COSMO SkyMed Constellation The driving Mission requirements for the constellation development are the following: Capability to serve at the same time both civil and military users through a integrated approach; Large amount of daily acquired images; Satellites worldwide accessibility ; All weather and Day/Night acquisition capabilities; Very short interval between the acceptance of the user request acquisition and the release of the remote sensing product; High image quality (e.g. spatial and radiometric resolution); COSMO-SkyMed constellation is a Space-Earth Observation Dual Use System devoted to provide products/services for the following purposes: environmental monitoring and surveillance applications for the management of exogenous, endogenous and anthropogenic risks; provision of commercial products and services. 1
COSMO SkyMed Constellation SPOTLIGHT 10 Km X 10 Km 1 m Resolution MULTI-MODE ACQUISITION STRIPMAP - HIMAGE 40 Km X 40 Km 3 m Resolution STRIPMAP PING PONG 30 Km X 30 Km 15 m Resolution SCANSAR WIDE 100 Km X 100 Km 30 m Resolution SCANSAR HUGE 200 Km X 200 Km 100 m Resolution 2
Process adopted Two processing chain to obtain DEMs Ingestion CSK Data www.csiricerca.org Co-registration The The IDC Delft Platform Institute of uses Earth artificial Observation neural networks and Space algorithms Systems of indelft orderuniversity to obtainof the digital Technology elevation has developed model that an is commonly referred Interferometric as CSI Synthetic Rierca Aperture & Ambiente Radar - Supervised (InSAR) processor Algorithm named (CSI-SV) Doris. This the chain platform applies isthe a property conventional of CSI algorithms Ricerca & Aambiente to generation (deposited the Digital Elevation patented Model. n. GE2013A000022 ). Interferometric products DEM GENERATION GEOCODING 3
Artificial neural networks algorithm Competitive growth of the Regions Two distinct phases: Translation and Aggregation for the reconstruction of the original image creating progressively increasingly large regions of pixels that have a difference in phase value less than π, until arriving to a unique region that includes all of the pixels of the image. To make homogeneous regions of pixels that have phase difference greater than π, are added, or removed, multiples of 2π, to make the values of the edges between adjacent regions as similar as possible. Composition Regions: List of pixels belonging to the region; List of pixels that are located on the edges of the region; List, for each pixel of the board of adjacent pixels belonging to other regions. 4
Artificial neural networks algorithm Competitive growth of the Regions Stages of the algorithm: 1. Each pixel of the image becomes a region with only one edge and with four adjacent pixels arranged in a cross. 1. For each region is carried out a translational motion, if possible, and an aggregation, if possible. 1. Explore all the regions you will get the number of combinations and translations carried out by the algorithm and the number of remaining regions. 2. Repeat the operations of aggregation and translation to the following conditions: the number of regions is equal to 1; are not carried out in a cycle operations of translation and aggregation. Expected results: reduction of the current processing times; higher quality of DEMs. Pixel/ region 5
Comparison and validation activity Comparison activity is performed by overlapped the two model (DORIS and CSI) with the a third DEM obtained using the GPS-RTK technique, capable to return DEMs with a precision of few tens of centimeters. Validation activity is executed by refers to the Italian standard CSIS described in the guide lines named Regole Tecniche sui Dati Territoriali delle Pubbliche Amministrazioni 6
Comparison activity Comparison and validation activity The comparison activity is based on the using the Helmert transformation. This transformation method is frequently used in geodesy to produce distortion-free transformations from one datum to another. where X T is the transformed vector; X is the initial vector. The parameters are: translation vector; Contains the three tranlations along the coordinate axes scale factor, which is unitless; rotation matrix. Consists of three axes The rotation matrix is an orthogonal matrix. The rotation is given in radians. 7
Comparison and validation activity Standard parameter adopted for Validation activity: Accuracy degree of closeness of observations of a magnitude with respect to the value assumed as a reference for size. It is expressed through the value of RMSE (Z) (Root Mean Square Error). Confidence Interval probability that the real value of the observation is contained in a certain interval [a, b]. LE95 is expressed as (Linear Error at 95% probability). Precision degree of closeness of the observations of a size compared to the estimated mean. It is expressed using SD (Z) (Standard Deviation). 8
Comparison and validation activity Pearson correlation which, defined as the covariance of two variables divided by the product of the standard deviations of the same, indicating whether the variables are: directly related to, or positively correlated; are not related; inversely correlated or negatively correlated. The coefficient always takes values between -1 and 1: : Weak correlation; : Moderate correlation; : Strong correlation. 9
Results obtained Tested areas Craco and Metaponto (Matera) Guidonia (Rome) E.G.U. General Assembly 2013 Wien - Austria - 07 12 April 13 10
