Physics-aware deep edge detection network

dc.contributor.authorPatel, Kinjal
dc.contributor.authorSleeman, Jennifer
dc.contributor.authorHalem, Milton
dc.date.accessioned2022-08-18T22:33:26Z
dc.date.available2022-08-18T22:33:26Z
dc.date.issued2021-09-12
dc.descriptionSPIE Remote Sensing, 2021, Online Onlyen_US
dc.description.abstractIn this paper, we describe an effort to build a new deep edge detection method designed to detect weather-related phenomena such as clouds and planetary boundary layer heights present in backscatter profile imagery. This method builds on the existing deep model called Holistically-Defined Edge Detection (HED), which was shown to perform better than other information theory and convolutional network techniques for edge detection. Though HED outperforms techniques such as Canny Edge detection, HED’s performance is based on it being trained on natural images with very little noise. Weather-related backscatter profiles, such as those generated from LIDAR-based ceilometers, often contain noise. In addition, there is often less of a difference in the pixel density between edges and non-edges, and due to atmospheric dynamics, continuous edges are not always detected in the images. Under these conditions when using HED, subtle but useful edges are lost from side outputs during the fusing process while the network is being trained. Canny Edge detection also does not perform well under these conditions, as it determines edges based on the differences in pixel density. We describe a new edge detection deep network developed specifically for overcoming these issues by applying physics-aware attention mechanisms to the side outputs of the HED learning process. We show how this method is able to learn the subtle edges as opposed to HED or Canny, when used to identify planetary boundary layer heights which involves distinguishing the mixing layer, residual layer, and nocturnal layer in addition to the cloud heights for ceilometerbased backscatter. Though the intent of this network is to learn planetary boundary layer heights and cloud heights, this method could be applied to other weather-related phenomena and applied to backscatter imagery generated from other sources such as satellites.en_US
dc.description.sponsorshipThis work has been funded by the following grants: NASA grant NNH16ZDA001-AIST16-0091 and NSF CARTA grant 17747724.en_US
dc.description.urihttps://www.spiedigitallibrary.org/conference-proceedings-of-spie/11859/1185908/Physics-aware-deep-edge-detection-network/10.1117/12.2600327.short?SSO=1en_US
dc.format.extent8 pagesen_US
dc.genreconference papers and proceedingsen_US
dc.identifierdoi:10.13016/m2jju7-zudo
dc.identifier.citationKinjal Patel, Jennifer Sleeman, Milton Halem, "Physics-aware deep edge detection network," Proc. SPIE 11859, Remote Sensing of Clouds and the Atmosphere XXVI, 1185908 (12 September 2021); https://doi.org/10.1117/12.2600327en_US
dc.identifier.urihttps://doi.org/10.1117/12.2600327
dc.identifier.urihttp://hdl.handle.net/11603/25499
dc.language.isoen_USen_US
dc.publisherSPIEen_US
dc.relation.isAvailableAtThe University of Maryland, Baltimore County (UMBC)
dc.relation.ispartofUMBC Computer Science and Electrical Engineering Department Collection
dc.relation.ispartofUMBC Faculty Collection
dc.relation.ispartofUMBC Student Collection
dc.rights©2021 Society of Photo-Optical Instrumentation Engineers (SPIE)en_US
dc.subjectUMBC Ebiquity Research Groupen_US
dc.titlePhysics-aware deep edge detection networken_US
dc.typeTexten_US

Files

Original bundle

Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
1185908.pdf
Size:
5.84 MB
Format:
Adobe Portable Document Format
Description:

License bundle

Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
2.56 KB
Format:
Item-specific license agreed upon to submission
Description: