Fig. 6: MNIST handwritten digital image classification demonstration.
From: Compact optical convolution processing unit based on multimode interference
a The network structure of the CNN, which contains an optical convolution layer and an electrically fully connected layer. b The confusion matrix of recognizing 10,000 digits in the MNIST test database, where the abscissa indicates the true labels and the ordinate indicates the recognition results. c The variation in simulation accuracy, experimental accuracy, and experimental cross entropy loss during 350 epochs of training.
