Water-land Classification Based On Waveform And Point Cloud Data Of Airborne LiDAR Bathymetry

Airborne LiDAR bathymetry(ALB)is a technique which has been proven to be an accurate,efficient,cost–effective,safe,and flexible method.The main goal of ALB is high accuracy bathymetry.ALB can also be used for retrieving of suspended sediment concentration by using laser waveforms and height biases of point clouds.ALB can realize integrated measurement of water and land,receive pulse returns of land and water,and save the laser waveforms after digitization.However,given the different properties and applications of pulse returns backscattered from water and land,it is necessary to classify the ALB waveforms.At present,there are some problems in ALB water-land classification.First,traditional classification methods using ALB waveform data offer high accuracy but present low efficiency and convenience in engineering applications.Second,the waveform methods can discriminate water and land effectively,but the accuracy is low in areas with complex water-land environment because the waveform characteristics are complex due to the interaction between the laser pulse and the environment.Point cloud elevations of inland waters and ocean waters are different.The traditional methods using ALB waveform features or 3D point cloud elevations have their own shortcomings.Thus,this article mainly includes the following two aspects: First,rapid water-land classification is realized based on ALB point cloud data.Second,the ALB waveform and point cloud data are intergrated to achieve water-land classification with high accuracy in areas with complex water-land environment.(1)A three-dimensional(3D)point cloud elevation method is proposed to realize rapid water-land classification.Firstly,based on the assumption of linear mean ocean surface,the RANdom SAmple Consensus(RANSAC)algorithm is used for rough water-land classification by using the point cloud data obtained by ALB infrared(IR)laser.Then,the water surface points obtained by rough water-land classification are used to determine the standard deviation of ocean surface and the trend model of mean ocean surface.Then,the elevation threshold of water surface points is obtained.Finally,the water-land classification is achieved by using the elevation threshold of water surface points.The method is applied to the water-land classification experiment using Optech CZMIL.In the experimental area,the overall accuracy of the proposed 3D point cloud elevation threshold method is 98.26 %.The combination of IR and green laser waveform features based on traditional support vector machine(SVM)has the highest overall accuracy of 99.11 %.In terms of classification efficiency,the time of the proposed 3D point cloud,IR laser waveform based on SVM method,green laser waveform based on SVM method,and IR and green laser waveform based on SVM method are 1.5 s,17 minutes,24 minutes,and 27 minutes,respectively.The proposed 3D point cloud method improves the efficiency of water-land classification.(2)A water-land classification method combining ALB waveform and point cloud data is proposed to realize water-land classification with high accuracy in areas with complex environment.First,the 3D point cloud elevations derived by IR laser are used as features to conduct ocean-land discrimination by using K-means clustering.Then,the features of IR and green laser waveforms are used to identify the inland waters from the land derived by the water-land discrimination.Finally,the effective classification of ocean,land,and inland waters is realized in areas with complex environment.In the experimental areas,the overall accuracy of the wavefrom culustering,point clustering,and proposed method are 93.7%,95.6%,and 99.2%,respectively.The results show that the proposed method combining ALB waveform and point cloud data improves the accuracy of water-land classification in areas with complex environment.