Receiver-operating curves have been used to examine a novel target-recognit
ion system using a number of knowledge-based techniques to automatically de
tect surface land mines present in 30 sets of thermal and multispectral ima
ges. A summary of results, graphed at a probability of detection greater th
an or equal to 96%, shows the false-alarm rates (FAR's) obtained using vari
ous combinations of fusing sensors and neural classifiers averaged over the
30 images. Results show that using two neural-network classifiers on the i
nput textural and spectral characteristics of selected multispectral bands,
we obtain FAR'S of approximately 3%. Using polarization-resolved images on
ly, we obtain FAR's of 1.15%. Fusing the best classifier output with the po
larization-resolved images, we obtain FAR's as low as 0.023%. This result h
as shown the large improvement gained in the fusion of sensors. Also, an im
provement is shown by comparing these results with those reported in an exi
sting commercial system published in an internal report.