Related Experiment Video
Updated: Oct 4, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.7K
Using Convolutional Neural Networks to Efficiently Extract Immense Phenological Data From Community Science Images
Rachel A Reeb1, Naeem Aziz1, Samuel M Lapp1
1Department of Biological Sciences, University of Pittsburgh, Pittsburgh, PA, United States.
Frontiers in Plant Science
|February 3, 2022
Summary
Deep learning models can accurately extract plant phenology data from community science images, matching human performance. Image quality, not the AI model, is key for reliable phenological research using platforms like iNaturalist.
Area of Science:
- Plant Ecology
- Computational Biology
- Citizen Science
Background:
- Community science platforms, such as iNaturalist, house vast, underutilized datasets for phenological research.
- Manual phenological annotation of these images is labor-intensive, limiting research scope and scale.
- Deep learning offers a potential solution for automating phenological data extraction from large image archives.
Purpose of the Study:
- To assess the accuracy and efficiency of deep learning models in annotating plant phenophases from iNaturalist images.
- To compare the performance of a convolutional neural network (CNN) against non-expert human annotators for phenological classification.
- To identify challenges and provide guidelines for improving image quality in community science for phenological analysis.
Main Methods:
- A convolutional neural network (CNN) was trained to classify phenophases in Alliaria petiolata images from iNaturalist.
- The CNN's classification accuracy was evaluated for two-stage (flowering/non-flowering) and four-stage (vegetative, budding, flowering, fruiting) phenologies.
- Model performance was compared to that of non-expert human annotators.
Main Results:
- The CNN achieved 95.9% accuracy for two-stage phenology and 86.4% accuracy for four-stage phenology.
- The CNN's overall accuracy was comparable to non-expert human annotators (p = 0.383), though performance varied by phenophase.
- Image quality issues (improper distance, manipulation, alteration) in up to 4% of images posed a significant challenge for both human and machine annotation.
Conclusions:
- Deep learning techniques are effective for extracting phenological information from community science image data.
- The primary limitation for automated phenological analysis stems from the quality of community science images, not the deep learning models themselves.
- Establishing photography guidelines for community scientists can enhance the utility of image data for phenological research.
