Spatio-temporal Receptive Fields in Dim Light - Evolution and Adaptation

Vision is an important sense for most species throughout the animal kingdom and even in the dimmest habitats animals have functional eyes. Different eye types are not adaptations to ecological conditions alone but eye design is also constrained by the physical nature of light. Vision in dim light is photon limited and the superposition eye in insects has evolved to utilise as much photons reaching the eye as possible. Some nocturnal insects, however, have apposition eyes, an eye type not optimal for vision at low luminances.

Spatial summation and temporal integration can enable a diurnal eye to improve vision in dim light. The aim of this study is to model and examine neural adaptations of receptive fields of interneurons under dim-light conditions. The model is based on natural image data. It is shown that receptive fields get spatially enlarged as images get noisier. Furthermore the effect of motion on the time response of photoreceptors is investigated. Noisier images require a prolonged integration time whereas motion causes a shortening of integration time. It is shown that spatial pooling and temporal integration deteriorate signal-to-noise ratio at high spatial and temporal frequencies in order to improve signal-to-noise ratio at low frequencies.

The spatial part of the receptive field is modeled by a semilinear neural net (Backpropagation algorithm). The temporal properties are modeled with a log-normal function and parameter optimisation is done by means of a direction set method (Powell's algorithm).