Vpliv stiskanja jpeg2000 z izgubami na klasifikacijo posnetkov worldview-2 ; Effects of lossy jpeg2000 compression method on worldview-2 image classification

Stiskanje z izgubami se vse pogosteje uporablja v daljinskem zaznavanju, eprav njegov vpliv na rezultate obdelav posnetkov e ni v celoti raziskan. V lanku so predstavljeni u inki stiskanja JPEG2000 z izgubami na klasifikacijo posnetkov zelo visoke lo ljivosti WorldView-2. Za klasifikacijo stisnjenih posnetkov smo prvi uporabili objektni metodi k-najbli ji sosed (k-nearest neighbor) in metodo podpornih vektorjev (support vector machine), ki smo ju tudi primerjali. Rezultati razkrivajo vpliv stiskanja na same posnetke, segmentacijo in kon no klasifikacijo. tudija dokazuje, da v splo nem stiskanje z izgubo ne vpliva negativno na klasifikacijo posnetkov, e ve , v nekaterih primerih ima klasifikacija stisnjenih posnetkov bolj e rezultate kot klasifikacija izvirnih posnetkov. Natan nost klasifikacij z metodo podpornih vektorjev ka e na mo nost stiskanja podob do razmerja 30 : 1 brez izgube natan nosti klasifikacije. Najbolj i rezultat z metodo k-najbli ji sosed smo pridobili z najvi jim kompresijskim razmerjem (100 : 1). V raziskavi smo ugotovili, da metoda podpornih vektorjev daje bolj e rezultate klasifikacije kot metoda k-najbli ji sosed ter se priporo a tudi za nadaljnje raziskave. Poleg metode klasifikacije ima pri natan nosti rezultatov pomembno vlogo tudi segmentacija posnetkov ; Lossy compression is becoming increasingly used in remote sensing, although its effect on the processing results has yet not been fully investigated. This paper presents the effects of JPEG2000 lossy compression on the classification of very high-resolution WorldView-2 imagery. For the first time, the k-nearest neighbor and support vector machine methods of the object based classification were used. The results explore the impact of compression on the images, segmentation and resulting classification. The study proves that in general lossy compression does not adversely affect the classification of images; moreover, in some cases classification of compressed images yields better results than classification of the original image. Classification accuracy of the support vector machine method indicates that compression ratios of up to 30:1 can be used without any loss of classification accuracy. The best result of the k-nearest neighbor method was obtained with the highest compression ratio (100:1). The support vector machine is recommended for further research. In addition to the classification method, image segmentation also plays an important role in the accuracy of the results.