3D ATR plus Motion

In the realm of SAR imaging, a critical challenge involves automatically identifying ground and surface targets in contested environments. This challenge, at a fundamental level, requires the development of an ATR system that can classify targets using image products from the SAR system. This has been difficult using conventional 2D SAR images because conventional images do not correctly represent the geometry of the targets. Consequently, the proposed work will investigate the potential to create 3D SAR images and exploit the more accurate and complete information in them to obtain better classification performance.ááTo that end, MTRI, Inc. and MTU will conduct the Phase I SBIR 3D SAR ATR. This project will investigate the feasibility of improving 3D SAR ATR for robust target classification and identification of multiple classes of ground and surface targets which may be stationary or moving. The Phase I SBIR will alsoágenerate simulated ground and surface target data representative of realistic radar data collections, support automated construction of 3D inputs for the 3D ATR system, and initiate the development of a performance prediction algorithm to facilitate choosing superior SAR data collection strategies. Phase I of the project will focus on improving classification results provided by a 3D neural network architecture.ááImprovements to the ATR performance may be made by improving the processing that converts the SAR data into 3D images. Further improvements can be made by exploiting modifications of the 3D neural network architecture to adapt to special characteristics of the SAR signals. Many neural network configurations work well on optical imagery because of the extensive availability of training data. New strategies need to be developed for SAR imagery for which the labeled training data is less abundant.ááThe SAR based ATR system will provide performance without being interrupted by unfavorable weather and illumination conditions. This supports continuous intelligence for defense operations.á