Emergency recognition method for rail transit based on TextCBAM network

Rail transit emergency command plays a crucial role in addressing unexpected incidents during operation and minimizing the impact of accidents. Traditional emergency command systems in rail transit face challenges such as communication errors caused by multi-layer information transmission and inefficiencies in emergency response plan matching. To address these issues, this paper proposes an intelligent rail transit emergency command system based on the TextCBAM network, which can accurately identify incident types from voice reports provided by the network or emergency personnel, thereby enhancing the efficiency of emergency plan matching. At the heart of this system lies an intelligent emergency recognition engine. This engine first utilizes the Whisper speech recognition engine to convert voice reports into text and then leverages the proposed TextCBAM network to classify the textual data. To evaluate the performance of the TextCBAM network in incident recognition, we compared it with the TextCNN and BiGRU algorithms. The results demonstrate that TextCBAM achieved accuracy rates of 0.9635 on the training set and 0.9550 on the validation set, outperforming TextCNN by 3.1% on the training set and 4.5% on the validation set. Compared to BiGRU, TextCBAM showed a 3.7% improvement in training accuracy and a 5.5% improvement in validation accuracy. Additionally, on the test set, TextCBAM achieved an average precision, recall, and an accuracy of 93.15%, 91.42%, and 95.33%, respectively, outperforming the next best model, TextCNN, by 4.1%, 4.4%, and 4.2%. Therefore, the TextCBAM-based rail transit emergency command system offers superior accuracy in identifying incidents reported via voice, facilitating more efficient emergency plan matching and response.