In the industrial Internet of Things, deep learning-based multimodal image damage detection has been widely used in production quality inspection and safety assurance. Aiming at the problems of low real-time performance, high cost and incomplete coverage in current subway tunnel monitoring methods, this paper presents an improved LSGAN model to generate GPR images of tunnel damage and integrate a detection algorithm to identify tunnel defects, so as to realize high-precision detection of underground structural targets and concealed diseases.To resist the risks of network attacks and data tampering in subway tunnel damage detection systems, an ISSA‑BiLSTM‑based network security situation prediction model is proposed to predict the system security state in a fixed future period and provide decision support for network managers.Experimental results show that the proposed damage detection method achieves an average accuracy of over 70%. In terms of network security situation prediction, ISSA‑BiLSTM has higher prediction accuracy than CNN‑DBO‑BiLSTM‑Attention, CNN‑IPSO‑BiLSTM‑Attention, IPSO‑BiLSTM‑Attention and BiLSTM models.
