Curation and Dissemination of Complex Multi-modal Data Sets for Radiation Detection, Localization, and Tracking

TL;DR

This work presents PANDAWN, a 13-node urban, multi-modal sensing network for radiological detection, localization, and tracking. It describes an automated labeling and curation pipeline that fuses radiation, environmental, and contextual data using edge computing and a cloud-backed data store, enabling scalable, ground-truth-rich datasets. The approach combines continuous edge calibration, NMF-based background modeling, and computer-vision–based contextual labeling (YOLOv10 and Norfair) with triggered acquisition for event-driven data capture, and demonstrates several studies that leverage the curated data for isotope identification, background adaptation, and network-wide data fusion. The datasets and labeled resources aim to accelerate development of radiological/nuclear analytics and broader nonproliferation insights, with data sharing planned through public repositories.

Abstract

The PANDAWN sensor network in Chicago, IL, is a state-of-the-art test-bed for networked, multi-modal sensing. It integrates AI/data science methods into its operation, from data acquisition to automated data labeling and curation workflows. The curation and dissemination of diverse multi-modal data sets will enable the development of new radiological/nuclear (R/N) detection, localization, and tracking algorithms, and methods relevant across the nonproliferation mission space. This paper first introduces the PANDAWN sensor network and the features that make it stand out from previous multi-modal data acquisition efforts. We then review the various data streams acquired on the PANDAWN nodes, and present the implementation of an automated data curation pipeline that includes the labeling of radiation and contextual data streams. We finally provide a short overview of different studies that leveraged the curated data sets.

Figures (11)

• Figure 1: Engineering drawing of a PANDAWN node (left), and map showing the deployment locations of 13 such nodes in Chicago, IL (right). Inserts show a node mounted on a traffic light pole (top), and a typical intersection field of view (bottom).
• Figure 2: Example time series of environmental data streams for meteorological data (top left), and precipitation and radon progeny proxy (bottom left). Typical static background data (top right), and example of radiation data including the presence of a positron-emitting radiotracer, e.g., $^{18}$F (bottom right).
• Figure 3: Illustration of the triggered acquisition scheme: upon an R/N anomaly detection (top, e.g., $^{60}$Co detection), high rate camera and LIDAR data (bottom left and right resp.) are synchronously acquired with radiation and environmental data.
• Figure 4: Schematic diagram of the automated labeling and curation workflow.
• Figure 5: Example calibration fit combining a detector/readout model and a series of simulated NORM templates.