A LIGHTWEIGHT CNN ARCHITECTURE FOR LAND CLASSIFICATION ON SATELLITE IMAGES

Abstract

A city might be thought of geographically as a patchwork of dense land uses.The logical and prudent method of allocating the land resources that are accessible is called land use, For example, settlements, arable fields, pastures, and managed woods.It is a method of making use of the land, which includes resource allocation, planning, and management.A piece of land's physical characteristics and its intended purpose are related.Despite the abundance of geographical data, there is a dearth of study that proves this connection.The purpose of this study is to connect a city's physical form to its purpose.Currently, satellite-based data can be found in vast collections of high-resolution photos from numerous satellites, including Sentinel, Landsat 8, and others.It is challenging to classify the land cover in these photographs because of the enormous amount of data and wide range of types.In this context, Deep Neural Networks can be extremely useful because they can classify these enormous amounts of data.Similar research in the field have relied on simpler models and a substantial number of manually created parameters, requiring subject-matter expertise.It is acknowledged that the majority of models are too superficial for such a nuanced picture.The Convolutional Neural Network (CNN) is a kind of Neural Network that is mostly utilised for image and speech recognition applications.Its built-in convolutional layer decreases picture dimensionality without sacrificing information.That is why CNNs are ideal for this application.Convolutional neural networks (CNNs) are neural networks with one or more convolutional layers that are primarily utilised for image processing, classification, segmentation, and other auto-correlated data.