Liquid crystal image recognizer, preparation method thereof and image recognition device

Abstract

The invention discloses a liquid crystal image recognizer, a preparation method thereof and an image recognition device. The liquid crystal image recognizer comprises a liquid crystal box and a reflecting mirror; the liquid crystal box comprises a first substrate and a second substrate which are oppositely arranged; the first rotation cholesteric liquid crystal layer and the second rotation cholesteric liquid crystal layer are located between the first substrate and the second substrate in a stacked mode; the first rotation cholesteric liquid crystal layer and the second rotation cholesteric liquid crystal layer are provided with a plurality of orientation structures which are distributed at equal intervals and are optimized by a deep learning algorithm; the reflector is located on the side, away from the second substrate, of the first substrate, and a preset distance is arranged between the reflector and the first substrate, so that a light beam turn-back framework is formed between the reflector and the liquid crystal box. The liquid crystal image recognizer provided by the invention can realize an independent image recognition function on circular polarization incident light with two kinds of chirality, and can also realize multifunctional image recognition of wavelength multiplexing through liquid crystal box cascading of devices, so that the functional diversity and multiplexing dimension of the all-optical image recognizer are improved.

Description

Technical Field

[0001] The present invention relates to the field of liquid crystal micro-nanostructure manipulation and optical computing technology, and in particular to a liquid crystal image recognizer and a preparation method thereof, and an image recognition device. Background Art

[0002] Artificial intelligence is an important driving force for the new round of scientific and technological revolution and industrial transformation. Artificial neural network is one of the most popular artificial intelligence frameworks and has been widely used in many fields to perform computing tasks such as machine vision, natural language processing, and medical diagnosis. However, as Moore's Law approaches its physical limit, mainstream electronic neural networks will encounter bottlenecks in processing speed and power consumption. Accelerating the development of a new generation of artificial intelligence has become a strategic issue. Optical neural networks are a strong competitor for the next generation of computing platforms. They use photons instead of electrons to achieve the mapping between input and output signals. Due to the high propagation speed, high parallelism and multiple degrees of freedom of optical information, optical neural networks have potential advantages such as fast, low energy consumption and high throughput. In recent years, scholars at home and abroad have proposed a variety of optical neural network frameworks for realizing different intelligent computing. Among them, the diffraction neural network realizes the interconnection of neurons at each layer through the diffraction of light, providing a scalable and cost-effective platform for optical computing.

[0003] In 2018, the Ozcan team at the University of California, Los Angeles proposed a 3D-printed all-optical diffraction neural network that can achieve classification and imaging functions in the terahertz band, which opened up a research boom in this field. Since then, the Ozcan team and teams from the University of Delaware, Tsinghua University, Southeast University, and others have used diffraction neural networks for various computing tasks such as image recognition, optical logic operations, and all-optical phase recovery. Among them, all-optical image recognition technology has always been a research hotspot in this field and is expected to have great potential in application scenarios such as autonomous driving and medical diagnosis. With the increasing demand for image recognition speed, accuracy, and versatility, multi-channel, high-efficiency, and easy-to-integrate image recognizers are urgently needed.

[0004] Traditional all-optical image recognizers are mainly based on 3D printing, spatial light modulators, metasurfaces and other systems, which can achieve high-accuracy image recognition. However, most of them have limitations such as unadjustable functions, single applicable wavelength, limited efficiency, and difficulty in integration, which restricts the development of all-optical image recognition technology. Summary of the invention

[0005] The embodiments of the present invention provide a liquid crystal image recognizer and a preparation method thereof, and an image recognition device. The liquid crystal image recognizer can realize independent image recognition functions for two types of chiral circularly polarized incident light. By cascading the liquid crystal boxes of the device, multifunctional image recognition with wavelength multiplexing can also be realized to improve the functional diversity and multiplexing dimension of the all-optical image recognizer.

[0006] According to one aspect of the present invention, there is provided a liquid crystal image recognizer, comprising a liquid crystal cell and a reflective mirror;

[0007] The liquid crystal box comprises:

[0008] A first substrate and a second substrate arranged opposite to each other;

[0009] A first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer are stacked between the first substrate and the second substrate, a first orientation layer is disposed on a side of the first substrate close to the first chiral cholesteric liquid crystal layer, and a second orientation layer is disposed on a side of the second substrate close to the second chiral cholesteric liquid crystal layer;

[0010] The first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have opposite chiral directions, and the first orientation layer and the second orientation layer are both provided with a plurality of orientation regions distributed at equal intervals, and within the orientation regions, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have different orientation structures, and the orientation structures are trained as weight parameters of a multi-layer diffractive neural network and are obtained through iterative optimization of a deep learning algorithm;

[0011] The reflector is located on a side of the first substrate away from the first alignment layer, and a preset distance is set between the reflector and the first substrate to form a beam return structure between the reflector and the liquid crystal box.

[0012] Optionally, the liquid crystal image recognizer includes at least two liquid crystal cells arranged in cascade, and the central wavelengths of the Bragg reflection bands of at least two of the liquid crystal cells are different, so as to form a wavelength multiplexing multifunctional liquid crystal image recognizer.

[0013] Optionally, the first alignment layer and the second alignment layer are provided with a plurality of alignment regions of the same size, and along the direction from the first substrate to the second substrate, the corresponding alignment regions in the first alignment layer and the second alignment layer are aligned;

[0014] For the first alignment layer or the second alignment layer, the alignment regions are distributed at equal intervals along a direction from the first substrate to the second substrate, and different alignment regions do not overlap.

[0015] Optionally, in different alignment regions, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer are provided with different alignment structures, and the total number of the alignment structures is equal to twice the number of the alignment regions;

[0016] Outside the alignment region, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have a disordered surface alignment;

[0017] The orientation structures are all optimized by a deep learning algorithm, which is based on a multi-layer diffraction neural network architecture, in which the input layer is an image to be identified, the output layer is the light intensity in multiple preset detection areas, the weight of the hidden layer is set to the phase, the number of hidden layers is equal to the number of times the light beam returns to pass through the orientation area, and the layers are connected by an angular spectrum diffraction formula; according to the geometric phase principle of cholesteric liquid crystal, for a cholesteric liquid crystal layer with a left-handed spiral structure, the orientation structure is equal to -0.5 times the weight of the corresponding hidden layer after iteration, and for a cholesteric liquid crystal layer with a right-handed spiral structure, the orientation structure is equal to +0.5 times the weight of the corresponding hidden layer after iteration.

[0018] Optionally, the first handed cholesteric liquid crystal layer is a left-handed spiral structure, and the second handed cholesteric liquid crystal layer is a right-handed spiral structure, or the first handed cholesteric liquid crystal layer is a right-handed spiral structure, and the second handed cholesteric liquid crystal layer is a left-handed spiral structure;

[0019] In the same liquid crystal cell, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have circularly polarized selective Bragg reflection bands with the same central wavelength.

