Land Classification Using Satellite Imagery

Abstract

Using information from remote sensing, the process of identifying different land cover types—such as plants, water bodies, and soil—is known as land cover classification. This is useful for anticipating future interactions between human activity and the environment as well as for discovering significant facts about the surface of the Earth. These expected interactions lead to the formation of sustainable land use practices and the preservation of natural resources. This study focuses on classifying land cover using both supervised and unsupervised methods. The supervised method for classification uses random forest classifiers, while the supervised approach for land cover detection uses CNN. The values of the evaluation parameters are calculated and compared for the input and output photographs.

Introduction

I. INTRODUCTION

The term "land cover" describes the natural and man-made characteristics that blanket the surface of the Earth. It consists of various land use forms, vegetation, and bodies of water. Understanding how this land cover is classified is a crucial first step towards comprehending Earth's biophysical system. Classified land cover is useful for estimating the distribution and availability of natural resources as well as for monitoring the health of ecosystems. A method for analysing the effects of climate change on ecosystems and determining an area's susceptibility to natural catastrophes like floods, landslides, and forest fires is the land cover classifier.[1] Classifying land cover using satellite multiband imagery with 12 bands is the aim of this study. Land, water, and vegetation are the three primary categories into which satellite data is divided. This would help the country's future development initiatives become safer and more effective.

II. LITERATURE SURVEY

A. Existing Work

Title: Deep Learning for Land-Cover Classification in Satellite Imagery: A Review,Authors: Liangpei Zhang, Qiangqiang Yuan, Xiao Xiang Zhu, Christiane Schmullius,Published: IEEE Geoscience and Remote Sensing Magazine, 2016,Summary: This paper provides a comprehensive review of deep learning techniques, including CNNs and RNNs, for land cover classification in satellite imagery. It discusses various architectures, methodologies, and applications in the field.

Title: Multiscale Recurrent Neural Networks for Hierarchical Conditional Random Fields-based Scene Labeling, Authors:      Shuai Zheng, Sadeep Jayasumana, Bernardino Romera-Paredes, Vibhav Vineet, Zhizhong Su, Dalong Du, Chang Huang, Philip H. S. Torr,Published: CVPR, 2015, Summary: This paper presents a framework that combines CNNs with multiscale RNNs for hierarchical scene labeling, which can be adapted for land cover classification in satellite imagery.

Title: Deep Recurrent Neural Networks for Hyperspectral Image Classification,Authors: Zhihui Wang, Dingwen Zhang, Fei Yan, Hengchao Li, Yubao Sun,Published: IEEE Transactions on Geoscience and Remote Sensing, 2017,Summary: This paper explores the use of deep recurrent neural networks, specifically Long Short-Term Memory (LSTM) networks, for hyperspectral image classification, which can also be applied to land cover classification in satellite imagery.

Title: DeepSat - A Learning framework for Satellite Imagery,Authors: Keiller Nogueira, Gustavo Batista, Adriano Veloso

Published: Pattern Recognition Letters, 2017,Summary: This paper introduces DeepSat, a deep learning framework that combines CNNs and RNNs for the classification of satellite imagery. It presents experiments on various datasets, including land cover classification tasks.

III. PROBLEM STATEMENT

In the pursuit of optimizing agricultural practices and resource allocation, there exists a critical need to accurately differentiate between cultivable and non-cultivable lands. This project aims to address this pressing challenge through the utilization of advanced models applied to satellite imagery analysis.

IV. PROPOSED WORK

A. Arhcitechture of the Model

B. Satellite  image  Acquisition

The  initial  phase  includes  obtaining  satellite  imagery  statistics  from  legitimate  resources  which  includes  Sentinel  and  Landsat.  This  records  encompasses  multispectral  imagery  capturing  various  wavelengths  of  light,  crucial  for  discerning  special  land  cover  types.

C. Data  Preprocessing

Following  acquisition,  the  satellite  imagery  undergoes  preprocessing  to  make  sure  quality  and  consistency.  tasks  include  correcting  geometric  distortions,  eliminating  atmospheric  outcomes,  and  resampling  to  a  standardized  decision.

D. Feature  Extraction

Feature  extraction  is  pivotal  for  discerning  relevant  information  from  satellite  imagery.  techniques  consisting  of  spectral  analysis,  texture  evaluation,  and  flora  indices  computation  are  employed  to  extract  discriminative  features  like  plant  life  density,  water  bodies,  and  urban  regions.

E. Model  architecture  layout

The middle  of  the  technique  includes  designing  a  hybrid  CNN-RNN  model  tailor-made  for  land  cover  type.  CNNs  are  utilized  for  spatial  feature  extraction  from  satellite  imagery,  at  the  same  time  as  RNNs  seize  temporal  dependencies  over  sequential  pictures,  permitting  dynamic  land  cover  analysis.

