Machine Learning in Image Mining: Nanoscale, X-ray, LiDAR and Satellite Data

a project with Dr. Varde

This work involves machine learning and data mining techniques in the area of image analysis for knowledge discovery. Early work here has been funded by a grant from NSF REU and supports undergraduate students from the tri-state area during summers. The focus of this grant is in the area of image processing and significant work occurs in the area of learning from image data at the nanoscale level. It entails proposing and implementing techniques for image mining useful in domain-specific decision-making. There is real data from researchers in Nanotechnology with applications having broader impact in health informatics. For example, the results of the learning are useful in finding a cheaper material instead of a more expensive material to develop a human body implant, if both materials yield similar results as evident from image similarity search. Publications from this work include a paper in SPIE 2010 conference, a presentation in ACM CCSC 2010 conference, and a paper in ICML 2010 Workshops. Further work in this general area includes decision support in automated detection of COVID-19 based on analysis of chest X-ray images on COVID-positive, pneumonia-positive and normal / healthy cases. This entails the deployment of computer vision models based on convolutional neural networks (CNNs) via transfer learning along with data augmentation. This part of the work has been published in IEEE Big Data 2020. Ongoing work includes image analysis in the area of earth science. These include satellite images (Landsat and Sentinel) for predictive analysis of of ocean salinity and vegetation in mangrove islands. They also involve analysis of LiDAR images captured from drones to analyze dune formation from various perspectives. Machine learning techniques such as unsupervised learning with clustering, supervised learning with classification (random forest, decision trees) and deep learning with CNNs are being explored here. The results would be useful to support domain-specific decision-making in earth science and fall in the broad realm of geo-informatics.

  • Ph.D. Student (Committee): Isamar Cortes (ongoing)
  • Ph.D. Student (Committee): Shane Daiek (ongoing)
  • M.S. Project Student: Divyadharshini Karthikeyan (Graduated Jan 2021)
  • Summer Research Student: Gregory Roughton (Completed: July 2009)
  • Summer Research Student: Daniel Jackowitz (Completed: July 2010)
  • Collaborators / Co-authors: Dr. Stefan Robila (CS, Montclair), Dr. Jianyu Liang (Materials Science, WPI, MA) Dr. Jorge Lorenzo-Trueba (EAES, Montclair), Dr. Weitian Wang (CS, Montclair)
  • Funding: NSF REU Grant (2009 to 2011) - PI: Dr. Stefan Robila;
  • TA from Computer Science;
  • NASA Fellowship - PhD Student Isamar Cortes (2019-2022)