V&V: the results about elaboration time Interferometric pair CSKS2_SCS_U_HI_03_VV_RD_SF_20120726165909_20120726165917 CSKS4_SCS_U_HI_03_VV_RD_SF_20120730165908_20120730165916 Craco and Metaponto (MT) Interferometric DEM coherence amplitude obtained phase Characteristics Unit Value Temporal baseline: [m]: 4 Perpendicular baseline: [m]: 303,7 Parallel baseline: [m]: 153,2 Horizontal baseline: [m]: 340,1 Vertical baseline: [m]: 1,5 Baseline: [m]: 340,1 Baseline orientation: [deg]: 0,2 Look angle: [deg]: 26,5 Incidence angle: [deg]: 29,3 Angle between orbits: [deg]: 0,000889521 Ambiguty height: [m]: 17,8 Elaboration time with multi-look 4 x 4 Elaboration time without multi-look factor DORIS: 23 min DORIS: 13 h 03 min CSI-Supervised: 3 min CSI-Supervised: 03 h 46 min E.G.U. General Assembly 2013 Wien - Austria - 07 12 April 13 11
V&V phase: imaging quality Craco and Metaponto (Matera) For benchmark was made a reference DEM with the technique GPS - RTK, so as to have points (Check Point - CP) determined with an accuracy (σ CP) of at least one order of magnitude smaller than the tolerances set, ie: σ CP, EN < 1/10 * T EN where T EN is planimetric tolerance. In order to ensure a good reception of satellite signals, the period of execution of the detection, for the entire duration of the mission, has had a value of Position Dilution Of Precision - PDOP less than 2 (the std is <= 5). The number of CPs considered are 2330. E.G.U. General Assembly 2013 Wien - Austria - 07 12 April 13 12
Craco and Metaponto (MT) Graphics and table Interferometric pair Height (meter) 510 480 450 420 390 360 330 300 270 240 210 180 150 120 90 60 30 0 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 1801 1901 2001 2101 2201 2301 Number of measurements RTK Doris CSI-SV Aster Heights (meters) Absolute Error Mean Standard Deviation Pearson Correlation Mean Standard Deviation GPS-RTK 194,56 120,61 Doris 186,94 173,25 0,9291 7,63 75,71 CSI-SV 199,23 119,08 0,9427-4,66 40,58 Aster 122,37 122,37 0,9436-3,82 40,84
V&V phase: the results about elaboration time Guidonia (Rome) Interferometric pair Interferometric DEM Coherence Amplitude obtained Phase CSKS4_SCS_U_HI_01_VV_RD_SF_20120917171631_20120917171638 CSKS1_SCS_U_HI_01_VV_RD_SF_20120921171629_20120921171636 Characteristics Unit Value Temporal baseline: [m]: 4 Perpendicular baseline: [m]: 51,8 Parallel baseline: [m]: 22,1 Horizontal baseline: [m]: 56,3 Vertical baseline: [m]: 0,9 Baseline: [m]: 56,3 Baseline orientation: [deg]: 0,9 Look angle: [deg]: 24,1 Incidence angle: [deg]: 26,6 Angle between orbits: [deg]: 0,00190694 Ambiguty height: [m]: 93,8 Elaboration time with multi-look 4 x 4 DORIS: 10 h 32 min CSI-Supervised: 0 h 46 min E.G.U. General Assembly 2013 Wien - Austria - 07 12 April 13 14
V&V phase: imaging quality Guidonia (Rome) For benchmark was made a reference DEM with the technique GPS - RTK, so as to have points (Check Point - CP) determined with an accuracy (σ CP) of at least one order of magnitude smaller than the tolerances set, ie: σ CP, EN < 1/10 * T EN In order to ensure a good reception of satellite signals, the period of execution of the detection, for the entire duration of the mission, has had a value of Position Dilution Of Precision - PDOP less than 2 (the std is <= 5). The number of CPs considered are 2190. E.G.U. General Assembly 2013 Wien - Austria - 07 12 April 13 15
Guidonia (Rome) Graphics and table Height (meter) Interferometric pair 1200 1000 800 600 400 200 0 1 51 101 151 201 251 301 351 401 451 501 551 601 651 701 751 801 851 901 951 1001 1051 1101 1151 1201 1251 1301 1351 1401 1451 1501 1551 1601 1651 1701 1751 1801 1851 1901 1951 2001 2051 2101 2151 RTK Doris Aster CSI-SV Number of measurements Heights (meters) Absolute Error Mean Standard Deviation Pearson Correlation Mean Standard Deviation GPS-RTK 246,85 176,14 Doris 150,00 150,73 0,7971 96,85 106,87 CSI-SV 243,52 206,69 0,9049 3,33 88,66 Aster 242,07 215,35 0,9010 4,78 95,01
Conclusion and remarks DEM interpolation by reference Artificial Neural Network Whereas the nature of the terrain have a strongly non-linear trend, it was necessary to investigate new methods of interpolation than the Taylor polynomial used by DORIS and based on adaptive algorithms. The aim is to have available a more specific reference DEM that would significantly improve the final result of the process of unwrapping. DEM filtering by reference Artificial Neural Network The raw output of a process of unwrapping can generate a three-dimensional image is not perfectly continuous and full of imperfections and steps. This phenomenon depends mainly on the quality of the image of wrap phase. Areas with low coherence (salt and pepper) or sudden changes in altitude can cause this type of artifacts. The purpose is that of treating the unwrap image (still in radar coordinates), or directly the three-dimensional image in meters, with an adaptive filter based on neural networks, in order to obtain a more continuous surface while still maintaining the characteristics and morphology of soil. E.G.U. General Assembly 2013 Wien - Austria - 07 12 April 13 17
Conclusion and remarks Quality-Map for the identification of areas of salt / pepper By the application of algorithms based on the properties of the trigonometric (sine and cosine) of the phase of a single pixel and its surroundings, the objective is to generate a map (similar to that of consistency) to identify areas of the image more difficult to process (lakes, sea, shadows created by mountains, etc...). Unwrapping assisted by the Quality-Map and interpolation ANN Using the map, as described above, as a mask to the image to be superimposed rolled the objective is to perform one unwrapping "intelligent" that goes to concentrate, initially, on the zones of high quality coefficient (Qj) and interpolate, or process to end, those with a low coefficient. E.G.U. General Assembly 2013 Wien - Austria - 07 12 April 13 18
Conclusion and remarks Example of interpolation 100 x 100 pixel Reference DEM CSI- Artificial Neural Network weighted average Wien - Austria - 07 12 April 13