Plant-CNN-ViT: Plant Classification with Ensemble of Convolutional Neural Networks and Vision Transformer.

Area of Science:

• Botany
• Computer Science
• Machine Learning

Background:

• Plant leaf classification is crucial for species identification.
• Machine learning models improve accuracy but require extensive training data.
• Data scarcity poses a significant challenge for many plant species.

Purpose of the Study:

• To develop an accurate and efficient plant leaf classification model.
• To overcome the limitations of data-dependent machine learning models.
• To leverage ensemble learning by combining diverse deep learning architectures.

Main Methods:

• Proposed a Plant-CNN-ViT ensemble model integrating Vision Transformer, ResNet-50, DenseNet-201, and Xception.
• Vision Transformer uses self-attention for feature focus.
• ResNet-50, DenseNet-201, and Xception employ residual, dense, and separable convolutions for efficient feature extraction.

Main Results:

• Achieved 100.00% accuracy on the Flavia and Folio Leaf datasets.
• Obtained 100.00% accuracy on the Swedish Leaf dataset and 99.83% on the MalayaKew Leaf dataset.
• Demonstrated the effectiveness of the ensemble approach in plant leaf classification.

Conclusions:

• The Plant-CNN-ViT ensemble model significantly enhances plant leaf classification accuracy.
• The model effectively addresses the challenge of limited training data.
• This approach offers a robust solution for botanical species identification.