UAV-based tree species classification using DenseNet121 with transfer learning on visible light images.

Area of Science:

• Environmental Science
• Computer Science
• Remote Sensing

Background:

• Traditional tree species classification methods face limitations in accuracy and scalability.
• Unmanned Aerial Vehicle (UAV)-based visible light imagery offers a promising alternative for detailed environmental monitoring.
• Deep learning architectures are increasingly vital for analyzing complex image datasets.

Purpose of the Study:

• To introduce and evaluate a novel approach for tree species classification using UAV-based visible light imagery.
• To leverage the DenseNet121 deep learning architecture for enhanced feature extraction in tree species identification.
• To assess the effectiveness of transfer learning in optimizing DenseNet121 performance for this specific task.

Main Methods:

• Utilized UAV-captured visible light imagery of diverse tree species.
• Employed the DenseNet121 deep learning model, known for its dense connectivity for superior feature extraction.
• Applied transfer learning to fine-tune the pre-trained DenseNet121 model on the custom tree species dataset.

Main Results:

• The DenseNet121 model with transfer learning significantly outperformed established models like ResNet, VGG, and MobileNet.
• Achieved exceptional classification accuracy, precision, recall, and F1-score metrics.
• Demonstrated the model's robustness and effectiveness in complex image classification tasks.

Conclusions:

• Combining DenseNet121 and transfer learning provides a highly effective and scalable solution for UAV-based tree species classification.
• This approach establishes a reliable, cost-effective baseline for forestry monitoring and ecological studies.
• The method is particularly valuable for environmental and ecological studies in data-scarce regions.