Posture Recognition Based on the Improved Optical Diffractive Neural Network

Optical computing has significant advantages in power consumption efficiency and computing speed. As an optical machine learning framework, diffractive optical neural networks have achieved good results in the application of feature detection and target classification. This paper introduces an optical neural network structure-express wavenet, which has a random shift wavelet pattern and a highway structure, which can greatly reduce the parameters generated in the modulation process of light and solve the problem of gradient disappearance during network training. The classification rate of express wavenet network for the 4 types of different posture and action image data sets reached 87.96%, which is a greater improvement than the 77.8% recognition rate of the diffraction deep neural network D2NN.