An assessment of vegetation cover of Mysuru City, Karnataka State, India, using deep convolutional neural networks

Abstract

Mysuru City is a unique place in India due to its culture, green cover, historical places, and pleasant weather. In the last few decades, the city was witnessed rapid urban growth. This present work is conducted to assess the decadal changes in Mysuru City vegetation cover using multispectral remotely sensed data of 2009 and 2019 within Mysuru City Corporation (MCC). The main objective of this work is to assess the vegetation cover of the city and generate the land use and land cover classes (LULC) map using the deep learning model. Therefore, convolutional neural network (CNN)-based multiple training round (CNN-MTR) deep learning model is proposed and used for the classification of remote sensing images. The classified results were analyzed to assess the vegetation cover changes in the city over one decade. Vegetation cover within the Mysuru City Corporation area was estimated in 2019 to be 39.09% as compared to 43.32% in 2009. These results indicate that over a decade vegetation cover of Mysuru City is decreased by 3.43%. The overall classification accuracy of the proposed CNN-MTR model was estimated to be 95.20% for 2009 and 94.17% for 2019 respectively.