In-Class Vehicle Categorization Using Standard Convolutional Neural Networks Via Fiber Bragg Grating Sensor Array

Internet of Things is a key innovation that will shape future systems, especially in future data-hungry applications. this mostly includes sensing, communication, biomedicine as well as smart buildings, military or automotive 4.0 industry, among other surging application areas. Internet of Things leverages the fact that inherently different physical systems are combined together, alongside with modern machine learning algorithms. To date, the IoT technology is ubiquitous for automotive domain due to its profound impact on vehicle classification, instant on-road control, or smart traffic management. In this work, we evaluate a fiber Bragg grating based dataset by using established open-access convolutional neural networks (AlexNet, GoogLeNet, and ResNet-50), enabling an effective vehicle categorization. In particular, we show that individual fiber Bragg grating technological platform can offer good detection accuracies for passing vehicles using standard convolutional neural networks. In this case, we obtained classification precisions between 34% - 62% for AlexNet, 51% - 77% for GoogleNet and 57% - 78% for ResNet-50 architectures. Those detection accuracy levels are sufficiently high to pave a promising way for effective car categorization using passive fiber Bragg grating technology coupled with standard and open-access neural networks.