[0020] Optionally, the liquid crystal box further comprises a spacer disposed between the first substrate and the second substrate to control the total thickness of the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer;

[0021] The extension length of the spacer is greater than or equal to 20 times the helical pitch of the liquid crystal molecules in the first chiral cholesteric liquid crystal layer.

[0022] According to another aspect of the present invention, there is provided a method for preparing a liquid crystal image recognizer, which is used to prepare the above-mentioned liquid crystal image recognizer, and the preparation method comprises:

[0023] Providing a first substrate and a second substrate, forming a first alignment layer on one side of the first substrate, and forming a second alignment layer on one side of the second substrate;

[0024] The first substrate and the second substrate are arranged opposite to each other, and a first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer having a plurality of equidistantly distributed alignment structures are prepared between the first substrate and the second substrate to form a liquid crystal box;

[0025] A reflector is arranged on a side of the first substrate of the liquid crystal box away from the first alignment layer to form the liquid crystal image recognizer;

[0026] Among them, the first orientation layer is located on the side of the first substrate close to the first chiral cholesteric liquid crystal layer, and the second orientation layer is located on the side of the second substrate close to the second chiral cholesteric liquid crystal layer. The orientation structures of the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer are different, and are both trained as weight parameters of a multi-layer diffraction neural network and obtained through iterative optimization of a deep learning algorithm.

[0027] Optionally, forming a first alignment layer on one side of the first substrate and forming a second alignment layer on one side of the second substrate includes:

[0028] Spin coating a solution containing an alignment material on one side of the first substrate, and after the spin coating is completed, annealing the first substrate to form the first alignment layer;

[0029] A uniform mixed solution containing an alignment material and a photoinitiator is spin-coated on one side of the second substrate. After the spin coating is completed, the second substrate is annealed to form the second alignment layer.

[0030] Optionally, a first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer having a plurality of equally spaced alignment structures are prepared between the first substrate and the second substrate to form a liquid crystal cell, comprising:

[0031] Performing linearly polarized ultraviolet exposure on an empty box consisting of the first substrate, the second substrate, the first alignment layer and the second alignment layer, and sequentially exposing the alignment structure corresponding to the second chiral cholesteric liquid crystal layer at equal intervals;

[0032] Filling a mixture of second chiral cholesteric liquid crystal and polymer monomer into the space between the first substrate and the second substrate to form the second chiral cholesteric liquid crystal layer with the alignment structure;

[0033] The formed second handedness cholesteric liquid crystal layer is subjected to surface-induced ultraviolet polymerization, and the molecules that have not undergone polymerization reaction are washed away with acetone, so that the second handedness cholesteric liquid crystal layer shrinks to form a second handedness polymer network support connected to the surface of the second substrate;

[0034] Performing a second round of linearly polarized ultraviolet exposure to sequentially expose the alignment structure corresponding to the first chiral cholesteric liquid crystal layer at equal intervals, and rewriting the alignment directions of the first alignment layer and the second alignment layer to be consistent with the alignment structure of the second round of exposure;

[0035] The first chiral cholesteric liquid crystal is filled into the space between the first substrate and the second chiral cholesteric liquid crystal layer to form the first chiral cholesteric liquid crystal layer with the alignment structure of the second round of exposure, and the second chiral cholesteric liquid crystal layer rebounds to form the liquid crystal box.

[0036] According to another aspect of the present invention, there is also provided an image recognition device, comprising a laser, a first polarizer, a first lens, a second lens, a first quarter wave plate, a beam splitter, a spatial light modulator, a second quarter wave plate, a second polarizer, a third lens, a fourth lens, a third quarter wave plate, the above-mentioned liquid crystal image recognizer and a detector;

[0037] The light beam output by the laser passes through the first polarizer, the beam expansion system composed of the first lens and the second lens, the first quarter wave plate and the beam splitter in sequence, and is modulated and reflected by the spatial light modulator. After being reflected by the beam splitter, it passes through the second quarter wave plate, the second polarizer, the 4f imaging system composed of the third lens and the fourth lens and the third quarter wave plate in sequence to obtain circularly polarized incident light of the image to be recognized, which is incident from one side of the first substrate of the liquid crystal image recognizer, and is reflected back and forth between the equally spaced orientation areas and the reflector. The reflected light that has undergone multiple geometric phase modulations is incident on the detector after diffracting a preset distance. The detector classifies the image to be recognized according to the cumulative light intensity in each detection area.

[0038] Rotating the third quarter wave plate changes the circular polarization chirality of the incident light to switch different functions of the liquid crystal image recognizer.

[0039] The liquid crystal image recognizer provided by the embodiment of the present invention comprises a liquid crystal box and a reflector; the liquid crystal box comprises: a first substrate and a second substrate arranged oppositely; a first-handed cholesteric liquid crystal layer and a second-handed cholesteric liquid crystal layer stacked between the first substrate and the second substrate, a first orientation layer is arranged on the side of the first substrate close to the first-handed cholesteric liquid crystal layer, and a second orientation layer is arranged on the side of the second substrate close to the second-handed cholesteric liquid crystal layer; wherein the first-handed cholesteric liquid crystal layer and the second-handed cholesteric liquid crystal layer have opposite hand directions, and the first orientation layer and the second orientation layer are both provided with a plurality of equally spaced orientation regions, and within the orientation regions, the first-handed cholesteric liquid crystal layer and the second-handed cholesteric liquid crystal layer have different orientation structures, and the orientation structures are trained as weight parameters of a multi-layer diffraction neural network and obtained by iterative optimization of a deep learning algorithm; the reflector is located on the side of the first substrate away from the first orientation layer, and the reflector is arranged at a preset distance from the first substrate to form a beam return structure between the reflector and the liquid crystal box. The technical solution of the embodiment of the present invention can independently perform multiple geometric phase modulations on two chiral circularly polarized incident lights, thereby realizing the image recognition function of circular polarization multiplexing. In addition, cholesteric liquid crystal also has wavelength selectivity, and only has reflection and geometric phase modulation effects on light with wavelengths within the Bragg reflection band, and has no phase modulation effect on light outside the band. The liquid crystal image recognizer provided by the embodiment of the present invention can further realize multifunctional image recognition with wavelength multiplexing by cascading multiple liquid crystal boxes with different central wavelengths of the Bragg reflection bands, effectively improving the limitations of traditional all-optical image recognizers such as non-adjustable functions, single applicable wavelength, limited efficiency, and difficulty in integration.

