Remote Sensing Interpretation Of Surface Structure Based On Machine Learning And Edge Detection Algorithm

Remote sensing,as a technical means of all-round observation of the earth,can effectively assist the field geological investigation work.With the progress of remote sensing technology,the spatial resolution and spectral resolution of remote sensing images have been greatly improved,and the research of geological information extraction and interpretation based on high-resolution remote sensing images has received wide attention.However,due to the complexity of natural features,the interpretation of geological bodies in the field is mostly done by manual visual interpretation,which consumes a lot of manpower and is not efficient at the same time,and the remote sensing geological intelligent interpretation method needs to be developed urgently.In this paper,we use Worldview-2 high-resolution stereo image pairs and Sentinel-2 multispectral satellite images to carry out a methodological study of remote sensing structure interpretation based on machine learning methods and image edge detection methods,and take the Tugerming anticline in the eastern Kuqa area,which is difficult to work in the field and relatively complex in structure,as an example.The main contents and conclusions are as follows(1)The stratigraphic classification work is carried out on the Sentinel-2 true color synthetic images of the Tugerming anticline using various remote sensing image classification methods such as maximum likelihood method,support vector machine,neural network,random forest and object-oriented respectively.We compare and analyze the classification effect,combine the relevant geological data,image classification results and visual interpretation,and complete the stratigraphic classification work,and draw the following conclusions: The stratigraphic classification still requires manual post-processing;the random forest method based on "integrated learning" is fast,with an overall classification accuracy of 78.9855%,and can reflect the boundaries of the characteristic strata;the object-oriented classification method can provide classification vector files to facilitate the visual interpretation work.(2)We extracted a high-precision Digital Elevation Model(DEM)based on photogrammetry principles and Worldview-2 high-resolution stereo image pairs,and overlaid it with Digital Orthophoto Map(DOM)to carry out 3D visual interpretation.Making full use of the rich detailed texture information of high-resolution images to interprete the stratigraphic deficiencies,unconformity contacts,Middle Jurassic to Oligocene cusps and faults exposed on the surface of the Tugerming anticline.(3)For the triangular facet,which is a faceted field geological body with significant texture information on remote sensing images,an interactive intelligent extraction method of triangular facet attitude based on image edge extraction algorithm is proposed.The triangular facet location is selected by human-computer interaction,and the rock triangular facet is extracted based on Canny edge extraction algorithm and Random Sample Consensus(RANSAC)algorithm,and the data quality is evaluated using regression.The QGIS plug-in platform of open source GIS software is used to write the intelligent extraction function module and to conduct the facet attitude extraction experiment.The following conclusions are drawn: the accuracy of the semiautomatic extraction method in this paper can reach 90.2%,which has the ability of objectivity,accuracy and data quality assessment,and the accuracy of automatic extraction is lower than that of the semi-automatic method.(4)The intelligent extraction module is used to extract the attitude information of the high and steep areas of the Tugerming anticline,and the stratigraphic classification,3D structure interpretation and attitude information are integrated to draw one remote sensing geological map and one structural section of the Tugerming anticline,and the tectonic and evolutionary history is discussed with related literature.The following conclusions are drawn: this paper's remote sensing interpretation method of surface structure based on machine learning and edge extraction methods can efficiently extract a large amount of field geological body information and provide information constraints for structure modeling.