Optimizing Deep Learning Models for Climate-Related Natural Disaster Detection from UAV Images and Remote Sensing...

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Climate change is increasing the frequency and intensity of natural disasters.
• Effective monitoring and early detection of these disasters are crucial for mitigation and response.
• Aerial imagery provides valuable data for disaster assessment, but automated analysis is challenging.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) models for detecting natural disasters, specifically flooding and desertification, from aerial images.
• To create and utilize a novel dataset, the Climate Change Dataset, for training and comparing deep learning models.
• To demonstrate the potential of AI in addressing environmental challenges and supporting climate change adaptation.

Main Methods:

• A new dataset, the Climate Change Dataset, was compiled with 6334 aerial images from UAVs and satellites.
• Four machine learning (ML) models, including convolutional neural network (CNN), DenseNet201, VGG16, and ResNet50, were trained.
• Model performance was compared, and DenseNet201 was selected for optimization. Cross-validation was performed.

Main Results:

• All four ML models demonstrated high performance in detecting natural disasters.
• DenseNet201 and ResNet50 achieved the highest testing accuracies at 99.37% and 99.21%, respectively.
• The models successfully classified images into three categories: Flooded, Desert, and Neither.

Conclusions:

• AI, particularly deep learning models, shows significant potential for accurate and efficient natural disaster detection from aerial imagery.
• The developed Climate Change Dataset and optimized ML models offer a valuable resource for environmental monitoring and research.
• This AI-driven approach can enhance disaster response, contribute to environmental sustainability, and aid in climate change adaptation strategies.