Optically realize convolution operation of microlens array

As a simple linear translation invariant operation, convolution has been widely used in various fields of image processing, and the convolutional neural network derived from it is brilliant in the field of artificial intelligence. In order to deal with the problem of limited computing power of AI reasoning chip in the post-Moore era, optical neural network came into being. As one of the important research hotspots, optical convolutional neural network plays an important role in promoting the development of optical neural network. An optical convolution system was designed, based on the uniform light path formed by micro lens array and lens, the image carried in the light place was convoluted in two-dimensions. The system can complete simple image smoothing and sharpening in the optical path. When the spatial light modulator is used to realize the convolution kernel and input surface, the system can realize three convolution forms of various step sizes, and can also realize multi-channel three-dimensional convolution through multiple projection or flattening, thus laying a foundation for the realization of optical convolution neural network for complex image processing tasks.