Journal of Engineering and Applied Sciences
ISSN: Online 1818-7803ISSN: Print 1816-949x
Abstract
The Hyperspectral Images (HSI) acquired by remote sensors are characterized by hundreds of contiguous channels with high spectral resolution. Hyperspectral image classification is the process of assigning land cover classes to pixels. Classifying remotely sensed data is a challenge because many factors such as complexity of landscape, image processing and classification approaches affect the success of classification. Convolutional Neural Networks (CNN) are gaining attention due to their capability to automatically discover relevant relative features in image classification problems. The proposed approach employs totally eight layers in which four convolutional layers, two pooling layers and two Fully Connected Network (FCN) hidden layers to extract features from hyperspectral images. This method is able to extract the features invariably for their location and distortion which leads to better classification accuracy. By extracting both spatial and spectral information, the performance of the model is a promising one for high dimensional HSI with few available training data. This method has been applied to University of Pavia and Indian pines datasets. The resultant of the model demonstrates very good classification accuracy within limited number of training epochs.
How to cite this article:
B.R. Shivakumar and J. Prakash. A Novel Method for Remotely Sensed Hyperspectral Image Classification
Based on Convolutional Neural Network.
DOI: https://doi.org/10.36478/jeasci.2020.13.22
URL: https://www.makhillpublications.co/view-article/1816-949x/jeasci.2020.13.22