A distributed network of autonomous environmental monitoring systems
Date
2018
Authors
Kinhal, Kiran Krishnamurthy, author
Azimi-Sadjadi, Mahmood R., advisor
Wilson, Jesse, committee member
Ghosh, Sudipto, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Acoustic wireless sensor networks have found applications in various areas including monitoring, assisted living, home automation, security and situational awareness. The process of acoustic detection and classification usually demands significant human involvement in the form of visual and audio examination of the collected data. The accuracy of the detection and classification outcome through this process is often limited by inevitable human errors. In order to overcome this limitation and to automate this process, we present a new fully decentralized decision-making platform referred to as Environmental Monitoring Station (EMS) for sensor-level detection and classification of acoustic airborne sources in national parks. The EMS automatically reports this information to a park station through two wireless communication systems. More specifically, in this thesis, we focus on the implementation of the communication systems on the EMS, and also on the design of 1/3rd octave filter bank that is used for onboard spectral sub-band feature generation. A 1/3rd octave filter bank was implemented on the ARTIX-7 FPGA as a custom hardware unit and was interfaced with the detection and classification algorithm on the MicroBlaze softcore processor. The detection results are stored in an SD card and the source counts are tracked in the MicroBlaze firmware. The EMS board is equipped with two expansion slots for incorporating the XBee as well as GSM communication systems. The XBee modules help to build a self-forming mesh network of EMS nodes and makes it easy to add or remove nodes into the network. The GSM module is used as a gateway to send data to the web server. The EMS system is capable of performing detection, classification, and reporting of the source events in near real-time. A field test was recently conducted in the Lake Mead National Recreation Area by deploying a previously trained system as a slave node and a gateway as a master node to demonstrate and evaluate the detection and classification and the networking abilities of the developed system. It was found that the trained EMS system was able to adequately detect and classify the sources of interest and communicate the results through a gateway to the park station successfully. At the time of writing this document, only two fully functional EMS boards were built. Thus, it was not possible to physically build a mesh network of several EMS systems. Thus, future research should focus on accomplishing this task. During the field test, it was not possible to achieve a high transmission range for XBee, due to RF interference present in the deployment area. An effort needs to be made to achieve a higher transmission range for XBees by using a high gain antenna and keeping the antenna in line-of-sight as much as possible. Due to inadequate training data, the EMS system frequently misclassified the sources and mis-detected interference as sources. Thus, it is necessary to train the detection and classification algorithm by using a larger and more representative data set with considerable variability to make it more robust and less prone to variability in deployment location.