Application of UAV and Deep Learning in Geological Survey–以辽宁兴城和甘肃北山地区为例

Author:Sang Xuejia

Supervisor:Xue Linfu


Degree Year:2018





The field geological survey work method has not been greatly improved,and it still stays in the field to explain the degree.The main reason is the lack of sufficient data collection methods to preserve the complex and diverse data in the field.This also limits the introduction of new technologies in geological survey work to a certain extent.As a new thing in recent years,drones have been introduced by more and more industries.The drone is a flexible,lightweight,terrain-independent,very good near-surface remote sensing data acquisition load platform.The maturity of photo modeling technology represented by SFM and MVS extends the application field of drones to the field of 3D modeling.Not only can it create beautiful 3D models,but its spatial relationships are also preserved intact.It can be said that the combination of drone and photo modeling technology can meet the requirements of geological field data collection.It can generate a variety of data such as 3D models,3D point clouds,orthophotos,and can provide very high resolution.However,as the amount of data increases,the analysis of data is also facing difficulties.Traditional data analysis methods are very bad for high-resolution images.It is also necessary to introduce artificial intelligence algorithms into the data processing flow.Artificial intelligence algorithms,especially deep learning algorithms,have developed very rapidly in the field of pattern recognition in recent years.Taking AlphaGo as an example,it achieves the victory of Li Shishi at a very fast speed through self-training and ultra-fast learning efficiency,and The AlphaGo Zero version completed the learning process from zero to over AlphaGo in just 36 hours.Artificial intelligence algorithms,especially deep learning algorithms,are well suited to solve the problem of massive data and massive judgment logic.As a result,photo-charging technology enables drones to generate more useful data,and artificial intelligence is a key technology that is indispensable for processing these data.These three complement each other.This thesis mainly studies the key technologies of photo modeling technology,drone and artificial intelligence in geology.Mainly from the data collection methods,data processing methods,data analysis methods and application cases to illustrate,these three technologies promote the digitalization of geological survey work.The main contents of this article are as follows:(1)Through the experiment,this paper has obtained a set of workflows for the use of drones for geological data collection.In this process,site surveys of the target area are required prior to flight to understand local terrain and airspace conditions.And prepare a number of emergency landing sites in advance for future use.Then,using drones to collect data on geological outcrops of different scales,geological data with different resolutions and different coverage ranges can be obtained.They are used to represent regional scales of surface fluctuations and rock formations,geological phenomena of outcrop scales,and fine models of hand specimen scales.In addition,ground surveys to confirm difficult lithology are also indispensable.(2)This paper introduces the main processing methods of geological data of drones,including image preprocessing,slicing,extraction of attitude parameters,SFM and MVS.After the flight of the drone is completed,through the processing of these algorithms,various geological thematic data such as three-dimensional models,three-dimensional point clouds,and orthophotos can be obtained.(3)This paper introduces several geological analysis methods,including CNNs algorithm,super pixel classification algorithm,analog flow direction algorithm,Kd-tree index algorithm,color difference algorithm and so on.These algorithms can effectively convert data generated by drones into secondary data that is easy to interpret.Under the premise of ensuring speed,these algorithms have better guarantees.(4)In this paper,in Xingcheng,the application of CNNs-SLIC algorithm in large-scale geological mapping,the application of simulated flow algorithm in conglomerate granularity statistics,the application in 3D structural analysis and GOCAD modeling application.(5)The SLIC-CNN method was applied in the Taili waterfront of Xingcheng City,Liaoning Province,China.In this study,the ability of the deep neural network to locate irregular objects was enhanced by combining the SLIC algorithm’s ability to identify texture edges to help geological filling.Figure.In addition,experts can modify the fusion rules to speed up the mapping process and improve the accuracy of geological mapping.The classification of the algorithm in this paper is 88.7% in Xingcheng Taili,Liaoning,and Kappa test result is k=0.8523,which provides a high accuracy of rock mass identification,which can greatly reduce the field geological survey work,and part of it.Replacing the work of manual mapping in the industry.(6)Using the Kd-tree algorithm to cluster the three-dimensional point cloud,the structural information extraction of the geological outcrops of Xingcheng Jiashan and Baimiaozi was realized.And experimented with three-dimensional plotting and vectorization on the 3D model,and imported the results into GOCAD software.(7)Experiments on the application of drones in green mineral geological surveys,and compared the differences in the degree of environmental damage caused by drone geological surveys and traditional geological surveys.It is concluded that the disturbance of the environment by the drone’s field geological survey is about one-twentieth of the traditional method.(8)We have set up a virtual field geological system.Combining many beautiful models collected and produced in the field,it realizes a geological system that is completely different from the previous virtual geoscience system.It not only provides a high degree of immersion,but also provides a platform for digital preservation of vulnerable geological outcrops.It can also be used as a course for geological teaching and geological internships.Through the above research,we have the following conclusions:(1)This paper proposes a geological survey method flow with drone,photo modeling and artificial intelligence improvement.It can be seen from the experiment in Xingcheng Taili Beach,Jiashan,Baimiaozi and other places.It works well.(2)This paper proposes a SLIC-CNN method,which can simplify the geological mapping process,and its automatic mapping accuracy rate is up to 88.7%,which can greatly reduce the internal labor of large-scale mapping and issue in less time.A large-scale geological sketch with a high classification accuracy.(3)The three-dimensional data generated by the UAV’s high-resolution image and photo modeling technology can help the geological survey to obtain more comprehensive information on the occurrence,lithology and structure,which is of great significance for the study of geological evolution.(4)Using the drone and photo modeling technology to record and preserve the multi-scale geological conditions in the field in a three-dimensional way,it can preserve the wild geological scene with rich details.(5)The virtual field geological system constructed using the real-life 3D model can provide a more immersive user experience than the traditional virtual system.(6)The use of drones in field geological surveys can significantly reduce the environmental disturbances of geological surveys.From the research point of view,taking the situation of motor vehicle rolling in geological survey work as an example,using drone-assisted geological survey can reduce soil and vegetation damage by more than 20 times.