Abstract In order to improve the recognition efficiency of tunnel drainage hole images based on traditional image detection, the artificial intelligence semantic segmentation method of the U-Net++ neural network model is used to process the blockage images of tunnel drainage holes, and the evaluation indicators of siltation degree are introduced to accurately classify drainage holes with different degrees of siltation. The results show that the loss curve of the proposed method shows a good convergence trend, and the accuracy on both the training and validation sets shows a stable improvement trend. It outperforms other commonly used image segmentation methods in terms of accuracy, recall rate and F1 score, reaching 96%, 95% and 95%, respectively. This model also performs well in terms of IoU and Dice coefficients, reaching 91% and 95% respectively. Furthermore, the model has certain adaptability to changes in lighting and different noise environments, and still performs well in different scenarios. The proposed intelligent recognition method based on U-Net++ neural network has shown high levels in terms of accuracy, robustness and adaptability, providing an effective solution for the detection of blockage of tunnel drainage holes.
Abstract：
In order to improve the recognition efficiency of tunnel drainage hole images based on traditional image detection, the artificial intelligence semantic segmentation method of the U-Net++ neural network model is used to process the blockage images of tunnel drainage holes, and the evaluation indicators of siltation degree are introduced to accurately classify drainage holes with different degrees of siltation. The results show that the loss curve of the proposed method shows a good convergence trend, and the accuracy on both the training and validation sets shows a stable improvement trend. It outperforms other commonly used image segmentation methods in terms of accuracy, recall rate and F1 score, reaching 96%, 95% and 95%, respectively. This model also performs well in terms of IoU and Dice coefficients, reaching 91% and 95% respectively. Furthermore, the model has certain adaptability to changes in lighting and different noise environments, and still performs well in different scenarios. The proposed intelligent recognition method based on U-Net++ neural network has shown high levels in terms of accuracy, robustness and adaptability, providing an effective solution for the detection of blockage of tunnel drainage holes.
WANG Yaodong DU Yaohui GAO Yue
.Detection Method for Blockage of Tunnel Drainage Holes Based on U-Net++ Network[J] MODERN TUNNELLING TECHNOLOGY, 2023,V60(4): 76-85
2023, Vol. 60 