[0040] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 A schematic diagram of the yz side structure of a liquid crystal image recognizer provided by an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of the yz side structure of another liquid crystal image recognizer provided by an embodiment of the present invention;

[0044] Figure 3A schematic diagram of a method for calculating the surface orientation structure of a liquid crystal layer of a liquid crystal image recognizer provided by an embodiment of the present invention;

[0045] Figure 4 A schematic flow chart of a method for preparing a liquid crystal image recognizer provided by an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of a process flow for preparing a liquid crystal cell in a liquid crystal image recognizer provided by an embodiment of the present invention;

[0047] Figure 6 A schematic diagram of spectrum characterization results of three liquid crystal image recognizers with different central wavelengths of Bragg reflection bands provided by an embodiment of the present invention;

[0048] Figure 7 A schematic diagram of the structure of an image recognition device provided by an embodiment of the present invention;

[0049] Figure 8 A schematic diagram of the detection area light spot and recognition result of the liquid crystal image recognizer provided in an embodiment of the present invention for multi-wavelength recognition of two types of images;

[0050] Fig. 9 A schematic diagram of confusion matrix results of multi-wavelength recognition of all test set images in two data sets by a liquid crystal image recognizer provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0053] Figure 1 A schematic diagram of the yz side structure of a liquid crystal image recognizer provided by an embodiment of the present invention.

[0054] The x and y directions are parallel to the plane where the first substrate is located, and the z direction is perpendicular to the plane where the first substrate is located. Figure 1 The liquid crystal image recognizer provided by the embodiment of the present invention includes a liquid crystal box 100 and a reflector 200; the liquid crystal box 100 includes: a first substrate 110 and a second substrate 120 arranged opposite to each other; a first-handed cholesteric liquid crystal layer 130 and a second-handed cholesteric liquid crystal layer 140 stacked and arranged between the first substrate 110 and the second substrate 120, a first orientation layer 150 is arranged on the side of the first substrate 110 close to the first-handed cholesteric liquid crystal layer 130, and a second orientation layer 160 is arranged on the side of the second substrate 120 close to the second-handed cholesteric liquid crystal layer 140; wherein the first-handed cholesteric liquid crystal layer 130 and the second-handed cholesteric liquid crystal layer 140 have opposite hands, and the first orientation layer 150 and the second orientation layer 160 are both provided with a plurality of orientation regions ( Figure 1 Three orientation regions are schematically shown, i.e., the regions corresponding to the dark gray in the orientation layer, which is not a limitation of the embodiments of the present invention). In the orientation regions, the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 have different orientation structures, which are trained as weight parameters of a multi-layer diffraction neural network and obtained through iterative optimization of a deep learning algorithm. The reflector 200 is located on the side of the first substrate 110 away from the first orientation layer 150, and a preset distance is set between the reflector 200 and the first substrate 110 to form a beam return structure between the reflector 200 and the liquid crystal box 100.

[0055] Among them, the first substrate 110 and the second substrate 120 can be flexible substrates, such as polyimide, or rigid substrates, such as quartz substrates or glass substrates. In specific implementation, the first substrate 110 and the second substrate 120 are both flat plates, and the reflector 200 can be a plane reflector. Optionally, the orientation material in the first orientation layer 150 and the second orientation layer 160 can be at least one of a photocrosslinking material, a photodegradable material, and a photoinduced cis-trans isomerization material, such as a photosensitive azo material SD1. These materials are photo-controlled orientation materials, which can undergo physical or chemical reactions under the irradiation of linear polarized ultraviolet light to generate anisotropic surface forces, thereby inducing the directional alignment of liquid crystal molecules. In addition to the orientation material, the second orientation layer 160 also contains a uniformly mixed photoinitiator, such as benzophenone, which can induce the polymerization reaction to start from the surface of the second orientation layer 160. Under the action of the alignment layer anchoring and the chiral agent, the first handed cholesteric liquid crystal layer 130 and the second handed cholesteric liquid crystal layer 140 both show a periodic spiral structure of a specific handedness. Optionally, the first handed cholesteric liquid crystal layer 130 is a left-handed spiral structure, and the second handed cholesteric liquid crystal layer 140 is a right-handed spiral structure, or the first handed cholesteric liquid crystal layer 130 is a right-handed spiral structure, and the second handed cholesteric liquid crystal layer 140 is a left-handed spiral structure. The structure can be selected according to actual conditions during implementation. In the same liquid crystal box 100, the first handed cholesteric liquid crystal layer 130 and the second handed cholesteric liquid crystal layer 140 have a circularly polarized selective Bragg reflection band with the same central wavelength. The first handed cholesteric liquid crystal layer 130 and the second handed cholesteric liquid crystal layer 140 have the same refractive index and pitch, and therefore have a circularly polarized selective Bragg reflection band with the same central wavelength. Optionally, in the orientation region, the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 can be oriented according to different orientation structures, and in the non-orientation region, the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 have a disordered surface orientation. In the Bragg reflection band, the cholesteric liquid crystal can reflect circularly polarized light having the same chiral orientation as the liquid crystal layer, and give it a geometric phase equal to ±2 times the surface orientation of the cholesteric liquid crystal layer, while transmitting circularly polarized light having the opposite chiral orientation. Light is reflected back and forth between the cholesteric liquid crystal layer and the reflector 200. The mirror reflection of the reflector 200 will flip the circular polarization rotation direction of the light. Therefore, the circular polarized incident light of the first rotation direction will be reflected by the first rotation direction cholesteric liquid crystal layer 130, the reflector 200, the second rotation direction cholesteric liquid crystal layer 140, the reflector 200, and the first rotation direction cholesteric liquid crystal layer 130 in sequence, and will be modulated by multiple corresponding orientation structures in sequence; the circular polarized incident light of the second rotation direction will be reflected by the second rotation direction cholesteric liquid crystal layer 140, the reflector 200, the first rotation direction cholesteric liquid crystal layer 130, the reflector 200, and the second rotation direction cholesteric liquid crystal layer 140 in sequence, and will be modulated by multiple corresponding orientation structures in sequence.By means of the back and forth reflection of light between the liquid crystal box 100 and the reflector 200, the technical solution of this embodiment can independently perform three-fold geometric phase modulation on two types of circularly polarized incident light, thereby realizing the image recognition function of circular polarization multiplexing.

[0056] Optional, continue to refer to Figure 1 The liquid crystal cell 100 further includes a spacer 170 disposed between the first substrate 110 and the second substrate 120 to control the total thickness of the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 .

[0057] The spacer 170 may be a quartz microsphere or a quartz column, and may be disposed at the boundary between the first substrate 110 and the second substrate 120 to support the first substrate 110 and the second substrate 120, and to form a filling space for the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140. Optionally, the extension length of the spacer 170 is greater than or equal to 20 times the helical pitch of the liquid crystal molecules in the first chiral cholesteric liquid crystal layer 130, so that the thickness of each of the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 may be greater than or equal to 10 times the helical pitch of the liquid crystal molecules in the layer, so that the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 may both have high reflectivity when circularly polarized selective Bragg reflection occurs. It is understandable that Figure 1 The figure only exemplarily shows the positional relationship of the spacer 170 used to support the first substrate 110 and the second substrate 120 rather than the actual size and proportion.

