A preview of the PDF is not available
Adaptif Komşuluk Seçimi ve Ağırlık Atama Yöntemleri ile Hiperspektral Görüntülerin Sınıflandırılması
Sparse representation based techniques are frequently used in the areas such as signal and image processing, computer vision and pattern recognition due to their performance. In recent years, sparse representation techniques have been used in the proposed methods related to classification of hyperspectral images and favorable results have been obtained. In this article, a joint sparse representation based classifier method that uses adaptive neighborhood selection and weighting processes together is proposed. First, instead of including all of the pixels in the fixed size window created around the test pixel in the classification process, pixels that are close and have similar spectral characteristics to the test pixel are included in the classification. In this way, neighboring pixels which are distant and unlike spectrally to the test pixels are discarded. Then, when determining the class label of the test pixel, the sparse coefficient matrix in the residual value to be calculated is multiplied by the weights determined for each class. Similarity between selected pixels and training dictionary of each class has been considered when the weights are determined. In this way, the probability of assigning the test pixel to the proper class has been increased.