Microscopic detection of nematodes in entomopathogenic nematode-enriched samples using a lightweight deep learning...

Area of Science:

• Agricultural Science
• Computer Science
• Nematology

Background:

• Manual counting of entomopathogenic nematodes (EPNs) is labor-intensive and variable.
• Automated detection is crucial for biological control research and nematode production.
• Existing computer vision methods lack architectural efficiency for microscopic imagery.

Purpose of the Study:

• To present LightDetectorMS, an ultra-lightweight, anchor-free object detection framework.
• To optimize automated detection of Steinernema feltiae infective juveniles in microscopic images.
• To evaluate the computational efficiency and detection reliability of the proposed model.

Main Methods:

• Developed an ultra-lightweight, anchor-free object detection framework (LightDetectorMS).
• Optimized the model for microscopic imagery of Steinernema feltiae.
• Evaluated performance using five-fold cross-validation and compared with manual expert counting.

Main Results:

• LightDetectorMS achieved high accuracy (mAP@0.5: 0.9119, mAP@0.5:0.95: 0.8207) with excellent precision and recall.
• The model is computationally efficient, with only 0.46 MB and processing at 152.5 FPS.
• Demonstrated statistical consistency with a coefficient of variation below 5% across metrics.
• Processed workloads thousands of times faster than manual expert counting.

Conclusions:

• LightDetectorMS offers a computationally efficient and statistically robust solution for EPN quantification.
• The framework supports semi-automated quantification in laboratory settings.
• Enables large-scale biological monitoring and production workflows through rapid, reliable detection.