[0058] Optional, continue to refer to Figure 1 The first alignment layer 150 and the second alignment layer 160 are provided with a plurality of alignment regions of the same size ( Figure 1 Medium-dark gray area) and unoriented area ( Figure 1 In the light gray area in the middle, along the direction from the first substrate 110 to the second substrate 120, the corresponding orientation regions in the first orientation layer 150 and the second orientation layer 160 are aligned; for the first orientation layer 150 or the second orientation layer 160, the orientation regions are evenly spaced, and along the direction from the first substrate 110 to the second substrate 120, there is no overlap between different orientation regions.

[0059] Optionally, in different orientation regions, the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 are provided with different orientation structures, and the total number of orientation structures is equal to twice the number of orientation regions; outside the orientation region (non-oriented region), the first chiral cholesteric liquid crystal layer 130 and the second chiral cholesteric liquid crystal layer 140 have a disordered surface orientation; the orientation structures are optimized by a deep learning algorithm, and the deep learning algorithm is based on a multi-layer diffraction neural network architecture, the input layer is an image to be identified, the output layer is the light intensity in multiple preset detection regions, the weight of the hidden layer is set to phase, the number of hidden layers is equal to the number of times the light beam returns to pass through the orientation region, and the layers are connected by an angular spectrum diffraction formula; according to the geometric phase principle of cholesteric liquid crystal, for a cholesteric liquid crystal layer with a left-handed spiral structure, the orientation structure is equal to -0.5 times the weight of the corresponding hidden layer after iteration, and for a cholesteric liquid crystal layer with a right-handed spiral structure, the orientation structure is equal to +0.5 times the weight of the corresponding hidden layer after iteration.

[0060] Optionally, the liquid crystal image recognizer includes at least two cascaded liquid crystal cells, and the central wavelengths of the Bragg reflection bands of the at least two liquid crystal cells are different, so as to form a wavelength multiplexing multifunctional liquid crystal image recognizer.

[0061] For example, a liquid crystal image recognizer includes three liquid crystal cells arranged in cascade. Figure 2 A schematic diagram of the yz side structure of another liquid crystal image recognizer provided by an embodiment of the present invention. Figure 2 The liquid crystal cell 100 of the liquid crystal image recognizer includes a first liquid crystal cell 101, a second liquid crystal cell 102, and a third liquid crystal cell 103. The central wavelengths of the Bragg reflection bands of the first liquid crystal cell 101, the second liquid crystal cell 102, and the third liquid crystal cell 103 are different, so as to form a wavelength multiplexing multifunctional liquid crystal image recognizer. In other embodiments, the number of cascaded liquid crystal cells can be set according to actual needs, and can be designed according to actual conditions during specific implementation.

[0062] For example, Figure 3 A schematic diagram of a method for calculating the surface alignment structure of a liquid crystal layer of a liquid crystal image recognizer provided by an embodiment of the present invention can be used to calculate the alignment structure required to realize the above-mentioned image recognition function. Figure 3 All orientation structures are optimized by a deep learning algorithm, where the deep learning algorithm is based on a multi-layer diffraction neural network architecture. The input layer is the image to be recognized, the output layer is the light intensity in multiple preset detection areas, the weight of the hidden layer is set to the phase, and the number of hidden layers is equal to the number of times the light beam returns to the orientation area ( Figure 3 The first training set ( Figure 3The first detection area can be optimized to obtain the corresponding first-layer phase weight φ1, second-layer phase weight φ2, and third-layer phase weight φ3. Figure 3 The example is 60,000 images in the Fashion-MNIST dataset) and the second detection area can be optimized to obtain the corresponding first-layer phase weight φ4, second-layer phase weight φ5 and third-layer phase weight φ6.

[0063] For example, Figure 3 The first hand direction is left-handed, and the second hand direction is right-handed (not limiting the embodiment of the present invention). Figure 1 The three orientation regions arranged from left to right in the figure are defined as the first orientation region, the second orientation region and the third orientation region respectively; in the first orientation region, the first chiral cholesteric liquid crystal layer 130 is provided with a first orientation structure, and the second chiral cholesteric liquid crystal layer 140 is provided with a second orientation structure; in the second orientation region, the first chiral cholesteric liquid crystal layer 130 is provided with a third orientation structure, and the second chiral cholesteric liquid crystal layer 140 is provided with a fourth orientation structure; in the third orientation region, the first chiral cholesteric liquid crystal layer 130 is provided with a fifth orientation structure, and the second chiral cholesteric liquid crystal layer 140 is provided with a sixth orientation structure. According to the geometric phase principle of cholesteric liquid crystal, for a cholesteric liquid crystal layer with a left-handed spiral structure, the orientation structure is equal to -0.5 times the weight of the corresponding hidden layer after iteration, and for a cholesteric liquid crystal layer with a right-handed spiral structure, the orientation structure is equal to +0.5 times the weight of the corresponding hidden layer after iteration. Therefore, the first orientation structure α1 = -0.5φ1, the fourth orientation structure α4 = 0.5φ2, the fifth orientation structure α5 = -0.5φ3, the second orientation structure α2 = 0.5φ4, the third orientation structure α3 = -0.5φ5, and the sixth orientation structure α6 = 0.5φ6. Among them, the first orientation structure, the third orientation structure, and the fifth orientation structure are equally spaced in the first-handed cholesteric liquid crystal layer 130, and the second orientation structure, the fourth orientation structure, and the sixth orientation structure are equally spaced in the second-handed cholesteric liquid crystal layer 140. It can be understood that the technical solution of this embodiment can independently perform three-fold geometric phase modulation on the circularly polarized incident light of two chiralities, thereby realizing the image recognition function of circular polarization multiplexing. For example, for the circularly polarized incident light of the first handed direction, the recognition function of handwritten digital images can be realized, and for the circularly polarized incident light of the second handed direction, the recognition function of fashion images can be realized.

[0064] It should be noted that Figure 3 The calculation method of the surface orientation structure of the liquid crystal layer is merely provided as an example, and is not intended to limit the present invention. In other embodiments, other calculation methods may be used according to specific requirements to obtain the surface orientation structure of the liquid crystal layer with similar functions.

[0065] The liquid crystal image recognizer provided by the embodiment of the present invention can independently perform multiple geometric phase modulations on two types of circularly polarized incident light by setting a first and a second cholesteric liquid crystal layer with opposite rotation directions and the same central wavelength of the Bragg reflection band, and forming a plurality of equally spaced orientation structures in the first and second cholesteric liquid crystal layers, thereby realizing the image recognition function of circular polarization multiplexing. In addition, the cholesteric liquid crystal also has wavelength selectivity, and only has reflection and geometric phase modulation effects on light with a wavelength within the Bragg reflection band, and is suitable for constructing multi-wavelength optical devices. The liquid crystal image recognizer provided by this embodiment can also further realize multifunctional image recognition with wavelength multiplexing by cascading multiple liquid crystal boxes with different central wavelengths of the Bragg reflection bands. The multifunctional liquid crystal image recognizer can effectively improve the limitations of traditional all-optical image recognizers such as non-adjustable functions, single applicable wavelength, limited efficiency, and difficulty in integration.

