CN118072097A - Correlation imaging target recognition method and device based on all-optical neural network - Google Patents

Abstract

The invention discloses a correlation imaging target identification method and device based on an all-optical neural network, wherein the target identification method comprises the steps of using a plurality of transmission or reflection layers to physically create the all-optical depth diffraction neural network; training and verifying the created full-light diffraction depth neural network, and determining a target full-light diffraction depth neural network structure; performing physical processing on a target all-optical diffraction depth neural network structure, and constructing an associated imaging light path system; extracting a barrel detector signal in an associated imaging optical path system; target identification is accomplished using the bucket detector signal. The invention carries out the associated imaging target recognition based on the full-optical diffraction depth neural network, solves the problem that the recognition speed of the target object in the traditional associated imaging field is limited due to the dependence on hardware, builds an associated imaging optical path system by combining the full-optical diffraction depth neural network which is physically created, and can realize the recognition classification task of the target object under the light speed by directly processing the light intensity information detected by the barrel detection array.

Description

Correlation imaging target recognition method and device based on all-optical neural network

Technical Field

The present invention belongs to the technical field of optical imaging, and in particular relates to a method and device for associative imaging target recognition based on an all-optical neural network.

Background technique

Optical neural network is a new type of artificial neural network formed by using optical technology, such as optical connection technology and optical device technology, with the ability of ultra-parallel processing and light-speed information transmission. It uses photons as physical carriers to construct the basic computing unit in the artificial neural network algorithm, thereby realizing a high-performance new computing architecture. It can collect incident light and outgoing light through sensors, and reverse the relevant parameters in the formula describing its physical characteristics from the input and output, so as to adjust the parameters and use the physical properties of light to complete some complex calculations, and can perform optical signal calculations without additional energy input. In recent years, with in-depth research, the all-optical neural network framework based on physical mechanisms has gradually realized the execution of various complex tasks: all-optical image analysis, feature detection, object classification, etc. The diffraction network in the all-optical neural network uses the diffraction of light through continuous transmission layers to calculate a given task. Each diffraction layer is usually composed of tens of thousands of diffraction neurons, which modulate the phase and amplitude of the incident light singly or simultaneously in a compound manner, and use deep learning tools such as error back propagation to optimize the modulation value of each layer, and map the complex input field containing the optical information to be processed to the desired output field, which has the advantages of high speed, parallelism and low power consumption.

Correlated imaging, also known as ghost imaging or quantum imaging, is an active illumination staring imaging method and a new imaging technology. Compared with traditional imaging technology that uses light field intensity information, correlated imaging uses the fluctuation characteristics of light field intensity for imaging, which makes it have the characteristics of lens-free imaging and strong anti-disturbance. The classic correlated imaging system contains two light paths. One beam is irradiated on the target, which is the signal light path. The total light intensity information of the reflection or transmission is collected and recorded by the bucket detector without spatial resolution. The other beam does not pass through the target, which is the reference beam. After freely propagating the same distance in space, it is received by the CCD with spatial resolution. Finally, the information received by the signal light path and the reference light path is correlated and calculated to restore the target image. It has the characteristics of object-image separation, breaking the diffraction limit, and strong anti-interference ability. It has important application prospects in military defense, medical imaging, remote sensing and other fields. However, the general CCD detection rate is about hundreds of hertz, which limits the sampling rate of correlated imaging, and the speed of target object recognition is limited by hardware.

Summary of the invention

The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a method and device for associative imaging target recognition based on an all-optical neural network, combined with a new all-optical diffraction deep neural network, which can realize the recognition and classification tasks of target objects at the speed of light by directly processing the light intensity information detected by the bucket detection array.

In order to achieve the above object, the present invention has the following technical solutions:

In a first aspect, a method for associative imaging target recognition based on an all-optical neural network is provided, comprising:

Using multiple transmissive or reflective layers to physically create all-optical deep diffractive neural networks;

Train and verify the created all-optical diffraction deep neural network to determine the target all-optical diffraction deep neural network structure;

Physically process the target all-optical diffraction deep neural network structure and build a related imaging optical path system;

Extracting the bucket detector signal in the correlation imaging optical path system;

Target identification is accomplished using the bucket detector signal.

