TMTCPT: The Tree Method based on the Taxonomic Categorization and the Phylogenetic Tree for fine-grained...

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Fine-grained categorization is a complex machine vision challenge.
• Convolutional Neural Networks (CNNs) have significantly improved classification accuracy.
• Existing methods require further enhancement for nuanced visual tasks.

Purpose of the Study:

• To introduce a novel framework, TMTCPT, for improved fine-grained visual categorization.
• To leverage taxonomic categorization and phylogenetic trees within a CNN framework.
• To enhance classification accuracy through hierarchical feature analysis.

Main Methods:

• Developed a tree-based method (TMTCPT) integrating taxonomic categorization, phylogenetic trees, and CNN classifiers.
• Employed hierarchical categorization with multiple classification levels, from general to highly similar features.
• Utilized phylogenetic information to represent organismal relationships.

Main Results:

• The TMTCPT method demonstrated high efficiency and robustness on the CUB-200-2011 dataset.
• Achieved an average classification accuracy of 88.34%.
• Outperformed all previously reported methods on the benchmark dataset.

Conclusions:

• The proposed TMTCPT framework effectively addresses fine-grained visual categorization challenges.
• The integration of taxonomic and phylogenetic structures with CNNs offers a powerful approach.
• This method significantly advances the state-of-the-art in fine-grained image classification.