In shield tunnels,water leakage often leads to short circuits in electrical systems,equipment corrosion,and structural deterioration.However,existing water leakage technologies are detected with low intelligence.Moreover,the precision and efficiency are insufficient to meet the requirements of shield safety monitoring.Therefore,this paper proposed an intelligent water leakage detection method based on point cloud intensity images.First,mobile laser scanning equipment was used to collect three-dimensional point cloud data in tunnel conditions with low light.In addition,a water leakage dataset was constructed and annotated.Then,a high-performance target detection network was designed to detect multi-scale water leakage regions.This network integrated a reconfigurable contextual information fusion module,which is specifically designed for handling multi-scale water leakage information.Then,a hard example mining loss function was explored for water leakage to enhance the capability to detect multi-scale and challenging targets.Moreover,an improved lightweight header structure was used to reduce the size and complexity of the model.Finally,through training and testing on the water leakage dataset,experimental results showed that the model achieved a high water leakage recognition rate of 93.59%,with an increase of 3.69%in the AP index compared to the original model.Additionally,the improved model reduced computational workload by 47.15%.Ablation experiments and comparative experiments further showed that the effectiveness of the proposed method is better than that of the comparison methods.Overall,this method significantly improved the efficiency and accuracy of detecting water leakage.It can be used for safety monitoring in shield tunnels.
维普
万方数据