Ministry of Education and Training HANOI UNIVERSITY OF MINING AND GEOLOGY DO THI HOAI

Transcription

1 Ministry of Education and Training HANOI UNIVERSITY OF MINING AND GEOLOGY DO THI HOAI RESEARCHING THE SOLUTIONS TO IMPROVE THE ACCURACY OF LARGE SCALE ORTHO IMAGE MADE FROM HIGH RESOLUTION SATELLITE IMAGES Department: Photogrammetry and Remote sensing Code : SUMMARY OF THESIS Hanoi, 2016

2 The thesis was completed at the Department of Photogrammetry and Remote Sensing - Faculty of Geodesy - Cartography and Land Administration, Hanoi University of Mining and Geology, Duc Thang Street, Bac Tu Liem Dictrict, Hanoi. SCIENTIFIC ADVISORS: 1. Asso.Prof.Dr Tran Dinh Tri 2. Dr. Nguyen Xuan Lam Reviewer 1: Dr Tran Van Anh Reviewer 2: Asso.Prof.Dr Pham Quang Vinh Reviewer 3: Dr Nguyen Du Khang This thesis presented at Scientific Council of Hanoi University of Mining and Geology, Duc Thang Street, Bac Tu Liem District, Hanoi. Time and date: / /2016 This thesis can be found at National Library of Vietnam Or at Library of Hanoi University of Mining and Geology

3 1 INTRODUCTION 1. Background In recent years, the high resolution satellite images have been used extensively for establishing and updating large scale map such as Geoeyes, IKONOS, QuickBird, OrbView-3, WordView-1 images... Ortho image made from remote sensing data, used for digitizing objects which are parts of the contents of map. In the fact, the high resolution satellite images has been used in Vietnam but their application results are limited, not yet widely applied in producing. The high resolution satellite images used in process and evaluated accuracy of the ortho image mostly are made by experience. As a result, the using of information on ortho image is ineffective economy and technology. So that we need to study this project "Researching solutions to improve the accuracy of large scale ortho image made from high resolution satellite images" 2. Researching purposes - To study processing technology and characteristics of high resolution satellite images; the sources of errors affect the accuracy of ortho images made from high resolution satellite image. According to that we propose the technical solutions to improve the geometric accuracy of ortho image large scale mapping by high resolution satellite images; - Improving application ability of high resolution satellite image for establishing and updating maps exactly and rapidly adapt in fact requirements; - To improve the knowledge for researching and applying for producing. 3. Researching objectives - The characteristics of high resolution satellite image; - Mathematical models using for differential original data; - The resources of error affect the accuracy of large scale ortho image made from high resolution satellite images. 4. Researching contents - Overview of researches about mathematical models for geometric correction in the world and in Vietnam; - Researching the characteristics of high resolution satellite

4 2 images; - Researching procedure of establishing ortho images from high resolution satellite images; - Researching about the factors affect geometric correction accuracy of high resolution satellite images during creating ortho images using to establish and update topographic map at suitable scale; - Implementing experiments and programming ortho image quality assessment software. 5. Methodology - Theoretical researching method: Collecting documents and finding information on the internet relate to high resolution satellite image over the world and in Vietnam. Base on that to evaluate the disadvantages of using high resolution satellite image for creating ortho image and propose solutions. These are contents of thesis. - Analysis method: Using synthetic document method, from that, choosing suitable mathematical model for solving satellite image rectification problems objectively. - Expert method: Exchanging and learning experiences of the experts in the remote sensing field in order to find optimal solutions in this research. - Experimental method: Doing experiments to clarify researching issues in this thesis. 6. Researching significance 6.1. Scientific significance - Contribute supplement and complete methodology of establishing and updating topographic map using high resolution satellite images; - Proposing technical solutions in order to improve the accuracy of ortho image made from high resolution satellite image which used for establishing and updating large scale topographic map Practical significance Proposing the solutions to improve the accuracy of ortho image made from high resolution satellite image for establishing and updating large scale topographic map. 7. Contentions

