节点文献

基于HHT的果蝇振翅鸣声特征提取及分类研究

【作者】 贾春花

【导师】 郭敏;

【作者基本信息】 陕西师范大学 , 计算机应用技术, 2011, 硕士

【摘要】 昆虫鸣声是其生命活动中与外界环境联系的重要媒介,生物学意义极其丰富。果蝇是一种严重危害农业发展的昆虫,主要危害瓜果蔬菜和多种植物的茎叶。果蝇鸣声可以反映种的特异性,对外部形态特征相似的近缘种,可以利用鸣声特征进行区分;果蝇鸣声还可以用于果蝇害虫的诱集捕杀以及科学工作者的实验研究,所以,果蝇鸣声的研究具有很大的实用价值。对果蝇鸣声的研究,大多数研究者分析的是果蝇鸣声中的求偶歌,而且多数研究是在鸣声信号的时域上进行。本文以果蝇振翅鸣声为研究对象,提出应用时频局部化分析方法——希尔伯特-黄变换(HHT)分析同种内两个不同品系果蝇振翅鸣声的时频特征。然后,用BP神经网络分类两个品系果蝇振翅鸣声。本文方法的应用研究可以为果蝇鸣声及其它昆虫鸣声的进一步研究提供更新的科学依据。本文的研究主要包括以下几个方面:(1)简单概括了昆虫与人类生活的关系、昆虫鸣声的发声机制、鸣声类型和功能以及鸣声研究的意义,重点介绍了果蝇对各种水果蔬菜的危害及果蝇鸣声的国内外研究现状。(2)深入研究了希尔伯特-黄变换(HHT)方法的基本原理,阐明了希尔伯特变换、瞬时频率以及固有模态函数的数学定义,给出了经验模态分解的具体分解步骤和流程图,分析了希尔伯特谱和边际谱,并用仿真信号实例和实际信号实例对希尔伯特-黄变换的整个流程作了更详细的说明。本文还详细介绍了希尔伯特-黄变换方法在生物医学领域、地球物理学领域以及其他领域的应用情况,充分说明了该方法在信号处理中的优势。(3)本文提出将希尔伯特-黄变换方法应用于果蝇振翅鸣声信号的特征分析和提取中。针对鸣声信号经验模态分解后各阶固有模态函数的幅值谱能量不同,提取了两个不同品系果蝇振翅鸣声信号的各阶固有模态函数与原信号的能量比值特征;针对鸣声信号的希尔伯特谱上能量分布的差异性,提取了两个不同品系果蝇振翅鸣声信号在希尔伯特谱上的相对能量特征和时频熵特征;针对鸣声信号的边际谱不同频段上幅值分布的不同,提取了两个不同品系果蝇振翅鸣声信号的边际谱幅值特征。(4)用BP神经网络分类两个不同品系果蝇振翅鸣声。简单介绍了人工神经网络的发展过程,并详细分析了BP神经网络的基本原理和网络的参数选择。将希尔伯特-黄变换方法提取的果蝇振翅鸣声的特征作为BP神经网络的输入向量,分类两个品系果蝇振翅鸣声,识别结果都可以达到86%以上,说明了本文特征提取和分类识别方法的可行性和有效性。

【Abstract】 Insect’s sound is the important media that can connect with outside environment in their life activities, which contains rich biological significance. Fruit fly is a kind of insect that has more serious harm for agricultural product, mainly damages vegetables, stem and leaves of various plants. Fruit fly’s sound can reflect species-specificity, especially for some allied species, outer morphological characteristics are extremely similar, using characteristics of their sound can distinguish them. Also fruit fly’s sound can be used to trap drosophila pests and experimental study for scientific workers. Therefore, the research of fruit fly’s sound has great practical value.For the study of fruit fly’s sound, most researchers analyzed the courtship song of drosophila and most analyses focused on the time domain of sound. The paper uses fruit fly’s wing vibration sound as research object, proposes that uses Hilbert-Huang Transform (HHT), a new method of time-frequency localization analysis, to analyze time-frequency characteristics of wing vibration sound of fruit fly of two different strains in the same species. Then, it uses BP neural network to classify two strains of fruit fly’s wing vibration sound. The method of the paper can provide scientific basis for further researches of fruit fly’s sound and other insects’sound. The study of paper mainly includes following aspects:(1) The paper simply summarizes the relationship of human life and insects, insects’sound mechanism, type and function of the sound, and significance of sound research, mainly introduces the fruit fly’s sound of this paper, analyzes harms of fruit fly to various fruits and vegetables, and describes the current state of fruit fly’s sound research at home and abroad.(2) The paper deeply studies the fundamental principles of Hilbert-Huang Transform, clarifies the mathematic definitions of Hilbert transform, instantaneous frequency and intrinsic mode function, gives the specific steps and flow chart of empirical mode decomposition, analyzes the Hilbert spectrum and marginal spectrum, and illustrates the entire process of the Hilbert-Huang transform in more detail with examples of simulation signal and actual signal. In addition, the paper introduces application of the method in biomedical, geophysics and other fields, which fully explains the advantages of the method in signal processing.(3) The paper proposes that applies Hilbert-Huang transform in feature analysis and extraction of fruit fly’s wing vibration sound. Because amplitude spectrum energy of each intrinsic mode function is different after EMD, the paper extracts ratios of each intrinsic mode function and signal total energy of two stains of fruit fly’s sound. Because the energy distribution in Hilbert spectrum of fruit fly’s wing vibration sound is different, the paper extracts the relative energy of HH spectrum and time-frequency entropy. Because the amplitude distribution of marginal spectrum in different frequency band is different, the paper extracts the amplitude characteristics of marginal spectrum of fruit fly’s wing vibration sound.(4) The paper uses BP neural network to classify two strains of fruit fly’s wing vibration sound. Firstly, it briefly introduces development process of artificial neural network, and analyzes the basic principle and parameter selection of BP neural network in detail. Then, the paper takes features extracted by Hilbert-Huang transform as input vector of BP neural network, classifies two strains of fruit fly’s wing vibration sound, and the recognition result can reach over 86%, which explains the feasibility and effectiveness of the method.

节点文献中: