line. Power consumption analysis and experimental results show that the proposed spiking neural network significantly reduces power consumption, while achieving detection accuracy higher than YOLOv8 and close to YOLOv5 Meanwhile, it can improve performance with increasing network depth.