5 3 Contention 1: High resolution satellite images absolutely have ability for establishing and updating large scale maps. Contention 2: To improve economic effect and technical solutions of using high resolution satellite images for establishing and updating maps, it has to use suitable mathematical model for each type of remote sensing image, original data and type of terrains. Contention 3: Geographic database (Geodatabase) fully used as an original data to rectify satellite images. 8. The new ideas - Defining scientific basis and proposing solutions to improve accuracy of geometric correction high resolution satellite. - Propose using original data for geometric correction to increase economic effects and accuracy of ortho images made from high resolution satellite images to serve establishing and updating large scale maps. - Completing theoretical basis and building ortho image quality assessment software. 9. References - The foreign and Vietnamese documents in the fields of geodesy, remote sensing and geographical information system. - The articles, researches, reports, theses, scientific journal of photogrammetry and remote sensing. - Some experimental results of Vietnam Institute of Geodesy and Cartography, National department of Remote sensing and other agencies relate to remote sensing field. - The author s researches. 10. Structure The thesis structure includes four chapters: - Chapter 1: Overview of research. - Chapter 2: Specifications of high resolution satellite image. - Chapter 3: Procedure of ortho satellite image. - Chapter 4: Solutions for improving accuracy of large scale ortho image made from high resolution satellite image. 11. Thanks Thanks to scientific advisors Asso.Prof.Dr Tran Dinh Tri and Dr Nguyen Xuan Lam, leaders and teachers of Department of

6 4 Photogrammetry and Remote Sensing - Faculty of Geodesy - Cartography and Land Administration, Hanoi University of Mining and Geology, leaders of Vietnam Institute of Geodesy and Cartography, colleagues and my family who helped me to complete this thesis. CONTENTS OF THESIS CHAPTER 1. OVERVIEW OF RESEARCH 1.1. Researches in the world Nowadays, remote sensing techniques are strongly development. The small satellites and high-resolution (better than 1m) has been researched and developed successfully such as WorldView, QuickBird, IKONO, KOMSAT... Beside the development of image technical acquisition, high resolution satellite image processing methods are also researched and developed serving for mapping and updating topographic map, establishing thematic maps (land cover map) and another applications from high-resolution satellite image. Procedure of mapping and updating topographic map from highresolution satellite image includes following steps: - Pre-processing image: adjusting spectral reflectance values and geometric error of satellite image. - Rectifying and transforming satellite image from image coordinate system into VN2000 coordinate system. - Enhancing quality of image. - Digitizing and editing the map. Scientists in the world produced technical processes to improve geometric accuracy of satellite image base on different satellite images. The technical processes to improve accuracy of highresolution satellite image are done experiments by scientists along with appearing new generation satellites. Researching results expressed by the mean fields as: The first: Adjusting errors cause by movement and structure of the sensor which affect to the geometric accuracy of the image. The movement of satellites defined by orbital parameters exactly, capture angle, threshold angle, construct and characteristic of the sensors will affect directly to the geometric accuracy of each pixel in the image. Second: Influence adjustment of the conditional acquisition of

