CNN Based Automated Land Use Classification from Remotely Sensed Images

EasyChair Preprint 6613

Abstract

A land use classification from a remotely sensed satellite images is one of the important applications which provides information on changes in land cover and land use over a period of time. It may also facilitate the assessment of environmental impacts on and potential or alternative uses of land. In the ever changing environment, the land use change has been occurring every day. For the continuous land use changes, there is a need for creating the current land use data. Therefore, a Convolutional Neural Network (CNN) architecture is proposed to classify the remotely sensed image. The proposed architecture contains two convolutional layers followed by a fully connected layer. The fully connected layer contains three hidden layers with a single neuron in the output layer. In order to train the network, the dataset has been created manually by collecting images from Google Earth. The dataset contains 201 images for two classes, namely, building and non- building areas. Out of 201 images 157 images are used for training the network and 44 images are  used for testing the network. Binary Cross Entropy (BCE) loss is used for measuring the performance of the network. The network parameters are randomly initialized and updated with Gradient descent with momentum optimization algorithm. The proposed architecture achieves 99% training accuracy and 93% testing accuracy.

Keyphrases: Binary Cross-entropy, CNN, Land use classification

@booklet{EasyChair:6613,
  author    = {S Preethi and C J Pravenkumar and K P Rahul and K Ramanath and R Ramyea and A Arun Prakash},
  title     = {CNN Based Automated Land Use Classification from Remotely Sensed Images},
  howpublished = {EasyChair Preprint 6613},
  year      = {EasyChair, 2021}}