An Approach for Detecting Mangrove Areas and Mapping Species Using Multispectral Drone Imagery and Deep Learning.

Area of Science:

• Ecology
• Remote Sensing
• Computer Science

Background:

• Mangrove ecosystems are vital for coastal biodiversity and ecological balance.
• Efficient and intelligent monitoring methods are crucial for mangrove conservation.
• Traditional methods for mangrove identification lack accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate novel deep learning models for mangrove detection and species identification.
• To integrate multi-scale spectral and spatial information for improved mangrove mapping.
• To enhance the accuracy and efficiency of mangrove monitoring using high-resolution drone imagery.

Main Methods:

• Developed MangroveNet for mangrove area detection, integrating spectral and spatial information.
• Developed AttCloudNet+ for mangrove species identification using high-resolution multispectral drone images.
• Incorporated spectral and spatial attention mechanisms into both models.

Main Results:

• MangroveNet achieved superior accuracy (99.13%) and generalization capabilities compared to other deep learning models.
• AttCloudNet+ demonstrated optimal performance in mangrove species identification (Kappa: 0.81, OA: 0.87) against traditional and machine learning methods.
• Both models confirmed effectiveness in real-time monitoring of mangroves and their species.

Conclusions:

• The developed deep learning models offer efficient and accurate solutions for mangrove monitoring.
• Dual attention mechanisms enhance the performance of models in processing spectral and spatial data.
• These models provide a valuable tool for the conservation and management of mangrove ecosystems.