胡蜂攻击蜂群的声学特征与模式识别研究

Acoustic Characteristics and Pattern Recognition of Wasp Attacking Bee Colony

胡蜂攻击蜂群是蜜蜂养殖过程中的主要敌害之一,通过声音手段进行蜂群的监测管理是智慧养蜂业的重要手段.本文以意大利蜜蜂为研究对象,通过实验过程中采集到的胡蜂攻击蜂群的声学数据,从声音信号的声谱图特征分析出发,选择了短时能量、短时过零率、短时平均幅度、声音复杂度等9个声音特征参数,利用支持向量机技术进行模式识别建模.结果表明,在径向基核函数的支持下选择9维参数参与建模,精度可达到84.70%;利用主成分变换后的6维参数进行建模,精度可达到86.23%.利用随机森林分析技术对选择的9维变量进行分析,得到短时过零率、频谱质心、音频均匀度指数、短时平均幅度这4个参数对建模获得的重要性最大.

Wasp attack is one of the main enemies in the process of bee breeding.Monitoring and management of bee colony by means of sound is an important means of intelligent beekeeping.In this research,Apis mellifera were taken as the research object.Based on the acoustic data collected during the experiment,nine acoustic characteristic parameters,such as short-term energy,short-term zero-crossing rate,short-term average amplitude and sound complexity,were selected from the spectral analysis of sound signals.The support vector machine analysis technology was used to model the pattern recognition.The results show that the accuracy of the model can reach 84.70%by choosing 9-dimensional parameters with the support of radial basis function,and 86.23%by using 6-dimensional parameters after principal component transformation.Random forest analysis technology was used to analyze the selected 9-dimensional variables,and it is found that such four parameters as short-time zero-crossing rate,spectral centroid,audio evenness index and stMeanAmp are most important for the modeling.