Neuro-computational processing of moving sonar echoes classifies and localizes foliage

Echoes from in situ tree trunks, similar to those observed by flying bats, are processed. A moving sonar converts echoes into spike sequences and applies neural-computational methods to classify objects and estimate passing range. Two classes of tree trunks act as retro-reflectors that generate strong echoes (SEs), identified by a locally dense spike pattern. Linear drive-by sonar trajectories cause SEs to follow hyperbolic curves specified by passing range. A glint is a collection of consecutive range readings matching expected values on a specific hyperbolic curve. Passing-range detectors compare successive SE data with expected values in a table and tally coincidences. Counters increment when coincidences occur and decrement when they do not. A glint terminates after tallying a sufficient number of coincidences and coincidence failure occurs in the maximum-count detector. Reflector roughness, deviations in sonar trajectory, and echo jitter necessitate a coincidence window to define matches. Short windows identify small glints over piecewise linear sonar trajectories, while long windows accommodate deviations in sonar speed and trajectory, and associate multiple glints observed with shorter windows. The minimum coincidence window size yielding glints classify smooth and rough retro-reflectors.