Multi-level information identification for civil aviation safety risks: A hierarchical multi-branch deep learning approach

Abstract

Deep learning techniques have been widely applied in the study of risk assessment and prediction in civil aviation safety, as they can effectively learn patterns and rules from aviation safety data. However, the task becomes more challenging when addressing the hierarchical structure of aviation safety risk identification. In this context, a hierarchical branching (HB) structure endows risk identification models with stepwise decision-making capabilities. This study proposes a hierarchical multi-branch deep learning approach which integrates Convolutional Neural Networks-Bidirectional Long Short-Term Memory (CNN-BiLSTM) blocks into HB to form the HB-CNN-BiLSTM (HCBL) model for identifying multi-level civil aviation safety risk information. The proposed method simultaneously facilitates safety hazards detection, hazard attribute identification, and risk level assessment, thereby capturing finer-grained risk patterns and relationships. Comparative experiments were conducted on different civil aviation safety datasets. Experimental results show that the combination is efficient and robust.