EcoTransLearn: an R-package to easily use transfer learning for ecological studies-a plankton case study

Area of Science:

• Ecology
• Computer Science
• Bioinformatics

Background:

• Deep Learning (DL) excels at image recognition but building models for ecological studies is challenging due to organism variability and high labeling costs.
• Transfer Learning (TL) offers a solution by leveraging large pre-trained datasets to improve classification with limited ecological data.

Purpose of the Study:

• To introduce EcoTransLearn, an R-package designed to automate image classification in ecological research.
• To provide an accessible tool for ecologists, reducing the technical barriers associated with DL model development.

Main Methods:

• The EcoTransLearn package utilizes Transfer Learning (TL) methods pre-trained on the ImageNet dataset.
• It integrates R with Python (using reticulate and tensorflow) for image classification tasks.
• The package supports images from various devices like FlowCam and ZooScan.

Main Results:

• EcoTransLearn enables automated classification of ecological images.
• It facilitates the application of advanced TL techniques in ecological studies.
• The package streamlines the process of species identification from various imaging sources.

Conclusions:

• EcoTransLearn addresses the need for user-friendly software in ecological image analysis.
• The package promotes wider adoption of DL and TL in ecological research.
• It enhances the efficiency and accuracy of classifying biological specimens from diverse image data.