Spikenet Research And Its Application In Image Recognition

With the development of image processing technology, image recognition and intelligence technology have become more and more important. Research on modern machine vision focuses on two aspects:the rate of recognition and the speed of recognition. In modern recognition technique, neural networks simulate the process of brain thinking with math methods, which represent image recognition technique in future.This paper researches the SpikeNet which was represented the next generation of Artificial Neural Networks. SpikeNet is a neural simulator for modeling large networks of integrates and fire neurons. It uses simple integrate-and-fire neurons which undergo step-like changes in membrane potential. With rank order coding, SpikeNet uses networks of asynchronous spiking neurons to process the information, which can improve the speed of processing so that it can recognize object in real time.The main contributions of this paper are as follows:(1)The architecture that we used was a very simple layer feed-forward neural network. Each layer was composed of a set of maps of different selectivity. The principle of the learning method is based on the intrinsic properties of integrate-and-fire neurons and Rank Order Coding.It results in that a neuron will fire a spike when the relative order of firing in its inputs will best match the order of the corresponding connections.(2)We use the face recognition and face positioning as an example to describe the image-oriented design method of SpikeNet. Among them, in the face positioning experiments, we use four-layer feed-forward network and use the image pixel value as the reference weight. And in the face recognition experiments, adopting the method of network training to get the weight by using three-layer feedback-forward network. In the related experiments, introducing this two network designing respectively and the experimental processes and results.(3)Aimed at the needs of identifying the road signs, this article applies SpikeNet to road signs classification and recognition, in the system design, using two sets of SpikeNet network. The former is mainly used for road sign image classification, through classification, the images was get into the respective neural network to identify. The main purpose of doing so is to improve the image recognition rate. The internal patterns are more similar among many images, and therefore need to first come through the identification of shape classification of road signs, that is, to judge it as warning signs, or the ban signs, or indication signs, and then through the mask to identify singly wiping off the same appearance. The latter one repeat the work of the former network, get the recognition results. The difference between the process of identification and classification is the final weights are made according to different sample training. So the recognition result is different.In summary, this article describes SpikeNet thought in the application of image processing methods and strategies, and for the first time to apply SpikeNet thought on constructing the road sign recognition system. We have 84 kinds of features of road signs in different light and noise, a total of 1680 pictures have been used in a series tests, and experimental results prove that our system can well overcome the impact of contrast and noise, to obtain comparative ideal experimental results.