A Hybrid Model Integrating CNN–BiLSTM and CBAM for Anchor Damage Events Recognition of Submarine Cables

Abstract

Strain and acceleration signals are essential for accurate event recognition along submarine cables. However, concise identification still poses challenges for the prompt recognition and classification of anchor smashing and hooking events with multiple types of sensors and multiple locations. For these reasons, this paper proposes a hybrid model that combines convolutional neural networks-bidirectional long short-term memory (CNN-BiLSTM) and a convolutional block attention module (CBAM) to instantaneously identify and organise anchor smashing and hooking events. Eight different categories of data were selected as data samples and collected from multiple Fiber Bragg Grating (FBG) strain sensors and accelerometers at four different locations. The results demonstrated that the recognition best accuracy of the method could reach 98.95%. This method has a better identification rate than the existing schemes used for the same purposes, confirming the validity and reliability of the proposed hybrid model.