7 5 satellite image include the conditions of solar illumination; effects of atmosphere, curvature and terrain elevation of the earth. Third: The main method to improve object position accuracy is geometric adjustment using different mathematical models and changing the number of GCPs to rectify image. Researching and improving the accuracy of rectifying satellite image shown in many documents but that is the basis information and it is hard to understand and implement in the fact Researches in Vietnam The applications of satellite image to establish topographic maps in Vietnam popularly but not yet systematically. There are a lot of researches shown that high resolution satellite image can satisfy regulation of position (horizontal) accuracy of topographic map at scale 1: and can satisfy regulations of position (horizontal) and height (vertical) accuracies at scales 1: and 1: Conclusions Base on researches over the world and in Vietnam can conclude that remote sensing data - multispectral images, multiple resolution images and multiple temporal images (especially ortho images) is necessary data to monitor natural resources and environment and for establishing and updating maps. The researchers shown that there isn t a general processing method for whole high resolution satellite images because it depends on the type of image, type of terrain... to select appropriate mathematical model and suitable number of GCPs to eliminate distortions and rectify images. Beside that each satellite has differential capturing methods, processing techniques define the sharpness of objects on the images which affect to accuracy of collecting point (GCPs, Tie points) during geometric adjustments. Thus, choosing suitable mathematical models for each type of terrains in Vietnam base on results of experimental researches. Researching results in the world and in Vietnam focus on processing spectrum, extracting information from imagine or using rectify methods such as image to image, or image to map. However, the principles of receiving high-resolution image, especially the better 1m resolution satellite images which didn t show in scientific

8 6 researches in Vietnam. These researches till focus on 2.5m resolution SPOT image for establishing and updating map. There isn t much experimental research about satellite images which have resolution better than 1m. To apply high resolution satellite images widely in fact, it needs experimental researches about principle of capturing high resolution satellite images and technical solutions for improving ortho image accuracy. Therefore, this thesis focuses on and solves basic problems: 1. Researching about technical characteristics of high resolution satellite image, especially for high resolution satellite image better 1m (panchromatic band). 2. Researching solutions to improve accuracy of ortho images which established from high resolution satellite images for establishing updating large scale topographic maps. CHAPTER 2. SPECIFICATIONS OF HIGH RESOLUTION SATELLITE IMAGE 2.1. The technical characteristics of high-resolution satellite The technical characteristics of high-resolution satellite image Currently, the high resolution satellite system rapidly developed with advancing technical characteristics such as: - The capturing resolution has raised up so much. The satellites have better 1m resolution (with panchromatic band) often have altitude orbit higher 650km; - Swath Width: 8 km ~ 17.6 km; - The sensor of high resolution satellite is still mainly pushbroom; - High resolution satellite system using Time Delay and Integration techniques for sensors; - All of high resolution satellites have stereo acquisition images mode; - Revisit time reduced, now, that period of time is from 01 day to 3 days Characteristics of satellite orbits The light is an important factor to achieve high quality and

9 7 stability while capturing images. The angle of sun light and orbit platform is unchanged in time (in a day or between seasons) because the orbits of high resolution satellites often synchronous with the sun orbit. The satellite orbit often flies North - South direction combine with rotating itself axis of the Earth (West and East) to satellite can always observe the Earth at the time when the sun has best illumination Sensor characteristics - CCD arrangement of sensors - Layout staggered CCD-line arrays - Time Delay and Integration sensors. With this layout, the TDI of high resolution satellite has created new advancements in signal processing techniques compared with conventional optical satellite. This enhances high spatial resolution of pixel, while reducing the geometrical errors on of high-resolution satellite imagery Geometric properties of high resolution satellite image Geometric distortion of the image The causes of the distortion error on the satellite image include: - Caused by detector; - Caused by the movement of the satellites and the orbital deviations, orbit speed change; - Caused by continuously recording orbital position and oblique angle of satellite; - Caused by the rotation, the curvature and topography of the Earth in the process of image acquisition. - Caused by atmospheric refraction; - Caused by projection used to adjust image. Analysts above show that the sources of distortion error can be divided into 2 main groups: - Internal error of sensor; - External error of sensor The resolution and the ability to extract information from satellite images a) Spatial resolution and related factors Spatial resolution of an image related to the distance or the

