Mangrove Species Classification from Unmanned Aerial Vehicle Hyperspectral Images Using Object-Oriented Methods Based...

Area of Science:

• Ecology
• Remote Sensing
• Computer Science

Background:

• Accurate spatial distribution of mangrove species is vital for ecological conservation.
• Hyperspectral imaging offers effective mangrove monitoring but faces challenges in fine classification due to spatial complexity and spectral redundancy.
• Spectral similarities among mangrove species hinder accurate classification using spectral information alone.

Purpose of the Study:

• To develop an object-oriented multi-feature combination method for the fine classification of mangrove species.
• To evaluate the effectiveness of various spectral, vegetation indices, fractional order differential, texture, and geometric features in mangrove classification.
• To compare the performance of different machine learning classifiers for mangrove species identification.

Main Methods:

• Hyperspectral images were segmented using multi-scale techniques for object extraction.
• A genetic algorithm was employed for feature selection from spectral, vegetation indices, fractional order differential, texture, and geometric features.
• Ten feature combination schemes were tested with K-nearest neighbor (KNN), support vector machines (SVM), random forests (RF), and artificial neural networks (ANN) classifiers.

Main Results:

• Support vector machines (SVM) using texture features achieved 97.04% accuracy.
• Artificial neural networks (ANN) utilizing a combination of raw spectra, first-order differential spectra, texture, vegetation indices, and geometric features reached 98.03% accuracy.
• Texture and fractional order differentiation were key features, with vegetation indices and geometric features further enhancing accuracy.

Conclusions:

• The proposed object-oriented multi-feature combination method significantly improves mangrove species classification accuracy and efficiency compared to pixel-based methods.
• This approach provides robust technical support for mangrove restoration and management by enabling precise species identification.
• The study highlights the importance of combining diverse features and employing object-based analysis for complex ecological monitoring.