Fast Detection of Compressively Sensed IR Targets Using Stochastically Trained Least Squares and Compressed Quadratic Correlation Filters
Online Publication Date2017-05-02
Print Publication Date2017-10
Permanent link to this recordhttp://hdl.handle.net/10754/626601
MetadataShow full item record
AbstractTarget detection of potential threats at night can be deployed on a costly infrared focal plane array with high resolution. Due to the compressibility of infrared image patches, the high resolution requirement could be reduced with target detection capability preserved. For this reason, a compressive midwave infrared imager (MWIR) with a low-resolution focal plane array has been developed. As the most probable coefficient indices of the support set of the infrared image patches could be learned from the training data, we develop stochastically trained least squares (STLS) for MWIR image reconstruction. Quadratic correlation filters (QCF) have been shown to be effective for target detection and there are several methods for designing a filter. Using the same measurement matrix as in STLS, we construct a compressed quadratic correlation filter (CQCF) employing filter designs for compressed infrared target detection. We apply CQCF to the U.S. Army Night Vision and Electronic Sensors Directorate dataset. Numerical simulations show that the recognition performance of our algorithm matches that of the standard full reconstruction methods, but at a fraction of the execution time.
CitationMillikan B, Dutta A, Sun Q, Foroosh H (2017) Fast Detection of Compressively Sensed IR Targets Using Stochastically Trained Least Squares and Compressed Quadratic Correlation Filters. IEEE Transactions on Aerospace and Electronic Systems 53: 2449–2461. Available: http://dx.doi.org/10.1109/taes.2017.2700598.
SponsorsThis work was supported in part by the National Science Foundation under Grant IIS-1212948 and Grant DMS-1412413.