SARImageSegmentation Based Onimmune Algorithm

Fulong Ma,Han Bao-jun,Licheng Jiao
2005-01-01
Abstract:Theimmunealgorithm-IA isproposed with analogies totheconcept andthetheory ofimmunity inbiotic science. Itinherits theGA'sadvantages andavoids thedete- rioration-phenomenon. TheIAneedsthefeature orknowl- edgeofproblem-space asthevaccine. Spatial matrix describes theprobabilities thatoneareaistheneighbor ofother areas. Different areas inSARimages havedifferent textural features andhavebigcontrast. Therefore thespatial matrix ofaSAR imagehasthecharacter thattheprobability oftheareasof samekindisthelargest andtheprobability oftheareasof different kinds isthesmallest. Using this feature asthevaccine this paperemploys IAtofind thebestsegmentation-threshold. Simulation results showthatthis methodiseffective forSAR imagesegmentation. I. INTRODUCTION Withthecharacteristics ofall-weather observation andhigh resolution, theSynthetic Aperture Radar (SAR)hasbecome animportant method forthecurrent remote sensing obser- vation, widely usedinareas ofresource, environment, ar- chaeology andmilitary affairs. Compared with theoptical image, theSARground object image hasclear profile, withbetter contrast andrich tex- tural features; andtheobjects ofvarious shapes andphysi- calcharacteristics showvarious textural features. Forex- ample, thesurfaces ofobjects suchasbuildings andbridges havethecutoff andjumping structure inthesight ofinci- dence radar energy; sothat thereflected powerishigh, and consequently, thegreyvalue ofimaging isalso high. There- fore, thelike artificial objects canbeeasily detected and recognized withSARimage. Weall knowthat theautomatic recognition isahotspot ofimage recognition. However, theSARimage suffers its multiplicative speckle noise, andthere arenoeffective noise reduction methods right now,causing diffilculties with theautomatic object recognition bycomputer, because itcan't bemadesegmentation withthetraditional thresh- olding algorithms.
What problem does this paper attempt to address?