One dimensional convolutional neural networks for improved training time and standardization in spectral classification

Physical Sciences Inc. (PSI) proposes to develop a deep learning based spectral target detection algorithm for identification and classification of opioids and explosives that will be executed on a portable hardware solution. The proposed algorithm is designed to be spectrometer agNstic and allows for rapid training on previously untrained spectrometer platforms. The proposed hardware solution is a modular system comprised of a larger training module, and an operational module with a SWAP footprint that is approximately four times less than the training module. The proposed algorithm will define a set of standard inputs required for training and operational deployment, and will include a simple user interface allowing an operator to train a machine-learning algorithm for a new spectrometer and/or to identify previously defined targets when in operation. In the Phase I effort, PSI will demonstrate the proposed algorithm feasibility using simulated datasets representative of potential IR spectrometer data, demonstrate the training and operational throughputs using the identified hardware solutions, standardize the required data types and formats that are user supplied, and develop a preliminary design for the user interface.