|
Signal processing of distributed optoacoustic sensors by means of neural networks in the automotive transport monitoring problem
P. A. Nazarenko, S. P. Levashkin, O. I. Zakharova, K. N. Ivanov, S. V. Kushukov Artificial Intelligence Research Lab, Povolzhskiy State University of Telecommunications and Informatics, Samara
Abstract:
Neural network models are used as a tool for an automotive transport monitoring. The solution of the problem of recognition of distributed optoacoustic sensor signals generated by
vehicles using neural networks is considered. Signals features and signals preliminary
processing are described. The neural network architecture for the vehicles generated signals recognition is selected. The architecture of the network of vehicles signal recognition,
including heavy tracks signals, has single layer with two hundred and one input and one or
two outputs. The neural network can be built with the Python programming language and
Scikit-Learn, Keras and NumPy libraries. The network training images, the training results
and the trained network practical application are described. The recommendations for further research in the field of using neural networks of various architectures for recognizing
vehicle signals using distributed optoacoustic sensors are given. The study results are important for road traffic monitoring, as well as other areas of the distributed optoacoustic sensor applications.
Keywords:
artificial neural networks, perceptron, machine learning, big data space dimensionality reduction, pattern recognition, training patterns, backpropagation algorithm, Python, Keras, TensorFlow.
Received: 07.08.2023 Revised: 25.10.2023 Accepted: 04.12.2023
Citation:
P. A. Nazarenko, S. P. Levashkin, O. I. Zakharova, K. N. Ivanov, S. V. Kushukov, “Signal processing of distributed optoacoustic sensors by means of neural networks in the automotive transport monitoring problem”, Matem. Mod., 36:3 (2024), 20–34
Linking options:
https://www.mathnet.ru/eng/mm4539 https://www.mathnet.ru/eng/mm/v36/i3/p20
|
Statistics & downloads: |
Abstract page: | 144 | Full-text PDF : | 2 | References: | 23 | First page: | 14 |
|