Multi-Class Classification of Aerial Imagery Using Deep Learning and Ensemble Models

Authors: Cibaca Khandelwal

DOI: https://doi.org/10.5281/zenodo.14838573

Short DOI: https://doi.org/g84g6p

Country: USA

Full-text Research PDF File:   View   |   Download

Abstract: The classification of aerial imagery is a critical task in various domains, including urban planning, agriculture, and disaster management. Recent advancements in deep learning have enabled the development of automated systems capable of accurately analyzing aerial images. Aerial imagery provides valuable insights into land use, environmental changes, and disaster mitigation strategies. This paper explores multi-class classification of aerial images using state-of-the-art deep learning models, emphasizing their potential applications and limitations. The study evaluates the performance of several pre-trained convolutional neural network (CNN) architectures, including ResNet50 [1], MobileNetV2 [2], EfficientNetB0, VGG16 [3], and DenseNet121 [4], on a benchmark aerial imagery dataset. DenseNet121 achieved the highest validation accuracy of 96%, outperforming other architectures. This paper highlights the importance of model selection, data augmentation, and stratified data splitting for effective aerial image classification. The results provide actionable insights for researchers and practitioners in adopting robust models for aerial image analysis.

Keywords: Aerial Image Classification, Deep Learning, Ensemble Models, Magnification-Aware Training, ResNet50, DenseNet121, EfficientNet, Land Use Analysis.

Paper Id: 232098

Published On: 2022-11-07

Published In: Volume 10, Issue 6, November-December 2022

Share this