A NOVEL TRANSFER LEARNING BASED DEEP MODEL FOR LAND CLASSIFICATION

Abstract

A systematic framework for comprehending the qualities and possibilities of different land portions is provided by land classification, which makes it easier to make well-informed decisions and implement sustainable land management techniques across a range of industries. LULC has numerous significant uses in a variety of fields such as Urban planning, agriculture, natural resource management, environmental assessment, infrastructure development, disaster management, Tourism and recreation, Transportation planning, water resource management, etc. Contemporary trends show deep learning technology has achieved very good results in land classification and thus the land classification has become very attractive prospect for research and development. However, having a stable model in terms of training and testing will be the need of an hour. In this proposed system, we have utilized the high performing deep learning model to address the uncertainty prediction in the model. We have used Bayesian model to tackle the uncertainty prediction in the model. Our proposed system is marginally compromised the training and validation accuracy; However, results have shown that the loss curve generated from training and testing a stable model is substantially important to ensure it’s learning rate stability as well as general confidence in real-time production of the deep learning model.