Image watermarking refers to covertly embedding information (i.e., a watermark) into a cover image. By designing deep learning methods such as specialized training schemes and novel layers, this project aims at adaptivity and robustness for image watermarking. One typical downstream application is to extract a watermark from camera-resampled marked images, and the end users can scan any cover image for more information. 

Labeling an image as a whole is much easier than labeling each object/segment inside the image. However, supervised learning can only learn from training data with exact labels. Hence, weakly supervised learning, more concretely, inexact supervision where the training data are given with labels but not as exact as desired, is the purpose of this project. Here, the given label will be the classification labels, and the desired output will range from object localization to detection to segmentation.

Foreign Object Debris is any substance alien to an airport system that can cause damage. We leverage multidisciplinary techniques by integrating machine learning, computer vision, small unmanned aerial technology, and traditional airport operations protocols to develop a system that helps overcome the high-cost, low-efficiency, and technical challenges of Foreign Object Debris detection that a great number of small-scale airports are facing. 

Funded by: 

• NASA Nebraska Space Grant

• Nebraska University (NU) Collaboration Initiative