10 8 minimum size of an object on the ground can be distinguished on the image. The influencing factors are: Instantaneous Field Of View (IFOV), focal distance, the shape of the target object and the effects of the atmosphere. Spatial resolution is limited by the distance sampling on the ground (the pixel distance defined by the size of CCD and ratio of image) because it cannot separate an object within a pixel b) The factors affect the resolution and quality of image - The natural conditions; - Acquisition factors; - Scanner; - The image data processing; - The technical elements of the satellite; - Sensors of satellite Establishing large scale map from high-resolution satellite image Ability of information extraction from high-resolution satellite image Normally, the relationship of ground sample distance (GSD) - the pixel size and scale of map (which will be established) can be expressed by following formula: GSD mm M bd (2. 1) Accuracy requirements of high resolution satellite imagery for mapping Requirements of high resolution satellite images for mapping include: - Adapting position accuracy regulations of mapping objects and terrain features on the images. - It is easy to interpret images for establishing maps Conclusions Through researches about general characteristics of high resolution satellite image, it can give some conclusions such as: - With characteristics as high resolution, wide coverage, multiple spectrum bands..., the high resolution satellite images facilitate establishing and updating maps. - High-resolution satellite images have short revisit time from 1

11 9 day to 3.5 days allow establishing and updating maps using multitemporal images and assessing the change of features on the Earth s surface rapidly and exactly. Thence, it can be used to detect sudden change of terrain, especially for monitoring natural disasters; - High-resolution satellite images provide coefficients of geometric correction, allow computing the systematic errors in the image. Thus, ortho images which built from high-resolution satellite images can be achieve high accuracy serve for establishing and updating topographic map. Chapter 3. PROCEDURE OF ORTHO SATELLITE IMAGE 3.1. Definition of ortho satellite image Ortho image is the general digital image products located (referenced) in the coordinate system of map which will be established, corrected effects of height, resample base on resolution of satellite images suit scale of map. It is extracted by each sheet of map relatively Procedure of ortho satellite image (Figure 3.1) Figure 3.1. Procedure of establishing high resolution ortho satellite image

12 10 Procedure shows factors affecting ortho image such as: - The resolution of satellite images - Rectifying methods, mathematic models - The number of GCP, their arrangement and accuracy of them. - DEM Methods of rectifying satellite images Satellite high resolution images affected by projection, angle, line scan, atmosphere conditions, curvature of the Earth, difference of terrain elevations These causes of geometric distortions on satellite image which can be eliminated by mathematical models. There are two geometric rectifications for satellite images: Using physical model for geometric rectification Physical models allow error elimination while capturing image if known position of projection center on the satellite orbit. To define the position of projection center base on using GCP which known coordinates in the ground reference coordinate system exactly and can be defined on the images easily. This procedure called modeling images. According to the theory (principles), to model image scenes only need enough the number of GCP to determine the necessary parameters. Arrangement of GCPs doesn t much affect accuracy of model, otherwise, it doesn t affect rectifying image accuracy Using polynomial model for geometric rectification - Polynomial model used to transform between image coordinates and object coordinates. The order of polynomial model defined base on the error of image distortion, the number of GCPs and terrain type. - Types of polynomial function model: Affine -3D, Direction linear transformation model, Project transformation model, Parameter model, and rational function model Factors affect accuracy of rectifying satellite images There are a lot of factors affect accuracy of rectified images such as resolution of satellite images, mathematical models or accuracy of GCPs, the number and layout of GCPs on the images as well as DEM accuracy Image fusion Image fusion means that is a combination technology using types

13 11 of image to create another image. New image includes more information to enhance the ability of analysis and recognize the features on that. Image fusion can be implemented between differential nature images as follow: - Combination of panchromatic images with multispectral images (same type of image - optical images) to create new images have high spectral resolution (same as multispectral images) and high spatial resolution (same as panchromatic images); - Combination of differential image types: using optical images and radar images allows associating the chemical information (optical images) with the physical information of features (radar images). - Mixing basic color methods: Modified IHS, Principal Component Analysis (PCA), Multiplicative, High Pass Filter (HPF), Brovey transform Criterions for assessing the accuracy of ortho images Criterions for assessing accuracy of ortho images include: - Position accuracy of GCPs: can be defined by comparing (X, Y) coordinates of GCPs with their position on the ortho image relatively. If Xg, Yg are coordinates of GCP using as Check Point and Xn, Yn are coordinates of Check Point measured on the ortho image then: X X X (3. 1) g g Y Y Y Thus root mean square error (RMSE) of determining position on the ortho image calculated by: n n m 1 S n 2 (3.2)

