Data Labeling and Curation at Scale (DLCS) for Machine Learning Algorithms

Homeland Security operations create image data from security screening machines. There is a strong interest in creating curated datasets (threats identified in images by bounding boxes and labels) from collected but not annotated data. Once available, the curated datasets will train Machine Learning (ML) models. The proposed project will develop the Data Labeling and Curation at Scale (DLCS) capability for creating high quality annotations from the available unlabeled data with minimal effort. The DLCS will be easy to use, versatile to curate data from different screening machines, easy to maintain, portable to different platforms, and capable of curating X-ray and CT-scan images. Available X-ray and CT-scan datasets will be used for demonstrating the full ML workflow and that high quality annotated data are generated with minimal effort. The proposed DLCS has applications in many areas of Homeland Security since screening inspection systems are used for luggage/package screening, container inspection, human activity detection and vehicle screening. The annotated data created by DLCS will enable thorough training of ML algorithms that will automate, expedite and expand inspection and reduce human involvement. DoD has similar needs for annotating images and using them for training ML capabilities. The same need exists in commercial self-driving vehicles. The proposed DLCS system will fulfill this gap. Finally, the DLCS supports the “Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence,” issued by The President of The United States on 10/30/23.