Intelligent beehive status monitoring in noisy environment
View fulltext online
Citation:Biswas, S. (2020). Intelligent beehive status monitoring in noisy environment. (Unpublished document submitted in partial fulfilment of the requirements for the degree of Master of Computing). Unitec Institute of Technology, Auckland, New Zealand. Retrieved from https://hdl.handle.net/10652/4930
Permanent link to Research Bank record:https://hdl.handle.net/10652/4930
Artificial neural network (ANN) based bee-hive monitoring algorithm does not perform very efficiently in noisy environments. Although an ANN based bee-hive monitoring algorithm using audio signals of a beehive could perform with very high accuracy in a noise-free environment. Such a beehive monitoring algorithm faces gap in classification accuracy when noise is mixed to the sound of the beehive. There are evidences of many research on building beehive monitoring algorithms but none of them investigated the noise tolerance capacity of the system. This study aims to demonstrate the effect of environmental noise on beehive monitoring algorithm’s performance and establish a solution for overcoming this problem. Building on existing work on beehive monitoring algorithms one could ask: Could the enhancement of beehive audio feature extraction method effectively increases the noise handling capability of a beehive monitoring algorithm? Based on the literature review included in this thesis, a beehive monitoring algorithm with an enhanced feature extraction method is proposed. In this context, Multilayer perceptron (MLP) artificial neural network is used for classifying the states of the beehive. New feature extraction method extracts the audio feature from the targeted frequency range with 3 overlapping sub-bands. These 3 feature sets from 3 overlapping sub-bands are later combined to form the final and enhanced feature sets. Performance of the proposed system is tested in a computer-simulated environment with 4 different noise levels. The system is designed to detect the queenless and queenright states. Analysis of the simulation data shows that the new feature extraction method is highly efficient and performs better in a noisy environment compared to the conventional feature extraction method.