14 Where: n is number of Check Point and S X Y - Accuracy of boundary features: 2 [ l ] (3.3) m2 2n Where: l : Deviation of feature on the 2 close to images at the mosaicking boundary n: Number of checking points at the boundary of 2 close to images Conclusion - Ortho image is an original product for establishing and updating topographic map and defining geographic features on the map. Thus, accuracy of map which will be established depends very much on ortho image. - To achieve high accuracy ortho image serves building map that should be pay attention to following factors: 1. Choosing suitable rectifying model for complex terrains such as mountain, hill, and mid areas. 2. Using suitable number of GCPs and designing suitable layout of GCPs on the image blocks. 3. Using DEM to rectify images for areas have elevation change largely. CHAPTER 4. SOLUTIONS FOR IMPROVING ACCURACY OF LARGE SCALE ORTHO IMAGES MADE FROM HIGH RESOLUTION SATELLITE IMAGE This chapter presents experimental researches about solutions improve accuracy of large scale ortho image from high resolution satellite images in particular cases as following: 1. Collecting the number and layout of GCPs on a satellite image scene. 2. Using geometric rectifying models for image block (apply for large areas include multiple images) 3. Accuracy of ortho image which established from satellite high resolution images using DEM. 4. Choosing fusion image methods sure the relatively between spectral values of original image with fusion image.

15 Introduction about experimental area Input data of experimental area - WorldView - 2 single image scene of Bac Giang city. This area has geodatabase at scale 1: Quickbird - image block for Dong Thap area. This area has geodatabase at scale 1: Other data (DEM, GCPs, ortho images, topographic map) extract from Projects Establishing geodatabase at scale 1: combining with DEM for whole country and Establishing geodatabase at scale 1: 2.000, 1:5.000 for urban, industrial zone, key economic area Experimental characteristics Single image scene - Bac Giang city WorldView 2 was processed into Geographic Lat/Lon WGS84 coordinate system, still affected by terrain elevation Image blocks - Dong Thap Dong Thap is a flat region, there are many wetland areas, and thus directly measuring GCP is difficult. And it includes multiple images. Therefore, GCP used in this context extracted from photogrammetric results in other project at scale 1: To assess dependent of large scale ortho image accuracy (which established from satellite images) with DEM accuracy by rectifying image block using DEM accuracy s 0.4 m Solutions for selecting the number and layout of GCPs The layout of GCPs Requirement of GCPs collection Solution to choose control points in this thesis: - Using horizontal (flat) control points, vertical (height) control points while producing geodatabase. These control points eliminated gross error, with all their attached information (name, position coordinates, diagrams ), high reliability because there no measurement error and identification error. - Using ortho images in geodatabase Requirement of number and layout of GCPs In this thesis, using at least 5 GCPs and largest is 30 GCPs to adjust images. The layout of GCPs shown in figure 4.1.

16 14 Figure 4.1. Layout of GCPs used in this thesis Accuracy assessment results Purpose of geometric rectifying in this experimental context is transforming geodetic coordinate system into VN-2000 coordinate system and limit position shift of pixels by the difference of elevation terrain. In this thesis using Affine - 3D polynomial model, Direction linear transformation model, 3D Project transformation model and Rational function model; differential options of number of GCPs and average elevation of these experimental areas. The experimental results are shown in (table 4.1).

