A Comparative Study of Hybrid Machine-Learning vs. Deep-Learning Approaches for Varroa Mite Detection and Counting.

Area of Science:

• Agricultural Science
• Computer Science
• Entomology

Background:

• Varroa destructor mites are a significant threat to honeybee colonies worldwide.
• Accurate and efficient detection of these mites is crucial for effective pest management and preserving bee health.
• Hyperspectral imaging offers a promising non-invasive method for mite detection.

Purpose of the Study:

• To comparatively evaluate traditional machine learning (ML) and deep learning (DL) approaches for detecting and counting Varroa destructor mites.
• To assess the performance of ML and DL models using hyperspectral images under varying data conditions.
• To provide practical guidance on selecting appropriate mite detection strategies based on resource availability and performance requirements.

Main Methods:

• A traditional ML pipeline using Principal Component Analysis (PCA), k-Nearest Neighbors (kNN), and Support Vector Machine (SVM) was implemented.
• A deep learning (DL) approach utilizing Faster R-CNN with ResNet-50 and ResNet-101 backbones was fine-tuned on the same dataset.
• Both methods were evaluated on hyperspectral images for mite detection and counting accuracy.

Main Results:

• The ML pipeline achieved high performance (precision = 0.9983, recall = 0.9947) with rapid training and inference on CPU.
• The DL models (ResNet-50/101) achieved lower precision (0.966/0.971) and recall (0.757/0.829) despite requiring GPU acceleration and longer training times.
• ML demonstrated superior robustness under limited-data conditions compared to DL.

Conclusions:

• Traditional ML methods are more suitable for Varroa destructor mite detection in hyperspectral images within resource-constrained environments.
• DL approaches, while powerful, present challenges in training time, computational resources, and reproducibility for this specific application.
• The study provides valuable insights for selecting optimal mite detection strategies in apiculture, balancing performance with practical implementation.