Land Use Classification of Kathgodam Region using Transfer learning-based approach

Abstract

Transfer learning is a new concept that has the potential to propel machine learning further in research and industry. One of the key reasons to use it is the absence of data on specific jobs, as gathering and classifying data can be costly and time-consuming. In machine learning, transfer learning is used to reprocess a formerly trained model on an original problem. In transfer learning, a machine uses knowledge from a previous assignment to improve the prediction about a new task. Transfer learning allows CNNs to learn with small amounts of data by transferring knowledge from models that have been pre-trained on big datasets. In this paper we have used various pretrained models like Inception v3, SqueezeNet, VGG-19 and VGG-16 to study the performance of pre-trained models to classify Agriculture land and non-Agriculture land in Kathgodam Region. The capabilities of the four pre-trained CNN models SqueezeNet, Inception V3, VGG16, and VGG19 for extracting features from images were explored by the authors. The accuracy of 98.5% is obtained using Inception V3.