Implementation of a Multilayer Perceptron and Wavelet-Neural Network on a Microcontroller with Ultra-Low Power Consumption in Control and Signal Analysis Systems
| Autor: | Daniil V. Ermolenko, Ivan A. Bogoslovskii, Albina V. Pomogalova, Andrey B. Stepanov |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Signal processing
Microcontroller ComputingMethodologies_PATTERNRECOGNITION Wavelet Quantitative Biology::Neurons and Cognition Artificial neural network Computer science Approximation error Multilayer perceptron Computer Science::Neural and Evolutionary Computation Electronic engineering Focus (optics) Perceptron |
| Zdroj: | 2019 III International Conference on Control in Technical Systems (CTS). |
| DOI: | 10.1109/cts48763.2019.8973340 |
| Popis: | The paper has a practical focus and is dedicated to the implementation of a multilayer perceptron and wavelet- neural network on an ultra-low power consumption microcontroller MSP430G2553. A description of the implementation technologies of these neural networks is given, taking into account the limitations of the element base when solving the approximation problem. Comparison of neural networks implemented on the microcontroller showed that the wavelet-neural network allows to obtain a smaller approximation error compared to the perceptron. The comparison results also showed that with an equal number of neurons and the same number of iterations, the training speed of a multilayer perceptron is 2,15 times higher than that of a wavelet-neural network. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|