Recognition of queen loss state of honeycomb based on multi-domain feature fusion

The loss of the queen is a critical signal indicating the need for human intervention in the honey bee breeding process. Sound serves as a vital source of information that reflects the state of the hive and can be used to detect the presence or absence of the queen. In this paper, we propose a method for recognizing the queen-loss state of the hive based on multi-domain feature fusion. The method extracts features from the time, frequency, and cepstral domains of hive sounds, and selects the optimal feature subset FBest through feature screening. One-dimensional sound signals are converted into two-dimensional time-frequency representations using Constant Q Transform (CQT), and deep features are extracted using shallow Convolutional Neural Networks (CNNs). A feature-level fusion strategy is employed to combine traditional acoustic features with deep features, and a support vector machine (SVM) model is used for classification. Validation results on a real hive sound dataset demonstrate that the proposed method achieves an accuracy of 96.67% in recognizing the queen-loss state, which is significantly higher than that of traditional methods. The findings provide essential technical support for the artificial breeding of honey bees.