Chris Baker (Committee Member), Fred Garber (Committee Member), Arnab Shaw (Advisor)
Master of Science in Engineering (MSEgr)
Some blind humans have developed the remarkable capability of echolocation, similar to the type used by mammals such as the bat, dolphin and whale. This population of human has shown the ability to classify targets based on their location, size, shape and material in diverse environmental conditions simply by listening to the reflected echoes of tongue clicks generated by their mouth. To date, much of the research into human echolocation has been confined exclusively to behavioral science and the analysis is inconsistent with the approaches used in engineering. The waveforms used in current radar systems appear different to those typical of mammal echolocation. It is speculated that the lack of robust success in radar target recognition may therefore be attributed to application of an inappropriate waveform. This research focuses on the analysis of human echolocation waveforms and their reflected echoes from different objects to investigate what properties of the waveform may carry target information. Results based on the analyses of echo data collected for various targets and their extracted features suggests that normalized target signatures cannot provide target classification in efficient manner. The normalized frequency spectrum has some potential for target classification, but it does not lead to confident classification results. The absolute difference between normalized frequency spectrum of transmit signal and normalized frequency spectrum of echoes performs much better than the two features discussed previously. It should be noted that the tongue click waveform performs much better at classifying objects made of hard materials from objects made of soft materials. However, they cannot be classified based on their shape or size by utilizing this feature. The chirp waveform provides superior classification performance for this feature, however, it is unclear which broad categories the targets can be put in for classification. The chirp, certainly, cannot classify all the targets as distinct from each other using this feature. From this analysis, it can be concluded that the frequency spectrum of echoes contain the most useful information for the task of target classification based on its material. Moreover, a moment invariant (MI) based automatic target recognition system has been developed that can potentially be implemented using efficient table-look up hardware. The phenomenon that echolocation experts utilize multiple perspectives of the objects to classify a target has inspired the development of this ATR algorithm. The main advantages of MIs are the translation and scale invariant properties and the possibility of implementation using table-look up hardware. The results of applying the theory of moment invariants on High Range Resolution profile data for the purpose of target classification are presented in work. For the developed ATR system, the performance is as high as 98% when the azimuth angle search range is 2 degrees while it reduces to approximately 58% for azimuth angle search range of 172 degrees.
Department or Program
Department of Electrical Engineering
Year Degree Awarded
Copyright 2013, all rights reserved. This open access ETD is published by Wright State University and OhioLINK.