基于鸣声的鸟类智能识别方法研究

发布时间：2018-01-09 15:01

本文关键词：基于鸣声的鸟类智能识别方法研究　出处：《西北农林科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：鸟类是湿地野生动物中最具代表性的类群，是湿地生态系统重要组成部分，也是监测湿地环境质量重要的生物指标。鸟的种类确定对湿地生物多样性和生态平衡提供了重要的依据。鸟类的鸣声和形态特征一样，具有物种的特性，是鸟类重要的生物学特征，也是识别鸟类的重要依据。本文针对我国经济开发和环境保护的矛盾突出，湿地资源遭受严重破坏的问题，在分析现有声音识别技术原理与系统结构的基础上，研究基于鸣声的鸟类智能识别方法，为实现湿地鸟类监测和迁徙规律提供技术支持。本文的主要工作和结论如下：（1）在分析当前已有研究成果的优点与不足的基础上，根据鸟类鸣声的特点，提出了分别处理鸣叫声和鸣唱声，采用双重高斯混合模型的设计方案。（2）综合考虑声音样本的多寡、地域的合理性、科目的差异和鸣声类型等因素，选择以陕西地区常见的黄臀鹎、矛纹草鹛、北红尾鸲、绿背山雀、方尾瀇、红嘴相思鸟、黄喉濦和淡尾瀇莺8种鸟类为研究对象，收集其鸣声，并使用Goldwave软件对声音样本进行除噪、裁剪等预处理。（3）在分析Mel倒谱系数（MFCC）理论基础和实现方案基础上，对MFCC进行了改进，提出一种新的特征参数MFCCA，并对其在正确识别率和识别效率上进行实验。实验结果表明，MFCCA特征参数具有更好的灵活性和鲁棒性，更适合作为鸟类鸣声的特征参数。（4）根据鸟类鸣声分为鸣叫声和鸣唱声的特点，打破以前采用单高斯混合模型（GMM）的传统，分别对每种鸟类建立鸣叫声GMM模型和鸣唱声GMM模型，并进行正确识别率和识别效率的实验。结果表明，当采用双重GMM模型时，识别效果最好，并且识别效率影响不大。并讨论不同阶数对与GMM模型的影响，结果表明当GMM模型的阶数为32时识别效果最好。（5）选取陕西地区50种鸟类，提取MFCCA特征参数，，建立双重GMM模型，并设计测试软件进行测试，结果表明，本文提出的鸟类智能识别方法的正确识别率能达到91.52%。
[Abstract]:Birds are the most representative group of wetland wildlife and an important part of wetland ecosystem. It is also an important biological index to monitor wetland environmental quality. Bird species determination provides an important basis for wetland biodiversity and ecological balance. It is an important biological characteristic of birds, and it is also an important basis to identify birds. In view of the contradiction between economic development and environmental protection in China and the serious damage to wetland resources, this paper analyzes the principle and system structure of existing sound recognition technology. In order to provide technical support for the monitoring and migration of wetland birds, the intelligent recognition method based on song sound is studied. The main work and conclusions of this paper are as follows: 1) on the basis of analyzing the advantages and disadvantages of the existing research results, and according to the characteristics of bird song, the design scheme of dealing with the singing sound and singing sound separately and adopting the mixed model of dual Gao Si is put forward. (2) considering the number of sound samples, the rationality of region, the difference of subjects and the type of song, the author chooses the common yellow-breasted Bulbul, spear-striped Babbler, Northern red-tailed robin, green-backed tit, square tail and so on. Eight species of birds, red mouth acacia, yellow larynx and light tail warbler, were collected, and the sound samples were pretreated with Goldwave software. 3) based on the analysis of Mel cepstrum coefficient and its implementation scheme, the MFCC is improved and a new characteristic parameter MFCCA is proposed. Experiments on the correct recognition rate and recognition efficiency show that MFCCA feature parameters have better flexibility and robustness, and are more suitable as the characteristic parameters of bird song. (4) according to the characteristics of birds' singing and singing, the tradition of single Gao Si mixed model was broken. GMM model and GMM model were established for each bird, and the correct recognition rate and recognition efficiency were tested. The results showed that the recognition effect was the best when using double GMM model. The effect of different order on GMM model is discussed. The results show that the recognition effect is the best when the order of GMM model is 32. 5) 50 species of birds in Shaanxi are selected, the characteristic parameters of MFCCA are extracted, the dual GMM model is established, and the test software is designed for testing. The results show that. The correct recognition rate of the intelligent bird recognition method proposed in this paper can reach 91.52%.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.34

【参考文献】