A Study Of Image Processing Using Spiking Neural Networks Based On GPU

Spiking neural network(SNN) has been widely used in various fields, and has made remarkable progress. In the field of image processing, SNNs have provided a new solution for artificial intelligence.Generally, in the study of SNN algorithm, researchers usually use a serial algorithm, computing on the CPU. As spiking neural networks are highly parallel and large scale, it would take long time to simulate, despite of the increasing power of conventional CPUs. CUD A has provided a new solution for this problem.CUDA (Compute Unified Device Architecture), which is a parallel computing engine in GPUs developed by NVIDIA, supports programmers to compute parallel using GPU’s memory and run thousands of lightweight threads. CUDA can be used to compute SNNs in general PC. With the development of technology, the computational capability of GPU is dramatically improved.In this paper, the CUDA programming mode and the programming mothed have been analysed in-depth. Based on the CUDA programming concepts and methods, we studied the similarities and differences between SNN parallel model and the CUDA parallel model, and analysed the similarities and differences between the SNN’s structure and the CUDA architecture. The SNN model and CUDA model have been fused together, and an algorithm of image processing using spiking neural networks based on GPU is proposed. In succession, this paper presents two GPU implementations of SNN for edge detection and skin detection using this algorithm. For comparison, the SNN models have also been implemented using C#and C. The results showed a great speedup using the proposed algorithm. This paper provides a quick way for simulation of a large-scale SNN on a personal computer.