1 B2X: Flare recognition in Solar EUV Images Applications to EUVI Véronique Delouille, Judith de Patoul Bogdan Nicula, Jean-François Hochedez 5th SECCHI Consortium Meeting, 4-8 March 2007
2 Outline Goal : Goal : –Real-time alerts & EUV flares Catalog Method : Method : –CRH (Cosmic Ray Hit) removal Median filter, Wavelet denoising Median filter, Wavelet denoising –Histogram processing Skewness monitoring Skewness monitoring Data set : Data set : –24 January 2007; –Images from EUVI-A and EUVI-B
3 Detection of brightness enhancement in EUV images Aim : Aim : –Process EUV images, and detect when a flare is happening. –In case of detection, give all characteristics such as localization, characteristics such as localization, size, intensity, time duration. size, intensity, time duration. –Catalog EUV flares Tool : Tool : –Look for increased skewness of the pixel intensity distribution Flaring or non flaring ? SOHO-EIT, 1 May 98, 18:19
4 Skewness : a measure of assymetry Sudden brightness enhancement increased skewness in image histogram Sudden brightness enhancement increased skewness in image histogram Skewness of difference image histogram Coronal Dimmings detection Skewness of difference image histogram Coronal Dimmings detection –Cfr NEMO-Dimming detection by Elena Podladchikova & David Berghmans Gaussian distribution Chi 2 Skewness = 0 Skewness > 0 Skewness: Mean:
5 B2X on secchi_prep’ed FSFR images: raw 2k 2 Goes B Flare A EUVI A - Skewness EUVI B - Skewness
6 Discussion of previous graphs Pre-processing needed against Cosmic Ray Hits (CRH) Pre-processing needed against Cosmic Ray Hits (CRH) Possible strategies: Possible strategies: –Rebinning –Median filter for CRH removal –Wavelet denoising –Exclude the far off-disc region (and hence off-disk CRH)
7 B2X on secchi_prep’ed FSHR images Images 2x2 rebinned 1k 2 Goes B Flare EUVI A EUVI B
8 Discussion : B2X on Images 2x2 rebinned 1k 2 Still lots of unwanted increase in the skewness Still lots of unwanted increase in the skewness –CRH cleaning : remove pxl > 3 (local median) –Wavelet denoising
9 B2X on secchi_prep’ed FSHR images : Median filter pre-proc for CRH removal Goes B Flare EUVI A EUVI B
10 ‘Problematic’ images 24 Jan’ 07 02:21:40 Pb in image acquisition 24 Jan’ 07 19:25:40 Post-flare activity EUVI A ICER 7 image EUVI B
11 CRH attenuation via Wavelet Denoising (1/2) Remove finest discrete detail coefficients and reconstruct Remove finest discrete detail coefficients and reconstruct (linear Wavelet denoising) (linear Wavelet denoising) Gibbs effect introduce small values near large value. Gibbs effect introduce small values near large value. –It reduces the asymmetry and hence the skewness 24 Jan’ 07 10:05:40
12 Remove first 2 levels of wavelet coefficients, and reconstruct (linear Wavelet denoising) Remove first 2 levels of wavelet coefficients, and reconstruct (linear Wavelet denoising) Gibbs effect introduce small values near large value. Gibbs effect introduce small values near large value. –It attenuates the asymmetry and hence the skewness On-disc flares less affected than off-disc Cosmic Ray Hits On-disc flares less affected than off-disc Cosmic Ray Hits 24 Jan’ 07 05:15:40 CRH attenuation via Wavelet Denoising (1/2)
13 B2X on secchi_prep’ed FSHR images : Linear Wavelet denoising Goes B Flare
14 Flare localisation EUVI-A 24 Jan :15:40 Mexican Hat continuous wavelet transform Scale a=10; find position of maxima Once an increase in skewness has been detected: -find the flare position by finding pixel with maximum intensity -Mask the flare and re-compute skewness to find possible other flare(s)
15 B2X- Flare detection quality: minimize False Detection Rate Correlation between flare candidate and AR position Correlation between flare candidate and AR position Barra, Delouille, Hochedez, ‘Segmentation of Extreme Ultraviolet Solar Images via Multichannel Fuzzy Clustering’, AdvSpRes, submitted. Segmentation of EUV disc -> localize Active Regions
16 Conclusion : B2X Combining: EUVI-A & EUVI-B images, pre-processing (CR removal, wavelet denoising) skewness measure, localisation information allows to detect with high confidence B to X flares.