Varroa destructor detection on honey bees using hyperspectral imagery

Abstract: Hyperspectral (HS) imagery in agriculture is becoming increasingly common. These images have the advantage of higher spectral resolution. Advanced spectral processing techniques are required to unlock the information potential in these HS images. The present paper introduces a method rooted in multivariate statistics designed to detect parasitic Varroa destructor mites on the body of western honey bee Apis mellifera, enabling easier and continuous monitoring of the bee hives. The methodology explores unsupervised (K-means++) and recently developed supervised (Kernel Flows - Partial Least-Squares, KF-PLS) methods for parasitic identification. Additionally, in light of the emergence of custom-band multispectral cameras, the present research outlines a strategy for identifying the specific wavelengths necessary for effective bee-mite separation, suitable for implementation in a custom-band camera. Illustrated with a real-case dataset, our findings demonstrate that as few as four spectral bands are sufficient for accurate parasite identification.

• The paper demonstrates that hyperspectral imaging combined with advanced clustering methods enables accurate detection of Varroa mites on honey bees.
• Methodology includes unsupervised (K-means++) and supervised (KF-PLS) clustering, leveraging PCA and wavelength selection to reduce spectral bands.
• Implications include developing cost-effective monitoring systems for bee colony health and aiding early detection of Colony Collapse Disorder.

Varroa Destructor Detection on Honey Bees Using Hyperspectral Imagery

Introduction

The paper "Varroa destructor detection on honey bees using hyperspectral imagery" investigates the application of hyperspectral imaging (HSI) for identifying Varroa destructor mites on the body of the western honey bee, Apis mellifera. The study focuses on developing spectral processing techniques to differentiate these mites from bees using multivariate statistical methods, including unsupervised clustering with K-means++ and supervised clustering with Kernel Flows - Partial Least-Squares (KF-PLS). The paper outlines the potential for enhanced monitoring of bee hives and identifies critical wavelengths necessary for effective bee-mite separation, with applications for custom-band multispectral cameras.

Figure 1: Bees mixed-up with detritus.

Methodology

Hyperspectral Imaging Setup

The research employs a Specim IQ portable hyperspectral camera for image acquisition, utilizing a multispectral unit with LEDs covering the spectral range of 400 nm to 1000 nm. Images are collected and processed to extract spectral signatures for clustering and classification purposes.

Figure 2: Used Specim IQ camera.

Mathematical Techniques

The study leverages Principal Component Analysis (PCA) for spectral data decomposition, selecting components highly correlated with bee-mite discrimination variables for spectral reconstruction. K-means++ is used for unsupervised clustering, employing spectral profiles for efficient cluster identification. For supervised clustering, KF-PLS utilizes kernel projections in a Reproducing Kernel Hilbert Space to handle non-linear relationships in the spectral data.

Wavelength Selection

Two PLS-based methods are used for selecting discriminative wavelengths: an $R^2$-based method and a modified Covariance Procedure (COVPROC). These methods reduce the wavelength set from 204 bands to those essential for discrimination between bees and Varroa mites.

Results

The research demonstrates that four spectral bands are sufficient for accurate parasite identification. Unsuspected clustering methods demand spectral pre-processing, with K-means++ requiring projection from highly correlated PCA components. KF-PLS provides efficient discrimination with lower preprocessing needs. The identified wavelengths critical for bee-mite separation include bands around 492.97 nm and 796.74 nm, aligning partly with prior multispectral studies, but identifying discrepancies that could lead to improved methodologies.

Figure 3: Cluster formation workflow.

Discussion

This study marks the first application of hyperspectral imaging for Varroa mite detection. It identifies specific spectral bands crucial for discrimination, posing significant implications for real-time hive monitoring. The findings could enable the development of systems utilizing monochromatic illumination for routine hive assessments. Future work will expand dataset diversity and methodology to include broader detritus analysis and refine spectral discrimination techniques.

Conclusion

The paper illustrates the potential for hyperspectral imagery in bee colony health monitoring, outlining effective clustering techniques for Varroa mite identification. By identifying essential wavelengths for discrimination, the study provides insights that can be applied in developing cost-effective monitoring systems, contributing to global efforts in managing bee health and addressing Colony Collapse Disorder (CCD).

The paper's innovation and methodologies are supported by extensive references in hyperspectral imaging, multivariate statistics, and bee health monitoring literature, acknowledging advancements in spectral analysis and computational techniques. The discussion encourages further exploration into real-time monitoring capabilities and potential improvements in hive health assessments using hyperspectral technologies.