University of Twente Student Theses

Login

Designing a small and low-energy wildlife tag for parakeets within an urban environment capable of tracking and online activity recognition

Berg, R.T. Van den (2019) Designing a small and low-energy wildlife tag for parakeets within an urban environment capable of tracking and online activity recognition.

[img] PDF
22MB
Abstract:This work is concerned with designing and building an online activity classifier and tracking device on an embedded system to monitor parakeets. This involves the design choices in localization, feature selection, classifying algorithm and optimization of required features and algorithm parameters. In order to reduce the number of inputs to a machine learning algorithm, Forward Selection was applied, which reduced the number of features from 20 to 7. The resulting feature sets were then applied to six learning algorithms; k-Nearest Neighbours, Naive Bayes, Neural Network, Linear Discriminant Analysis, Decision Tree, and Support Vector Machine. In order to evaluate the machine learning algorithm, data from a parakeet was used for training and testing using a 70% to 30% ratio, creating a generic classifier which could accurately recognise the activity of a parakeet. The models were evaluated and compared according to three main metrics, namely performance, battery usage, and ease of implementation, as well as other metrics such as variance in performance, usability and training effort. It was found that the combination of the decision tree classifier with seven timedomain features from the accelerometer’s 3D vector magnitude comprised the best compromise between the evaluated metrics. The decision tree parameters were tuned such that its performance could be maintained while minimizing the tree size. A window size of two seconds and a 50% window overlap was used to yield an excellent compromise between computation and performance. The accuracy of classification varied between 87 and 90%. Next, satellite-, ground-based- and radar tracking methods were compared in terms of energy consumption and accuracy. A localization system has been proposed based on the received signal strength of Bluetooth Low Energy beacons. These beacons were detectable from a 40-meter distance. Finally, this tracking device and online activity classifier have been implemented on the AKMW-iB001M beacon, and the performance will be tested in the future on wild parakeets in Málaga, Spain.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:https://purl.utwente.nl/essays/79501
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page