Cellular Neural Network on Digital Signal Processor: An Algorithm for Object Recognition

Abstract

In this article, an attempt is made to pictorially analyze certain images pertaining to partial discharge in a high-voltage insulation system. The data obtained is in pictorial form, which can be represented in its equivalent pixel value and is stored in a file. These files are categorized and labeled. A cellular neural network is used to analyze these labeled files, and the convergence obtained before and after training is tabulated. Convergence will be faster after training, and the cellular neural network learns from each sample; hence, better classification is possible after training. A digital signal processor is used to simulate the cellular neural network, as the architecture of the digital signal processor is “multiply and accumulate,” which is ideal for all matrix-based calculations. A fixed-point digital signal processor is used (TMS320C542, Texas Instruments, Dallas, Texas, USA) for the present analysis. All programming for the cellular neural network is done in C language and converted to assembly language using Code Composer Studio (Texas Instruments, Dallas, Texas, USA).