Learning process in a data classification device comprising a neural network and classification device for the implementation of this process

摘要

Data classification device comprising a neural processor (21) with (N+1) inputs and K outputs and notable in that it also includes: (A) a memory (10) for storing the V prototype vectors of dimension N of a learning base, associated, for defining the order of presentation of the said prototype vectors to the neural processor, with a sequencer (11) controlled by a master processor (12) receiving the said base's defining information; (B) between the said memory and the said neural processor (21), a preprocessing processor (20) for converting the said prototype vectors into vectors of dimension (N+1); (C) at the output of the said neural processor, a memory (30) for storing, for each of the prototype vectors, the K output activations subsequently provided to a stochastic processor (31) itself followed by a memory (32) for the classes thus determined, and then, at the output of this memory, by a processor (33) for determining the adaptation or deactivation decision to be provided to the neural processor. Application: image segmentation. IMAGE