Foliage is very common feature in the habitats of echolocation bats and thus its echoes constitute the major input of bats' sensory systems. Acquiring useful information from vegetation echoes facilitates the bats significantly in the navigation and foraging behaviors. To better understand the foliage echoes, in this dissertation, a computer model was constructed to simulate foliage echoes with following simplifications: approximating leaves as circular disks, leaving out shading effects between leaves, and distributing leaves uniformly in the space. Then one tree can be described with three parameters in the model, leaf radius, orientation, and leaf density, where the first two determine the beampattern of each leaf. Compared with echoes collected from real trees, the simulation echoes are qualitatively accurate, i.e., they match in waveforms and also first-order statistics. Since the ground truth is known in the model, the three parameters were estimated with lasso model by selecting 40 features from each echo. The results have shown that estimation of one parameter with the other two known is usually successful with coefficient of determination close to one, and the classification still has reasonable accuracy when the number of known parameter is reduced to one. Besides, the three simplifications were examined with both experimental and simulation approaches. To assess the acoustic impact of leaf geometry on individual leaves, experiments were carried out by ensonifying leaves from both a single and different species. How the leaves' impulse responses change according to their equivalent radii was investigated. The simulation model of disks fits the experiments done with real leaves within one species and across species reasonably well. Shading effect is found to exist locally when two disks were 25 cm apart and were both in pulse direction. In addition, the inhomogeneous distribution of leaves was introduced by using the branching patterns of L-system. The evaluation of inhomogeneity in echoes produced with two distributions shows that there is always inhomogeneity in echoes, and L-system model does bring more inhomogeneity but not to the same extent as changes in the relative orientation between sonar beam and foliage do. / Ph. D. / Echolocating bats use ultrasonic waves to navigate and forage at night in the forest. They constantly emit pulses and analyze the returning echoes to perceive the surroundings. Foliage echoes are common and important input of their sensory systems, yet what accessible information is contained in foliage echoes for bats is not fully understood. Hence, this dissertation has built an efficient computer model to compute vegetation echoes. To simplify the problem, leaves with various shapes were approximated as circular disks. Besides, every leaf was assumed to be “visible” to sonar, in other words, even if one leaf was shaded by another in the pulse direction, it can still interact with sonar as if the front leaf didn’t exist. Then the leaves were uniformly distributed in the space. With the simplifications above, foliage can be described with three parameters, mean leaf radius, orientation, and leaf density. By varying the three parameters to match features of different trees, a large amount of echoes can be calculated efficiently. Compared with measured echoes from real trees, the simulation echoes are similar with them in terms of waveforms and probability density functions. If producing echoes with two parameters fixed and the third randomly chosen from certain range, the random parameter can be estimated with a linear model, lasso regression model, by extracting features from the echoes as inputs. The estimation is accurate. But if varying one of the two known parameters, the accuracy of estimation is largely reduced. Besides, the three simplifications were examined if they have impact on the simulation results. Impulse responses from leaf specimens were measured with a bio-mimetic sonar head in the anechoic chamber where noise and unwanted reverberations are largely weakened. Experiments were also carried out for two disks of same size by aligning them in the direction of sound emission to quantify shading effect, which shows that shading effect exists locally. Then branching patterns were introduced to the simulation model using L-system that consists of a set of rules to determine how branches grow. The results demonstrates that the simplifications do affect the model accuracy but the influence may be compensated.
| Identifer | oai:union.ndltd.org:VTETD/oai:vtechworks.lib.vt.edu:10919/78825 |
| Date | 07 September 2017 |
| Creators | Ming, Chen |
| Contributors | Mechanical Engineering, Mueller, Rolf, Zhu, Hongxiao, Taylor, John E., Kurdila, Andrew J., Leonessa, Alexander |
| Publisher | Virginia Tech |
| Source Sets | Virginia Tech Theses and Dissertation |
| Detected Language | English |
| Type | Dissertation |
| Format | ETD, application/pdf |
| Rights | In Copyright, http://rightsstatements.org/vocab/InC/1.0/ |
Page generated in 0.0025 seconds