Accurate Target Identification Using Multi-look Fusion of Low Quality Target Signatures
DEFENCE RESEARCH AND DEVELOPMENT CANADA OTTAWA
Pagination or Media Count:
Single-look identification procedure using one input at a time has been the principal mode of analysis used in the target recognition community for much of the past two decades. But it has been realized that the single-look approach can only achieve modest correct identification performance and this is not adequate for many target identification applications. Furthermore, in order for the single-look procedure to perform well, good target data quality is required. In this report, a multi-look method known as score-level fusion is investigated. It permits a significant improvement in the identification performance. Moreover, it does not require good quality target signature data multiple signature samples of marginal quality may be used instead. Results from analysis using measured radar target data have shown quantitatively that the correct identification rate can be improved very dramatically. Furthermore, the score-level fusion method is very efficient the identification performance increases exponentially with an increase in the number of samples used in the multi-look sequence. A qualitative characterization of the score-level fusion method based on the principle of averaging is given. It provides an intuitive understanding of this fusion process and how improvement in identification accuracy is achieved. The identification results obtained in this study suggest that quantity can be used to replace quality of the data to improve target identification accuracy. This has an interesting practical implication. With the advent of sensor technologies, large quantity of data of marginal quality can be captured routinely. This quantity over quality approach could maximize the exploitation of available data to provide reliable and robust identification. Thus, the score-level fusion method could have the potential of being developed into a disruptive technology for target identification and pattern recognition problems.
- Target Direction, Range and Position Finding