In-Field Detection of American Foulbrood (AFB) by Electric Nose Using Classical Classification Techniques and...

Area of Science:

• Apiculture and Bee Health
• Sensor Technology and Data Analysis
• Veterinary Diagnostics

Background:

• American foulbrood is a highly contagious and lethal bee disease impacting colony survival.
• Accurate and timely diagnosis is crucial for disease management and prevention.
• Current diagnostic methods can be time-consuming and require laboratory access.

Purpose of the Study:

• To evaluate the efficacy of an electronic nose device (Beesensor V.2) in differentiating American foulbrood-affected bee colonies from healthy ones.
• To assess the performance of various machine learning algorithms for classifying bee colony health status based on sensor data.
• To determine the potential of the electronic nose as a tool for rapid, on-site diagnosis of American foulbrood.

Main Methods:

• Field experiments were conducted on 18 bee colonies (9 infected, 9 healthy).
• The Beesensor V.2, equipped with TGS gas sensors, collected volatile organic compound data over 40-minute sessions per colony.
• Data analysis involved classical machine learning (k-NN, Naive Bayes, SVM, Random Forests) and sequential neural networks, with various data preparation techniques.

Main Results:

• The electronic nose achieved classification accuracies between 65-75%, demonstrating practical potential.
• Both classical and sequential neural network methods showed comparable performance.
• Visualizations confirmed the separability of data from healthy and infected colonies.

Conclusions:

• The Beesensor V.2, coupled with appropriate classification algorithms, can serve as a valuable tool for facilitating rapid diagnosis of American foulbrood in field conditions.
• This sensor-based approach offers a promising alternative for early detection and management of bee diseases.
• Further research and validation are warranted to optimize the system for widespread apicultural use.