SpikeYOLO-NDFEB: Integrating NeuroDynamic Enhancement and Dynamic Pruning for Advanced Energy-Efficient Spiking Object Detection

Spiking Neural Networks (SNNs) offer energy efficiency and bio-plausibility advantages over traditional Artificial Neural Networks (ANNs). However, their application in complex tasks like object detection is limited by inadequate feature extraction, insufficient temporal processing, and high computational demands. To address these challenges, we present SpikeYOLO-NDFEB, which integrates a novel NeuroDynamic Feature Enhancement Block (NDFEB) and employs Dynamic Pruning and Adaptive Pruning (DPAP) techniques. Experiments on our proprietary dataset demonstrate that Spike YOLO-NDFEB significantly improves detection performance and computational efficiency compared to the baseline SpikeYOLO, while maintaining high accuracy. SpikeYOLO-NDFEB effectively enhances SNNs for real-time, energy-constrained applications such as auto-nomous driving and edge computing.