Session S4: Feature Recognition: Needs and Techniques ===================================================== Organizers: M.Aschwanden, R.Bush Time: Thursday, Jan 16, 3:15-4:45 pm Program: -------- 3:15 10 min Presentation Serge Zharkov: - automated detection of sunspots in WL and - magnetic field inversion lines and filaments - solar feature catalogue EGSO 10 min Discussion 3:35 10 min Presentation Greg Slater: - active Region Boundary Determination From Soft X-ray Images - application to SXI images 10 min Discussion 3:55 10 min Presentation Andrzej Fludra: - automated detection of small-scale variability events - examples from analysis of CDS database 10 min Discussion 4:15 10 min Presentation Markus Aschwanden: - tracing and automated detection of coronal loops in EUV - examples from TRACE data, application to STEREO/EUVI mission 10 min Discussion 4:35 10 min Presentation and Discussion Rock Bush: - Questionnaire about Need and Demands of Feature Event Catalogs 4:45 Break Related topics (presented at AIA/HMI conference) ------------------------------------------------ Sami Solanki (& Wiegelmann): - Magnetic field modeling constrained by EUV loop tracing [? presentation in Session C2/M7, Tue morning] Andrzej Fludra: - automated CME detection algorithms [presentation in Session C1] Session S4 (DeRosa, Gurman, Bush) - The Sun Today, events, logs, real-time data Need for feature catalogs Haimin Wang: (participant cancelled) - automated solar filament detection, disappearance, Ha/BBSO Andrei Zhukov: - automated detection of EIT waves, EUV dimmings Hochedez, Jean-Francois: - Image compression (with Poisson recoding) wavelet analysis of EIT images Scott McIntosh, Joe Gurman: - EUV bright point statistics Summary writeup of discussion in Session S4: FEATURE RECOGNITION: NEEDS AND DEMANDS (write-up by M.Aschwanden) ============================================================= "Feature Recognition in solar images" has become a prominent research field. A good overview on recent developments can be glanced in the new SOLAR PHYSICS issue 228 (May 2005), which contains 24 invited reviews on the topics, based on the recent (2nd) Solar Image Processing Workshop (held in Nov 2004 in Annapolis), ranging from magnetic field analysis, automated detection of photospheric, chromospheric, and coronal features, to feature databases and catalogs. A 3rd workshop on the topic "Solar Image Processing WOrkshop III: Challenges of New Instrumentation" is planned for Sept 6-8, 2006, at Trinity College Dublin, Ireland, which will address applications to new ground based instrumentation such as the ATST (Advanced Technology Solar Telescope), SVST (Swedish Vacuum Solar Telescope) and DOT (Dutch Open Telescope), as well as future space instrumentation on board STEREO, Solar-B, and SDO. In the planning of Session S4, the organizers asked the primary question: Who is the user community for a feature data base? And distilled from that: What do we need to recognize automatically? Which features require, or benefit form, observer assistance? Who develops software? What is needed to test and validate the software? How is the data integrated into the event logs and search engines? What external data are required (SDO, other space missions, NOAA, NSO, ...)? These tasks are useful and perhaps even necessary for realtime observation planning, such as CME and flare detections. A secondary group of applications of feature recognition techniques is off-line data analysis of any AIA/HMI science, such as for instance "fingerprinting of EUV images" to aid 3D magnetic field reconstruction, or fractal analysis of images to provide scaling laws, etc. Techniques of pattern recognition for scientific data analysis used in solar physics so fare include: - wavelet techniques, Trous wave filtering - fractal dimension, multi-fractal - segmentation techniques - automated filament detection - neural network techniques - difference image enhancing Of course, such science tools can be applied anytime and do not require realtime processing nor completeness of databases. Serge Zharkov (from Sheffield Univ.) gave than a presentation entitled "Solar Feature Catalogues", with contributions from Valentina Zharkova, (Bradford Univ.), S.S.Ipson, A.Benkhalil, N.Fuller, and J.Aboudarham. (see ppt document) He reported on the European Grid of Solar Observatory (EGSO) specifications, which provides comprehensive information on solar features detected on a particular observation, provides a cropped image library, and creates an option to search data by any feature characteristics. A prototype catalog on automated feature detection exists for the years 1996-2005, containing data from Meudon (daily Ca II K3 and H-alpha images) and SOHO/MDI (white light images and magnetograms). Zharkov showed examples how the area of active regions is detected from white-light images, the center of gravity of active region areas, the area size, diameter, the umbra size, the bounding box, intensity statistics, magnetic info, filament skeletons, center, elongation, curvatures, and magnetic inversion lines. So far, over 10,000 observations have been processed, including 370,000 sunspots and 100,000 active regions. Techniques for filament recognition use "trained neural network algorithms". The URL of the EGSO group is http://solar.inf.brad.ac.uk, where also publications can be found. In the following discussion it was mostly emphasized that such solar feature recognition software