As a preferred solution, in the step of using multiple transmission or reflection layers to physically create an all-optical deep diffraction neural network, each point on each layer in the transmission or reflection layer is set to transmit or reflect the incoming light, each point represents a neuron, which is connected to other neurons in the next layer through optical diffraction, and the transmission or reflection coefficient of each neuron is used as a multiplicative bias term. The physical mechanism of light propagation is used to simulate the weights and biases in the neural network training process, and then iteratively adjusted through the error back propagation method.

As a preferred solution, the steps of training and verifying the created all-optical diffraction deep neural network include: performing deep learning on the created all-optical diffraction deep neural network, inputting training data into the input layer, iteratively adjusting the phase values of neurons in each layer of the diffraction network through the output of the all-optical network to perform the recognition and classification problem of the target object, and respectively training and verifying the all-optical diffraction deep neural network with a training set and a test set of the target data set to be classified.

As a preferred solution, after the created all-optical diffraction deep neural network is trained and verified, the target all-optical diffraction deep neural network structure design is fixed and the phase values of neurons in each layer are determined. At this time, for the coherent diffraction network that performs pure phase modulation on the input light, each layer is approximated as an optical element, thereby materializing the target all-optical diffraction deep neural network structure through physical processing.

As a preferred solution, the associated imaging optical path system includes a light source, a target full-optical diffraction deep neural network structure, an object to be identified, a lens and a bucket detector array in sequence according to the direction of light propagation. After the light is emitted by the light source, it passes through the target full-optical diffraction deep neural network structure, the object to be identified and the lens in sequence, and is finally received in the bucket detector array containing multiple bucket detectors to obtain multiple bucket detector signals.

As a preferred solution, in the step of extracting the bucket detector signal in the associated imaging optical path system, the bucket detector array contains 20 bucket detectors, and the light intensity signals received by the 20 bucket detectors form a light intensity signal data set s=[s ₀ ,s ₁ ,…s ₁₉ ].

As a preferred solution, the step of using the bucket detector signal to complete target recognition includes: performing differential processing and analysis on the light intensity signal data in the light intensity signal data set s = [s ₀ , s ₁ , ... s ₁₉ ] to obtain differential signal values, and then processing the differential signal values with a SoftMax function to find the maximum value after processing, which is the classification category of the object to be recognized;

The difference calculation expression is as follows:

In the formula, i=0,1,…,9.

In a second aspect, a correlation imaging target recognition system based on an all-optical neural network is provided, comprising:

A neural network physical creation module for physically creating a full-optical deep diffractive neural network using multiple transmission or reflection layers;

A network structure training module is used to train and verify the created all-optical diffraction deep neural network and determine the target all-optical diffraction deep neural network structure;

The module for building a correlation imaging optical path system is used to physically process the target all-optical diffraction deep neural network structure and build a correlation imaging optical path system;

A bucket detector signal extraction module, used for extracting the bucket detector signal in the associated imaging optical path system;

The bucket detector signal processing module is used to complete target recognition using the bucket detector signal.

According to a third aspect, an electronic device is provided, including:

A memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the associative imaging target recognition method based on the all-optical neural network.

In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the associated imaging target recognition method based on the all-optical neural network is implemented.

Compared with the prior art, the present invention has at least the following beneficial effects:

Based on the all-optical diffraction deep neural network ( ^D2NN ), the correlation imaging target recognition is carried out, which solves the problem that the target object recognition speed in the traditional correlation imaging field is limited by hardware. The all-optical deep diffraction neural network created physically is combined to build a correlation imaging optical path system. It only needs to extract the barrel detector signal in the correlation imaging optical path system, and the target recognition can be completed by processing the barrel detector signal. The method of the present invention provides a new idea for the object image recognition and classification in the correlation imaging field. After setting the network parameters, the pure phase-modulated all-optical diffraction deep neural network is trained and verified, the target all-optical diffraction deep neural network structure is determined, and it is physically processed by 3D printing or lithography. After being materialized, it is placed in the correlation imaging optical path system. By directly processing the light intensity information detected by the barrel detection array, the recognition and classification task of the target object can be realized at the speed of light. The experimental results show that the recognition and classification accuracy of the target object using the method of the present invention can reach 90.89%. This method maps the complex-valued input field containing the optical information to be processed to the desired output field. It has the advantages of high speed, parallelism and low power consumption. It makes up for the deficiency that the speed of target object recognition in the field of associative imaging is limited by hardware, and provides a new idea and a new method for the recognition and classification of target objects in the field of associative imaging.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

FIG1 is a flow chart of a method for associative imaging target recognition based on an all-optical neural network according to an embodiment of the present invention.

