부산대학교 도서관

The proposed method extracts nonlinear, classifier, and invariant deep features from HSIs using a variety of convolution and convolution features. Both target recognition and image classification can gain from these characteristics. In order to address the normal problem of imbalance with large dimensions and a dearth of training sets for the identification of HIS, a few approaches, such as L2 regularisation and dropout, are being investigated to minimise dimensionality in class data modelling. A 3-D Fox news FE model with linked regularisation is also provided by study to efficiently extract spectral and spatial properties from hyperspectral data. Finally, a strategy for augmenting virtual samples is proposed to further improve performance. The suggested techniques are put to the test using three well-known hyper spectral data sets: “Indian Pines, Universidad of Pavia, and Nasa Space Centre”. Unwanted scattering from other objects might cause the object of interest's spectral characteristics to change.