Abstract

Hyperspectral Imaging (HSI) is a powerful technology for remotely inferring the material properties of the objects in a scene of interest.  In HSI classification, however, the dimensionality of hyperspectral images cause redundancy in information, especially in spatial-spectral feature domain, which reduces the classification and detection accuracy of the objects. To overcome the above mentioned problems, proposed work consists of three major contributions for hyperspectral image samples .In the first work, multiple feature selection is carried out through a Firefly Algorithm (FA) with Kullback Leibler Divergence - Local-Fisher’s Discriminant Analysis (LFDA) (FA-KLD-LFDA) is proposed for dimensionality reduction to remove irrelevant feature, select most important features and also Multiple Kernels Learning (MKL) based Support Vector Machine (SVM) is proposed for classification of hyperspectral image samples to enhance classification accuracy. In the second work, Restricted Bipartite Graphs (RBG) is applied for Target Detection of Hyperspectral Images. Gaussian Firefly Algorithm (GFA) for multiple feature selection and LFDA are applied for dimensionality reduction.  Hybrid Genetic Fuzzy Neural Network (HGFNN) classifier is applied for labeled and unlabeled samples further to isolating the labeled samples from each other into different classes. In the third work, Kernalized Generalized Learning Vector Quantization (KGLVQ) is carryout for anomaly detection. Modified Fish Swarm Algorithm (MFSA) with LFDA based schema for feature selection. Modified Adaptive Neuro-Fuzzy Inference System (MANFIS) algorithm for classification of hyperspectral images is also discussed. The experimentation proposed schemas are also experimented to hyperspectral images of Coimbatore samples with region of interests.

Paper Details