Artificial Intelligence-Based Tools for Automated Genus-Level Identification of Plant-Parasitic Nematodes.

Area of Science:

• Agricultural Science
• Computer Vision
• Nematology

Background:

• Accurate identification and quantification of plant-parasitic nematodes (PPN) are vital for effective crop management.
• Current PPN identification methods (morphology, molecular markers) are often time-consuming and resource-intensive.
• Developing automated detection systems is crucial for timely intervention against nematode infestations.

Purpose of the Study:

• To develop and validate advanced computer vision tools for automated PPN detection.
• To achieve accurate, reproducible, and rapid identification of economically important PPN genera in potato crops.
• To assess the performance of an AI-based object segmentation algorithm for nematode classification.

Main Methods:

• Microscopic images of five nematode groups (root lesion, root-knot, stubby root, other PPN, non-parasitic) were captured (n=8,654).
• Images were preprocessed, annotated, and split into training (75%), validation (15%), and testing (10%) datasets.
• An object segmentation algorithm (YOLOv11-seg) was trained and evaluated on unseen images.

Main Results:

• The AI model achieved high accuracy: 92.4% (validation) and 88.6% (testing).
• Strong performance was observed for key PPN genera (RKN, RLN, SRN) with F1-scores >0.92 and AUC >0.93.
• The 'PPN-OTHERS' group showed poor performance (43.9% accuracy), often misclassified.

Conclusions:

• Artificial intelligence-based computer vision shows significant potential for automated PPN identification.
• The developed tools offer a faster and more reproducible alternative to traditional nematode detection methods.
• Further refinement is needed to improve classification accuracy for diverse PPN groups.