Application Of Memristor-based Spiking Neural Network In Image Edge Extraction

With the development of modern bio-technology and electronic technology, the human’s demand of brain-like, efficient and intelligent information processing system continues to grow. Whether it is in the fields of intelligence, information processing, automatic control, or in the fields of computer science, robotics, pattern recognition, artificial neural network has been widely used. Currently, the third generation of artificial neural network--spiking neural network is one of the most popular intelligent algorithms, and the spiking neural network has become an important topic of current research to solve complex real-world intelligence. Memristors are like synapses and can be manipulative. They are small and have a natural ability to store information, now known closest synaptic function circuit elements. It is ideally suited as electronic synapses of a neural network system. By combining the memory characteristic of memristors with high-efficient processing ability of spiking neural network, a memristor-based and bionic intelligent spiking neural network structure is constructed, which is expected to improve information processing capabilities of neural network system.This paper studies the basic theory and the nature of spiking neural network and memristors, focusing on the physical model of the HP memristor and their similarity with biological synapses and deduces the formula of memristors about updating weights as synapses. With memristors as synapses, the most basic is based on non-linear electrical properties of synaptic transmission which has similar characteristics. Through the structure comparison and theoretical analysis, it confirms that memristors can effectively achieve STDP learning rules in biology. By simulating biological synapses with memristors according to the function and principle of biological visual system, a three-layer spiking neural network model for image edge extraction is constructed, in which the image edge information is represented by the variation of the memristor conductance. The edge extraction result with this approach has the characteristics of continuity, smoothness, low false leak detection and edge positioning accuracy. Since the processing mechanism of this neural network conforms to the biological counterpart, it offers a new idea for the bionic implementation of biological visual system.