Mingzhou Bai

Mingzhou Bai

Auto-generated profile, based on publicly available information

Mingzhou Bai


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 June 2024 in Measurement
Verified
Journal Article
Closed access
HLHongyu Liu
Hongyu Liu
·B·mIngzhou Bai
·mIngzhou Bai
YLYanjun Li
Yanjun Li
LYLing Yang
Ling Yang
HSHai Shi
Hai Shi
YQYanli Qi
Yanli Qi

1 January 2026 in Advances in Civil Engineering
Verified
Journal Article
Open access
ZCZhuangzhuang Cui
Zhuangzhuang Cui
MBMingzhou Bai
Mingzhou Bai
YLYanjun Li
Yanjun Li
HZHongyue Zhan
Hongyue Zhan
JRJun Ren
Jun Ren
In complex metamorphic rock mountain tunnels, the evaluation of surrounding rock integrity is critical for tunnel construction. Traditional methods for identifying rock mass fractures at tunnel faces are characterized by low efficiency, heavy reliance on manual experience, and susceptibility to complex environmental conditions. This study proposes an intelligent identification and quantification method based on an enhanced deep learning model to address these limitations. A sample library comprising 1400 high‐resolution images was constructed, and multiangle shooting combined with data augmentation strategies such as geometric transformation, noise injection, and light equalization processing was employed. These techniques effectively mitigated issues related to uneven illumination, background noise interference, and diverse fracture morphologies in tunnels. To account for the small size and complex shapes of fractures, innovations were introduced by integrating residual connections and dual attention mechanisms (channel and spatial) into the YOLOv8 and Unet models. In the Unet model, residual blocks were incorporated to optimize gradient propagation paths, thereby enhancing the recognition accuracy of fracture edges. For the YOLOv8 model, anchor box parameters were fine‐tuned, and background interference was minimized through adaptive anchor box matching and cross‐scale attention mechanisms. Pixel‐level quantification of crack geometric parameters was achieved using the orthogonal skeleton line method and Delaunay triangulation network. By incorporating image scale mapping, a correlation with actual sizes was established to control the error rate. Experimental results demonstrated that the improved YOLOv8 model significantly outperformed the Unet model in complex fracture scenarios, achieving an average recognition accuracy of 78%. Notably, the recall rate for compressive fractures with widths under 0.5 mm was substantially enhanced. The introduction of image scale calibration and multiscale training strategies greatly improved the consistency between the fracture geometric parameters output by the model and actual engineering data, thereby verifying the practicality and robustness of the proposed method in complex tunnel environments.