![no-image silhouette](/themes/custom/ornl/images/no-profile-img.png)
Sally S Ghanem
Machine Learning and Signal Processing Researcher
Bio
Sally Ghanem received the M.Sc. and PhD degree in electrical and computer engineering from North Carolina State University, Raleigh, NC, USA in 2021. Her research interests include the areas of computer vision, digital signal processing, image processing and machine learning.
Professional Experience
Oak Ridge National Laboratory June 2021- Current
Machine Learning and Signal Processing Researcher
- Devised a digital twin for dynamic energy and cost prediction in water reuse processes using SUMO software to track the state of full-scale and pilot-scale water reuse processes in real-time. The digital twin has the ability to tune model parameters and thus mature the models to reflect more accurately the particular physics of the underlying water reuse system.
- Analyzed different color mappings by training on versions of a curated dataset collected in a controlled campus environment to investigate the effect of color space selection for visual inputs in vehicle recognition tasks.
- Constructed an efficient and robust wheel detector that precisely located and selected vehicular wheels from vehicle images. The associated hubcap geometry was utilized to extract fundamental signatures from vehicle images and exploit them for vehicle re-identification. Experiments showed that our approach could detect vehicular wheels accurately for 99.41% of the vehicles in the dataset.
- Developed a principled framework for decision fusion utilizing features extracted from vehicle images and their detected wheels. Siamese networks were exploited to extract key signatures from pairs of vehicle images. The proposed approach examined the extent of reliance between signatures generated from vehicle images to robustly integrate different similarity scores and provide a more informed decision for vehicle matching.
North Carolina State University (VISSTA Lab) January 2015-May 2021
Research Assistant
- Developed a framework, based on a non-parametric metric, to quantify the inherent separation between classes of high-dimensional imagery data and evaluate various common feature descriptors in the relevant feature spaces.
- Employed a distribution-free metric to obtain bounds for the k-Nearest Neighbor (k-NN) classification accuracy. The proposed bounds were exploited for selecting the least number of features that would retain sufficient discriminative information.
- Evaluated the efficacy of different pre-processing techniques and selected the least number of features, that would achieve the desired classification performance, using the developed accuracy bounds.
- Implemented a multimodal data fusion algorithm, relying on robust subspace recovery, to prospect the structure of each data modality and obtain a new representation of the time series, respecting an underlying Union of Subspaces model.
- Devised a multi-modal approach to vehicle classification and identification using an ensemble of sensors.
- Established a multimodal feature fusion approach that exploits the structural dependencies between heterogeneous data modalities to cluster the associated target objects. Our approach is competitive with other state-of-the-art subspace clustering methods.
- Optimized a Convolutional Neural Network autoencoder to fuse multimodal data by extracting key features from each data modality and subsequently combine those features to generate a common discriminative feature.
Awards
- Awarded Laboratory Directed Research and Development funding for the proposal titled “Vehicular Pattern of Life Analysis from Uncontrolled Multiple Views”, October 2023.
- Received the best national laboratory collaboration award from National Nuclear Security Administration (NNSA), June 2019.
- Awarded Electrical and Computer Engineering Department merit-based fellowship, NC State University, 2014 & 2015.
Education
North Carolina State University, Raleigh, NC GPA: 4.0/4.0
Doctor of Philosophy in Electrical and Computer Engineering August 2017-May 2021
North Carolina State University, Raleigh, NC GPA: 4.0/4.0
Master of Science in Electrical and Computer Engineering August 2014-May 2016
Professional Affiliations
Institute of Electrical and Electronics Engineers (IEEE)
Trademarks and Patents
- S Ghanem, RA Kerekes, RA Tokola, “Vehicle reidentification” - US Patent App. 18/139,172, 2023 (https://patents.google.com/patent/US20230343065A1/en)
- H Krim, S Roheda, S Ghanem, “Volterra neural network and method” - US Patent 11,842,526, 2023 (https://patents.google.com/patent/US11842526B2/en)
Publications
Other Publications
S. Ghanem, S. Roheda, and H. Krim. "Latent Code-Based Fusion: A Volterra Neural Network Approach." Intelligent Systems with Applications no. 18, Elsevier, 2023.
S. Ghanem, and H. Krim. "Scaling Subspace-Driven Approaches Using Information Fusion.", IntechOpen, 2023.
S. Ghanem, and R. A. Kerekes. "Robust Wheel Detection for Vehicle Re-Identification." Sensors 23, no. 1, 2022.
S. Ghanem, R. A. Kerekes, and R. Tokola. "Decision-Based Fusion for Vehicle Matching." Sensors 22, no. 7, 2022.
S. Ghanem, A. Panahi, H. Krim, and R. A. Kerekes. “Robust Group Subspace Recovery: A New Approach for Multi-Modality Data Fusion". IEEE Sensors Journal, 2020.
S. Ghanem, A. Panahi, H. Krim, R. A. Kerekes, and J. Mattingly. “Information subspace-based fusion for vehicle classification". In 26th European Signal Processing Conference (EUSIPCO), IEEE, 2018.
S. Ghanem, H. Krim, H. S. Clouse, and W. Sakla. “Metric Driven Classification: A Non-Parametric Approach Based on the Henze-Penrose Test Statistic". In IEEE Transactions on Image Processing, no. 12, IEEE, 2017.
S. Ghanem, E. Skau, H. Krim, H. S. Clouse, and W. Sakla. “Non-parametric bounds on the nearest neighbor classification accuracy based on the Henze-Penrose metric". International Conference on Image Processing (ICIP), IEEE, 2016.