AIpollen: An Analytic Website for Pollen Identification Through Convolutional Neural Networks.

Area of Science:

• Computational biology
• Botany
• Artificial intelligence

Background:

• Deep learning (DL) excels in complex tasks like computer vision.
• Accurate pollen identification is crucial in various scientific fields.
• Existing methods may lack the precision and efficiency needed for large-scale analysis.

Purpose of the Study:

• To develop a high-precision deep learning system for pollen identification.
• To leverage DL for efficient and accurate classification of pollen grains across diverse genera.

Main Methods:

• A dataset of pollen images from 36 genera was constructed.
• A pre-trained ResNet34 network was fine-tuned for pollen classification.
• Training incorporated Adam optimizer, cross-entropy loss, ELU activation, data augmentation, learning rate decay, and early stopping.

Main Results:

• The model achieved 97.01% accuracy on the test set and 99.89% on the training set.
• An F1 score of 95.9% demonstrated good balance and robustness across categories.
• A user-friendly web interface was developed for easy image uploading and genus prediction.

Conclusions:

• A highly accurate and robust deep learning model for pollen identification was successfully developed and validated.
• The system provides an efficient tool for researchers, aiding in pollen analysis and classification.
• The developed web interface enhances accessibility and usability for pollen identification tasks.