A2TTA: Real-time Isotope Identification and Quantification using YOLO-based Neural Network for Advanced Atom Trap Trace Analysis

The ATTA technique makes it possible for DTRA to accurately monitor nuclear activities in a real-time fashion. However, such a capability should be provided in varying contexts of low/high S/N images, spurious events, vibrations, and cosmic radiations, operated in the ocean floater/UAV/orbital vehicles. For this purpose, InfoBeyond advocates A2TTA (Real-time Isotope Identification and Quantification using YOLO-based Neural Network for Advanced Atom Trap Trace Analysis) to offer accurate learning capability for isotope identification and quantification in a real-time fashion. A2TTA presents an innovative design resorting to (i) YOLO (“You Only Look Once”)-based deep learning, (ii) self-attention, and (iii) optimization technologies. Superior to existing approaches, A2TTA can extract and fuse the image features at different objective fields, i.e., abundant uncertainty features that are necessary for atom identification in an uncertain context. The quantification process further implements deep learning of the abstracting hyper-dimensional relations to quantify the number of atoms in the presence. In order to achieve high accuracy, all the extracted image features, fused image features, full image, detected ROI, and their statistical features are considered in deep learning. In addition to atomic identification and quantification, the result from A2TTA in data streaming can be converted into real-time concentration/radioactivity for advanced nuclear monitoring.