Learning with imprecise and uncertain data

In remote sensing, it is often challenging to acquire or collect a large dataset that is accurately labeled. This difficulty is usually due to several issues, including but not limited to the study site’s spatial area and accessibility, errors in the global positioning system (GPS), and mixed pixels caused by an image’s spatial resolution. Multiple instance learning (MIL) can overcome the need to have precise training labels. MIL only requires the labeling of positive and negative bags, which are groupings of pixels. Each bag may contain many pixels, but a bag is labeled positive if at least one of the pixels belongs to the target class. This framework alleviates the need to have accurate labels which are inherently challenging to collect.

Current work

We developed an approach, with two variations, that estimates multiple target signatures from training samples with imprecise labels: Multi-Target Multiple Instance Adaptive Cosine Estimator (MTMI-ACE) and Multi-Target Multiple Instance Spectral Match Filter (MTMI-SMF). The proposed methods address the problems above by directly considering the multiple-instance, imprecisely labeled dataset. They learn a dictionary of target signatures that optimizes detection against a background using the Adaptive Cosine Estimator (ACE) and Spectral Match Filter (SMF). Both simulated and real hyperspectral target detection experiments show that the proposed algorithms are effective at learning target signatures and performing target detection.

Papers

Meerdink, S., Bocinsky, J., Zare, A., Kroeger, N., McCurley, C., Shats, D., & Gader, P. (2021). Multitarget Multiple-Instance Learning for Hyperspectral Target Detection. IEEE Transactions on Geoscience and Remote Sensing, doi: 10.1109/TGRS.2021.3060966