Single pixel imaging method and system

Abstract

The present invention relates to the field of optical detection imaging technology, and in particular to a single-pixel imaging method and system. The method includes obtaining a light signal to obtain a reference image and target light field information and using them as a label set and a signal set respectively, establishing a neural light field model and a feature extraction network; randomly sampling the corresponding data set into a training set, a test set and a verification set according to the same random seed function; training with a neural light field model, and outputting the model file when the minimum value of the loss function obtained after training is less than a set threshold; using a verification set to verify the trained model file, and obtaining an accuracy greater than a first threshold range is qualified, otherwise a new random seed function is used to repeat iterative model training and verification. The present invention has no complex optical-mechanical structure and spatial modulation method, and can use a single-pixel sensor to achieve high-precision imaging of an object. At the same time, a semi-supervised deep learning algorithm is used, which is conducive to improving the training speed and saving hardware resources.

Description

Technical Field

[0001] The present invention relates to the field of optical detection imaging technology, and in particular to a single-pixel imaging method and system. Background Art

[0002] Traditional optical imaging widely uses array pixel detectors (such as CCD and CMOS) based on silicon or indium gallium arsenide as photosensitive materials. The spectral range is concentrated in the visible and near-infrared bands (400-1700nm). For other bands (mid- and far-infrared, ultraviolet, terahertz, etc.), there are few reports of high-resolution and low-cost array detectors in the market.

[0003] In recent years, single-pixel detectors and imaging applications have developed rapidly. They have the advantages of high sensitivity, high-speed response, and low production cost compared to array-type detectors. This imaging technology usually uses spatial modulation, that is, global sampling is used to encode the spatial information of the object, and then the encoded information is collected in sequence by a single-pixel detector, and the object image is reconstructed through computational decoding. The commonly used technology is ghost imaging technology. This technology usually uses random patterns as spatial modulation patterns to achieve intensity modulation of the target, record the random pattern distribution and detect the corresponding return light intensity, and calculate and decode to achieve object image reconstruction. The light modulation modes include random speckle, Hadamard mode and Fourier mode, but this method has a large amount of data collection, is time-consuming, and has low imaging efficiency. Summary of the invention

[0004] The present invention aims to at least improve one of the technical problems existing in the prior art. To this end, the present invention proposes a single pixel imaging method and system.

[0005] A single pixel imaging method, comprising:

[0006] Step S1, obtaining light signals of multiple targets after being illuminated by a light source, and obtaining a reference image and target light field information according to the light signals;

[0007] Step S2, taking the reference image as a label set and the target light field information as a signal set, establishing a neural light field model and a feature extraction network according to the reference image and the target light field information, and performing prediction according to the input target light field information to obtain a predicted image, wherein the predicted image is:

[0008] ,

[0009] in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image, is the target light field information;

[0010] Step S3, according to the same random seed function, randomly sampling the corresponding data set consisting of the label set and the signal set in a first range ratio into a training set, a test set and a validation set;

[0011] Step S4, training the neural light field model, the loss function is as follows:

[0012] ,

[0013] in, is the number of training sets, Z is the total number of training sets, is the target light field information of the i-th training set, is the pixel matrix of the reference image of the i-th training set;

[0014] When the minimum value of the loss function obtained after training the training set is less than the set threshold, the output is used as the model file;

[0015] Step S5, use the verification set to verify the trained model file, if the accuracy is greater than the first threshold range, it is qualified, otherwise use a new random seed function to repeat steps S3 to S4 for iterative model training and verification until the accuracy is greater than the first threshold range.

[0016] In a possible technical solution, further, in step S1, the reference image is discretized into a P×Q pixel matrix, defined as ,

[0017] Target light field information It is expressed as follows:

[0018] ,

[0019] Among them, x and y are the corresponding coordinates of the light spot occupied by the light reflected or transmitted by the object through the optical component and converged on the photosensitive film of the single-pixel detector; u1, u2, ..., u M are different bias voltage values; V1, V2, V3, ..., V N is the signal strength detected by different electrodes on a single pixel sensor.

[0020] In a possible technical solution, further, in step S1, the acquired optical signal is converted into a voltage signal, which can be expressed as:

[0021] ,

[0022] Among them, x and y are the corresponding coordinates of the area where the light reflected or transmitted by the object is concentrated on the photosensitive film of the single-pixel detector through the optical component, and the area where the light spot occupies is ; and is the electrode position; is the Seebeck coefficient of the hole; γ is the wavelength of the incident light The divergence relationship correlation factor of; u is the bias voltage; is the voltage temperature drift rate; T is the film temperature; is the bias voltage at room temperature 25°C; is the average incident light intensity; is the reflectivity of the film; d is the thickness of the film; is the film absorption rate; is the beam spot radius.

[0023] In a possible technical solution, further, in step S2, the neural light field model establishes a relationship between the light spot distribution characteristics displayed by each voltage intensity distribution and the focal length f of the optical component.

[0024] ,

[0025] Where n is the encoding focal length encoding dimension, Decoded values ​​for different positions;

[0026] The encoded focal length of the spot distribution feature is then input into a small multilayer perceptron with only 3 hidden layers and 256 neurons in each hidden layer.

[0027] In a possible technical solution, further, in step S2, the feature extraction network adopts a multilayer perceptron, the multilayer perceptron includes 12 hidden layers, 1024 neurons and is activated by a rectified linear unit, and the feature dimension is set to 302.

