A Time-Delay Feedback Neural Network for Discriminating Small,\n Fast-Moving Targets in Complex Dynamic Environments

Abstract

Discriminating small moving objects within complex visual environments is a\nsignificant challenge for autonomous micro robots that are generally limited in\ncomputational power. By exploiting their highly evolved visual systems, flying\ninsects can effectively detect mates and track prey during rapid pursuits, even\nthough the small targets equate to only a few pixels in their visual field. The\nhigh degree of sensitivity to small target movement is supported by a class of\nspecialized neurons called small target motion detectors (STMDs). Existing\nSTMD-based computational models normally comprise four sequentially arranged\nneural layers interconnected via feedforward loops to extract information on\nsmall target motion from raw visual inputs. However, feedback, another\nimportant regulatory circuit for motion perception, has not been investigated\nin the STMD pathway and its functional roles for small target motion detection\nare not clear. In this paper, we propose an STMD-based neural network with\nfeedback connection (Feedback STMD), where the network output is temporally\ndelayed, then fed back to the lower layers to mediate neural responses. We\ncompare the properties of the model with and without the time-delay feedback\nloop, and find it shows preference for high-velocity objects. Extensive\nexperiments suggest that the Feedback STMD achieves superior detection\nperformance for fast-moving small targets, while significantly suppressing\nbackground false positive movements which display lower velocities. The\nproposed feedback model provides an effective solution in robotic visual\nsystems for detecting fast-moving small targets that are always salient and\npotentially threatening.\n