RUI: Multilayer Neural Network With Multi-Valued Neurons, its Application to Image Recognition and Processing and Incorporation of the Research Results into the Educational Process

Texas A&M University-Texarkana, Texarkana TX

Investigators

Abstract

Abstract "This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5)" The project focuses on the development of a new tool for solving pattern recognition problems. This tool is based on the application of an original artificial neural network ? multilayer neural network with multi-valued neurons. This network has a simple and productive learning algorithm, which allows to solve recognition and classification problems that are difficult to approach using other techniques. An example is multiple-class classification problems. The project involves solving multiple-class image recognition problems, such as texture classification, textural segmentation, blurred images recognition, and intelligent edge detection. The research involves study of nonlinear phenomena of a multilayer neural network based on multi-valued neurons (MLMVN). A multi-valued neuron (MVN) is a complex-valued neuron whose inputs and output are located on the unit circle. A multilayer neural network based on this neuron has a derivative-free self-adaptive learning algorithm. It outperforms other techniques in terms of training speed and classification/prediction rates. The following problems are considered in this project. The relationship between the topology of MLMVN and the quality of multiple-class classification is investigated. MLMVN is used for texture classification, textural segmentation, and as an edge detector for noisy images. The Fourier phase spectrum is used as a feature space for blurred images recognition using MLMVN. A hardware implementation of MVN and MLMVN is also considered. These studies will involve undergraduate students, who participate in scientific research as part of their education at Texas A&M-Texarkana. Hence, in addition to obtaining answers to fundamental questions related to artificial neural networks and pattern recognition, this project also is closely tied to educating students who will become scientists and engineers.