Gabor Filter-based Event Stream Feature Enhancement and Event Camera Object Recognition
Gabor filter-based event-driven convolution is a widely used feature extraction method in bio-inspired hierarchical spiking neural networks for event camera object recognition.To enhance the feature extraction performance of event camera ob-jects,a Gabor filter-based event stream feature enhancement algorithm was proposed and applied to a spiking neural network event camera object recognition system that used reward-modulated STDP rule.The proposed algorithm first segmented event streams in-to fixed-length fragments,then extracted features from each fragment and enhanced features in fragments with high event density.The system used reward-modulated STDP rule to help the spiking neural network learn diagnostic features.The proposed algorithm outperformed the spiking neural network without the event stream feature enhancement algorithm on the MNIST-DVS dataset,and can classify short event stream input data effectively.The proposed event stream feature enhancement algorithm improves the fea-ture extraction and classification performance of Gabor filter-based event-driven convolution for event camera object recognition.
NETL
NSTL
万方数据
维普
