Real-time identification of borehole rescue environment situation in underground disaster areas based on multi-source heterogeneous data fusion

To reduce the risk for rescue workers to dredge the collapsed tunnel and explore the disaster area during the underground rescue, as well as improve the survival rate of trapped personnel, the paper takes borehole rescue technology as the research object, and develops a borehole rescue command and decision system based on multi-sensor fusion convolutional neural network, which realizes the detection of key information in the underground rescue. The results show that the SD, SSIM_u, EN, Q^AB/F and VIFF of human pose fusion image recognition algorithm are 90.872, 0.874, 4.892, 0.169 and 1.465, respectively, which are higher than the image fusion algorithms such as LLF–IOI, NDM, PA–PCNN, TA–cGAN and U2fuse. The multi-source heterogeneous data fusion model of borehole rescue based on deep learning could accurately identify the risk of disaster areas, with an accuracy of 98.85 %, which is 16.15 % higher than that of the feedforward neural network model and 35.26 % higher than that of the SVM model. The gas, audio, video, personnel positioning and so on four kinds of sensors, through the experiment of roadway, 16.09 % higher than that of single sensor sensing accuracy, 10 % higher than that of two sensors. The borehole rescue command and decision system has been realized real-time acquisition, transmission and online command of various sensor data in disaster rescue, and the reliability of the system has been verified by industrial applications. The study provides scientific rescue methods and system equipment for rescue, and is beneficial to ensure the safety of trapped people.