would be desirable to organize the AIA and HMI database, especially for active region catalogues, besides obtaining automated statistics for scientific analysis. Greg Slater then gave a presentation on "Coronal boundaries of Active Regions derived from Soft X-ray Images" (see ppt document). He described the differences of coronal boundary detections versus photospheric boundary detections. Since coronal emission is extended in 3D dimension, rather than two, the boundary definition is more complex and aspect-angle dependent. Moreover, coronal emission in EUV and soft X-rays is optically thin, and thus subject to more confusion from foreground and background emission. Third, coronal emission is highly variable, demanding time-dependent boundary definitions. A prototype software has been developed to determine active region boundaries in Yohkoh, GOES/SXI, and He II 304 A images, which specifies in the feature catalogue the center of brightness, the radial and angular extent of each feature, the brightness moment distribution, and region number. The prototype tests included 29,000 images from Sept-Nov 2001. In the following discussion the suggestion was made that the following product would be useful for feature search and quick-looks of the AIA database: Create (14-day long) lightcurves from the 8 most prominent active regions visible on the disk at any time in all AIA EUV channels. This would allow the customer to monitor easily the activity in individual active regions, rather than from the full disk (as currently provided by GOES or SXI light curves). Active region based feature catalogues seem to be a natural organization for the AIA database, which will keep only selected data for the permanent database. Andrzej Fludra reported on tests of automated detection algorithms for small-scale variability events. These events include short-lived (1-2 min) and small-scale (less than 100 arcsec^2) brightenings in EUV and soft X-rays. There are hundreds of them at any given time and it is yet to be decided whether they should even be catalogues individually. The goal is to derive statistical frequency distributions of their parameters (sizes, durations, peak intensities, energies) without assigning events to physical classes. In the following discussion two options were considered: (1) event-based catalogues, vs. (2) statistical distributions. Event-based catalogues could grow quite extensive at low flux levels, since the cumulative number of events grows by about a factor of 10^(0.8)~6 for every decade of lower flux or energy threshold (since the differential frequency distribution is approximately N(E)~E^(-1.8), (Aschwanden & Parnell, 2002, ApJ 572, 1048). At the solar cycle maximum, about 20 flares/day are registered (with an energy of >10^28 erg, corresponding to about the GOES C-class level, while the number of nanoflares detectable in EUV (with an energy of >10^24 erg) is expected to be about 20*10^(4*0.8)~ 30,000 nanoflares/day. A more compact form of documenting the nanoflare activity would be to run an automated detection code in real-time and to store only the frequency distributions as function of time and AIA channel, say in hourly intervals, but continuously over the years of operation. This would allow quickly to monitor variations and deviations of coronal activity from standard parameters. There was also a short description of faculae detection in white-light by Michael Turmon (JPL). Turmon also published a paper on "Statistical Pattern Recognition for Labeling Solar Active Regions: Application to SOHO/MDI Imagery" (Turmon, Pap, & Mukhtar 2002, ApJ 568, 396). The abstract describes it as follows: "This paper presents a new application of statistical methods for identifying the various surface structures on the Sun that may contribute to observed changes in total and spectral solar irradiance. These structures are divided for our purposes into three types: quiet Sun, faculae, and sunspots (umbra and penumbra). Each region type is characterized by the observed data present at pixels of that type. Statistical models characterizing these observables are found from expert identification of a sample set of regions or unsupervised clustering. Information about the spatial continuity of regions is incorporated into the model via a prior distribution on the label image; the contribution of the prior can be interpreted as a regularizing term. Once the parameters defining the models are fixed, the inference procedure becomes to maximize the probability of an image labeling given the observed data. This allows objective and automated classification of a large set of images. We describe the application of these procedures to computing labelings from synchronized full-disk high-resolution magnetic-field and light-intensity maps from the Michelson Doppler Imager experiment on the Solar and Heliospheric Observatory." The following discussion emphasized that this type of feature recognition would be of interest for AIA to have a means of characterizing and monitoring the irradiance variations of the Sun during the solar cycle. Rock Bush concluded the discussion by summarizing the benefits of feature event catalogs based on white-light and HMI magnetogram data, which can easily specify active region areas and sunspot areas, useful for predicting the evolution of active regions and the associated irradiance variations. ________________________________________________________________________