Application of memristive device based model in image processing using CNN

Abstract

The circuit theory was governed by the three basic elements namely, resistors, inductors and capacitors until 1971. In that year, Leon Chua, a professor working at the University of California, Berkeley, noticed that a fourth element is missing. He proposed the existence of another fundamental element which would establish the relationship in between the flux and charge and thus complete the symmetry. He called this device memristor since it is a resistor with memory. Memristor is a nano dimensional device that shows distinctive I-V characteristic of pinched hysteresis loop. They were just a theoretical postulation until Dr. Stanley Williams of HP labs successfully fabricated a device which mimicked the I-V characteristics of the postulated memristor. Memristors and their devices are electrical resistance switches which have an ability to preserve a state of internal resistance depending on the previously fed voltages and currents. So, they are also called resistive random-access memory (RRAM). After this postulation in theory, a significant amount of research dedicated in realizing the memristor and its applications started. Over the last ten years, one of the attracting and happening researches is over the development of a general memristive model to simulate the nervous system of human beings since it has the ability to overcome the issues in the artificial neural network (ANN) field of study. Here, a generic, nonlinear analytical model of a memristive system has been utilized for the automatic diagnosis of COVID-19 from chest X-ray images (CXIs) using tunable Q-wavelet transform (TQWT) and convolutional neural network (CNN). Also, a new activation function has been introduced based on the window function of this model. Both of these applications have given better accuracy than the conventional methods.