Highly accurate recognition of handwritten digits by using photomemristive properties of graphene/diamond heterojunctions

Abstract This work applied graphene/diamond photomemristors, a special type of memristor with optically-controllable nonvolatile memory states, as highly accurate image recognition devices. These photomemristors were able to function as high-performance physical reservoirs, based on the competition between a high photoconductivity response and a short signal decay that induced complicated dynamics. The accuracy with which handwritten numerals (0–9) could be recognized was found to depend on the reduction ratio for images during preprocessing. A high recognition accuracy of approximately 97% was possible in the absence of reduction or with minimal reduction. Higher reduction ratios substantially decreased the recognition accuracy to approximately 80%. It is evident that careful selection of the reduction ratio for such images is vital. The present photomemristors provide both photo-sensing and processing capabilities, and so may have applications in novel neuromorphic devices capable of information processing in sensors with lower power consumption and reduced calculation costs.