FIG. 2 is a schematic diagram showing the working principle of correlation imaging.

FIG. 3 is a schematic diagram of the working principle of the correlation imaging optical path system constructed in an embodiment of the present invention.

FIG4 is a schematic diagram of a pure phase-modulated diffraction deep neural network trained with a target data set to be classified and identified;

FIG. 5( a ) is a visualization diagram of a handwritten number “2” input and a corresponding output signal according to an embodiment of the present invention.

FIG5( b ) is a visualization diagram of the handwritten number “3” input and the corresponding output signal according to an embodiment of the present invention.

FIG5( c ) is a visualization diagram of the handwritten number “9” input and the corresponding output signal according to an embodiment of the present invention.

FIG6( a ) is a curve diagram showing the change of the Loss value of the target data set to be classified and identified during training and verification when Layer number=9 and Epoch=30 according to an embodiment of the present invention.

FIG6( b ) is a curve diagram showing the variation of the Accuracy value of the target data set to be classified and identified during training and verification when Layer number=9 and Epoch=30 according to an embodiment of the present invention.

FIG6( c ) is a curve diagram showing the change of the Loss value of the target data set to be classified and identified during training and verification when Layer number=5 and Epoch=50 according to an embodiment of the present invention.

FIG6( d ) is a curve diagram showing changes in the Accuracy value of the target data set to be classified and identified during training and verification when Layer number=5 and Epoch=50 according to an embodiment of the present invention.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, ordinary technicians in this field can also obtain other embodiments without making creative work.

In the existing correlation imaging field, the general CCD detection rate is about hundreds of hertz, which limits the sampling rate of correlation imaging. However, by using a phase, amplitude single or compound modulation device to modulate coherent light as a light source, the distribution of coherent light after the phase or amplitude modulation device is known, and the speckle field formed by diffraction can be calculated by the light field propagation function. Therefore, the reference optical path can be omitted in the correlation imaging system, and the traditional correlation imaging system can be transformed into a computational correlation imaging system with only a signal optical path. The speed of computational correlation imaging is greatly improved due to the diverse and controllable intensity distribution, field of view and lateral coherence length of the speckle field. With the in-depth study of computational correlation imaging, the emergence of neural networks has further improved the imaging speed and imaging quality of computational correlation imaging systems, which provides a new research perspective for computational correlation imaging.

Referring to FIG. 1 , the associated imaging target recognition method based on the all-optical neural network according to the embodiment of the present invention mainly includes:

(1) Build an all-optical diffraction deep neural network;

(2) Train and validate the all-optical diffraction deep neural network;

(3) 3D print the trained mask template and build the optical path;

(4) extracting bucket detector signals in the bucket detector array;

(5) Differential processing of the bucket detector signals in the bucket detector array is performed to complete target recognition.

In a possible implementation, in step (1) when building an all-optical diffraction deep neural network:

First, an all-optical deep diffraction neural network is physically created by using multiple transmission or reflection layers. Each point on each layer of the transmission or reflection layer is set to either transmit or reflect the incoming light. Each point represents an artificial neuron, which is connected to other neurons in the next layer through optical diffraction. Similar to a standard deep neural network, the transmission or reflection coefficient of each point or neuron can be regarded as a multiplicative bias term, that is, the physical mechanism of light propagation is used to simulate the weights and biases in the neural network training process, and then iteratively adjusted through the error back propagation method.

In one possible implementation, in step (2), when training and verifying the all-optical diffraction deep neural network:

The all-optical diffraction deep neural network uses deep learning to input training data into the input layer, and iteratively adjusts the phase values of neurons in each layer of the diffraction network through the output of the all-optical network to perform the recognition and classification of target objects. The neural network is trained and verified using the training set and test set of the target data set to be classified.

In a possible implementation, when the trained mask is 3D printed and the optical path is constructed in step (3):

After the training and verification phases are completed, the network design is fixed and the phase values of the neurons in each layer are determined. At this time, for the coherent diffraction network that performs pure phase modulation on the input light, each layer can be approximated as a thin optical element. Therefore, the all-optical neural network can be physically processed using 3D printing or photolithography to perform the specific task it is trained for at the speed of light. In the present invention, the trained neural network with fixed parameters is materialized by 3D printing technology. After 3D printing is completed, the device is built to complete the experimental part. As shown in Figure 3, the associated imaging optical path system includes a light source 1, a target all-optical diffraction deep neural network structure 2, an object to be identified 3, a lens 4, and a barrel detector array 5 in sequence according to the direction of light propagation. After the light is emitted by the light source 1, it passes through the target all-optical diffraction deep neural network structure 2, the object to be identified 3 and the lens 4 in sequence. Finally, it is received in the barrel detector array 5 containing multiple barrel detectors 7 to obtain multiple barrel detector signals. At this time, the printed mask plate, that is, the target all-optical diffraction deep neural network structure 2, performs pure phase modulation on the input visible light. The light source 1 and the target all-optical diffraction deep neural network structure 2 can be roughly equivalent to a light source that can identify the target object.

