Biosonar signal classification

Bats can distinguish objects by emitting a series of ultrasound signals (chirps) that generally sweep covering frequencies from 22 to 100 kHz. Inspired by the bat biosonar system, we have utilized ultrasonic sensing techniques for mobile robots (biomimetic robots) and tried to classify different textures and landmarks using received echo signals. We use a sonar head system consisting of three ultrasound transducers, one for emission chirp signals (Polaroid 7000), two for reception (Polaroid 6000) and tried to classify three different trees as landmarks (Fig.1). The aim of our research is to develop a classification system based on the bat neurophysiology.

The topics involved in this project are as follows:
  • Signal processing, filter design and preprocessing according to the bat neurophysiology.
  • Classification of the preprocessed signals and develop new kernels for extraction the similarities between reflected echoes.

Fig.2 shows the block diagram of the preprocessing steps for reflected echoes.

Fig. 1: Biosonar head and trees.
Fig. 2: Block Diagram of the preprocessing steps for the reflected echoes.

References

A novel kernel-based method for local pattern extraction in random process signals
Majid M. Beigi and Andreas Zell
In 15th European Symposium on Artificial Neural Networks, Bruges, Belgium (ESANN 2007) .
 
Object Detection in Biosonar Based Robot Navigation
Majid M. Beigi and Andreas Zell
In Worhshop on Planning, Perception and navigation fot Intelligent Vehicles, IEEE International Conference on Robotics and Autmation (ICRA2007).
 
Time-Resolved Spectrum Kernel for Biosonar Target Classification
M. M. Beigi, M. Wang and A. Zell
In The Fourth IASTED International Conference on Signal Processing, Pattern Recognition, and Applications (SPPRA 2007) .

Contact

Majid Beigi , Tel.: (07071) 29-78983, majid.beigi at uni-tuebingen.de