[0066] Figure 4 A schematic diagram of a process for preparing a liquid crystal image recognizer provided in an embodiment of the present invention, wherein the preparation method is used to prepare the liquid crystal image recognizer provided in the above embodiment, with reference to Figure 4 , the preparation method comprises:

[0067] S110 , providing a first substrate and a second substrate, forming a first alignment layer on one side of the first substrate, and forming a second alignment layer on one side of the second substrate.

[0068] The first substrate and the second substrate may be flexible substrates or rigid substrates with high light transmittance (greater than or equal to 85%). For example, the first substrate and the second substrate may be made of glass, and the thickness of the substrate may be 1 mm to 2 mm. The first orientation layer is located on the side of the first substrate close to the second substrate, and the second orientation layer is located on the side of the second substrate close to the first substrate.

[0069] Optionally, forming a first alignment layer on one side of the first substrate and forming a second alignment layer on one side of the second substrate includes:

[0070] Spin coating a solution containing an alignment material on one side of a first substrate, and after the spin coating is completed, annealing the first substrate to form a first alignment layer;

[0071] A uniform mixed solution containing an alignment material and a photoinitiator is spin-coated on one side of the second substrate. After the spin coating is completed, the second substrate is annealed to form a second alignment layer.

[0072] For example, in this embodiment, the orientation material is photosensitive azo material SD1, and the spin coating solution without photoinitiator contains 0.35% SD1 and 99.65% dimethylformamide; the photoinitiator is benzophenone, and the spin coating solution containing photoinitiator contains 0.35% SD1, 0.15% benzophenone and 99.5% dimethylformamide.

[0073] Exemplarily, the spin coating process may include: first adjusting the rotation speed to 600-900 rpm, controlling the first stage spin coating time to 5s-10s, so that the material is evenly distributed on the substrate surface; then adjusting the rotation speed to 2500-3500 rpm, controlling the second stage spin coating time to 30s-50s, so that the material is coated according to a specific thickness. Optionally, the thickness of the first alignment layer and the second alignment layer can be 30nm-50nm.

[0074] Exemplarily, the annealing process may include: an annealing atmosphere of air, an annealing temperature of 80° C. to 120° C., and an annealing time of 8 min to 12 min.

[0075] It should be noted that the above-mentioned spin coating solution composition, spin coating parameters, and annealing parameters are only exemplary descriptions, and in other embodiments, they can be adjusted according to actual needs.

[0076] S120, disposing a first substrate and a second substrate opposite to each other, preparing a first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer having a plurality of equidistantly distributed alignment structures between the first substrate and the second substrate, to form a liquid crystal box.

[0077] Wherein, when the first substrate and the second substrate are opposite to each other, the first orientation layer and the second orientation layer are arranged relative to each other. In specific implementation, a second chiral cholesteric liquid crystal layer with a second orientation structure, a fourth orientation structure and a sixth orientation structure can be prepared between the first substrate and the second substrate, and then a first chiral cholesteric liquid crystal layer with a first orientation structure, a third orientation structure and a fifth orientation structure can be prepared between the first substrate and the second chiral cholesteric liquid crystal layer. The second orientation structure, the fourth orientation structure and the sixth orientation structure are evenly spaced and are all optimized by a deep learning algorithm. The first orientation structure, the third orientation structure and the fifth orientation structure are evenly spaced and are also optimized by a deep learning algorithm, and are aligned with the second orientation structure, the fourth orientation structure and the sixth orientation structure. The first orientation structure, the second orientation structure, the third orientation structure, the fourth orientation structure, the fifth orientation structure and the sixth orientation structure are all the same size, but have different structures.

[0078] Optionally, when the first substrate and the second substrate are arranged opposite to each other, a first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer having a plurality of equidistantly distributed alignment structures are prepared between the first substrate and the second substrate to form a liquid crystal cell, the method further includes:

[0079] A spacer is disposed between the first substrate and the second substrate.

[0080] The spacer may be a quartz microsphere or a quartz column, and may be disposed at the boundary between the first substrate and the second substrate to support the first substrate and the second substrate to form a filling space between the first chirality cholesteric liquid crystal layer and the second chirality cholesteric liquid crystal layer. In a direction perpendicular to the first substrate and the second substrate, the extension length of the spacer must be greater than or equal to 20 times the pitch of the liquid crystal molecules in the first chirality cholesteric liquid crystal layer.

[0081] S130, disposing a reflector on a side of the first substrate of the liquid crystal box away from the first alignment layer to form a liquid crystal image recognizer.

[0082] The reflector and the first substrate are arranged at a preset distance to form a light beam folding structure between the reflector and the liquid crystal box.

[0083] Figure 5 A schematic diagram of a process flow for preparing a liquid crystal cell in a liquid crystal image recognition device provided by an embodiment of the present invention. Figure 5 Optionally, a second chiral cholesteric liquid crystal layer having a second orientation structure, a fourth orientation structure and a sixth orientation structure is first prepared between the first substrate and the second substrate, and then a first chiral cholesteric liquid crystal layer having a first orientation structure, a third orientation structure and a fifth orientation structure is prepared between the first substrate and the second chiral cholesteric liquid crystal layer, comprising:

[0084] Performing linearly polarized ultraviolet exposure on an empty box composed of a first substrate, a second substrate, a first alignment layer, and a second alignment layer, sequentially exposing the second alignment structure, the fourth alignment structure, and the sixth alignment structure of the cholesteric liquid crystal layer corresponding to a second handedness (taking right handedness as an example, which is not a limitation of the embodiment of the present invention) at equal intervals, so that the alignment directions of the first alignment layer and the second alignment layer are consistent with the three alignment structures;

[0085] Filling a mixture of a second chiral cholesteric liquid crystal and a polymer monomer into the space between the first substrate and the second substrate to form a second chiral cholesteric liquid crystal layer with a second alignment structure, a fourth alignment structure and a sixth alignment structure;

[0086] The formed second handed cholesteric liquid crystal layer is subjected to surface-induced ultraviolet polymerization, and the molecules that have not undergone polymerization reaction are washed away with acetone, so that the second handed cholesteric liquid crystal layer shrinks to form a second handed polymer network support connected to the surface of the second substrate;

[0087] Perform a second round of linear polarized ultraviolet exposure to sequentially expose the first alignment structure, the third alignment structure, and the fifth alignment structure of the cholesteric liquid crystal layer corresponding to the first hand direction (taking left-handed orientation as an example) at equal intervals, so as to rewrite the alignment directions of the first alignment layer and the second alignment layer to be consistent with the three alignment structures exposed in the second round;

[0088] The first chiral cholesteric liquid crystal is filled into the space between the first substrate and the second chiral cholesteric liquid crystal layer to form the first chiral cholesteric liquid crystal layer with the first orientation structure, the third orientation structure and the fifth orientation structure of the second round of exposure, and the second chiral cholesteric liquid crystal layer rebounds.

