Detection of leaf miner in sweet potato crops through image analysis using machine learning-based models

Area of Science:

• Agricultural Science
• Computer Science
• Image Analysis

Background:

• Leaf miner (Liriomyza huidobrensis) poses a significant threat to sweet potato crops.
• Traditional visual inspection methods are slow and subjective, hindering efficient pest management.

Purpose of the Study:

• To develop and evaluate deep learning models for automated detection of the leaf miner pest.
• To compare the performance of YOLOv8s and YOLOv11s architectures for pest identification.

Main Methods:

• A dataset of 751 images (
• camote_minador
• ") was created and annotated.
• Dynamic data augmentation was applied to enhance training variability.
• YOLOv8s and YOLOv11s models were comparatively evaluated using standardized hyperparameters.

Main Results:

• The YOLOv11s model achieved a Precision of 73.20%, Recall of 66.72%, and mAP@50 of 71.63%.
• YOLOv11s demonstrated superior performance in discriminating pest galleries from background noise compared to YOLOv8s.
• A functional mobile prototype using TensorFlow Lite for mid-range devices was developed.

Conclusions:

• The YOLOv11s architecture is a technically superior and effective solution for automated leaf miner detection.
• This technology offers an accessible and scalable approach to enhance agricultural phytosanitary monitoring.
• Deep learning-based image analysis provides a robust alternative to traditional pest inspection methods.