High Speed High Accuracy Artificial Neural Networks for UAV based Target Identification

The machine learning and artificial intelligence community has recently garnered much attention for ground breaking performance of novel neural network architectures for self-driving cars. One of the machine learning methods used in self-driving cars is semantic segmentation. In this fashion each pixel in an image is label with a class, allowing for contour-based image segmentation which is different than the traditional sliding windows technique used in convolutional neural networks. UHV has previously developed a contour-based image segmentation technique for several of its commercially available products which include machinery in industrial recycling. In this Phase I effort UHV Technologies will develop innovative methods for target identification for UAV and UAS with state of the art machine learning methods in semantic segmentation, by modifying their existing ML/AI software, then demonstrate the technology at one of the nation’s six major test sites for UAVs and UASs with UTARI. The Phase II work will include optimizing SWaP parameters in addition to customizing the algorithms for specific applications.