Recurrent and Concurrent Neural Networks for Objects Recognition

F. Cecconi and M. Campenní (Italy)


Artificial neural networks; recurrent neural networks; ob ject recognition; artificial vision; autonomous robotics.


A system based on a neural network framework is consid ered. We used two neural networks, an Elman network [1][2] and a Kohonen (concurrent) network [3], for a cate gorization task. The input of the system are objects derived from three general prototypes: circle, square, polygon. We varied the size and orientation of the objects in a continuous way. The system is trained using a new algorithm, based on recurrent version of backpropagation and Kohonen rule. The system achieves the capacity to predict the shape of the objects with a remarkable generalization [4]. We compare our results with the results using a classical Elman network. The model is implemented by a Matlab/Simulink environ ment.

Important Links:

Go Back