Colour Detection Using Memristive Neural Network

Abstract

The research on memristors and memristive devices significantly increased after the physical realization of memristor in 2008 at HP Labs. Its potential can be utilized in different applications such as memory technology and analog and digital logic circuit implementations. Since the memristor has the ability to act as a synapse, it has many uses in the field of neuromorphic engineering and machine learning. In this research work, memristor based neural network has been used for colour detection which is usually a primary stage in most of the image processing applications such as road signs detection, face detection, skin colour detection, object detection etc. The memristor bridge synapse which consists of four memristors is very effective in implementing the weights of neural networks at hardware level. SPICE is one of the best software available to simulate very large scale integration circuits. In this report we have built memristor based neural network architecture in HSPICE. Multiple simulations are performed to select the suitable weight update pulse of the memristor bridge synapse. The results of these simulations are discussed in detail in this report. The ideal memristor model is used to implement the memristor bridge synapse. We first normalized the data set of Red, Green and Blue (RGB) values between 0 and 1. Then we trained the memristive neural network to detect the colour in the RGB space. For the purpose of comparison, we have also detected colour using artificial neural network model in python. The comparison of the memristive bridge synapse with the other existing analog synapses and the advantages of memristor bridge synapse over them is discussed as well. The power calculation of the memristive neural network is also presented in this report.