Insect Bio-inspired Neural Network Provides New Evidence on How Simple Feature Detectors Can Enable Complex Visual Generalization and Stimulus Location Invariance in the Miniature Brain of Honeybees
Fig 5
Schematic representation of the models. The pattern processing stages for the type A and type B lobula large-field orientation-sensitive neurons (LOSN) and their connectivity to the mushroom body Kenyon cells.
(a) Each simulated eye perceives one half of the test image (left eye shown). Lamina: converts a given pattern image into a binary (black/white) retinotopic representation. Medulla: extracts edges resolvable by honeybees and determines the length of all orientations (0°-180°) within the upper and lower image halves. Lobula: within the upper and lower image regions, the LOSN firing rates for the type A and type B neurons are calculated (see Fig 6). The same process is repeated for the right eye producing in total eight LOSN responses. These are then passed to the appropriate 10,320 (DISTINCT model) or 5,160 (MERGED model) mushroom body Kenyon cells. (b) Firing rate responses of our theoretical LOSNs (type A: orange, type B: blue) to a 280 pixel edge at all orientations between 0°–180°; tuning curves adapted from honeybee electrophysiological recordings [13]. (c) Scale factor applied to the LOSN firing rates dependent on the total edge pixel length in each pattern quadrant, nonlinear scaling factor derived from dragonfly neuronal responses to oriented bars with differing bar lengths [14].