论文部分内容阅读
昆虫的运动、取食、鸣叫都会发出声音,这些声音存在种内相似性和种间差异性,因此可用来识别昆虫的种类。基于昆虫声音的昆虫种类自动检测技术对协助农业和林业从业人员方便地识别昆虫种类非常有意义。本研究采用了语音识别领域里的声音参数化技术来实现昆虫的声音自动鉴别。声音样本经预处理后,提取梅尔倒谱系数(Mel-frequency cepstrum coefficient,MFCC)作为特征,并用这些样本提取的MFCC特征集训练混合高斯模型(Gaussian mixturemodel,GMM)。最后用训练所得到的GMM对未知类别的昆虫声音样本进行分类。该方法在包含58种昆虫声音的样本库中进行了评估,取得了较高的识别正确率(平均精度为98.95%)和较理想的时间性能。该测试结果证明了基于MFCC和GMM的语音参数化技术可以用来有效地识别昆虫种类。
Insects exercise, feeding, tweets will make a sound, these sounds within the species similarity and interspecies differences, it can be used to identify the types of insects. Insect-based insect-type automatic detection techniques make sense to help agricultural and forestry practitioners easily identify insect species. This study uses the voice parameterization technique in the field of speech recognition to realize the automatic identification of insects. After the acoustic samples were preprocessed, the Mel-frequency cepstrum coefficient (MFCC) was extracted as a feature and the Gaussian mixture model (GMM) was trained using the MFCC feature set extracted from these samples. Finally, GMMs obtained by training are used to classify the unknown types of insect sound samples. The method was evaluated in a sample library containing 58 species of insect sounds and achieved high recognition accuracy (average accuracy of 98.95%) and better time performance. The test results demonstrate that speech parameterization techniques based on MFCC and GMM can be used to effectively identify insect species.