In a possible implementation, when extracting the bucket detector signal in step (4):

The bucket detector array 5 of the embodiment of the present invention includes 20 bucket detectors 7 , and the light intensity signals received by the 20 bucket detectors 7 form a light intensity signal data set s=[s ₀ , s ₁ , . . . s ₁₉ ].

In a possible implementation, when step (5) performs differential processing on the barrel detector signals in the barrel detector array to complete target recognition: perform differential processing and analysis on the light intensity signal data set s = [s ₀ , s ₁ , ... s ₁₉ ] in pairs to obtain differential signal values, and then use the SoftMax function to process the differential signal values to find the maximum value after processing, which is the classification category of the object to be identified (3); the differential calculation expression is as follows:

In the formula, i=0,1,…,9.

FIG2 is a schematic diagram of the principle of correlation imaging. The light source 1 in FIG2 is a device for emitting light; the spatial modulator 6 is a device for phase modulation of light; the object to be identified 3 is the target to be identified; the lens 4 is to allow the light passing through the object to be better received by the barrel detector 7; the barrel detector 7 is a signal receiving device. The coherent light is modulated by a phase or amplitude modulation device as the light source 1 for correlation imaging. After being emitted by the light source 1, the light passes through the spatial light modulator 6, the object to be identified 3 and the lens 4 in sequence, and is finally received by the barrel detector 7. The modulated coherent light forms a speckle field in the far field through Fraunhofer diffraction. The distribution of the coherent light after the phase or amplitude modulation device is known. Therefore, the speckle field formed by diffraction can be calculated by the propagation function of the light field. Therefore, in this correlation imaging optical path system, the speckle generated by the spatial modulator 6 and the signal received by the barrel detector 7 are correlated and calculated, and the image of the object to be identified 3 can be restored.

Referring to FIG3 , the present invention first builds an all-optical diffraction deep neural network, simulates the neural network to perform pure phase modulation on light in the visible light band, sets each point on each layer to either transmit or reflect the incoming light, and each point represents an artificial neuron, which is connected to other neurons in the next layer through optical diffraction. Similar to the standard deep neural network, the transmission or reflection coefficient of each point or neuron can be regarded as a multiplicative bias term, that is, the physical mechanism of light propagation is used to simulate the weights and biases in the neural network training process, and then iteratively adjusted by the error back propagation method. The network is set to be a fully connected network of 8cm*8cm, each neuron size is 400μm*400μm, and the target data set to be classified and identified is used for training. The embodiment sets its batch size to 32, that is, based on the target data set to be classified and identified, 60,000 photos of the training set are used as a group of 32 photos for training 1875 times, and then 10,000 photos of the test set are used as a group of 32 photos for verification 313 times. Then set different neural network layers and training times, and by analyzing the changes in the accuracy and loss values of the test data set and the verification data set, try to find the training cycle number and neural network layer number that best suits the model. Through comparative analysis, the embodiment chooses to set a 5-layer diffraction neural network. Therefore, the number of neurons is [200*200]*5=200,000. Due to the characteristics of light propagation, a fully connected network is used. At this time, the number of connections is (200*200)2*5=8 billion. Finally, after training and verification, a 5-layer all-optical diffraction deep neural network for specific target object recognition and classification can be obtained, as shown in Figure 4. After the training and verification stages are completed, the network design is fixed and the phase values of each layer of neurons are determined. For a coherent diffraction network with pure phase modulation, each layer can be approximated as a thin optical element. Therefore, 3D printing technology is used for physical processing to materialize the trained neural network. After completing 3D printing, we build the optical path according to the experimental device diagram designed in Figure 3.

