Detecting a target object using an expanded neocognitron

This paper proposes an expanded neocognitron which can recognize both the shape and the location of an object. To construct such an expanded neocognitron as a robot vision system, we present an improved mechanism of local feature extraction and competitive learning. Moreover, we introduce an expanded network architecture. This system can classify the patterns on an input screen in just one feed forward processing. In the computer simulation, it is assumed that the image is to be input to the system each time from a single camera attached to the end-effector of a robot manipulator. Through this experiment, the ability of the proposed system to detect and maintain attention on a target object is demonstrated.