针对围岩风化程度判定存在主观性强、准确率低的问题，开展基于图像识别和机器学习的围岩风化程度定量化判别技术研究。依托大广复线扩容工程搜集隧道施工期间掌子面的图像，采用数字图像处理技术对掌子面图像颜色信息、表面纹理信息、节理信息进行自动化提取。基于提取的指标参数，借助机器学习方法进行指标筛选，并建立围岩风化程度预测模型。通过研究可以得到： 1）对预测模型准确率贡献度由大到小的指标分别为颜色、完整性和纹理，各单一指标预测模型准确率分别为71.91%、70.78%、47.19%； 2）多指标组合模型相较单一指标模型判定准确率更高，采用颜色、纹理、完整性3个指标的组合模型，准确率达到86.52%； 3）对13个二级指标进行特征筛选，剔除5个贡献度较低的纹理特征指标，采用剩余的8个关键指标构建预测模型，预测精度仍能达到85%，满足工程需求。研究结果表明：
掌子面图像的颜色、纹理、完整性信息能够有效表征围岩风化程度，建立在这些指标之上的机器学习预测模型能够稳定有效地对掌子面的风化程度进行判定。
Onsite determination of the weathering
degree of the surrounding rock has disadvantages such as strong subjectivity
and poor accuracy. Therefore, a quantitative identification technology based on
image recognition and machine learning is developed. Based on the images of the
tunnel working surface collected from the DaqingGuangzhou
doubleline expansion project, the digital image
processing technology is applied to automatically extract the color, texture,
and joint information of the tunneling face. Next, a model for determining the
weathering degree of the surrounding rock is established by applying machine
learning to filter the extracted parameters. Conclusions drawn are as follows:
(1) The indices contributing to the accuracy of the prediction model from large
to small are color, integrity, and texture, corresponding to the prediction
accuracies of 71.91%, 70.78%, and 47.19%. (2) Compared with the single index
prediction model, the multiindex prediction model has
a higher prediction accuracy. The prediction model with three indices of color,
texture, and integrity exhibits a prediction accuracy of 86.52%. (3) Eight key
secondary indices are filtered from 13 indices, based on which a new prediction
model is constructed with a prediction accuracy of 85%, meeting the engineering
requirements. Results show that color, texture, and integrity information can
effectively represent the weathering degree of the surrounding rock.
Additionally, the prediction model based on these indices can stably and
effectively determine the weathering degree of the tunneling face.
