,
SRM Institute of Science and Technology , Chennai , India
,
SRM Institute of Science and Technology , Chennai , India
,
SRM Institute of Science and Technology , Chennai , India
,
SRM Institute of Science and Technology , Chennai , India
SRM Institute of Science and Technology , Chennai , India
Terrain Type Identification is an essential aspect of environmental monitoring, urban planning, and resource management. This study discusses a comparative investigation of the effectiveness of three Convolutional Neural Network (CNN) architectures Simple CNN, ResNet-50, and the Enhanced Efficient CNN Model which is Efficient Net in training with the Euro SAT dataset of 27,000 geo-referenced Sentinel-2 satellite images in 10 land cover classes. This paper also examines the role of preprocessing techniques including image resizing, data augmentation, and normalization in improving the overall performance of the models. The effectiveness of each model was closely observed by looking at how they recognized different land cover types, ensuring balanced performance across all classes. Experimental results indicate that the Efficient Net model has the highest classification accuracy of 97.5%, followed by Simple CNN with the accuracy of 94% and ResNet-50 with the accuracy of 89%. Furthermore, simple scaling preserves RGB features better than grayscale conversion. The study's results emphasize the need for selecting suitable model structures and preprocessing methods for successful terrain type identification through the use of remote sensing imagery, which concludes with the creation of efficient deep learning models for remote sensing.
This is an open access article distributed under the Creative Commons Attribution Non-Commercial License (CC BY-NC) License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
0
The statements, opinions and data contained in the journal are solely those of the individual authors and contributors and not of the publisher and the editor(s). We stay neutral with regard to jurisdictional claims in published maps and institutional affiliations.