2019/09/07 by Susan Meerdink, James Bocinsky, Meerdink, Susan +11 · 1 citation
Engineering · Computer Science · #Remote-Sensing Image Classification #Image Retrieval and Classification Techniques #Advanced Image and Video Retrieval Techniques
paper · pdf · doi:10.48550/arxiv.1909.03316
In remote sensing, it is often challenging to acquire or collect a large\ndataset that is accurately labeled. This difficulty is usually due to several\nissues, including but not limited to the study site's spatial area and\naccessibility, errors in the global positioning system (GPS), and mixed pixels\ncaused by an image's spatial resolution. We propose an approach, with two\nvariations, that estimates multiple target signatures from training samples\nwith imprecise labels: Multi-Target Multiple Instance Adaptive Cosine Estimator\n(Multi-Target MI-ACE) and Multi-Target Multiple Instance Spectral Match Filter\n(Multi-Target MI-SMF). The proposed methods address the problems above by\ndirectly considering the multiple-instance, imprecisely labeled dataset. They\nlearn a dictionary of target signatures that optimizes detection against a\nbackground using the Adaptive Cosine Estimator (ACE) and Spectral Match Filter\n(SMF). Experiments were conducted to test the proposed algorithms using a\nsimulated hyperspectral dataset, the MUUFL Gulfport hyperspectral dataset\ncollected over the University of Southern Mississippi-Gulfpark Campus, and the\nAVIRIS hyperspectral dataset collected over Santa Barbara County, California.\nBoth simulated and real hyperspectral target detection experiments show the\nproposed algorithms are effective at learning target signatures and performing\ntarget detection.\n