Target Detection over Temperature Profiles

Abstract

This paper investigates the performance of hyperspectral target detection methods for target rediscovery on temperature profiles extracted from longwave infrared (LWIR) hyperspectral images (HSI). The targets in the experiments are selected as the three military vehicles in the captured scenes at different times. These targets are placed in the scene with different angles, at 0, 90, 135 degrees, respectively. The scope of the performed experiments includes the investigation and comparison of target detection performances, (i) with respect to sampling period in a day, (ii) time and temperature difference between reference and test days (iii) time window in a day. We utilized adaptive coherence estimator (ACE) target detection method on temperature profiles extracted over HSI. The experiments indicated that the sampling period during the day should be less than 1 hour for a robust target detection. Secondly, the target detection performance decreases as the distance between the days increases, but can be increased dramatically if similar weather temperatures are observed despite a long duration between the data recordings. Third, it is observed that the time zones with intense temperature changes such as sunrise and sunset are suitable for target detection.