Using tri-training to exploit spectral and spatial information for hyperspectral data classification

A semi-supervised classification method for hyperspectral data using a joint spectral and spatial analysis is proposed. In the method, the dimensionality reduction process is followed by the computation of textural features via the gray level co-occurrence matrices (GLCM) and markov random field (MRF). Three classifiers are used based on the labeled samples from the spectral data and two spatial features, respectively. These classifiers are refined using the unlabeled samples in the tri-training process, and an improvement in the final classification accuracy is achieved. Experiments on two hyperspectral data sets indicate that the proposed method can effectively integrate the information from the spectra and texture, labeled and unlabeled samples for classification.