[0089] After the above-mentioned secondary photo-orientation, surface induced photopolymerization, washing and refilling and other process steps, a first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer with opposite chiral directions and the same central wavelength of the Bragg reflection band are prepared between the first substrate and the second substrate, and a plurality of equally spaced orientation structures are respectively formed in the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer to form a liquid crystal box in a liquid crystal image recognizer.

[0090] Optionally, the two rounds of linear polarized UV exposure include: using a digital micromirror projection system to synchronously control the exposure pattern and the polarizer angle according to the exposure order, and sequentially performing UV exposure treatment on the first orientation region, the second orientation region and the third orientation region of the first orientation layer and the second orientation layer, so that the first orientation layer and the second orientation layer form an orientation structure with equal spacing.

[0091] Exemplarily, the cholesteric phase liquid crystal can be a mixture of a nematic phase liquid crystal and a chiral agent. In this embodiment, the nematic phase liquid crystal uses mixed crystal E7, and the chiral agent uses left-handed chiral agent S5011 or right-handed chiral agent R5011. By changing the proportion of the chiral agent, the central wavelength of the Bragg reflection band of the liquid crystal image recognizer can be arbitrarily controlled. In the mixture of the second handed cholesteric phase liquid crystal and the polymer monomer, the polymer monomer uses RM257, and the mass proportion of the polymer monomer is selected to be 20%.

[0092] It should be noted that the above-mentioned cholesteric liquid crystal components and proportions are merely exemplary descriptions, and are not limitations on the embodiments of the present invention. In other implementations, other components and proportions may be used according to specific requirements.

[0093] For example, Figure 6 A schematic diagram of the spectrum characterization results of three liquid crystal image recognizers with different central wavelengths of Bragg reflection bands provided by an embodiment of the present invention. Figure 6 , the horizontal axis represents the wavelength of the incident light, in nanometers (nm), and the vertical axis represents the reflectivity; the reflection spectrum corresponding to left-handed circularly polarized light is represented by a dotted line, and the reflection spectrum corresponding to right-handed circularly polarized light is represented by a solid line. Figure 6It can be seen that the liquid crystal image recognizer in this embodiment can simultaneously reflect left-handed circularly polarized light and right-handed circularly polarized light within the Bragg reflection band, and has a high reflectivity. This is because the prepared first-handed cholesteric liquid crystal layer and the second-handed cholesteric liquid crystal layer have opposite handedness and the same central wavelength of the Bragg reflection band, and the thickness of the first-handed cholesteric liquid crystal layer and the second-handed cholesteric liquid crystal layer can both meet the Bragg reflection condition. Exemplarily, Figure 6 The three curves in the figure correspond to Figure 2 Among the three liquid crystal cells, the central wavelength of the Bragg reflection band of the first liquid crystal cell is about 485 nm, the central wavelength of the Bragg reflection band of the second liquid crystal cell is about 560 nm, and the central wavelength of the Bragg reflection band of the third liquid crystal cell is about 650 nm.

[0094] It should be noted that Figure 6 The corresponding Figure 2 In other embodiments, other Bragg reflection band positions can be selected for the liquid crystal image recognizer according to actual needs.

[0095] The liquid crystal image recognizer prepared by the embodiment of the present invention comprises a first-handed cholesteric liquid crystal layer and a second-handed cholesteric liquid crystal layer with a stacked arrangement and multi-region orientation; wherein the first-handed cholesteric liquid crystal layer and the second-handed cholesteric liquid crystal layer are provided with different orientation structures, and each orientation structure is trained as a weight parameter of a multi-layer diffraction neural network and is obtained by iterative optimization of a deep learning algorithm. By using processes such as secondary light orientation, surface induced photopolymerization, washing and refilling, a first-handed cholesteric liquid crystal layer and a second-handed cholesteric liquid crystal layer with opposite hand directions and the same central wavelength of the Bragg reflection band can be prepared, and a plurality of equally spaced orientation structures are formed in the first-handed cholesteric liquid crystal layer and the second-handed cholesteric liquid crystal layer. In the Bragg reflection band, the cholesteric liquid crystal can reflect circularly polarized light with the same hand direction as the liquid crystal layer, and give it a geometric phase related to the surface orientation structure of the cholesteric liquid crystal layer, while transmitting circularly polarized light with opposite hand directions. Therefore, by means of the back and forth reflection of light between the cholesteric liquid crystal layer and the reflector, the liquid crystal image recognizer provided by the embodiment of the present invention can independently perform multiple geometric phase modulations on two types of circularly polarized incident light, thereby realizing the image recognition function of circular polarization multiplexing. In addition, cholesteric liquid crystal also has wavelength selectivity, and only has reflection and geometric phase modulation effects on light with wavelengths within the Bragg reflection band, and has no phase modulation effect on light outside the band. The liquid crystal image recognizer provided by the embodiment of the present invention can further realize multifunctional image recognition with wavelength multiplexing by cascading multiple liquid crystal boxes with different central wavelengths of the Bragg reflection bands, effectively improving the limitations of traditional all-optical image recognizers such as non-adjustable functions, single applicable wavelength, limited efficiency, and difficulty in integration.

[0096] Figure 7 A schematic diagram of the structure of an image recognition device provided by an embodiment of the present invention. Figure 7 The image recognition device comprises a laser 1, a first polarizer 2, a first lens 3, a second lens 4, a first quarter wave plate 5, a beam splitter 6, a spatial light modulator 7, a second quarter wave plate 8, a second polarizer 9, a third lens 10, a fourth lens 11, a third quarter wave plate 12, a liquid crystal image recognizer 13 provided in the above embodiment and a detector 14; the light beam output by the laser 1 passes through the beam expansion system composed of the first polarizer 2, the first lens 3 and the second lens 4, the first quarter wave plate 5 and the beam splitter 6 in sequence, is modulated and reflected by the spatial light modulator 7, is reflected by the beam splitter 6, and passes through the second The 4f imaging system composed of the quarter wave plate 8, the second polarizer 9, the third lens 10 and the fourth lens 11 and the third quarter wave plate 12 obtains the circularly polarized incident light of the image to be identified, which is incident from the first substrate side of the liquid crystal image recognizer 13, and is reflected back and forth between the equally spaced orientation areas and the reflector. The reflected light that has undergone multiple geometric phase modulations is incident to the detector 14 after diffracting a preset distance. The detector 14 classifies the image to be identified according to the cumulative light intensity in each detection area; the third quarter wave plate 12 is rotated to change the circular polarization chirality of the incident light to switch different functions of the liquid crystal image recognizer 13.