17 15 Table 4.1. RMSExy of check point position of geometric rectification models (m) Mathematic Models 3D 1 st order 5 GCPs 6 GCPs 7 GCPs 9 GCPs 13 GCPs 20 GCPs 21 GCPs 25 GCPs 30 GCPs nd order GCPs 3 th order Direction linear transformation (DLT) D Project 1 st order transformation 2 nd order P2=P4 3 th order st order D Project transformation 2 nd order P2 P4 3 th order 0.22 Rational function 1 st order Standard deviation nd order th order From results in (table 4.1) it can conclude: * Number of GCPs: - The least number of GCPs depends on the parameters of each mathematical model. - If the number of GCPs from 5 to 7, it should use 1 st order Affine - 3D model and can get 1 pixel accuracy. - If the number of GCPs from more than 9 to 20, it should use DLT model and can get 0.5 pixel accuracy. The maximum number of GCPs should use for a single image scene with 16.4 km size is 30 because the accuracy can be archived 0.22m adapts requirement of ortho image at scale 1: If using more than 30 GCPs, the Defining scientific basis and proposing solutions to improve accuracy of geometric correction high resolution satellite accuracy does not improve. * Layout of GCPs

18 16 The GCPs should be layout whole the images regularly. From results in table 4.3 shows the layout of GCPs without center point (six point layout) has error varies from 0.61 to 2.65m with standard deviation is 1.161m. If layout of GCPs with center point can archive standard deviation of check point s root mean square errors is from 0.175m to 0.33m. * Selecting mathematical model Recommend use Rational function model because it can get high accuracy if increase the number of GCPs Choosing mathematical model for image block The experimental results as follow: This thesis apply coefficient of 3 th order Rational function model, achieved results as: - The GCPs maximum error is ±1.12pixel; - The worst root mean square error of check points is ± 1.32 pixel, - Root mean square error of all GCPs and check points after adjusting is smaller than ± 1 pixel, the largest value is ± 0.93pixel and minimum value is ± 0.41 pixel. After rectifying image, the quality of images should be checked: - Checking position of features on the each panchromatic image scene and multispectral image scene. - Checking position of feature at boundary in the image cover areas. - Checking position errors of GCPs and check point on the rectified images Accuracy of ortho image established from high resolution satellite image using DEM The detail levels of DEM used to rectify single image - Worldview 2 as follow: - DEM with spacing grid is 25m 25m (DEM10) 1m accuracy built from aerial photogrammetry which used to establish topographic map 1:10.000; - DEM with spacing grid is 10m 10m (DEM5) 0.5m accuracy extracted from Lidar DEM which used to establish topographic map 1:5.000; - DEM with spacing grid 2m 2m (DEM2) 0.5m accuracy is

19 17 original DEM built from Lidar data; - DEM with spacing grid 1m 1m (DEM1) is original DEM built from Lidar data. In the (figure 4.2) shows RMSExy corresponding to differential GCPs and differential DEMs as follow: Figure 4.2. RMSExy of GCPs using differential DEMs From figure 4.2 can see: - When use DEM10 with differential GCPs is 5, 21, 30 can achieve RMSExy = ±1.912m, RMSExy = ±1.531m and RMSExy = ±1.322m relatively. - When use DEM5 with differential GCPs is 5, 21, 30 can achieve RMSExy = ±1.347m, RMSExy = ±1.201m, and RMSExy= ±0.781m relatively. - When use DEM2 with differential GCPs is 5, 21, 30 can achieve RMSExy = ±0.901m, RMSExy = ±0.673 and RMSExy= ±0.392m relatively. - When use DEM1 with differential GCPs is 5, 21, 30 can achieve RMSExy = ±0.811m, RMSExy = ±0.512m, and RMSExy= ±0.380m relatively. Conclusion: - Elevation accuracy of DEM depends on slope of terrain as well

