Dynamic granular neural networks for remote sensing image classification

Document Type

Conference Article

Publication Title

International Geoscience and Remote Sensing Symposium (IGARSS)


Land use and land cover classification of remote sensing data is useful in taking relevant decisions of an area. There are various types of automated methodologies which are used for the classification of remote sensing images. Neural networks (NNs) are such systems which are known for their ability to classify the data with highly overlapping classes. But, NNs face problems like uncertainty in the data during the classification. Granulation of data handle uncertainty present in the remote sensing data, and neural networks which are used to classify granulated data are called granular neural networks (GNNs). In order to accommodate the incoming labelled pattern, the size of granules and the number of granules has to evolve. This specify the necessity of systems that can compute dynamic granules. In the present study, we propose dynamic granular neural networks (DGNNs) for the classification of remote sensing images. Architecture of DGNNs evolves according to the incoming data, which posses better classification ability compared with similar types of model. Performance of the proposed model has been tested with Indian remote sensing linear imaging self scanner (IRS LISS III) and hyper spectral remote sensing (HSRS) reflective optics system imaging spectrometer (ROSIS) datasets. Superiority of DGNNs over similar type of models has been verified with performance metrics like overall accuracy (OA) and kappa coefficient (KC).

First Page


Last Page




Publication Date


This document is currently not available here.