基于多传感器的声景大数据在线采集与分析系统研究

声景包含重要的生态信息，具有实时性强、信息密度高的特点，有重要研究价值。现有的声景研究中，音频及相关环境参数采集和分析仍需要大量的人工作业，耗时耗力。基于多传感集成、边缘计算和深度学习技术，建立了一套声景大数据在线采集与分析系统，包括边缘计算节点和中心计算服务器。并通过3个实验站点，进行了近1年的技术验证，实现了声景大数据的自动化在线采集、传输和分析。该系统能适应户外恶劣的自然环境，能根据任务需求持续不断地进行声景大数据在线采集和分析，稳定性好。声学指数可以反映声景变化，但因指数侧重点不同，不同的声学指数之间变化特征差异较大，需要组合使用。通过声纹特征图能直观地识别出不同发声源，对物种的快速识别、声源的分类等具有较强的借鉴意义。系统借助VGGish网络提取的高维声景特征图能很好地识别不同站点和不同时间的声景变化，在不同站点和昼夜上具有较高的区分精度，有快速和直观地反映不同生态系统的类型特征、生态系统动态变化的潜力。丰富声纹特征库、优化声景特征分析神经网络、建设声景长期监测共享网络，有助于扩展系统在物种识别、生物多样性快速分析、生物与环境相互作用机制方面的应用。研究为声景大数据的在线采集...

The soundscape big data online capturing and analysis system based on multi-sensor

The soundscape contains important ecological information, with features of real-time and high information density. It has significant research value and has been paid more and more attention by ecologists in recent years. Nowadays, in the soundscape research field, capturing and analyzing the audio and the related environmental parameters is still time and labor-consuming work, which throttles further research on the soundscape. We built a soundscape big data online capturing and analysis system based on state-of-the-art technology, such as multi-sensor, edge computing, and deep learning to remove the barrier. The system consists of edge computing units and central computing servers. In order to verify the stability and reliability of this system, we performed one-year technical confirmation by setting up three field research sites to automate the capturing, transmission, and analysis of soundscape big data. The system could stably and continuously complete the scheduling task, including online capturing and analysis of soundscape big data. The system can still work correctly in a harsh environment. The calculation results of the acoustic index show that the acoustic index can reflect the soundscape change. The variation pattern of these acoustic indexes was different in the same acoustic environment, caused by the different emphasis of the acoustic indexes. We suggest the researcher combine these acoustic indexes to explain the soundscape variation. The voiceprint map extracted from the audio file transmitted back by the edge computing unit can directly identify different sound sources and is beneficial for rapid species identification and sound source classification. The high dimensional soundscape features that the analysis system extracted from audio using VGGish-Net are commendably able to distinguish the change of soundscape in different times and locations, which has the potential to quickly and intuitively reflect the types and dynamic changes of diverse ecosystems. Especially with the extracted high dimensional soundscape features with the help of random forest classifier has high distinguished degree at different research sites and day-night scale. Enriching the voiceprint feature database, optimizing the neural network for soundscape feature analysis, and building a shared network for long-term soundscape monitoring will be beneficial for expanding the application of the system in species identification, rapid biodiversity analysis, and the interaction mechanism exploration between creatures and environment. This study provides a detailed reference for the big data online capturing and analysis of soundscape.