DML-TSCRT Algorithm for Long-Tailed Distributed Anchor Damage Event Recognition in Submarine Cables

In submarine cable safety monitoring, anchor damage events are rare, leading to a long-tailed distribution that challenges event identification. To address this issue, this study introduces the Deep Mutual Learning-Time Series Classifier Re-Training (DML-TSCRT) model, leveraging Fiber Bragg Grating (FBG) sensing technology., and proposes a deep mutual learning-time series classifier re-training (DML-TSCRT) model algorithm. This end-to-end, high-performance model features low computational cost and integrates time-series data augmentation, classifier re-training, and mutual learning. The effectiveness and superiority of the proposed method are validated through comprehensive experiments and comparative tests. In the training set with a head-tail sample quantity ratio of 12:1, the best accuracy reached 97.69<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula>, surpassing the baseline model, bidirectional gated recurrent unit (Bi-GRU), trained on the complete training set with an accuracy of 97.30<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula>. Its accuracy was also significantly improved compared to other models in the cases of 24:1 and an extremely small-sample head-tail ratio of 60:1. This study provides a new solution for accurately identifying submarine cable anchor damage events in small and extremely small-sample scenarios, thereby enriching the theory and methods of small-sample event identification.