Deep Automatic Target Recognition (DATR) Technology/R-DEX Systems, Inc.

R-DEX Systems proposes to adapt its commercial discrimination technology used in industrial automation to develop Deep Automatic Target Recognition (DATR) Technology. DATR utilizes revolutionary deep learning processing (deep belief networks or DBNs, restricted Boltzmann machines RBMs, and convolutional neural networks or CNNs) to automatically identify hidden, nonlinear features that are not identifiable by traditional, shallow methods. These deep, hidden features are used to discriminate between different target types (classes). DATR training can be performed offline, generating a set of weights for each node in the neural network. Only these weights are required for classification; template databases are not required for comparison and there is no metric calculations over multiple target types and poses. The classifier itself is computationally efficient, capable of real-time implementation in applications with limited onboard computational resources. Deep Learning, like all Machine Learning paradigms, requires large amounts of "good" data for train purposes. The amount of data required is a complex issue depending upon the complexity of the data and the complexity of the algorithm. Adapting the techniques of nonlinear fractal data analysis, R-DEX Systems will develop a process and algorithm to estimate the amount of training data required to meet a specified level of performance.