Sequential vector classifier based on SOM and feedback Hebbian network
Yusuke Araga, Zuiko Rikuhashi, Hiroomi Hikawa · 2012
This paper proposes a new type of hybrid network that can classify the dynamic temporal behavior of vectors. The proposed system consists of Self-organizing map (SOM) and a supervised learning network with feedback. The SOM performs stimulus classification and a supervised network identifies the dynamic behavior of input vectors. The supervised network is trained by using Hebbian learning. To demonstrate the feasibility of the proposed network, it is applied to recognize dynamic hand gestures. Experimental results show that the system can recognize nine gestures with the accuracy of 93%.