Deep Learning for Plant Species Classification Using Leaf Vein Morphometric.

Area of Science:

• Computer Science
• Botany
• Artificial Intelligence

Background:

• Automated plant identification systems are crucial for botanists and the public.
• Deep learning excels at extracting detailed features from images, aiding identification.

Purpose of the Study:

• To propose and evaluate a novel Convolutional Neural Network (CNN)-based method, D-Leaf, for automated plant species identification.
• To compare the performance of D-Leaf against pre-trained and fine-tuned AlexNet models and traditional morphometric methods.

Main Methods:

• Leaf images were pre-processed and features extracted using three CNN models: pre-trained AlexNet, fine-tuned AlexNet, and D-Leaf.
• Extracted features were classified using five machine learning techniques: Support Vector Machine (SVM), Artificial Neural Network (ANN), k-Nearest-Neighbor (k-NN), Naïve-Bayes (NB), and CNN.
• A conventional morphometric method served as a benchmark for performance comparison.

Main Results:

• The D-Leaf model achieved a testing accuracy of 94.88%, comparable to fine-tuned AlexNet (95.54%) and pre-trained AlexNet (93.26%).
• CNN models significantly outperformed traditional morphometric measurements, which achieved 66.55% accuracy.
• Features extracted by CNNs demonstrated strong compatibility with the Artificial Neural Network classifier.

Conclusions:

• The D-Leaf model shows significant potential as an effective automated system for plant species identification.
• Deep learning-based feature extraction offers superior performance compared to traditional morphometric analysis.
• The integration of CNNs with classifiers like ANN provides a robust approach for botanical identification tasks.