The human brain receives, integrates, and processes sensor data and various forms of metadata. It detects, recognizes, and tracks objects of interest. It communicates with other brains. The brain has motor control over its host body. On an abstract level, the brain and ATR have a lot in common. They have to solve similar computational tasks. This leads to similarities in design.
Any network whose neurons send feedback signals to each other is a recurrent neural network (RNN). The human brain is an RNN with many feedback loops. RNNs can learn to process sequential data not easily learned by other types of neural networks. A recurrent ATR is suitable for processing still frame, video, and various kinds of temporal signals.
Section 5.2 discusses brain versus ATR hardware design. Section 5.3
covers algorithm/software design. A strawman (reference) design is provided, but with no claim that this is the only way to construct a next-generation ATR. The strawman design should be thought of as a brainstormed, simple, draft proposal intended to generate discussion of its advantages and disadvantages, and to trigger the generation of new and better proposals. The strawman is not expected to be the final creation. It should be kicked around and refined until a finished model is obtained that meets a project’s key performance goals. The final ATR design can be very different from the strawman design.
Online access to SPIE eBooks is limited to subscribing institutions.