[0097] Among them, the first lens 3 and the second lens 4 are used as a laser beam expansion system to generate an incident light beam that matches the format of the spatial light modulator 7. The first polarizer 2, the first quarter wave plate 5, the second quarter wave plate 8, and the second polarizer 9 are used to change the polarization state of light. Combined with the phase modulation type spatial light modulator, spatial light intensity modulation can be achieved to generate arbitrary light intensity images. A first plane mirror 15 and a second plane mirror 16 are also provided in the optical path to change the direction of the light beam. The third lens 10 and the fourth lens 11 are used as a 4f imaging system to image the generated arbitrary light intensity image on the input surface of the liquid crystal image recognizer 13. The third quarter wave plate 12 is used to change the polarization state of the input light beam, and is set at an angle of ±45° with the second polarizer 9. By switching the circular polarization chirality of the input light beam, the different functions of the liquid crystal image recognizer 13 can be switched. In this embodiment, it is schematically shown that the liquid crystal image recognizer includes three cascaded liquid crystal boxes, and the three cascaded liquid crystal boxes are used to achieve multifunctional image recognition at different wavelengths. For example, the spacing between the centers of the devices is set to 0.5 cm. The reflector in the liquid crystal image recognizer is arranged parallel to and opposite to the cascaded liquid crystal cells, so as to make the light beam reflect back and forth between the cholesteric liquid crystal layer and the reflector. It can be understood that Figure 7 The relative position relationship of each optical path element is only shown for example rather than the actual size and proportion. In other embodiments, the optical elements can be designed according to actual needs. For example, the lens can be replaced by a lens group, and the reflector can be increased or decreased.

[0098] For example, in this embodiment, a supercontinuum laser with continuously adjustable output wavelength is selected. According to actual needs, 485nm, 560nm, and 650nm lasers are output, and the incident angle of the laser to the liquid crystal image recognizer 13 can be 5° to 15°.

[0099] Exemplarily, in this embodiment, a CMOS camera is selected as the detector of the output surface. According to the preset detection area, the cumulative light intensity in each detection area can be calculated to classify the image to be identified.

[0100] For example, Figure 8 A schematic diagram of the detection area light spot and recognition result of the liquid crystal image recognizer provided in an embodiment of the present invention for multi-wavelength recognition of two types of images, Fig. 9 A schematic diagram of the confusion matrix results of the liquid crystal image recognizer provided by the embodiment of the present invention performing multi-wavelength recognition on all test set images in two data sets. Figure 8 When a handwritten digital image in the MNIST data set is input and the incident light is set to a left-handed circular polarization state (the arrow in the lower right corner of the figure shows the polarization direction), the three cascaded liquid crystal cells in the liquid crystal image recognizer in this embodiment can accurately recognize the input handwritten digital image at three wavelengths of 485nm, 560nm, and 650nm. According to the cumulative light intensity in each detection area, it can be seen that the cumulative light intensity in the detection area corresponding to category 4 is the strongest at the three wavelengths, so the handwritten digital recognition results at the three wavelengths are all category 4 (all correct); when a fashion image in the Fashion-MNIST data set is input and the incident light is set to a right-handed circular polarization state, the three cascaded liquid crystal cells in the liquid crystal image recognizer in this embodiment can accurately recognize the input fashion image at three wavelengths of 485nm, 560nm, and 650nm. According to the cumulative light intensity in each detection area, it can be seen that the cumulative light intensity in the detection area corresponding to category 1 is the strongest at the three wavelengths, so the fashion recognition results at the three wavelengths are all category 1 (all correct).

[0101] refer to Fig. 9, at three wavelengths of 485nm, 560nm, and 650nm, the liquid crystal image recognizer in this embodiment is used to recognize all test set images in the MNIST data set and the Fashion-MNIST data set in turn, and the confusion matrix between the predicted label and the real label and the recognition accuracy of the two types of images at three wavelengths are obtained. For the MNIST test set (the first test set), the matching of the predicted label and the real label is extremely high, and the recognition accuracy at the three wavelengths of 485nm, 560nm, and 650nm is 91.09%, 91.11%, and 91.18% respectively; for the Fashion-MNIST test set (the second test set), the matching of the predicted label and the real label is generally high, and the label matching of category 2 and category 6 is relatively poor, and the recognition accuracy at the three wavelengths of 485nm, 560nm, and 650nm is 82.29%, 82.42%, and 82.61% respectively.

[0102] It should be noted that the embodiment of the present invention only exemplarily shows the dual-function image recognition effect at three wavelengths of 485nm, 560nm, and 650nm, and is not a limitation on the liquid crystal image recognizer provided by the present invention. In other embodiments, according to actual needs, the pitch or orientation structure of the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer can be adjusted to make the liquid crystal image recognizer provided by the present invention suitable for different wavelengths or have different image recognition functions.

[0103] The technical solution of the embodiment of the present invention can independently perform multiple geometric phase modulations on two chiral circularly polarized incident lights by setting a first and a second cholesteric liquid crystal layer with opposite rotation directions and the same central wavelength of the Bragg reflection band, and forming a plurality of equally spaced orientation structures in the first and second cholesteric liquid crystal layers. In addition, the cholesteric liquid crystal also has wavelength selectivity, and only has reflection and geometric phase modulation effects on light with a wavelength within the Bragg reflection band, and is suitable for constructing multi-wavelength optical devices. The liquid crystal image recognizer provided in this embodiment can realize the image recognition function of circular polarization multiplexing, and can further realize multifunctional image recognition with wavelength multiplexing by cascading multiple devices with different central wavelengths of the Bragg reflection bands, which effectively improves the limitations of traditional all-optical image recognizers such as non-adjustable functions, single applicable wavelength, limited efficiency, and difficulty in integration.

[0104] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A liquid crystal image recognizer, characterized in that: including a liquid crystal cell and a reflector; The liquid crystal box comprises: A first substrate and a second substrate arranged opposite to each other; A first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer are stacked between the first substrate and the second substrate, a first orientation layer is disposed on a side of the first substrate close to the first chiral cholesteric liquid crystal layer, and a second orientation layer is disposed on a side of the second substrate close to the second chiral cholesteric liquid crystal layer; The first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have opposite chiral directions, and the first orientation layer and the second orientation layer are both provided with a plurality of orientation regions distributed at equal intervals, and within the orientation regions, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have different orientation structures, and the orientation structures are trained as weight parameters of a multi-layer diffractive neural network and are obtained through iterative optimization of a deep learning algorithm; The reflector is located on a side of the first substrate away from the first alignment layer, and a preset distance is set between the reflector and the first substrate to form a beam folding structure between the reflector and the liquid crystal box.

