Research On Monitoring Technology And Platform Construction Of Highway Slope Micro-deformation

China’s vast territory,complex terrain and geology,highway slope disasters occur frequently,bringing great challenges to transportation safety and protection of life and property.In recent years,China’s highway investment construction is still increasing,the demand for highway slope stability monitoring is increasing day by day,and the rapid deformation of cracks is an important factor leading to slope instability.At present,highway slope deformation monitoring is mainly based on manual inspection,which is costly,subjective and difficult to ensure the safety of the monitor,and the automatic detection technology is not perfect.With the continuous improvement of camera hardware quality and the development of deep learning,the deformation monitoring technology based on computer vision has become an efficient and feasible method.Based on computer vision technology,this thesis explores highway slope cracking as the research object,explores highway slope micro deformation monitoring technology,enriches highway slope deformation monitoring means,and carries out practical application in Yunnan Du Xiang Expressway,realizing the breakthrough of pushing theory to practical application.The main study and findings of this thesis are as follows.(1)A consistent resolution,multi-species,geotechnical expert-approved highway slope crack segmentation dataset SFSDS and highway slope crack classification dataset SCCDS are constructed.standard,high-quality highway slope crack dataset is an important foundation for subsequent slope microdeformation monitoring.Given that there is no open source road slope crack dataset,this thesis starts from the characteristics of road slope cracks that lead to road slope disasters,based on the road slope crack data provided by China Capital Data Co.,Ltd.and independently photographed road slope crack data,under the guidance of professional geologists for road slope crack image annotation and data enhancement,the expert-approved road slope crack image segmentation dataset SFSDS is formed.then Then,the SFSDS dataset was used as the basis for manual primary classification to form the initial road slope crack classification dataset including four types of crack images,including horizontal tension cracks,vertical cracks,mesh cracks and diagonal cracks,and migration learning was used for secondary verification classification of the dataset to form the road slope crack classification dataset SCCDS.(2)A highway slope crack segmentation network based on the channel attention mechanism is designed.Firstly,based on the constructed SFSDS dataset,the performance of classical image segmentation models UNet,Deep Labv3+,PSPNet,and DANet networks in highway slope crack segmentation is explored.Then,SFSNet,a road slope crack segmentation network,is designed by fusing crack information of different scales with the idea of jump connection and codec structure and enhancing crack feature expression by combining the channel attention mechanism.The experimental results show that SFSNet network can extract road slope crack features well,and its average cross-merge ratio is as high as 87.86%,which has high segmentation accuracy and provides a good basis for subsequent The average intersection ratio is 87.86%,which has high segmentation accuracy and lays the foundation for the subsequent research on crack deformation metrics.(3)The strategy of road slope crack deformation measurement is proposed.The severity of deformation of different types of cracks has different influence on slope stability.In view of this,firstly,the classification algorithm of road slope cracks based on external rectangular classifier is designed using the constructed SCCDS dataset to realize the classification of road slope cracks,which has good interpretability while ensuring high classification accuracy and low computational power.Then,a crack geometric parameter calculation method based on the connectivity domain analysis and a crack actual geometric parameter calculation method based on the tessellation lattice mapping are proposed to complete the measurement of crack length,width,area,actual length,actual width and actual area geometric parameters and realize the crack deformation metric,and the experiment verifies that the overall error of the calculation method is small and has high reliability.Finally,a practical application study was conducted in Yunnan Du Xiang Expressway,which showed the practical feasibility of the road slope micro deformation monitoring technology proposed in this thesis.(4)A highway slope micro-deformation monitoring platform integrating machine vision technology was constructed.The platform is developed with SSM framework and has the functions of real-time display of highway slope monitoring screen,dynamic display of crack deformation curve and historical data query,which can realize the monitoring and change tracking of highway slope micro deformation.At present,the platform has been deployed in China Capital Data Co.,Ltd.for half year trial run,and the platform has good stability.