The light source 1 in Figure 3 is a device that emits light; the target full-light diffraction deep neural network structure 2 is a device that phase modulates light; the object to be identified 3 is the target to be identified; the lens 4 is to allow the light passing through the object to be identified 3 to be better received by the barrel detector array 5; the barrel detector array 5 is a signal receiving device. After being emitted by the light source 1, the light passes through the target full-light diffraction deep neural network structure 2, the object to be identified 3 and the lens 4 in sequence. The light intensity information is finally received in the detection array 5 containing 20 barrel detectors 7, and 20 barrel detector signals are obtained in sequence. At this time, the printed mask plate is the target full-light diffraction deep neural network structure 2. The input visible light is purely phase modulated by the target full-light diffraction deep neural network structure 2. The light source 1 and the mask plate can be roughly equivalent to a light source that can identify the object to be identified 3.

At the end of the experiment, the bucket detector signal is extracted, and the light intensity information received by the bucket detector array 5 is measured to obtain the signal value s = [s ₀ ,s ₁ ,…s ₁₉ ]. These 20 signal values are differentially processed and analyzed in pairs, and then the differential signal value is processed by the SoftMax function. According to the pre-trained network model, it can be known that the maximum value of the signal is the classification category of the target object, thereby realizing the recognition and classification of target objects based on correlation imaging under the all-optical neural network. The simulation before the experiment and the experiment itself proved the reasoning ability of the framework. This new method has become a new way to recognize and classify target objects in the field of correlation imaging.

See Figure 5(a), Figure 5(c), and Figure 5(d), which show the visualization of the input signal (handwritten digits "2", "3", "9") and the output signal (bucket detector array signal after differential and SoftMax function processing) in the experiment. The red box in the figure is the maximum value of the final light intensity information, that is, the classification category of the target classification. After the trained diffraction deep neural network is obtained and materialized, the bucket detector array signal value is measured experimentally. After differential and SoftMax function processing, the target object can be identified and classified from the maximum value of the processed bucket signal.

Figures 6(a), 6(b), 6(c), and 6(d) are curves showing the changes in the Loss value and the Accuracy value of the target data set to be classified and recognized during training and verification when the computer trains the all-optical diffraction deep neural network at Layer number = 9, Epoch = 30 and Layer number = 5, Epoch = 50, respectively. Layer number represents the number of layers of the all-optical neural network, Epoch represents the number of training times (epochs), Loss represents the loss value, and Accuracy represents the accuracy. From the changes in the Loss value shown in Figures 6(a) and 6(c), it can be seen that the training process is well-fitted, and there is no overfitting or underfitting. From the changes in the Accuracy value shown in Figures 6(b) and 6(d), it can be seen that the classification accuracy tends to be stable after a certain number of traversal learning, and the classification accuracy can reach 90.89% and 88.99% at this time.

Another embodiment of the present invention further provides a correlation imaging target recognition system based on an all-optical neural network, comprising:

A neural network physical creation module for physically creating a full-optical deep diffractive neural network using multiple transmission or reflection layers;

A network structure training module is used to train and verify the created all-optical diffraction deep neural network and determine the target all-optical diffraction deep neural network structure;

The module for building a correlation imaging optical path system is used to physically process the target all-optical diffraction deep neural network structure and build a correlation imaging optical path system;

A bucket detector signal extraction module, used for extracting the bucket detector signal in the associated imaging optical path system;

The bucket detector signal processing module is used to complete target recognition using the bucket detector signal.

Another embodiment of the present invention further proposes an electronic device, comprising: a memory storing at least one instruction; and a processor executing the instructions stored in the memory to implement the associative imaging target recognition method based on the all-optical neural network.

Another embodiment of the present invention further proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, an associated imaging target recognition method based on an all-optical neural network is implemented.

Exemplarily, the instructions stored in the memory may be divided into one or more modules/units, which are stored in a computer-readable storage medium and executed by the processor to complete the associative imaging target recognition method based on an all-optical neural network of the present invention. The one or more modules/units may be a series of computer-readable instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the server.

The electronic device may be a computing device such as a smart phone, a notebook, a PDA, and a cloud server. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the electronic device may also include more or fewer components, or a combination of certain components, or different components, for example, the electronic device may also include an input/output device, a network access device, a bus, etc.

The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

The memory may be an internal storage unit of the server, such as a hard disk or memory of the server. The memory may also be an external storage device of the server, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the server. Furthermore, the memory may include both an internal storage unit of the server and an external storage device. The memory is used to store the computer-readable instructions and other programs and data required by the server. The memory may also be used to temporarily store data that has been output or is to be output.

