Abstract: Beehive sound-based monitoring the states of a honeybee colony, including health and preswarming, is an important means for precise beekeeping. However, the present classifiers of a beehive sound are mostly developed according to the multiple features of the beehive sound using complex algorithms of machine learning. In the study, the "artificial swarming" method to get the beehive swarming was used. The beehive sounds of Apis cerana were recorded and analyzed under preswarming and non-preswarming conditions, and the power spectral density and the sub-band power ratio of the sounds were extracted. The results showed that the frequency ranges of the non-preswarming beehive sound were mainly within 0-200 Hz, while those of the preswarming beehive sound within 200-400 Hz. Based on the sub-band power ratio of the beehive sound, a machine learning classifier was developed for predicting the preswarming state of a honeybee colony with a CART decision tree algorithm, and a higher prior probability of 99.04% was achieved. This study would provide a new feature for developing the classifiers of a beehive sound.

Key words: Apis cerana, beehive sound, swarming, sub-band power ratio, machine learning, decision tree