Central Attention and a Dual Path Convolutional Neural Network in Real-World Tree Species Recognition

Area of Science:

• Botany
• Computer Science
• Artificial Intelligence

Background:

• Accurate plant identification is crucial for professionals and the public, but current deep learning models struggle in real-world scenarios.
• Convolutional Neural Networks (CNNs) show promise for plant recognition but require improvements for in-field application.

Purpose of the Study:

• To develop an advanced deep learning framework for accurate tree species recognition, addressing limitations of existing methods.
• To enhance the focus on target trees by reducing background interference in image recognition.

Main Methods:

• Proposed a dual-path CNN deep learning framework integrating a central attention model with an InceptionV3-based CNN.
• Employed a shared classification layer for joint learning of the central attention and CNN models.
• Created a comprehensive tree image database featuring whole-tree information.

Main Results:

• The proposed dual-path approach significantly outperformed single-path models and existing methods in tree species recognition.
• The central attention mechanism effectively focused on the tree, mitigating background distractions.
• The system demonstrated superior performance in identifying tree species from diverse images.

Conclusions:

• The novel central attention concept integrated into a dual-path CNN framework offers a robust solution for in-field tree species identification.
• The developed system, including a dedicated image database and mobile platform, provides real-time, accurate identification capabilities.
• This research advances automated plant recognition technology for broader accessibility and application.