Dear Ladies and Gentlemen, My name is Martin Söderström and I’m working as a Research Scientist at VERIFIN. My area of responsibility is database development, among other things. First would like to thank the organizers for an opportunity to give this presentation here today. My subject is ”Development of Anlaytical Database for the Chemical Weapons Convention.
Panchromatic and spectral band filters: The ADS40 captures strips of imagery with linear CCD arrays, as opposed to discrete aerial images with a square or rectangular CCD sensor. Panchromatic imagery is captured looking forward, nadir and backward from the aircraft. Near infrared information is also captured close to nadir, whereas the red, green and blue color bands are forward looking
Absorption Filters: Absorption filters absorb some parts of the spectrum and transmit others. In the diagram above and on the right, transmission of typical narrow band absorption filters for Blue, Green and Red are shown. Absorption filters are never perfect and narrow banded. Transition from full absorption to full transmission is always in a curve. This is also shown in the diagram above. e.g. the Green filter does not fully absorb Blue and Red. A perfect filter would have a very sharp peak centered in the Green region that trails off to zero transmission at non-green wavelengths. This is all but impossible to accomplish with absorption filters available in the real-world. Interference Filters: These filters differ from absorption filters in the fact that they reflect and destructively interfere with unwanted wavelengths as opposed to transmitting them fully. The term dichroic arises from the fact that the filters appear one color under illumination with transmitted light and another with reflected light. The dichroic filters used in the patented Trichroid RGB beam splitter of the ADS40 are of novel design and are manufactured using multi-layered thin metallic coatings that are deposited onto optical-grade glass using vacuum deposition for a total of 176 layers. Total thickness of the 84 layer Blue beam splitter part of the Trichroid is about 6.5 microns. Dichroic filters are far more accurate and efficient in their ability to block unwanted wavelengths when compared to absorption filters .
Use Of Ads40 Pushbroom Imagery For Tree Species
Use of ADS40 pushbroom imagery for tree species classification in Finland Felix Rohrbach University of Helsinki
About the project <ul><li>Idea of testing the ADS40 in Scandinavian forests (2007) RGBNIR: ADS40 4 km 40 cm GSD, UCD 112 cm Spectral bands: ADS40 narrow, UCD overlapping Calibration: ADS40 absolute, UCD relative </li></ul><ul><li>Qs: 1) Gain from using reflectance images for tree SP? </li></ul><ul><ul><li>Separation of effects by GSD and medium </li></ul></ul><ul><ul><li>Effects of tree age / site fertility in spectra </li></ul></ul><ul><li>2) Gain from adding LiDAR data and image features? </li></ul><ul><ul><li>Birch problematic in LiDAR intensity, in NIR? </li></ul></ul><ul><ul><li>Orthogonality LiDAR and image data </li></ul></ul><ul><li>3) Comparison of different digital airborne sensors? </li></ul>
Hyytiälä experiment <ul><li>Over 16 000 reference trees for SP-detection tests </li></ul><ul><li>Aerial images from 1946–2004, 2006-2009 UltraCAM D 2006, 2007, (UltraCAM Xp 2010) ADS40-SH52 2008* (1, 2, 3, and 4 km) Z/I DMC 2009* * NASA, IN-SITU, SMEAR </li></ul><ul><li>LiDAR data 2004, 2006-2008 (12-15 p/m2) </li></ul>
Narrow Multispectal bands for Remote Sensing Overlapping bands Only interference filters are suitable for remote sensing applications where response in non-overlapping narrow bands is evaluated Non-overlapping narrow bands nm nm
Workflow of research Q1. <ul><li>Image aqcuisition, in-situ measurements, geometric post-processing </li></ul><ul><li>Radiometric corrections (ATSR, ATM, BRDF, BRDF-ATM) </li></ul><ul><li>Feature extraction for the reference trees </li></ul><ul><li>Feature selection and analysis </li></ul><ul><li>Classification trials </li></ul>
What happened until today <ul><li>Flight on August 23, 2008 over Hyytiälä (clouds) </li></ul><ul><li>Four altitudes: 1, 2, 3 and 4 km </li></ul><ul><li>In-situ observations using a goniospectrometer for reflectance tarps & natural targets, also at-sensor radiance & nadir-reflectance </li></ul><ul><li>Aerosol optical depth ( j , j=1..8 ) with AERONET sun-photometer </li></ul><ul><li>Main processing (Imagery & DSO) by Leica Geosystems 115 Gbytes of Level-0 data </li></ul><ul><li>Sensor model implementation in KUVAMITT using Leica-SDK </li></ul>
Where are we now <ul><li>Validating of radiometrically processed images with in-situ data (@FGI, Lauri Markelin, Eija Honkavaara, Juha Suomalainen) </li></ul><ul><li>(ATSR, ATM, BRDF, BRDF-ATM) production using XPro 4.2 (with Dr. Ulrich Beisl @ Leica), L1 data (epipolar resampling) </li></ul><ul><li>Debugging KUVAMITT for L0/L1-capabilities </li></ul><ul><li>Cloud-screening </li></ul><ul><li>Shadow- and occlusion determination for crown points </li></ul>
Future progress / aims <ul><li>Feature extraction for ~ 12000 pine, spruce, birch trees in ~90 4 images (cluster of KUVAMITT-PCs) </li></ul><ul><li>Mono- and biangular RGBNIR-data - Quest for strong, invariant features - How do they change with GSD 10-40 cm - How the features behave with varying age of trees - Using a combination of features, classification trials, comparison of radiometric correction methods ... And? NN-imputation, RF? </li></ul>
” Expected Conclusions” <ul><li>We hope to be able to quantify the effects of GSD and atmosphere, and find some invariant features </li></ul><ul><li>We get image features to be plugged in together with LIDAR features </li></ul><ul><li>We set baseline for camera comparisons </li></ul>