Overhead Detection: Beyond 8-bits and RGB
This work addresses building detection accuracy in satellite imagery for remote sensing applications, but is incremental in nature.
The study established performance baselines for building detection in satellite imagery using the SpaceNet dataset, finding that using 13-bit data instead of 8-bit data improved detection accuracy by over 32% with the R-FCN algorithm. It also analyzed accuracy trends by building size and scene density, and evaluated methods for integrating additional spectral bands, which showed no significant performance improvement.
This study uses the challenging and publicly available SpaceNet dataset to establish a performance baseline for a state-of-the-art object detector in satellite imagery. Specifically, we examine how various features of the data affect building detection accuracy with respect to the Intersection over Union metric. We demonstrate that the performance of the R-FCN detection algorithm on imagery with a 1.5 meter ground sample distance and three spectral bands increases by over 32% by using 13-bit data, as opposed to 8-bit data at the same spatial and spectral resolution. We also establish accuracy trends with respect to building size and scene density. Finally, we propose and evaluate multiple methods for integrating additional spectral information into off-the-shelf deep learning architectures. Interestingly, our methods are robust to the choice of spectral bands and we note no significant performance improvement when adding additional bands.