RESEARCH ON AUTOMATED DATA PROCESSING METHOD OF PCI GEOGRAPHIC IMAGING ACCELERATOR (GXL) SYSTEM FOR BIG DATA OF SUPERVIEW-1 REMOTE SENSING IMAGE

The Chinese government attaches great importance to the development of remote sensing satellites and has now formed satellite series such as meteorological satellites, ocean satellites, resource satellites and environmental disaster reduction satellites. At the same time, many commercial space programs have been put forward or implemented in China in recent years. As the first commercial-oriented multi-means and high-resolution optical remote sensing satellite constellation in China, "Superview-1" satellite constellation is undoubtedly one of the outstanding representatives. In the implementation of digital orthophoto map project, how to deal with large data of remote sensing image in large area quickly and accurately is a difficult problem we are facing to be solved urgently. PCI geographic imaging accelerator (GXL) system plays an important role in the production of digital orthophoto image by virtue of its advantages in mass image processing. This paper briefly introduces the structure and advantages of "Superview-1" satellite constellation, introduces in detail the production process and the key technologies of "Superview-1" satellite constellation image by GXL system, and carries out research on computational efficiency comparison and production test by GXL system. In this paper, the advantages of this software system in the production of tremendous amount of satellite remote sensing image data are analyzed. At the same time, in the existing engineering practice, due to the limitation of current data scale and production mode, the software system still has some obscure shortcomings. Under the background of natural resources, the amount of remote sensing data will continue to increase substantially in the future. By analyzing these shortcomings, this paper combines the traditional production mode with the background of remote sensing data in the new era, and puts forward some reasonable and useful suggestions. 2*Corresponding author: GAO Jingnan, E-mail: jingnan Gao@spaceview.com. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15–17 November 2019, Guilin, Guangxi, China This contribution has been peer-reviewed. https://doi.org/10.5194/isprs-archives-XLII-3-W10-691-2020 | © Authors 2020. CC BY 4.0 License. 691


INSTRODUCTION
In the 1960s, remote sensing technology rose in the  In the actual image production work, we can find that different software has different performance characteristics, but the same is that the number of images processed each time is very limited. And too much rely on manual intervention, automation is far from enough, which greatly affects the efficiency of our actual image production work. The PCI geographic imaging accelerator (GXL) system plays an important role in the production of Digital Orthophoto Image by virtue of its advantages in automatic processing of massive images. This paper briefly introduces the structure and advantages of "Superview-1" satellite constellation. According to the actual needs of the image data of "Superview-1" remote sensing satellite and the special features of GXL software system. At the same time, GXL software system is used to carry out the research on calculation efficiency comparison and production test, and the advantages and disadvantages of GXL software system in the production of Orthophoto images of massive remote sensing data are analyzed. In the context of natural The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15-17 November 2019, Guilin, Guangxi, China resources, the amount of remote sensing data will continue to increase substantially in the future. Based on the analysis of these advantages and disadvantages, this paper combines the traditional production mode with the background of satellite remote sensing data of natural resources in the new period, and puts forward some reasonable and useful suggestions. Because "Superview-1" satellite has high agility, it can set multiple acquisition modes such as continuous strip, multi-strip splicing, shooting by target, and stereo acquisition. "Superview-1" can take 60 km×70 km images at a single shot. It has five core advantages:

INTRODUCTION OF
ultra-high resolution, optimized spectral band setting, wide shooting, outstanding agility and excellent acquisition ability.

Introduction of GXL system
PCI geographic imaging accelerator (GXL) system is the flagship product of Canada's PCI Geomatics Company. It integrates remote sensing image processing, GIS/spatial analysis, cartography and desktop digital photogrammetry system, and is an independent production platform. and aerial remote sensing image processing, but also in geophysical data image, medical image, radar data image and optical image processing.

THE PROCEDURE AND METHOD OF SATELLITE IMAGE MAKING
The basic idea of using GXL system to process Superview-1 satellite image production experiment is: first, control point acquisition, regional network adjustment and ortho-correction of 0.5 m high-resolution image of experimental data source, then use the corrected 0.5 m high-resolution image to register 2 m multi-spectral image, after registration is completed. The purpose of this production experiment has been achieved by image fusion and mosaic of high-resolution and multi-spectral images. But in the routine production work, we still need to continue to color, cut and slice the mosaic image. Because the follow-up work is greatly affected by human intervention, so we will not express it in this paper. Figure 1 shows the experimental process of processing "Superview-1" satellite image using GXL system.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15-17 November 2019, Guilin, Guangxi, China

Data preparation
First of all, the original data analysis is needed.
Because the original "Superview-1" satellite image GXL cannot be recognized, it needs to be transformed into GXL recognizable format through image import.
Using other programs, it is first converted into *. Pix format which can be recognized by GXL software, and then other operations are carried out. in different resolution image sources should be noted.

