| Industrial storage tanks are widely used in petroleum,chemical industry,metallurgy,and process industries for storing raw materials,finished products and intermediate products.They are essential for ensuring industrial safety production and product storage.In addition,the spatiotemporal distribution information of industrial storage tanks is of great significance for monitoring industrial development and environmental monitoring in China.With the recent advancement of remote sensing technology,computer vision and image processing,it is possible to automatically detect industrial storage tanks from optical remote sensing images.Using satellite remote sensing technology to detect industrial storage tanks at urban scale has the advantages of large-scale,periodic,and effective monitoring.This thesis mainly studies the automatic detection of industrial storage tanks in urban areas using deep learning based algorithm,aiming to help people manage and monitor the urban environment and resources.As the industrial storage tanks in the whole city have the characteristics of large volume and dispersed distribution,both the detection scope and the detection difficulty are a great challenge to the detection process.To address this challenge,this research aims to improve the accuracy of industrial storage tank detection in urban-scale by utilising high-resolution remote sensing images and deep learning algorithms.The first step in this thesis is to construct a city-scale industrial storage tank object detection dataset based on high-resolution remote sensing images.This dataset explores the effect of deep learning object detection algorithm optimization on industrial storage tank detection at urban scale was explored on the constructed data set of industrial storage tank.By comparing and analyzing the mainstream object detection algorithms,and combining techniques such as Hough transform,Res Net50 network,FPN and NMS,the detection accuracy of a single industrial storage tank in different cities is further improved,enhancing the robustness and detection performance of the model as a whole.Moreover,analysis of the experimental results shows that traditional deep learning algorithms are not well-suited for detecting ground objects in urban scale remote sensing image.To improve the accuracy of the detection process,this thesis constructs multiple urban industrial storage tank regional datasets based on high-resolution remote sensing images.The datasets are enriched by combining POI data and land use data of each city,creating an urban spatial multi-source data fusion dataset.By combining with the urban spatio-temporal multisource data fusion dataset with a regional detection model of industrial storage tank based on the combination of convolutional neural network and random forest,the research provides an accurate and efficient solution for detection industrial storage tanks in urban areas.Finally,this thesis develops a complete urban-scale industrial storage tank detection process by combining the optimized deep learning object detection algorithm with the regional detection model.The effectiveness of the model is validated through experiments.Builing upon the research presented,a prototype system for urban-scale industrial storage tank detection platform has been developed.The platform includes functions of managing urban spatiotemporal datasetand training models for industrial storage tank detection in remote sensing images at the urban scale.It is demonstrated that the prototype system of urban-level industrial storage tank detection platform can support the research of industrial storage tank detection in remote sensing images at urban scale,and has good practical application value. |
