Unsupervised feature learning based on sparse coding and spectral clustering for segmentation of synthetic aperture radar images

摘要: Synthetic aperture radar (SAR) image segmentation is fundamental for the interpretation and understanding of these images. In this process, representation SAR features plays an important role. Spectral clustering method making it possible to combine cues. This study presents a new spectral using unsupervised feature learning (UFL). method, primarily processed by non-negative matrix factorisation (NMF) algorithm then containing spatial structure information are extracted. Afterwards, extracted learned sparse coding increase discrimination power features. Sparse which finds patterns or high-level semantics data. Ultimately, operation performed applying on learns simultaneously creates similarity function required in through production coefficients. Therefore avoids Gaussian function, has problem with scale parameter adjustment that one drawbacks methods. The results demonstrate that, compared wavelet GLCM features, NMF manage obtain more meaningful provide better result. have also demonstrated significantly improved unlearned experimental indicate effect UFL segmentation.