IoT-Based Bee Swarm Activity Acoustic Classification Using Deep Neural Networks.

Area of Science:

• Agricultural technology
• Bioacoustics
• Machine learning

Background:

• Acoustic monitoring is crucial for beekeeping, especially for remote hive management.
• Classifying bee swarm activity from audio signals is an emerging application.

Purpose of the Study:

• To propose and evaluate a deep neural network (DNN) Internet of Things (IoT)-based system for acoustic bee swarm classification.
• To compare audio formats (WAV vs. MP3) and analyze DNN parameter impacts on classification accuracy.

Main Methods:

• Utilized audio recordings from the Open Source Beehive project.
• Extracted Mel-frequency cepstral coefficients (MFCCs) as features.
• Implemented and analyzed DNN models for classification.

Main Results:

• Achieved a maximum classification accuracy of 94.09% with uncompressed WAV audio.
• MP3 audio compression resulted in a >10% decrease in DNN accuracy.
• The proposed DNN system demonstrated superior performance compared to previous Hidden Markov Models (HMM).

Conclusions:

• Deep neural networks offer a viable IoT-based solution for acoustic bee swarm classification.
• Audio compression significantly impacts classification accuracy, highlighting the need for lossless formats in IoT applications.
• The DNN approach represents a significant advancement over HMMs for monitoring bee activity.