A Novel Method for the Separation of Overlapping Pollen Species for Automated Detection and Classification

Area of Science:

• Botany
• Computer Science
• Environmental Science

Background:

• Automated pollen identification is crucial for palynology applications like climate change studies, allergy calendars, and forensics.
• Overlapping pollen in microscopic images presents a significant challenge for automated analysis.

Purpose of the Study:

• To develop an automated system for capturing and classifying 12 species of pollen from Mexico's Lagunera Region.
• To address the challenge of overlapping pollen through advanced segmentation techniques.

Main Methods:

• A novel method was developed to segment overlapping pollen images.
• Pollen separation was achieved using mean shift process and H-minima erosion based on the Fibonacci series.
• Pollen characteristics (shape, color, texture) were used to train and evaluate Random Tree Forest, Multilayer Perceptron, and Bayes Net classifiers.

Main Results:

• The system achieved 100% segmentation accuracy for overlapping pollen.
• Classification performance exceeded 96.2% recall and 96.1% precision using the Multilayer Perceptron classifier with twofold cross-validation.

Conclusions:

• The developed system effectively segments and classifies overlapping pollen species.
• High accuracy in automated pollen identification has significant implications for various scientific fields.