Related Experiment Video
Updated: Jan 18, 2026
05:16
Flying Insect Detection and Classification with Inexpensive Sensors
Published on: October 15, 2014
25.7K
Utilizing CNNs for classification and uncertainty quantification for 15 families of European fly pollinators
Thomas Stark1, Michael Wurm1, Valentin Ştefan2,3,4
1German Aerospace Center (DLR), German Remote Sensing Data Center (DFD), Oberpfaffenhofen, Germany.
Plos One
|September 10, 2025
Summary
Automated monitoring of European pollinating flies (Diptera) using Convolutional Neural Networks (CNNs) achieved 95.61% accuracy. This AI approach enhances biodiversity and food security by improving pollinator identification and reducing misclassifications.
Area of Science:
- Ecology
- Computer Science
- Entomology
Background:
- Pollination is vital for biodiversity and food security, with flies (Diptera) being key European pollinators.
- Traditional monitoring methods for pollinators are labor-intensive and costly.
- Automated monitoring has largely overlooked flies due to identification challenges.
Purpose of the Study:
- To investigate the efficacy of Convolutional Neural Networks (CNNs) for classifying European pollinating fly families.
- To quantify classification uncertainty using Monte Carlo methods.
- To assess the impact of image cropping on classification performance and confidence.
Main Methods:
- Utilized three CNN architectures: ResNet18, MobileNetV3, and EfficientNetB4.
- Developed a dataset of diverse Diptera images, including wing morphology and body habitus.
- Employed Monte Carlo dropout and test-time augmentation to estimate aleatoric and epistemic uncertainty.
Main Results:
- Achieved an overall classification accuracy of up to 95.61% for 15 European fly families.
- Cropping images to Diptera bounding boxes improved accuracy by a mean of 5.58% and increased prediction confidence by 8.56%.
- Demonstrated CNNs' effectiveness in distinguishing fly families and reducing misclassifications.
Conclusions:
- CNNs offer a powerful tool for accurate and automated identification of pollinating flies.
- Image preprocessing, such as bounding box cropping, significantly enhances model performance and reliability.
- This AI-driven approach advances pollinator monitoring, supporting biodiversity and food security initiatives.