2. The liquid crystal image recognition device according to claim 1, characterized in that: The liquid crystal image recognizer comprises at least two liquid crystal cells arranged in cascade, and the central wavelengths of the Bragg reflection bands of at least two liquid crystal cells are different, so as to form a wavelength multiplexing multifunctional liquid crystal image recognizer.

3. The liquid crystal image recognition device according to claim 1, characterized in that: The first alignment layer and the second alignment layer are provided with a plurality of alignment regions of the same size, and along the direction from the first substrate to the second substrate, the corresponding alignment regions in the first alignment layer and the second alignment layer are aligned; For the first alignment layer or the second alignment layer, the alignment regions are distributed at equal intervals along a direction from the first substrate to the second substrate, and different alignment regions do not overlap.

4. The liquid crystal image recognition device according to claim 1, characterized in that: In different alignment regions, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer are provided with different alignment structures, and the total number of the alignment structures is equal to twice the number of the alignment regions; Outside the alignment region, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have a disordered surface alignment; The orientation structures are all optimized by a deep learning algorithm, which is based on a multi-layer diffraction neural network architecture, in which the input layer is an image to be identified, the output layer is the light intensity in multiple preset detection areas, the weight of the hidden layer is set to the phase, the number of hidden layers is equal to the number of times the light beam returns to pass through the orientation area, and the layers are connected by an angular spectrum diffraction formula; according to the geometric phase principle of cholesteric liquid crystal, for a cholesteric liquid crystal layer with a left-handed spiral structure, the orientation structure is equal to -0.5 times the weight of the corresponding hidden layer after iteration, and for a cholesteric liquid crystal layer with a right-handed spiral structure, the orientation structure is equal to +0.5 times the weight of the corresponding hidden layer after iteration.

5. The liquid crystal image recognition device according to claim 1, characterized in that: The first handed cholesteric liquid crystal layer is a left-handed spiral structure, and the second handed cholesteric liquid crystal layer is a right-handed spiral structure, or the first handed cholesteric liquid crystal layer is a right-handed spiral structure, and the second handed cholesteric liquid crystal layer is a left-handed spiral structure; In the same liquid crystal cell, the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer have circularly polarized selective Bragg reflection bands with the same central wavelength.

6. The liquid crystal image recognition device according to claim 1, characterized in that: The liquid crystal cell further includes a spacer disposed between the first substrate and the second substrate to control the total thickness of the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer; The extension length of the spacer is greater than or equal to 20 times the helical pitch of the liquid crystal molecules in the first chiral cholesteric liquid crystal layer.

7. A method for preparing a liquid crystal image recognizer, characterized in that: Used to prepare the liquid crystal image recognizer according to any one of claims 1 to 6, the preparation method comprising: Providing a first substrate and a second substrate, forming a first alignment layer on one side of the first substrate, and forming a second alignment layer on one side of the second substrate; The first substrate and the second substrate are arranged opposite to each other, and a first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer having a plurality of equidistantly distributed alignment structures are prepared between the first substrate and the second substrate to form a liquid crystal box; A reflector is arranged on a side of the first substrate of the liquid crystal box away from the first alignment layer to form the liquid crystal image recognizer; Among them, the first orientation layer is located on the side of the first substrate close to the first chiral cholesteric liquid crystal layer, and the second orientation layer is located on the side of the second substrate close to the second chiral cholesteric liquid crystal layer. The orientation structures of the first chiral cholesteric liquid crystal layer and the second chiral cholesteric liquid crystal layer are different, and are both trained as weight parameters of a multi-layer diffraction neural network and obtained through iterative optimization of a deep learning algorithm.

8. The method for preparing a liquid crystal image recognizer according to claim 7, characterized in that: Forming a first alignment layer on one side of the first substrate and forming a second alignment layer on one side of the second substrate, comprising: Spin coating a solution containing an alignment material on one side of the first substrate, and after the spin coating is completed, annealing the first substrate to form the first alignment layer; A uniform mixed solution containing an alignment material and a photoinitiator is spin-coated on one side of the second substrate. After the spin coating is completed, the second substrate is annealed to form the second alignment layer.

9. The method for preparing a liquid crystal image recognizer according to claim 7, characterized in that: A first chiral cholesteric liquid crystal layer and a second chiral cholesteric liquid crystal layer having a plurality of equidistantly distributed alignment structures are prepared between the first substrate and the second substrate to form a liquid crystal cell, comprising: Performing linearly polarized ultraviolet exposure on an empty box consisting of the first substrate, the second substrate, the first alignment layer and the second alignment layer, and sequentially exposing the alignment structure corresponding to the second chiral cholesteric liquid crystal layer at equal intervals; Filling a mixture of second chiral cholesteric liquid crystal and polymer monomer into the space between the first substrate and the second substrate to form the second chiral cholesteric liquid crystal layer with the alignment structure; The formed second handedness cholesteric liquid crystal layer is subjected to surface-induced ultraviolet polymerization, and the molecules that have not undergone polymerization reaction are washed away with acetone, so that the second handedness cholesteric liquid crystal layer shrinks to form a second handedness polymer network support connected to the surface of the second substrate; Performing a second round of linearly polarized ultraviolet exposure to sequentially expose the alignment structure corresponding to the first chiral cholesteric liquid crystal layer at equal intervals, and rewriting the alignment directions of the first alignment layer and the second alignment layer to be consistent with the alignment structure of the second round of exposure; The first chiral cholesteric liquid crystal is filled into the space between the first substrate and the second chiral cholesteric liquid crystal layer to form the first chiral cholesteric liquid crystal layer with the alignment structure of the second round of exposure, and the second chiral cholesteric liquid crystal layer rebounds to form the liquid crystal box.

10. An image recognition device, characterized in that: The device comprises a laser, a first polarizer, a first lens, a second lens, a first quarter wave plate, a beam splitter, a spatial light modulator, a second quarter wave plate, a second polarizer, a third lens, a fourth lens, a third quarter wave plate, a liquid crystal image recognizer according to any one of claims 1 to 6, and a detector; The light beam output by the laser passes through the first polarizer, the beam expansion system composed of the first lens and the second lens, the first quarter wave plate and the beam splitter in sequence, and is modulated and reflected by the spatial light modulator. After being reflected by the beam splitter, it passes through the second quarter wave plate, the second polarizer, the 4f imaging system composed of the third lens and the fourth lens and the third quarter wave plate in sequence to obtain circularly polarized incident light of the image to be recognized, which is incident from one side of the first substrate of the liquid crystal image recognizer, and is reflected back and forth between the equally spaced orientation areas and the reflector. The reflected light after multiple geometric phase modulations is incident on the detector after diffracting a preset distance, and the detector classifies the image to be recognized according to the cumulative light intensity in each detection area; Rotating the third quarter wave plate changes the circular polarization chirality of the incident light to switch different functions of the liquid crystal image recognizer.

Images