It should be noted that the information interaction, execution process and other contents between the above-mentioned module units are based on the same concept as the method embodiment. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera device/terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), electrical carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims (10)

1. A method for associative imaging target recognition based on an all-optical neural network, comprising: Using multiple transmissive or reflective layers to physically create all-optical deep diffractive neural networks; Train and verify the created all-optical diffraction deep neural network to determine the target all-optical diffraction deep neural network structure; Physically process the target all-optical diffraction deep neural network structure and build a related imaging optical path system; Extracting the bucket detector signal in the correlation imaging optical path system; Target identification is accomplished using the bucket detector signal. 2. According to claim 1, the associative imaging target recognition method based on the all-optical neural network is characterized in that, in the step of using multiple transmission or reflection layers to physically create an all-optical deep diffraction neural network, each point on each layer in the transmission or reflection layer is set to transmit or reflect the incoming light, each point represents a neuron, which is connected to other neurons in the next layer through optical diffraction, and the transmission or reflection coefficient of each neuron is used as a multiplicative bias term. The physical mechanism of light propagation is used to simulate the weights and biases in the neural network training process, and then iteratively adjusted through the error back propagation method. 3. According to the method for associative imaging target recognition based on all-optical neural network in claim 1, it is characterized in that the step of training and verifying the created all-optical diffraction deep neural network comprises: The created all-optical diffraction deep neural network is subjected to deep learning. The training data is input into the input layer. The phase values of the neurons in each layer of the diffraction network are iteratively adjusted through the output of the all-optical network to perform the recognition and classification of the target objects. The all-optical diffraction deep neural network is trained and verified using the training set and test set of the target data set to be classified. 4. According to claim 3, the associative imaging target recognition method based on the all-optical neural network is characterized in that after the created all-optical diffraction deep neural network is trained and verified, the target all-optical diffraction deep neural network structure design is fixed and the phase value of each layer of neurons is determined. At this time, for the coherent diffraction network that performs pure phase modulation on the input light, each layer is approximated as an optical element, thereby materializing the target all-optical diffraction deep neural network structure through physical processing. 5. According to claim 1, the associative imaging target recognition method based on the all-optical neural network is characterized in that the associative imaging optical path system includes a light source (1), a target all-optical diffraction deep neural network structure (2), an object to be identified (3), a lens (4) and a barrel detector array (5) in sequence according to the light propagation direction. After the light is emitted by the light source (1), it passes through the target all-optical diffraction deep neural network structure (2), the object to be identified (3) and the lens (4) in sequence, and is finally received in the barrel detector array (5) containing multiple barrel detectors (7) to obtain multiple barrel detector signals. 6. The associative imaging target recognition method based on an all-optical neural network according to claim 5 is characterized in that, in the step of extracting the bucket detector signal in the associative imaging optical path system, the bucket detector array (5) contains 20 bucket detectors (7), and the light intensity signals received by the 20 bucket detectors (7) constitute a light intensity signal data set s=[s ₀ ,s ₁ ,…s ₁₉ ]. 7. According to claim 6, the method for associative imaging target recognition based on an all-optical neural network is characterized in that the step of using the barrel detector signal to complete target recognition includes: performing differential processing and analysis on the light intensity signal data in the light intensity signal data set s = [s ₀ , s ₁ , ... s ₁₉ ] to obtain differential signal values, and then processing the differential signal values with a SoftMax function to find the maximum value after processing, which is the classification category of the object to be identified (3); the differential calculation expression is as follows: In the formula, i=0,1,…,9. 8. An associative imaging target recognition system based on an all-optical neural network, characterized by comprising: A neural network physical creation module for physically creating a full-optical deep diffractive neural network using multiple transmission or reflection layers; A network structure training module is used to train and verify the created all-optical diffraction deep neural network and determine the target all-optical diffraction deep neural network structure; The module for building a correlation imaging optical path system is used to physically process the target all-optical diffraction deep neural network structure and build a correlation imaging optical path system; A bucket detector signal extraction module, used for extracting the bucket detector signal in the associated imaging optical path system; The bucket detector signal processing module is used to complete target recognition using the bucket detector signal. 9. An electronic device, comprising: a memory storing at least one instruction; and A processor executes instructions stored in the memory to implement the associative imaging target recognition method based on the all-optical neural network as described in any one of claims 1 to 7. 10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the associated imaging target recognition method based on an all-optical neural network as described in any one of claims 1 to 7.