Abstract In response to the subjectivity and inefficiency issues in the analysis methods for advance forecasting images of adverse geological bodies in tunnels, this paper propose an intelligent image analysis algorithm called IR-TAG.This algorithm includes a feature encoding structure based on multi-cross attention, which effectively compensates for the inherent local induction bias of convolutional neural networks, making it difficult to fully extract global contextual information. Then, it introduces EffcinetNet-v2 as the backbone network, which has good classification performance and efficiency, to enhance the model′s ability to extract features of adverse geological bodies. The results show that in terms of detection accuracy, the mAP and F1 of IR-TAG are 84.09% and 83.63%, respectively, higher than other commonly used deep learning models. In terms of detection efficiency, IR-TAG has a smaller model size(73.5 MB) and faster image processing speed (38.87 f/s), making it suitable for intelligent and rapid detection of advance forecasting images of adverse geological bodies in tunnel construction.
Abstract：
In response to the subjectivity and inefficiency issues in the analysis methods for advance forecasting images of adverse geological bodies in tunnels, this paper propose an intelligent image analysis algorithm called IR-TAG.This algorithm includes a feature encoding structure based on multi-cross attention, which effectively compensates for the inherent local induction bias of convolutional neural networks, making it difficult to fully extract global contextual information. Then, it introduces EffcinetNet-v2 as the backbone network, which has good classification performance and efficiency, to enhance the model′s ability to extract features of adverse geological bodies. The results show that in terms of detection accuracy, the mAP and F1 of IR-TAG are 84.09% and 83.63%, respectively, higher than other commonly used deep learning models. In terms of detection efficiency, IR-TAG has a smaller model size(73.5 MB) and faster image processing speed (38.87 f/s), making it suitable for intelligent and rapid detection of advance forecasting images of adverse geological bodies in tunnel construction.
JIANG Yuan WANG Hailin CHEN Zhao
.Intelligent Image Analysis Algorithm for Advance Forecasting of Adverse Geological Bodies in Tunnels Based on Deep Learning[J] MODERN TUNNELLING TECHNOLOGY, 2024,V61(3): 148-156
2024, Vol. 61 