Three-dimensional Change Detection Of Buildings In Mining Area Using UAV Imagery And DSM

Coal resources are the main component of China ’s energy consumption.The exploitation of underground coal resources plays an important role in improving the economy and promoting scientific and technological progress.However,mining causes the movement and deformation of the overlying strata of the ore body and the corresponding ground,resulting in geological or environmental problems such as surface subsidence,cracking,landslide,collapse,and house damage.The buildings around the mining area are mostly self-built houses with brick-concrete structure,and the anti-interference ability is weak.The underground mining disturbance is easy to cause different degrees of influence on the structure of the buildings around the mining area,which poses a great safety hazard to the production activities of the mining area and the daily life of the surrounding residents.Therefore,efficient and accurate acquisition of building changes in mining areas is of great significance for environmental disaster assessment and management in mining areas.Most of the traditional building extraction methods take urban frame buildings as the research object,and cannot be directly applied to the extraction of brick-concrete buildings in mining areas.Aiming at the characteristics of buildings in mining area,this thesis proposes a three-dimensional change detection method of buildings in mining area based on UAV image and Digital Surface Model(DSM).This method can further obtain the plane change information and elevation change information of buildings in mining area on the basis of accurate extraction of buildings in mining area.Firstly,the dense matching point cloud and digital orthophoto model(DOM)of different phases are generated by UAV image three-dimensional reconstruction technology,and the normalized digital surface model(nDSM)is obtained by combining the point cloud filtering algorithm of cloth simulation(CSF)and interpolation operation to process the point cloud data,and the consistency of data coordinate datum of different phases is guaranteed by ground control points.Then,the DOM and nDSM are fused,and the classification feature selection and label making of the buildings in the mining area are completed.The random forest(RF),support vector machine(SVM)and U-Net network are used to classify and extract the buildings in the mining area,and the morphological operation is used to optimize the initial building extraction area to obtain the final result.Finally,the building change area is extracted by means of differential comparison,image feature matching and regional differential DSM analysis,and the results of building plane change and elevation change are analyzed and evaluated.The main results of this thesis are as follows:(1)A mining area building extraction method combining image and DSM height information is proposed.In this thesis,according to the characteristics of brick-concrete buildings in the mining area,the image spectral features and DSM height information are combined.RF,SVM and U-Net were used to extract buildings in the mining area before and after mining,and the extraction results of buildings in the mining area with high accuracy are obtained.Compared with SVM and U-Net classification extraction methods,RF method has higher accuracy.The accuracy,completeness and F1 score of the two extraction results are all above 90%.The highest accuracy can reach more than94%,the highest completeness is close to 93%,and the highest F1 score is about 93.6%.(2)The influence of nDSM height feature on the extraction accuracy of mining buildings is compared and analyzed.In the process of building extraction in mining areas under different time phases and different classification methods,A comparative test without nDSM height feature participation is set up and the results are analyzed.In the case of nDSM height feature participation,the accuracy of building extraction results of RF,SVM and U-Net has been greatly improved.The average increase of accuracy,integrity and F1 score is more than 20%,and the maximum increase is 26%,46% and 38% respectively.It is verified that nDSM height feature participation can effectively improve the accuracy of building extraction in mining areas.(3)On the basis of building with/without change,the 3D change detection of building plane and elevation in mining area is completed and verified.The surface of a mining area in Huaibei,Anhui Province before and after mining was selected as the observation object,and two UAV images of the experimental mining area were obtained.Based on the manual extraction results,the information of building changes in the area was extracted and analyzed.The results show that the proposed method can efficiently and accurately obtain the 3D change information of the buildings in the mining area.The detected elevation change information,displacement information and area change information are very close to the change information obtained by the manual method,which verifies the feasibility of the proposed method for three-dimensional change detection of buildings in the mining area.