Kernel one-class weighted sparse representation classification for change detection

ABSTRACT

In this letter, improved methods based on one-class sparse representation classifier (OCSRC) are proposed for change detection with multi-temporal multi-spectral remote sensing images. By adopting the weighted regularization and the kernel method, kernel one-class sparse representation classifier (K-OCSRC), one-class weighted sparse representation classifier (OCWSRC) and its kernel version, kernel one-class weighted sparse representation classifier (K-OCWSRC) are proposed. Performances of the OCSRC, K-OCSRC, OCWSRC, K-OCWSRC methods are tested with the flood dataset. Results show that the weighted methods (OCWSRC and K-OCWSRC) are less sensitive to the regularization parameter in the optimization process, and the kernel methods (K-OCSRC and K-OCWSRC) can distinctively improve change detection accuracies by solving the problem in the projected higher-dimensional space. Overall, the K-OCWSRC achieves the best change detection result as it can more accurately locate the flood affected areas while bringing in least undesirable false alarms.

Additional information

Funding

This work was supported by National Natural Science Foundation of China under Grant No. [61501017, 61571033, 61302164].