水封洞库Q分值围岩分级深度学习优化

Deep Learning Optimization of Q-Value Method of Water-Sealed Cave Library

水封洞库现场实地调研发现Q值法受施工人员、环境、设备等因素影响较大,分级工作流于形式,论文利用图像处理手段对其进行优化提高分级结果准确性。首先,对水封洞库现场地质素描围岩分级结果整理,将围岩等级和岩体发育程度分别划分为四梯度,节理面强度和赋存条件分别划分为三梯度;然后,将采集到的围岩图像批量裁剪与筛选,并借用数据增强技术完成现场图像分类数据构建;最后将训练集和验证集输入到分类网络(EfficentNet_B3)中进行训练,训练结束后再对模型进行验证。

The field investigation of the water-sealed cave warehouse found that the Q-value method was greatly affected by the construction personnel, environment, equipment and other factors, and the grading workflow was in form. The paper used image processing to optimize it to improve the ac-curacy of the grading results. Firstly, the grading results of the surrounding rock of the geological sketch at the site of the water-sealed cave reservoir were sorted out, and the surrounding rock grade and rock mass development degree were divided into four gradients, and the joint surface strength and occurrence conditions were divided into three gradients. Then, the collected sur-rounding rock images are clipped and screened in batches, and the data enhancement technology is used to complete the construction of on-site image classification data. Finally, the training set and validation set are input to the classification network (EfficentNet_B3) for training, and the model is verified after training.