Machine learning approach for automatic recognition of tomato-pollinating bees based on their buzzing-sounds

Area of Science:

• Agricultural Science
• Bioacoustics
• Machine Learning

Background:

• Bee pollination significantly enhances tomato fruit yield and quality.
• Accurate identification of efficient bee pollinators is crucial for maximizing agricultural economic returns.
• Traditional bee identification methods are labor-intensive and require expert knowledge.

Purpose of the Study:

• To evaluate the efficacy of Machine Learning (ML) algorithms in automatically recognizing bee species visiting tomato flowers.
• To compare ML performance using Mel Frequency Cepstral Coefficients (MFCC) versus fundamental frequency analysis of bee sounds.
• To determine the relevance of sonication versus flight sounds for taxonomic identification.

Main Methods:

• Collected buzzing sound recordings from bees visiting tomato flowers.
• Applied ML algorithms, including Support Vector Machines (SVM), with MFCC and fundamental frequency features.
• Compared classification accuracy for species and genera recognition based on sonication and flight sounds.

Main Results:

• ML algorithms, particularly SVM with MFCC, showed superior performance in taxonomic recognition compared to frequency-based methods.
• Sonication sounds provided more relevant information for bee species recognition than flight sounds alone.
• ML achieved higher accuracy in recognizing bee genera from flight sounds, though overall accuracy (max 73.39%) requires improvement.

Conclusions:

• ML techniques show potential for automating the identification of bee pollinators for tomato and other buzz-pollinated crops.
• Further research with larger datasets and unsupervised learning may enhance classification accuracy.
• Automated bee identification can support farmers in optimizing pollination and increasing crop yields.