Abstract
Many echolocating bat species are capable of navigating through highly cluttered environments, such as dense foliage—apparently without difficulty. Since interpreting foliage echoes must hence be key to the animals' navigation capabilities, simulating such echoes would be a major step towards understanding bat navigation. This is a difficult task, as vegetation echoes are highly complex and stochastic signals. Prior work on modeling foliage echoes frequently relied on physical approximations, since full physics models are computationally infeasible. For simulating the echo from a leaf, a round disk approximation has been used, since it’s echo has a known physical solution. These disks have been distributed according to the locations of leaves in a tree using a Lindenmayer System (L-System), thus creating a full tree echo model. Successful prior work in using a machine learning approach for echo based tree classification, shows that a data driven approach may also be useful in modeling echoes. Hence, future research will be using machine learning methods, such as auto-encoders and generative adversarial networks, to simulate the echoes of leaves and trees. The echo data to be fed into these methods will be of leaves recorded in an anechoic chamber using a biomimetic sonar head.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have