Key Technologies Of Data Processing And Applications Of Single-wavelength Airborne Bathymetric LiDAR

As the junction of land and sea,the coastal zone is an important space foundation for supporting the development of the national marine economy.Affected by land and sea interaction and human activities,this area has unique environmental characteristics and dynamic changes.Therefore,coastal surveying and mapping has always been the focus and difficulty of surveying and mapping tasks.Airborne Li DAR Bathymetry(ALB)technology is a new type of marine measurement technology developed in the past 60 years,which can be combined with other measurement technologies to form complementary advantages.With obvious advantages such as high efficiency,high precision,strong maneuverability,operational safety,and integration of sea and land,ALB has quickly become an ideal choice for coastal surveying and mapping.Therefore,key technologies of data processing and applications of airborne bathymetric Li DAR is of great significance to meet the urgent needs of oceans,surveying and mapping,water conservancy,transportation,and navy.This paper uses the full waveform data recorded by ALB as an entry point to study the key technologies of ALB data processing,and integrates multi-source data for coral reef identification research.The main research content and related conclusions are as follows:(1)The dissertation first briefly describes the research background and significance of airborne bathymetric Li DAR data processing and application.The current research status of ALB technology,including the development of relevant equipment at home and abroad,development of data processing technology and existing problems are summarized.The research contents and the organization structure of the dissertation are determined.(2)The dissertation introduces the system composition,principle of working geometry and full waveform data of the ALB system,and summarizes the error sources.(3)Aiming at the problem of single-wavelength ALB signal detection,the paper proposes a new signal detection method,FWST,based on summarizing the limitations and deficiencies of existing methods.The waveform types are refined through statistical analysis,and a novel waveform classification model based on the deep convolutional neural network framework,which learns the local and global features of the full waveform data and fully combines context information,is constructed to achieve fast and accurate echo signals recognition.Aiming at the problem of signal detection,it is divided into two tasks: waveform segment identification and signal detection.The integrated OS-SAPSO-ELM algorithm is proposed to improve the classification performance of the ELM classification model through Online Sequential learning and parameter optimization,and to recognize the target waveform fragments.A flexible peak detection combination model is constructed to realize the task of waveform signal detection in different scenes.The experimental results demonstrated that compared with the Generalized Gaussian model optimized by the Levenberg-Marquardt algorithm(LM-GGM),the new model has higher signal detection accuracy while reducing the time cost by 3 times.(4)Based on the actual characteristics of airborne bathymetric Li DAR data,a coarseto-fine ALB strip mosaic approach based on weight distribution is proposed.In order to avoid sacrificing the accuracy of the land point cloud in the process of searching for the global optimum,the data is divided into water and land regions,and different registration models are constructed for the data in different regions.The deviation between strips guides the distribution of weights to solve the uncontrollable limitation in registration.The RANSAC-ICP registration model based on weight assignment is adopted for land areas,and the ICP-improved non-rigid ICP registration model based on weight assignment is constructed for underwater areas.In order to weaken the interference of outliers to registration,a topological constraint term of spatial structure is added to the loss function of non-rigid ICP.Experiments show that the new model can better eliminate the correction residual errors to obtain high-precision digital terrain products.After strip registration,the RMSE reaches 0.12 m and the maximum error is0.36 m,which performs better than the traditional ICP registration model.(5)In order to realize the fusion of island and costal zone point clouds and satellite images,a registration method(FFAR)of ALB and satellite images based on feature fusion is proposed,which is capable of solving the problem of inaccurate geometric correction of satellite images due to the limitation of no control or less control.In addition,the model can well solve the registration problem caused by single feature,unobvious feature,and fuzzy underwater features in the scene.On the basis of fully taking into account the underwater feature information,a feature fusion model based on the deep learning framework and the pyramid model is constructed to fuse the advantages of different levels of data sources.The cluster analysis is fused by the SIFTRANSAC coarse registration model to eliminate interference features,and improve the correct rate of corresponding points.Then,a new approach fused with modal transformation and range constraints is proposed to weakens the problem of local overregistration and achieve precise registration.The proposed registration method obtained a good performance in Wuzhizhou Island ALB data and Worldview-2multispectral image registration experiment.Experiments show that the RMSE of the new model is 2.21 m,which is better than the traditional SIFT-RANSAC registration model(RMSE is 5.91m),fully verifying its effectiveness and advantages.Finally,the images of Wuzhizhou Island in 2013 and 2019 are automatically registered by new model,and the shoreline and building changes are analyzed and summarized.(6)The dissertation summarizes the advantages and limitations of different remote sensing technologies for coral reef identification and proposes a coral reef identification model that integrates multi-source data including ALB,MBES and multi-spectral image data.Based on the research results of(3)-(5),high-precision and seamless terrain data and registration results are obtained.The unique features corresponding to different data sources are extracted and an improved random forest model is constructed to determine the optimal model parameters and feature combinations through iterative calculations to achieve accurate and efficient identification of coral reefs.Taking Ganquan Island as the research area for experiments,the accuracy of coral reef identification by fusion of multi-source data reaches 94%,and the accuracy of using multi-spectral image data or terrain & waveform features are 83% and 79%,respectively.