Multimode Fiber Specklegram Analysis for Temperature Sensing with a Genetic-Based Evolving Neural Network

The temperature variations of the surrounding environment of Multimode Fiber (MMF) affect the specklegram pattern captured at the end of the MMF. This paper proposes the prediction of temperature by analyzing specklegrams. An Artificial Neural Network (ANN) is presented to estimate temperatures using layers and neurons optimized with genetic algorithms. The research employs a 3- and 4-layered NN architecture initially and introduces Genetic Algorithm (GA) optimization to achieve a 2-layered ANN with reduced complexity. The proposed approach has shown promising results in properly predicting temperature. This is with a test Mean Absolute Error of 0.6393, a test Root Mean Square Error of 0.8063, and a test R-squared value of 0.9459 with 2-layered ANN. The percentage improvement in the Root Mean Square Error from existing reported work 1.42° C is approximately 43.2 %. These results demonstrate the effectiveness of GA-optimized ANN in temperature sensing through MMF specklegram analysis, presenting a novel and efficient methodology for optimizing neural network architectures in optical fiber sensing applications. Furthermore, these results provide a strong indication that the proposed approach can be used to provide accurate temperature sensing over a wide range of temperatures.