A High-performance Service For Potential Illegal Buildings Along High-speed Railway Detection Based On Massive Remote Sensing Images

With the continuous expansion of the scale of China’s high-speed railway network,the operation safety of high-speed railway has been widely concerned.The high-density,high-speed and high-reliability operation requirements of high-speed railway train groups make the impact of the surrounding environment on the train operation safety more and more intense,and the potential safety problems caused by the external environmental factors of the railway become more and more prominent.A large number of illegal buildings along the high-speed railway may cause problems such as track foreign body intrusion,foundation settlement and collapse,which seriously threaten the operation safety of the high-speed railway.Therefore,it is imperative to carry out regular detection of potential illegal buildings along the high-speed railway.However,high-speed rail lines generally have a wide geographical span,and the traditional manual inspection mode is time-consuming and labor-consuming,which has been difficult to meet the needs of the increasingly normalized hidden danger investigation.In recent years,advanced information technologies such as high-resolution remote sensing,deep learning and distributed computing have developed vigorously,providing a new solution for the detection of potential illegal buildings with higher efficiency and lower cost.Firstly,remote sensing images can provide effective data support for largescale and low-cost detection of illegal buildings along high-speed rail;Secondly,deep learning technology can provide effective technical means for efficient,intelligent and automatic identification of buildings based on remote sensing images along the highspeed railway.However,in the process of practical application,how to efficiently store and manage the massive remote sensing images along the high-speed railway needs to be solved,and how to improve the detection efficiency when applying the deep learning technology to such massive remote sensing images is also a challenge.The distributed storage and computing technology has the characteristics of high reliability and high scalability,which can provide an effective solution for the efficient storage of massive images along the high-speed railway and the efficient detection of potential illegal buildings.Therefore,this thesis combines the deep learning technology with the remote sensing images along the high-speed railway,and makes full use of the distributed storage and computing technology,aiming to build a set of high-performance detection services for potential illegal buildings along the high-speed railway,and provide an automated,intelligent and efficient illegal building detection scheme.Specifically,the research and results of this thesis can be summarized as follows:(1)A distributed storage scheme for remote sensing images along the high-speed railway is constructed.Firstly,the original remote sensing images belonging to massive large files are stored in the HDFS distributed file system.Secondly,an image pyramid model is constructed for the distributed slices of massive remote sensing images along the high-speed railway by combining the geotrellis geographic data processing framework and spark distributed computing framework,Then the image tiles with the same amount of data but a small single file are distributed stored in the HBase distributed database and spatial indexes are established for them,so as to realize the efficient storage and management of massive remote sensing image data along the highspeed railway.(2)A distributed detection method for illegal buildings along high-speed railway is proposed.Firstly,the deep learning model is transformed into torchscript format for algorithm integration;Secondly,the image slice in the image pyramid is used as the detection data source to avoid the high requirements of hardware configuration for directly reading the whole large-scale remote sensing image in the traditional method;Finally,based on the spark distributed computing framework,the multi node image slice data is used for parallel partition detection to realize the efficient detection of potential illegal buildings along the high-speed rail.(3)A set of high-performance detection prototype service for potential illegal buildings along the high-speed railway based on massive remote sensing images is built.Taking the detection of illegal buildings along the Zhengzhou Xi’an high-speed railway as an application scenario,the performance of distributed image storage and distributed detection of illegal buildings in the service is tested and analyzed.Experiments show that the prototype service in this thesis can improve the slicing efficiency by nearly 4times compared with Arc GIS Server,and the detection performance of illegal buildings is about 3.6 times faster than that of single node detection.The detection of potential illegal buildings within 100 meters on both sides of the Zhengzhou Xi’an high speed railway can be completed within 1 hour,and the service performance can be further improved through the horizontal expansion of service nodes.In general,the high-performance detection service for potential illegal buildings along the high-speed railway constructed in this thesis can realize the efficient storage of massive remote sensing images along the high-speed railway and the efficient detection of potential illegal buildings.To a certain extent,it can provide help for the staff at all sections of the high-speed railway to check the illegal buildings along the line,optimize the work process of checking illegal buildings,and improve work efficiency.