Investigation on new Mel frequency cepstral coefficients features and hyper-parameters tuning technique for bee sound recognition

Due to the important role of honey bees in life, there have been increasingly advanced techniques introduced to support apiarists in taking the best care of their bees. Recently, machine learning (ML) methods emerge as a powerful tool among these techniques and have great contributions to automated beehive monitoring systems with low cost and better performance. By analyzing the data collected from beehives, ML algorithms are able to solve a number of crucial problems in monitoring the beehives such as early detecting the phenomena of swarming, identifying the bee queen’s absence, and recognizing pest infestations. In this study, we suggest several techniques to enhance the performance of the machine learning models applied to monitoring the honey beehives. Particularly, we apply an advanced technique for tuning hyper-parameters of the ML models and investigate the new Mel frequency cepstral coefficients (MFCCs) features. The obtained results show that our proposed methods can improve significantly the accuracy of these ML-based models in recognizing and classifying the bee buzzing from other ambient noises, making them even better than plural deep learning algorithms suggested in the literature. In addition, we introduce a new dataset of bee sound samples and we verify the efficiency of our proposed models on the new dataset.