F. Training  the  model

The  model  is  skilled  the  use  of  categorised  satellite  imagery  data,  with  ground  truth  annotations  for  unique  land  cover  training.  Supervised  mastering  strategies  are  carried  out  to  optimize  version  parameters  and  improve  category  accuracy.

G. Model  evaluation

The  trained  model's  overall  performance  is  evaluated  the  use  of  popular  metrics  including  typical  accuracy,  precision,  recall,  and  F1  score.  additionally,  visual  inspection  of  classification  maps  and  confusion  matrices  aids  in  assessing  classification  performance  and  identifying  regions  for  development.

H.  Integration  and  application

Upon  successful  assessment,  the  skilled  model  is  incorporated  into  GIS  structures  or  remote  sensing  software  for  practical  software.  it  could  be  used  for  duties  which  include  land  cover  mapping,  environmental  tracking,  urban  planning,  and  natural  aid  management.  normal  updates  and  refinement  ensure  the  version's  persevered  relevance  and  effectiveness  in  actual-global  situations.

Conclusion

The CNN architecture proposed in this paper can be used to extract scene information from satellite photos. The majority of alternative models that have been put forth in the literature start with a basic CNN model and include elements unique to satellite imagery. The AI community does not typically have access to such domain knowledge. The model has fewer parameters but is still developed along the lines of VGG. Additionally, our model outperformed all other architectures with an accuracy of 99.84 and 99.47 on the SAT4 and SAT6 satellite image datasets, respectively, thanks to the approaches of batch normalisation and dropout. Compared to other methods, the training and testing phase only takes thirty epochs, which is a very little amount of time. A further benefit is that a large number of photos can be batch processed without the requirement for preprocessing. This model will be used to process the state\'s data and determine its land resources.

References

[1] Title: Deep Learning for Land-Cover Classification in Satellite Imagery: A Review,Authors: Liangpei Zhang, Qiangqiang Yuan, Xiao Xiang Zhu, Christiane Schmullius ,Published: IEEE Geoscience and Remote Sensing Magazine, 2016 [2] Title: DeepSat - A Learning framework for Satellite Imager Authors: Keiller Nogueira, Gustavo Batista, Adriano Velos, Published: Pattern Recognition Letters, 2017 [3] Title: Deep Recurrent Neural Networks for Hyperspectral Image Classification, Authors: Zhihui Wang, Dingwen Zhang, Fei Yan, Hengchao Li, Yubao Sun,Published: IEEE Transactions on Geoscience and Remote Sensing, 2017 [4] Title: Multiscale Recurrent Neural Networks for Hierarchical Conditional ,Authors: Shuai Zheng, Sadeep Jayasumana, Bernardino Romera-Paredes, Vibhav Vineet, Published: CVPR, 2015 [5] Chaudhuri, S., C. H. Chen, C. S. Chen, and S. H. Hu. \"Remote sensing image scene classification: Benchmark and state of the art.\" Proceedings of the IEEE 101.10 (2013): 2108-2121. [6] Ma, L., Tan, X., Wang, J., Xie, H., & Wu, H. (2020). Land cover classification using deep learning: A review. Remote Sensing, 12(1), 135. [7] Pal, M., & Mather, P. M. (2005). Support vector machines for classification in remote sensing. International Journal of Remote Sensing, 26(5), 1007-1011. [8] Li, W., Fu, J., & Sun, Y. (2020). Deep learning in remote sensing applications: A meta-analysis and review. ISPRS Journal of Photogrammetry and Remote Sensing, 162, 212-234. [9] Pal, M., & Mather, P. M. (2005). Support vector machines for classification in remote sensing. International Journal of Remote Sensing, 26(5), 1007-1011 [10] Mertes, C., & Eskofier, B. M. (2018). Deep learning: A prerequisite for the real-time processing of high-resolution satellite data? Remote Sensing, 10(12) [11] Yuan, Xiao Xiang Zhu, Christiane Schmullius ,Published: IEEE Geoscience and Remote Sensing Magazine, 2016 [12] Title: DeepSat - A Learning framework for Satellite Imager Authors: Keiller Nogueira, Gustavo Batista, Adriano Velos, Published: Pattern Recognition Letters, 2017 [13] Title: Deep Recurrent Neural Networks for Hyperspectral Image Classification, Authors: Zhihui Wang, Dingwen Zhang, Fei Yan, Hengchao Li, Yubao Sun,Published: IEEE Transactions on Geoscience and Remote Sensing, 2017 [14] Title: Multiscale Recurrent Neural Networks for Hierarchical Conditional ,Authors: Shuai Zheng, Sadeep Jayasumana, Bernardino Romera-Paredes, Vibhav Vineet, Published: CVPR, 2015 [15] Lu, D., & Weng, Q. (2007). A survey of image classification methods and techniques for improving classification performance. International Journal of Remote Sensing, 28(5), 823-87.

Copyright

Copyright © 2024 Krish Pathak, Tanvi Thopte, Romir Maji, Jaishnu Kotian, Mrs. Shweta Yadav. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.