A device for image recognition based on a full-analog photonic neural network (FAPNN), comprising a first subsystem, a second subsystem, a third subsystem, and an electronic digital controller, wherein the first subsystem, the second subsystem, and the third subsystem share one laser and comprise an optical block for Fourier convolution with taking into account scaling, rotation, and phase factors, the first subsystem and the second subsystem share training transparent holographic matrices, and the second subsystem and the third subsystem share a three dimensional (3D) holographic memory matrix;
- the electronic digital controller is configured to control, by using deflectors and a beam emitted by the laser and a training analog image dataset as inputs, the first subsystem to create the training transparent holographic matrices corresponding to various discrete image categories in the training analog image dataset;
- and
- control, by using the deflectors and the beam emitted by the laser and an output of the first subsystem as inputs to the second subsystem, the second subsystem to generate different separate holographic cells on the 3D holographic memory matrix, wherein the holographic cells corresponding to the various discrete image categories;
- and, record, with an optical change of the training transparent holographic matrices, the various image categories in different holographic cells of the 3D holographic memory matrix at different angles of a reference beam and an object beam;
- the third subsystem is configured to receive an analog image to be recognized, convert an image recognition result into an analog signal, and output the signal according to the analog image to be recognized corresponding to the value of the holographic cell recorded in the 3D holographic memory matrix;
- wherein, a number of the training transparent holographic matrices corresponds to a number of the recorded categories of images, wherein each of the training transparent holographic matrices writing the images from one category of images;
- and a number of the holographic cells in one of the training transparent holographic matrices corresponds to the number of the recorded images of one category;
- the first subsystem further comprises the laser, a open shutter, two translucent reflectors, a reflector, a spatial light modulator, a first beam expander, a second beam expander, a first optical block, the training transparent holographic matrices, and a first deflector;
- a first end of the spatial light modulator is connected to the laser, through the open shutter, the reflector, the two translucent reflectors, and the first beam expander, a second end of the spatial light modulator connected to the first optical block, a first end of the first optical block is connected to a first end of the first deflector, and a second end of the first deflector is connected to the electronic digital controller;
- wherein, the laser and the first beam expander can be replaced by a laser matrix;
- wherein,
- the controlling of the first subsystem to create the training transparent holographic matrices corresponding to the various discrete image categories in the training analog image dataset, specifically comprising:
- receiving the object beam emitted by the laser and each image from the training analog image dataset by the spatial light modulator, transforming the object beam through an optical Fourier convolution operation by the first optical block, and directing the transformed beam to the first deflector to direct at a next cell of the training transparent holographic matrices from the first deflector;
- and
- expanding and directing the reference beam emitted by the laser to the training transparent holographic matrices, and fixing Fourier images in the cells of the training transparent holographic matrices in the form of a hologram;
- the second subsystem further comprises the laser, the second beam expander, a second optical block, a second deflector, a third deflector, the training transparent holographic matrices, and the 3D holographic memory matrix;
- wherein, the laser and the second beam expander can be replaced by the laser matrix;
- wherein,
- a first end of the second beam expander is connected to the laser, a second end of the second expander is directed to the training transparent holographic matrices and then is directed to the second optical block after output of the training transparent holographic matrices, and after the second optical block the beams are directed by the second deflector to the cells of the 3D holographic memory matrix;
- a first end of the third deflector is directed to the 3D holographic memory matrix and a second end of the third deflector is connected to the laser through the two translucent reflectors and the third deflector is connected to the electronic digital controller;
- wherein,
- the recording of the various image categories in different holographic cells of the 3D holographic memory matrix at different angles of the reference beam and the object beam, comprising:
- directing the object beam emitted by the laser to recorded training transparent holographic matrices and passing it through the second optical block, making simultaneously a Fourier convolution of an image set modulated in the training transparent holographic matrices occur in the second optical block, focusing the object beam into one beam carrying information about a plurality of Fourier images recorded in the training transparent holographic matrices, and then deflecting the beam by the second deflector to corresponding cells in the 3D holographic memory matrix, each cell corresponding to a single category of images;
- directing the reference beam emitted by the laser through the third deflector to the same cell towards which the object beam pointing as a result of two Fourier transforms, and repeating the directing operations as the object beam for other images recorded on another training transparent holographic matrix, until all the training transparent holographic matrices are stored into the 3D holographic memory matrix;
- and
- the third subsystem further comprises the laser, a third beam expander, a fourth beam expander, an analog camera, an optical block, the 3D holographic memory matrix, a detector matrix, an analog-to-digital converter, and an electronic analog calculator, wherein the detector matrix has an input end connected to the 3D holographic memory matrix and an output end connected to the analog-to-digital converter and the electronic analog calculator;
- wherein, when the third subsystem performs the image recognition, each image in the detector matrix is represented by a set of cells with different intensities, and all the cells are displayed by the analog-to-digital converter and under control of computation by the electronic analog calculator;
- wherein, the laser and the third beam expander can be replaced by the laser matrix.