Analyzing fiber specklegrams for speed and weight recognition of toy model vehicle using deep learning method

Abstract

In this paper, we present a fiber specklegram study-based application of deep learning in the field of traffic management. By leveraging the capabilities of AlexNet, a well-known deep convolutional neural network (CNN), we propose a straightforward method to classify the average speeds and weight of a toy model vehicle through multimode fiber (MMF) specklegram analysis. We have accumulated and trained the dataset comprising a myriad of specklegram images that are collected for 30 and 18 fiber bends with 632.8 nm and 532 nm wavelength of He-Ne laser respectively. We have achieved a classification accuracy of 100 % in both average speeds and weights. Furthermore, to establish the robustness of the model, we have mixed and shuffled these datasets of the aforementioned fiber bends and laser sources. We have achieved an optimal classification accuracy of 98.2 % for the average speed range and 100 % for weight classification which signifies the efficacy of the model in recognizing the speeds and weights in complex randomized situations. In addition, we have also included an analysis to classify the simultaneous weights and speed ranges of the toy vehicle where a classification accuracy of 98.8 % is achieved. This real-time efficacious study lays the groundwork for AI-assisted vehicular speed and weight detection in real-world applications for traffic management.