Nian Zhang
University of the District of Columbia, DC·Computer Science
Detecting explosives and other dangerous chemicals from a safe distance, without touching or sampling them, is a pressing security problem, made urgent by improvised explosive devices on the battlefield. One approach shines an eye-safe infrared laser at a surface and records the light scattered back with infrared sensors or imaging arrays.
- Fields
Computer Science
Engineering
- Topics
standoff detection
explosives detection
threat chemicals
- Grants
- 1 known grant, through Aug 2027
What they work on
- Develop a data preprocessing method for infrared backscatter hyperspectral images that handles class imbalance, where threat-chemical pixels are rare, and unlabeled data.
- Extract endmember bundles, the sets of pure spectral signatures within mixed pixels, using a multiobjective global optimization that combines particle swarm optimization with an evolutionary algorithm.
- Build a deep learning semantic segmentation algorithm, which labels every pixel by category, to suppress laser speckle noise and improve feature extraction from hyperspectral images.
- Integrate preprocessing, optimization, and classification into one framework that distinguishes explosives from the substrates they rest on at standoff distances.
More about the research
The hard part is the analysis. A useful system must be safe around people and must pick up trace amounts. It must also tell a threat chemical apart from harmless background chemicals and from the surface it sits on.
Why it matters
Better standoff detection could help soldiers and security personnel spot explosives and other hazardous chemicals from a safe distance, without contact or sampling. The new methods for hyperspectral image classification and endmember extraction may also be useful wherever hyperspectral imaging is used, including geospatial mapping, agriculture, and healthcare.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Brant Robertson
University of California-Santa Cruz, CA·Astronomy & Astrophysics
Incorporate the Morpheus deep learning framework, which classifies images pixel by pixel, into the Rubin Science Platform for analyzing Rubin Observatory data.
Lingwei Chen
Rochester Institute of Technology, NY·Computer Science
Repurpose pretrained transformer models for cyberattack detection through cross-modal adversarial reprogramming, which adds patch-level changes to inputs so few new parameters are needed when labeled data are scarce.
Salim Sazzed
Georgia Southern University, GA·Computer Science
Characterize the structural inconsistencies and variability of protein secondary structures, such as helices and sheets, in medium-resolution cryo-electron microscopy (cryo-EM) maps.
The school
University of the District of Columbia: 2 labs on PI Pages, $9M a year on research, 8 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.