20 18 as density of terrain data using for building DEM. - If using very high resolution DEM to rectify image, it did not improve accuracy of rectified image, otherwise, the pixels on the image will be distorted Selecting fusion methods for high resolution satellite image This thesis tested methods of image fusion such as: Principal Component Analysis (PCA) method; Multiplicative method; Modified Intensity Hue - Saturation (IHS) method; High Pass Filter (HPF) method; Brovey Transform method. Assess the quality of the image spectrum after fusion is analyzed through 03 methods: Assessing quality of image fusion results by eyes (Figure 4.3) a) Multispectral images b) Panchromatic images c) PCA method d) Multiplicative method e) Modified IHS method f) Brovey method Figure 4.3. Results of image fusion methods g) HPF method

21 Histogram analysis a) Original multispectral image b) PCA method c). Multiplicative method d). Brovey method e) Modified IHS method f) HPF method Figure 4.4. Histogram results analysis of image fusion methods Analysis of spectral quality assessment indexes The indexes used to assess spectral quality of image fusion as following: - Bias Index is the differential value of the mean value of original image and fused image. To achieve the best quality then Bias value is approximate zero. DP AT AT Bias 1 DP DP TB TB TB TB TB (4.1)

22 20 Where: DP TB, AT TB, are respectively mean values of original image and fusion image. They can be calculated by formula as following: TB 1 m n mn i j j i - STD Index is the standard deviation computed for each image band using express: x ij n 1/2 2 i, j TB (4.2) i 1 1 STD DP DP n 1 - Entropy Difference Index is the difference between entropy on the original image and on the fusion image: n 1 n HCL( x) p i log p i log p i 2 2 i 1 pi (4.3) i 1 Where: p (i) is probability of i value - ERGAS Index (Relative dimensionless global error in synthesis) stands for quality of image fusion base on standardized average error of each band of fusion image. Increasing ERGAS Index means the quality of fusion image decreased. EAGAS Index computed by following formula: 1/2 n 2 RMSE 2 (4.4) i 1 dh 1 ERGAS 100 dl n TB Where dh/dl is ratio the pixel sizes of panchromatic image and multispectral image and N is number of bands. RMSE is root mean square error computed as:

23 RMSE B bias B STD B I (4.5) - The correlation coefficient is given by Corr i Amn TBA Bmn TBB m n A 2 2 mn TBA Bmn TBB (4.6) m n i Table 4.2. The values of spectral quality assessment indexes of fusion image Method Bias H CL (x) ERGAS PCA Multi Modified HIS HPF Brovey Table 4.3. Correlation coefficient (Corr. coeff.) of the image before the fused images Method Band 1 Band 2 Band 3 Medium (Near IR) (Red) (Green) PCA Multi Modif HIS HPF Brovey

24 22 Base on the values of spectral quality assessment indexes of fusion image, it recommend using image fusion method is Modified IHS because the color and histogram of original image and fusion image is relatively, ERGAS Index is smallest in all results of experimental methods and correlation coefficient approximately equal to Programming ortho image quality assessment software Block diagram of software Start program Input n DataPoint: DP[n]; i = 0;Sum2_dX = 0;Sum2_dY = 0; Sum2_dS = 0; isstpx = 0;iSSTPy = 0;iSSTPs = 0; F i <n? T Imput check point X = GetPoint().X; Y = GetPoint().Y; dx = fabs(x DP[i].X); dy = fabs(y DP[i].Y); ds = sqrt(pow(dx, 2) + pow(dy, 2)); Sum2_dX += pow(dx, 2); Sum2_dY+= pow(dy, 2); Sum2_dS += pow(ds, 2); i = i + 1; Continuous? T F isstpx = sqrt(sum2_dx / i); isstpy = sqrt(sum2_dy/ i); isstps = sqrt(sum2_ds / i); Exporting image rectification error report Exporting report of error of boundary features End program Figure 4.5. Block diagram of software Software description - The software tools are programmed in MDL code named DanhGiaChatLuongBinhDoAnh.ma. It can only directly run on

25 23 Microstation V7 environment Functions of software - Supporting users assess image rectification error and error of boundary features on the image; - Allow users import coordinate text file at format [TEXT] [X] [Y] [Z] into Microstation software (*.dgn); - Allow users recheck coordinates of rectified image pixels; - Allow users exporting report of image rectification error and error of boundary feature on the rectified image. - Allow modifying the values of False X, Y parameters in seed files which is active file. CONCLUSIONS AND RECOMMENDATONS a. Conclusions - Accuracy of ortho image depends much on mathematical models, quantity and layout of GCPs. Therefore, choosing the number and layout of GCPs must be suitable for each rectification model. - Nowadays, the ortho image is a very important product which widely used to establish and update maps. Thus, to achieve ortho image adapted geometric accuracy require applying following solutions: 1. Number and layout of GCPs - The GCPs should be regularly distributed on the image. It has to have a control point at center image. - The number of GCPs depends on the number of parameters of each mathematical model. However, the maximum GCPs should not exceed 30 points for an area 16.4 x 16.4km. 2. Rectifying image blocks for area cover by multiple images Defining tie points on the trips, scenes to connect image blocks together and adjusting for each block or for all of them. 3. Choosing mathematical model The high resolution satellite image is usually calibrated radiation

26 24 and converted to geodetic coordinate system WGS84. Therefore, the mathematical model should use is rational polynomial model (3 th order RFM) using RPC coefficients which are provided in metadata of satellite image or calculated from the certain number of GCPs. The 3 th order RFM gives better and more stable results then other ones. 4. The using of 1:2.000 and 1:5.000 databases including ortho image, DEM, DSM, GCPs and tie points is completely satisfy creating large scale ortho image (1:5.000) from high resolution satellite image. It should be used DEM has grid space equal to 4 times of expected image pixel size. It is optimal for economy (the expensive of building DEM) and technical aspect (ensure accuracy of rectified image). 5. It should choose suitable image fusion method for each type of image data. Using spectral quality assessment indexes and histogram to assess the quality of image fusion method is objective. 6. Ortho image quality assessment software Software is a convenient program for assessing ortho image quality. It can be used for producer to check and accept ortho image which created from remote sensing image or air photos. b. Recommendations In order to reduce the time and improve the application of geodatabase for rectifying satellite image, it should have to be added into geodatabase these information packages as follows: + Coordinates, altitude and notes of GCPs; + Photogrammetric triangulation result files. + Ortho image.

27 THE AUTHOR S SCIENTIFIC RESEARCHES AND PAPERS RELATE TO THIS THESIS 1. Tran Đinh Tri, Do Thi Hoai, (2008), The ability of improving accuracy of mosaicking rectified images to create ortho image. Journal of Land Administration, No 4-8/2008, pp 41-43, Hanoi. 2. Do Thi Hoai, Tran Thi Trang, Pham Hong, Nguyen Thi Ngoc Hoi, (2010), Combining aerial photography, topographic maps and remote sensing image to build geographical database (geodatabase), Journal of Geodesy and Cartography No 5-9/2010 pp 23-28, Hanoi. 3. Do Thi Hoai, (2010), Combining aerial photography, topographic maps and remote sensing image to build geodatabase of Quang Ngai province, Scientific research project, Hanoi University of Mining and Geology. 4. Do Thi Hoai và Tran Thanh Ha, Tran Thi Hoa (2010), Creating ortho image for establishing and updating maps, Proceeding of 19 th Scientific workshop, Hanoi University of Mining and Geology, pp 67-70, Hanoi. 5. Do Thi Hoai, (2014), Researching on satellite image rectification using gedatabase, Proceeding of the conference on science and technology about geodesy and cartography for international integration, pp , Hanoi. 6. Do Thi Hoai, (2016), Selecting image fusion methods for high resolution optical satellite image (WorldView - 2), Journal of Geodesy and Cartography No 28-6/2016 pp 22-28, Hanoi.