Related Experiment Video
Updated: Jun 29, 2026
04:21
Sieving Fruit Pulp to Detect Immature Tephritid Fruit Flies in the Field
Published on: July 28, 2023
Tephritid26: A standardized, multi-angle image dataset of quarantine-significant true fruit flies for deep
Zitao Li1,2, Xingkai Wang1,2, Zhuojie Wu3
1State Key Laboratory of Agricultural and Forestry Biosecurity, College of Plant Protection, China Agricultural University, Beijing, 100193, China.
Scientific Data
|June 27, 2026
Summary
A new multi-angle image dataset, Tephritid26, aids agricultural biosecurity by enabling rapid identification of quarantine-significant tephritid fruit flies using deep learning. This resource significantly advances automated pest detection tools.
Area of Science:
- Agricultural Entomology
- Computer Vision
- Biosecurity
Background:
- Accurate identification of quarantine-significant tephritid fruit flies is crucial for global agricultural biosecurity.
- Current deep learning applications are hindered by a lack of comprehensive public image datasets for these species.
Purpose of the Study:
- To introduce Tephritid26, a novel multi-angle image dataset designed to address the scarcity of data for training deep learning models for tephritid identification.
- To establish a benchmark for developing automated identification tools in phytosanitary applications.
Main Methods:
- Assembled Tephritid26 dataset comprising 38,081 images from 1,473 specimens across 26 tephritid species, seven genera, and two subfamilies.
- Employed a novel specimen mounting protocol and a rotational imaging setup to capture multi-angle perspectives, simulating real-world inspection conditions.
- Trained various deep learning models (ResNet-50, ConvNeXt-B, Vit-Small, Swin-Tiny) for species identification using the Tephritid26 dataset.
Main Results:
- Deep learning models achieved high species-level accuracy, with Macro-Averaged F1-scores exceeding 96.75%.
- Gradient-weighted Class Activation Mapping (Grad-CAM) analysis indicated that models focused on taxonomically informative morphological regions for identification.
- The Tephritid26 dataset proved effective for training accurate deep learning models.
Conclusions:
- Tephritid26 is a valuable resource for advancing automated identification tools in agricultural biosecurity and phytosanitary applications.
- The dataset facilitates the development and validation of deep learning models for rapid and accurate tephritid identification.
- This work supports enhanced global pest management strategies through improved AI-driven detection systems.
