SBIR-STTR Award

Physical Sciences Inc. (PSI) proposes to develop a suite of artificial intelligence algorithms designed to discriminate airborne chemical/biological warfare agent plumes from battlefield clutter in standoff LIDAR data. The AI-assisted LIDAR clutter mitigation (ALCM) system will track all plume-type objects within the LIDAR field of regard, and employ a two-stage classification algorithm to quantify the probabilistic threat level of each plume. The ALCM will utilize a convolutional neural network to identify and characterize plumes in each LIDAR scan based on shape and concentration profile, and additional confidence refinement will be achieved through characterization of plume properties such as airborne mass and dissipation rate by performing temporal analysis of subsequent LIDAR scans with DisperseNET, PSI’s real-time dispersion modeling algorithm. The ALCM system is designed to quantify threat/non-threat confidences for each plume-like object, provide these outputs to the user in real-time, and achieve a greater than 90% threat classification probability at an operationally relevant false classification rate of 1 in 240 hours. The Phase I program will develop the CNN plume classification model, integrate the CNN model outputs to DisperseNET, and culminate in the performance characterization of the prototype ALCM system using government provided historical LIDAR data

Physical Sciences Inc. proposes to mature a suite of artificial intelligence algorithms developed to discriminate airborne chemical/biological warfare agent plumes from battlefield clutter in standoff light detection and ranging (LIDAR) data. The AI-assisted LIDAR clutter mitigation (ALCM) system tracks all plume-type objects within the LIDAR field of regard, and employs a two-stage classification algorithm to quantify the probabilistic threat level of each plume. The ALCM utilizes a convolutional neural network to identify and characterize plumes in each LIDAR scan based on shape and concentration profile. Additional confidence refinement is achieved through characterization of plume properties such as airborne mass and dissipation rate by performing temporal analysis of subsequent LIDAR scans using DisperseNET. The DisperseNET algorithm is PSI’s real-time dispersion modeling algorithm. The prototype ALCM system demonstrated a 92% probability of threat detection and classification at an average of 301 LIDAR scans between false track output in the Phase I. The primary objective of the proposed effort is to improve and operationalize the ALCM capability through enhancement of the DisperseNET simulation engine, expansion of the training and validation datasets, and demonstration of the ALCM capability through the production of a ruggedized processing module for LIDAR integration