Research On Tne Identification Method Of Dam Derects Variation Based On Multi-temporal UAV-borne Images

By the end of 2022,China had built more than 98,000 reservoirs and 430,000 kilometers of various river embankments.Cracks and seepage in dams pose a serious threat to their safety and long-term operation.UAV aerial photogrammetry has emerged as the preferred method for rapid dam inspection owing to its high efficiency,flexible operation,and high resolution of captured images.Based on multi-temporal images collected by periodic UAV inspections,automatic detection of apparent defects such as cracks and seepage in dams and their variations trends is essential to ensure the safe operation of dams.However,due to the fluctuations in UAV flight,variability in inspection paths,and changes in dam environment,defects in multitemporal UAV images may show significant deviations in angle,position,size,etc.This makes it difficult to capture variations in defects through intuitive visual comparison and existing automatic identification methods.Therefore,the study on accurate and efficient dam defects variation identification method has become an urgent need for dam safety detection and health status assessment,which has important academic significance and practical engineering value.The main research work of this paper is as follows:(1)In order to address the challenges of effectively identifying variations in cracks due to their small size,complex backgrounds,and varying perspectives in visible images collected by UAVs,this study proposes a multi-temporal dam crack registration and variation identification method based on object-aware enhancement.The method designs a regional focus module and introduces the Spatial Transformer Layer to extract crack features from multi-temporal UAV images and predict their spatial mapping relationships.Furthermore,a data augmentation technique based on Poisson blending and random projective transformations is investigated,and a multi-temporal dam cracking dataset is established.Finally,a comprehensive comparison is conducted to validate the effectiveness of the proposed method for identifying dam crack variation.(2)Seepage,which can be obscured by vegetation and challenging to identify in visible images,is further complicated by its mobility,unclear boundaries,and interference in infrared images,making it difficult for existing methods to accurately align and identify seepage variation with limited spatial feature points.To overcome these challenges,a multi-temporal dam seepage registration and variation identification method based on multi-modal image is proposed.The method designs a multi-modal feature fusion module and constructs a dualchannel transformation parameter prediction network to extract local and global spatially invariant features of multi-temporal seepage defects and predict their spatial mapping relationships.Finally,a multi-temporal dam seepage dataset is established,and a comprehensive comparison is conducted to verify the effectiveness of the proposed method for identifying dam seepage variation.(3)A management and control platform for intelligent inspection of dam defects by UAV has been developed based using PyQt5 and PyCharm.This software platform is designed with modules for function selection,data transmission,model training,and variation identification.It enables automatic and intelligent identification of defects variation,as well as visualization of the operation process and identification results.As a result,the efficiency of dam defect inspections has been significantly improved.