Improving object classification with biomimetic sonar

This paper presents improvements to a previously described biomimetic syste m to make it more robust in terms of localizing objects and increasing the types of objects that can be classified. Time-of-flight determination emplo ys a large threshold and searches the echo envelope backward in time to fin d a smaller threshold crossing point, rather than using a single small thre shold. This change eliminates small nuisance echoes, which can cause instab ilities in object positioning. Second, the echo envelope is also searched f orward in time to find the threshold crossing point, thus defining a window containing significant echo data. This changes the analysis from a fixed-d uration data window to one that is variable and allows echo duration to be an object feature. Third, separate feature vectors were extracted from righ t and left receivers rather than employing a concatenated feature vector. T he new system can then classify objects that are not isolated. The major si gnificance of these changes is that it allows a greater variety of objects to be classified without degrading the performance of the original system. Experimental results compare original and new methods. Since biological son ars encounter similar problems, it is possible that they exploit principles similar to those described in this paper. (C) 2001 Acoustical Society of A merica.