云南高原湿地生态环境音分类技术研究

项目名称： 云南高原湿地生态环境音分类技术研究

项目编号： No.61462078

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 张雁

作者单位： 西南林业大学

项目金额： 45万元

中文摘要： 生态环境音频与各种生物的生存环境、人类的生活环境密切相关。研究生态环境音频数据分类、识别技术，对于生态环境安全与监控、分析和研究生态环境、预测生态环境的变化都具有重要意义。云南高度重视对湿地资源的监测、保护与研究，而湿地环境音频数据的分类识别将为该地区的湿地生态环境的研究提供新的视角。 项目将依托云南高原丰富多样湿地资源，研究湿地生态环境的音频信号采集，信号平稳特征和非平稳特征的提取方法，多类型特征优化选择与融合技术；探讨半监督学习和主动学习集成的分类模型，以及基于多类音频数据信号特征构造多视图下的多分类模型；研究在训练样本较少的情况下，有效地提高环境音分类的精度和分类模型的泛化性。 通过项目研究，完成对选择湿地区域采集的环境音分类识别，初步建立云南高原湿地生态环境音的数据库。为探索云南湿地生态环境声音与生态环境变化之间的联系，湿地生态环境监测，预测生态环境变化提供有效音频分类信息。

中文关键词： 湿地生态环境音；非平稳特征；特征优化；半监督+主动学习；多视图多分类器

英文摘要： The ecological environment audio signals are closely related to living environment of creatures and human beings. Research of classification and recognition on the ecological environment of audio data, for the ecological environment security and the monitoring, the analysis and study on the ecological environment and predicting the change of the ecological environment, has important significance. Since Yunnan attaches great importance to the protection of wetland resources and related research, the classification of ecological environment audio signals of wetland will provide a new perspective for this field. Relying on Yunnan plateau wetland resources, the project will carry on research including the technologies and methods of acquisition of wetland ecological environmental audio signals, the extracting of the stationary and non-stationary features of environmental audio signals，as well as multi-type features optimization ,selection and fusion technology. In order to ,at the backgroud of decreasing the numbers of training labeled samples,effectively improve the accuracy of classification of environmental audio signals and the generalization of classification model,the project will extensivelly explore the technologies of classification.That is how to effectively build the ensemble of the semi-supervised learning and active learning classification model.In additional, based on the many kinds of features extracted from wetland environmental audio signals, the classification model with multiple views and multiple classifiers will be researched. The classification result of wetland enviromantal audio signals will serve for establishing database of wetland ecological environment sound. The outcomes of this project is to provide the effective classifications of wetland ecological environment audio signals in Yunnan to extand our understanding the connection between the environment sound and the ecological environment change, to improve the wetland ecological environment monitoring, to help us give effective forecast the change of the wetland ecological environment.

英文关键词： wetland ecological environmental sound;non-stationary feature;feature optimization;Semi-supervised+Active learning;multi-view multi-classifier