Description

Hyperspectral images show similar statistical properties to natural grayscale or color photographic images. However, the classification of hyperspectral images is more challenging because of the very high dimensionality of the pixels and the small number of labeled examples typically available for learning. These peculiarities lead to particular signal processing problems, mainly characterized by indetermination and complex manifolds. There are several algorithm has been proposed to classify the hyperspectral image. In our project a new novel method has been introduced that is Harmonic Analysis based classification such as HA-PSO-SVM approach. This new approach accurately classify the cluster band with respect to their amplitude and phase. Harmonic Analysis (HA) is introduced to extract the feature from hyperspectral image. Amplitude and phase features has been obtained by derived HA. Then select best feature among extracted feature by Particle Swarm Optimization. Finally, classify the respective band with related cluster which is performed with the help of Support Vector Machine (SVM). This classifier accurately classify the band to respective cluster form. In prior work, instead of HA, used MNF, PCA and ICA could extract features and also for combination of PSO-SVM could use CV-SVM and GA-SVM. HA is first proposed to transform the pixels from spectral domain into frequency domain expressed by amplitude, phase and residual, yielding more functional and discriminative features for classification purpose. The Reconstructed Image is adopted with PSO to optimize the penalty parameter C and the kernel parameter for SVM. Finally, the extracted features are classified with the optimized model. The main novelty and contributions of our work consist of twofold. By acting HA on a single pixel, a smooth curve in frequency domain represented by amplitude, phase, and remainder can be obtained, which is considered to be more functional and discriminative for classification. Different from the traditional feature extraction techniques, HA can better capture the functional relationships between spectral bands taking into account the adjacent bands to extract the frequency features.