Block adjustment
Block adjustment is to control the relative accuracy and absolute accuracy of images by collecting the connecting points between the overlapping areas of adjacent images and calculating the adjustment of the area with the control points of the previous acquisition step.
The control point mainly affects the absolute accuracy, while the connection point affects the relative accuracy.
Because the absolute accuracy of "Superview-1" image itself is high, the matching between "Superview-1" image and reference image is generally good, so the relative accuracy of image is particularly important. Only when the relative accuracy is well controlled, can the whole large data area of remote sensing image reach a higher precision level. In the adjustment process of all adjacent images, special attention should be paid to the setting of adjustment parameters such as search radius, number of connection points (TP) and minimum score of TPs.

Ortho-rectifying
With the help of digital elevation model (

Mosaic Pretreatment
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15-17 November 2019, Guilin, Guangxi, China Orthophoto correction mainly focus on the following aspects: map projection, panchromatic pixel output size, multi-spectral pixel output size and pixel resampling type.

Image Pansharpen fusion
Due The image fusion algorithm adopted by GXL is Pansharpen, which is an original and precise algorithm of GXL. Based on the mature least squares method, it seeks the best approximate gray value relationship among the original multispectral, panchromatic and fused images to achieve the best color performance.
After integration, attention should be paid to whether the results spill over (ghosting).

Mosaic pretreatment and Mosaic
When the research area exceeds the coverage of a single remote sensing image, two or more images need to be stitched together to form one or a series.
Columns cover larger images of the whole region, and this process is image mosaic. The GXL system performs all necessary pre-mosaic processing required to produce high-quality image mosaics, including color equalization, wiring generation, image standardization and image sequencing. Actually create the final output mosaic file.

Experimental results
In the experiment of processing large data of satellite remote sensing images with GXL system, the configuration parameters of the machine running GXL Through the experimental results of GXL image processing, we can see that GXL system has obvious advantages in the process of processing large satellite image data, whether in terms of production data volume or production efficiency. There is no limitation on the number of images when GXL processes massive images. At present, the limitation is the configuration of machines and hard disk. Storage space. The greatest advantage of GXL is that it is automatic processing workflow from batch data import, automatic control point acquisition, automatic adjustment, automatic fusion, automatic orthographic, automatic mosaic, to automatic DEM extraction, and so on. It submits tasks by one key without manual operation.

Problems in Practical Production
In the actual project production application, the GXL system needs further improvement in some aspects: The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15-17 November 2019, Guilin, Guangxi, China (1) The GXL system is not flexible and convenient enough to read data format. In the data reading link, the original file format of image data source needs to be converted to. pix file. In the Block adjustment link, the geodetic coordinates of control points need to be converted to longitude and latitude format.
(2) The non-uniformity of mining points may occur in GXL automatic acquisition control points, which is sometimes caused by control data. Sometimes, when facing individual topographic features, such as mountain areas, dams and other special topographic features, the accuracy will be poor due to fewer control points, and when large-scale changes occur, the accuracy will also be poor. Be careful, Processing method can use Erdas and other software to do manual dotting, but also pay attention to controlling the timeliness of data and so on.
(3) When GXL ortho-rectification is carried out, it is impossible to expand the image according to the user's needs. Occasionally, there will be dissatisfaction with the whole scene rectification in the operation.

CONCLUSIONS
(1) Based on the existing control data, it is reliable and efficient to use GXL system distributed processing technology and automatic processing to produce Orthophoto Image Based on "Superview-1" remote sensing image, which provides a good technical support for the rapid processing of large area remote sensing big data.
(2) GXL system is a powerful image processing system. It can satisfy the automatic production requirement of big data of mass remote sensing image for users.. The WEB user interface is simple and easy to operate. It supports multi-CPU, GPU acceleration and distributed processing. For the automatic processing system of massive satellites, it is not limited by the size of satellite data, and supports massive data. According to the automatic batch generation, the automation degree is higher, and user-defined workflow can be realized. This method breaks the traditional mode of operation, greatly reduces the time of manual intervention, and makes it possible to process large remote sensing data quickly.
These advantages are incomparable to most remote sensing software at present.
(3) Prospects in the context of natural resources. With the establishment of the Ministry of Natural Resources, it is faced with natural resources such as land, minerals, forests, grasslands, wetlands, water and oceans.
Whether the production of basic data or the investigation, analysis and evaluation work, the demand for remote sensing image will increase day by day. The amount of satellite remote sensing image data that surveying and mapping engineers will face in the future will also increase geometrically. In the era of big data, there is still room for improvement in the production mode of remote sensing image. It can better serve the work of natural resources system.