Implementação de redes neurais artificiais utilizando VHDL para aplicações em reconhecimento e classificação de padrões

Abstract

The scope of this paper is the development of artificial neural network in hardware, to solve problems of pattern recognition and classification, using VHDL. Two different networks were implemented, with the objective of showing two topologies and different activation function. Firstly, the project was developed and trained in software using MATLAB®. After the training of both networks and the acquisition of the parameters of weight and bias, started the implementation in hardware, using the same processing flow as in software. The circuit description of both ANNs was made using the trough parallel concept. After the implementation, simulations and comparative tests were made between the sequential implementation in MATLAB® and the parallel in VHDL. The two described networks had the same hit rate as those in software, but with a faster execution time and with little hardware resource utilization.