Scientists can better understand wetlands environments by collecting data they are interested in via sensor networks. However the deployment of these sensor nodes manually can be disruptive to these sensitive environments. We develop a set of algorithms for autonomously differentiating land from water via aerial imagery using an unmanned aerial vehicle (UAV). The UAV takes a picture of the area, clusters, classifies, defines regions, and then communicates the regions to other UAVs responsible for deploying the sensor nodes. These UAVs run an algorithm to determine the optimal locations for sensor nodes such that they completely cover the regions and allow for communication between the nodes in the sensor network.
Our classifier training algorithm identifies the best classifier using clusters and we compare its successful classification rate to a pixel-based approach and we see classification rates of 89.6%. This classifier feeds into our online algoorithm that the UAV successfully uses to classify the Calaveras River in California. In our simulations to determine the most effective algorithm for determining where the place the sensor nodes in a sensor network, we found Triangular Geometric Tessellation was the optimal algorithm, able to achieve 91.5% coverage in concave areas and 88.2% coverage in convex areas with relatively low computational complexity.
Identifer | oai:union.ndltd.org:pacific.edu/oai:scholarlycommons.pacific.edu:uop_etds-5022 |
Date | 01 January 2017 |
Creators | Medeiros, Thomas |
Publisher | Scholarly Commons |
Source Sets | University of the Pacific |
Detected Language | English |
Type | text |
Format | application/pdf |
Source | University of the Pacific Theses and Dissertations |
Page generated in 0.0016 seconds