Dense connected convolution neural network for land cover classification

Abstract

Hyperspectral Imaging is employed to monitor the earth regions on basis of spectral continuous data ranges initializing from visible wave infrared region to short wave infrared region of the electromagnetic spectrum. It authorizes the detailed recognition and classification of land cover on account of spectral feature space. Hyperspectral images seemed to be presented by employing traditional unsupervised and supervised classifier with regards to classification. Various problems seemed to cause Hughes phenomenon as it represents the curse of dimensionality issues. In spite of mitigating those challenges, a deep ensemble classification model seemed to be proposed in this work. It process the data features using various convolution layers of the network along modelling the activation function as a simple structure for classification of the hyperspectral data based on the spectral values using Softmax layer and error function to minimize the losses. Dense Connected Convolution Neural Network projected in this work as it has high potential to effectively classify the spectral features with learnt weights from one individual convolution layer to convolution layers. The main idea of Dense Convolution Neural Network is to produce discriminative classification results and to enhance the accuracy and diversity of a classifier simultaneously.