Neural network classification of EEG during camouflaged object identification

Abstract

A generalized regression neural network (GRNN) was trained to discriminate between EEGs recorded while subjects identified a camouflaged target object (picture condition) from EEGs recorded during a visually matched control task (control condition). In the picture condition subjects, three female and two male right handers, ages 23-47, viewed images depicting camouflaged target objects and signaled identification by blinking. In the control condition subjects viewed a neutral screen and blinked at will. EEGs made immediately preceding and following the blink were band-pass filtered at 2–8 Hz. The network achieved a marked increase in discriminability in the final 250 ms preceding target identification, with chance level of discrimination before and after. Network performance using scrambled data was also at chance level.