为解决土压平衡盾构掘进过程中存在的操控手检测渣土流塑性费时费力、受主观状态影响较大的问题，提出一种渣土流塑性自动检测方法。首先通过裁剪、恢复、增强等技术对实时监控视频图像进行预处理，构建改良渣土样本数据集，然后建立CSD （conditioned soil detection）网络模型，识别、分类监控视频中渣土流塑性，实现改良渣土流塑性的自动识别。基于实际案例，提出的改进Faster R-CNN网络方法，该方法检测3类改良渣土的平均正确率达91.55%，平均精确率达86.83%，平均召回率达87.42%，检测效果较好，将该模型运用在地铁其他线路中均有较好的检测效果，具有较好的应用推广价值。
The flow plasticity inspection by the worker is time-consuming
and labor-intensive during earth pressure balance
shield tunneling; thus, the results are heavily influenced by subjective state.
As a result, an automatic method for detecting the flow plasticity of the
conditioned soil is proposed. First, preprocessing techniques, such as cutting,
recovering, and strengthening are used to construct a conditioned soil sample
data set from real-time monitoring video images.
Second, a conditioned soil detection network model is established to identify
and classify the flow plasticity of the conditioned soil in the monitoring
video, enabling automatic identification of the flow plasticity of the
conditioned soil. Finally, the recommended method is applied in a project, and
the proposed improved Faster R-CNN neural network
method shows a good detection effect in three kinds of conditioned soils. The
average accuracy, precision, and recall rates are 91.55%, 86.83%, and 87.42%,
respectively, indicating that the recommended method should be widely used.
