DLeN: Deep Learning for standoff detection of Special Nuclear Material

Deep Learning for standoff detection of Special Nuclear Material (DLeN) applies the same deep learning techniques that allow computers to beat human performance in image recognition and the game of Go to detecting Special Nuclear Material. Spectral analysis and signal processing can in some cases be augmented by the use of much larger neural nets that conduct much deeper analysis of features of the sensor data. This may enable the extraction of information indicating presence of SNM from a standoff distance and with a shorter amount of time. Training a deep neural net is very computationally intensive and requires specialized hardware. Execution is very computationally inexpensive and can easily happen in JVM even with very modest CPU and memory.Phase I of the project decidedly proved feasibility by training an ensemble of deep neural nets to analyze gamma spectral data, resulting in a substantial improvement of range of detection. In Phase II, we propose to adopt the models for real-world use and get them to deployment.