Biosonar features extracted from natural frequencies for natural landmark classification.

Abstract

To navigate in a natural environment, it is practical for an artificial agent to use vegetation as landmark. This is the way bats have done this for its long history. For an outdoor biosonar system based on digital signal processing, it is ideal to extract characteristic features directly from natural landmarks. Any physical structure can be modeled as a number of springs, masses, and dampers. Vegetation with trunks, stems, and foliage in open air always sways in a certain degree; it is possible to extract features from a series of characteristic natural frequencies with electrical biosonar system. In this work, a biosonar system is utilized to study features extracted from several plants for mobile agent navigation. Experimental results indicate that there is always enough air turbulence in the outdoor environment and these biosonar features are efficient in classifying plants. Besides using natural frequency features with the same wind condition, a laboring calibration should be made to improve this method robustness in different wind environments. [Work supported by NSFC.]