Detection of Deployed Targets Using Airborne Hyperspectral Image Data: A Case Study

The evolving hyperspectral imaging technology relies on the information captured in large number of contiguous bands at fine spectral resolutions for detection and identification of objects of interest. In the present work, different hyperspectral data analysis methods were applied to airborne hyperspectral image data for detection of objects/targets of size comparable to size of image pixel. Keeping in view the non-availability of synchronized in-situ and image data, different types of targets (i.e. wood panel, green painted wood panel, fabric etc.) of different dimensions were placed over homogenous known backgrounds. Airborne hyperspectral imaging sensor in short wave infra-red (SWIR) region (910 ─ 1690nm) was flown over the area of interest to capture images of the background and targets. The in-situ spectral signatures of targets and backgrounds were also acquired. The raw frame data of airborne sensor was processed for necessary hypercube formation along-with radiometric and geometric corrections before analysis. A comparison of image and in-situ spectra for homogenous large features in the image lead to root mean square error of 0.04 with band-to-band correlation of 0.89. Four different methods i.e. Spectral angle mapper (SAM), Matched filtering (MF), Constrained energy minimization (CEM), and Adaptive coherence estimator (ACE), were applied on the hyperspectral image data to detect the presence of targets deployed at known field locations. Threshold criteria was tuned to increase the probability of detection of targets and to evaluate the performance of the methods on image. The challenges involved and results obtained are presented in the present work which may form basis for selection of suitable thresholds for target detection.