Materials Classification And 3D Reconstruction Of Ancient Building Wood Components Based On Hyperspectral LiDAR

Wooden building is one of the traditional Chinese ancient buildings.Due to the natural environment,human damage and other factors,a lot of damaged wood components need urgent protection.In order to preserve the original information of ancient buildings as much as possible,the principle of minimal intervention should be adopted to collect and extract information on wood components.Therefore,non-contact technology is essential to obtain wood component information for ancient buildings conservation.This thesis uses hyperspectral LiDAR(HSL)to obtain spectral-space point cloud data of wood components,HSL hyperspectral data are used to study the wood component material of species,moisture content,and moldy processes.For the problem of sparse distribution of point clouds obtained by HSL,3D scanner point clouds and HSL data are fused to obtain dense point clouds containing multi-channel spectral information.The fusion data are used to classify wood components with different materials or different moisture contents,and achieve the 3D reconstruction of classification results.The main work of this thesis as follows:(1)Improving adaptive band selection(ABS)algorithm,the improved ABS algorithm combined with random forest(RF)for species classification of wood component materials.The HSL hyperspectral data of six different species of wood component materials were collected.The feature bands were selected by the improved ABS algorithm,which were input into RF classifier to classify species.The classification accuracy of the improved ABS was compared with RF classification based on different spectral bands.Due to wood components are susceptible to moisture and mold,HSL hyperspectral data of the same sample in moldy process(normal-moist-mold)for spectral analysis and states classification.The results show that HSL hyperspectral data can classify species of wood component material and states of the moldy process,and the improved ABS algorithm can improve the classification accuracy of species from 88% to 98%.(2)The competitive adaptive reweighted sampling-successive projections algorithm(CARS-SPA)and partial least squares regression(PLSR)were used to predict the moisture content of wood component materials.Experimentally,300 sets of HSL hyperspectral data with different moisture contents were collected,the CARS-SPA algorithm was used to select the feature bands as input parameters for PLSR to establish a moisture content prediction model,which was compared with the PLSR models based on CARS or SPA.The results show that HSL hyperspectral data can predict the moisture content of wood component materials,and the model of CARS-SPA-PLSR has the best prediction results,which can reach 90.(3)Proposing a 3D reconstruction method based on spectral classification.Firstly,the data from HSL and 3D scanner are fused by principal component analysis-iterative closest point(PCA-ICP)algorithm and Euclidean distance nearest point method.Then the feature bands are selected by the improved ABS algorithm and CARS-SPA algorithm,and the RF algorithm is used for wood components classification.Finally,the classification results are mapped to 3D coordinates by different color labels,and3 D reconstructions of the wood component are based on classification results.The spectral and point cloud data of three wood components with different materials and two wood components with different moisture contents were collected,the results show that this method can classify wood components of different materials or different moisture contents,and achieve the 3D reconstruction of the classification results.Figure[36] Table[4] Reference [58]...