Non-Intrusive System for Honeybee Recognition Based on Audio Signals and Maximum Likelihood Classification by Autoencoder

Abstract

Artificial intelligence and Internet of Things are playing an increasingly important role in monitoring beehives. In this paper, we propose a method for automatic recognition of honeybee type by analyzing the sound generated by worker bees and drone bees during their flight close to an entrance to a beehive. We conducted a wide comparative study to determine the most effective preprocessing of audio signals for the detection problem. We compared the results for several different methods for signal representation in the frequency domain, including mel-frequency cepstral coefficients (MFCCs), gammatone cepstral coefficients (GTCCs), the multiple signal classification method (MUSIC) and parametric estimation of power spectral density (PSD) by the Burg algorithm. The coefficients serve as inputs for an autoencoder neural network to discriminate drone bees from worker bees. The classification is based on the reconstruction error of the signal representations produced by the autoencoder. We propose a novel approach to class separation by the autoencoder neural network with various thresholds between decision areas, including the maximum likelihood threshold for the reconstruction error. By classifying real-life signals, we demonstrated that it is possible to differentiate drone bees and worker bees based solely on audio signals. The attained level of detection accuracy enables the creation of an efficient automatic system for beekeepers.

Keywords: anomaly detection; artificial intelligence; autoencoder neural network; beehive monitoring; power spectral density; signal processing; smart beehives.

Figure 1

Honeybees Apis mellifera L.: male drone (left) and female worker bee (right). Photography: Beltsville Agriculture Research Center, public domain.

Figure 2

Signal processing diagram.

Figure 3

Diagram of PSD estimation by MUSIC.

Figure 4

The MFCC extraction diagram.

Figure 5

The GTCC extraction diagram.

Figure 6

General structure of autoencoder neural network.

Figure 7

Example of poor class separation: worker bee (training set), drone bee (test set). Histograms of the MSE loss produced by autoencoder neural network with marked threshold values T1, T2, T3, and T*.

Figure 8

Example of excellent class separation: worker bee (training set), drone bee (test set). Histograms of the MSE loss produced by the autoencoder neural network with marked threshold values T1, T2, T3, and T*.

Figure 9

Practical illustration of maximum likelihood approach to the classification with the optimal threshold T* separating the decision areas.

Figure 10

Accuracy of classification of worker bees and drones by autoencoder neural networks with 1, 2, 3, or 4 encoder hidden layers for threshold values: T1, T2, T3.

Figure 11

F1-score for classification of worker bees and drones by autoencoder neural networks with 1, 2, 3, or 4 encoder hidden layers for threshold values: T1, T2, T3.

Figure 12

A novel approach to autoencoder neural networks: transformation of high-dimensional classification problem based on feature vectors to a one-dimensional one based on autoencoder reconstruction error.

Figure 13

Our evaluation board prototype with LTE module and SIM card slot, designed for data acquisition, signal processing, and wireless communication with a beehive.