[0028] A single-pixel imaging system, used to implement the above-mentioned single-pixel imaging method, comprising:

[0029] A light source, used for emitting a broadband light source;

[0030] An optical component for converging and splitting the reflected light or transmitted light irradiating the target object;

[0031] An acquisition module, used to acquire light signals of multiple targets after being illuminated by a light source, and obtain target light field information according to the light signals;

[0032] A reference unit, configured to obtain a reference image according to the optical signal;

[0033] The module is used to establish a neural light field model and a feature extraction network based on the reference image and the target light field information, and to predict the input target light field information to obtain a predicted image, wherein the predicted image is:

[0034] ,

[0035] in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image, is the target light field information;

[0036] A sorting module is used to randomly sample the corresponding data set according to the same random seed function and divide it into a training set, a test set and a validation set in a first range ratio;

[0037] The training module is used to train the neural light field model. The loss function is as follows:

[0038] ,

[0039] in, is the number of training sets, Z is the total number of training sets, is the target light field information of the i-th training set, is the pixel matrix of the reference image of the i-th training set;

[0040] When the minimum value of the loss function obtained after training the training set is less than the set threshold, the output is used as the model file;

[0041] A verification module is used to verify the trained model file. If the accuracy rate is greater than the first threshold range, the model file is qualified. Otherwise, a new random seed function is used to repeat steps S3 to S4 to iteratively train and verify the model until the accuracy rate is greater than the first threshold range.

[0042] The image generating unit generates an image of a target object.

[0043] In a possible technical solution, further, the acquisition module includes a single-pixel detector for detecting converged light and obtaining information of the target light field, wherein the single-pixel detector is a single photosensitive film with a thermoelectric effect, has N electrodes for receiving electrical signals, N>3, and is modulated by a bias voltage with a preset step value, and the modulation amount is M.

[0044] In a possible technical solution, further, the target light field information includes light field information generated by the target object based on the converged light, and the light field information is an M×N matrix composed of M modulation quantities and N electrodes receiving electrical signals.

[0045] In a possible technical solution, further, the optical component includes one or more optical elements for converging and splitting the reflected light or transmitted light irradiating the target object, and the field of view angle and aperture of the lens group are designed according to the illumination mode of the object and the pixel size of the single-pixel detector.

[0046] In a possible technical solution, further, the image generating unit includes:

[0047] The multi-electrode signal amplification circuit is used to detect the signals of multiple electrodes at the same time and amplify them uniformly with a certain amplification resistor;

[0048] The signal acquisition and processing circuit is used to acquire the signal after amplification of each electrode, and the circuit processes the acquired signal to obtain an image of the object.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention does not require complex optomechanical structures and spatial modulation methods, and can use a single-pixel sensor to achieve high-precision imaging of objects. At the same time, it uses a semi-supervised deep learning algorithm, which is beneficial to improving training speed and saving hardware resources.

[0051] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. 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 paying creative work.

[0053] Figure 1 is a flow chart of a single pixel imaging method according to an embodiment of the present invention;

[0054] Figure 2 is a schematic diagram of a transmissive single-pixel imaging system according to an embodiment of the present invention;

[0055] Figure 3 is an example diagram of single-pixel imaging of a digital image according to an embodiment of the present invention;

[0056] Figure 4Schematic diagram of a reflective single-pixel imaging system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The embodiments of the present invention are described in detail below. The embodiments described with reference to the accompanying drawings are exemplary. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0058] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element at the same time.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0060] The terms "first", "second", "third", etc. in the specification and claims of the present application and the drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a series of steps or units are included, or optionally, steps or units not listed are included, or optionally, other steps or units inherent to these processes, methods, products or devices are included.

[0061] Only the part relevant to the present application is shown in the accompanying drawings, but not all of the content. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processing or methods depicted as flow charts. Although the flow chart describes each operation (or step) as a sequential process, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of each operation can be rearranged. When its operation is completed, the process can be terminated, but it can also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0062] The terms "component", "module", "system", "unit", etc. used in this specification are used to represent computer-related entities, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a unit can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or distributed between two or more computers. In addition, these units can be executed from various computer-readable media having various data structures stored thereon. Units can communicate through local and / or remote processes, for example, based on signals having one or more data packets (e.g., data from a second unit interacting with another unit in a local system, a distributed system, and / or a network. For example, the Internet interacts with other systems via signals).

[0063] In some embodiments of this application, please refer to Figure 1 , is a flow chart of a single pixel imaging method proposed in a possible embodiment of the present application, wherein the proposed method comprises:

[0064] Step S1, obtaining light signals of multiple targets after being illuminated by a light source, and obtaining a reference image and target light field information according to the light signals;

[0065] Step S2, taking the reference image as a label set and the target light field information as a signal set, establishing a neural light field model and a feature extraction network according to the reference image and the target light field information, and performing prediction according to the input target light field information to obtain a predicted image, wherein the predicted image is:

[0066] ,

[0067] in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image, is the target light field information;

[0068] Step S3, according to the same random seed function, the corresponding data set consisting of the label set and the signal set is randomly sampled in a ratio of 4:3:3 and divided into a training set, a test set and a validation set;

[0069] Step S4: training with the neural light field model, the loss function is as follows:

[0070] ,

[0071] in, is the number of training sets, Z is the total number of training sets, is the target light field information of the i-th training set, is the pixel matrix of the reference image of the i-th training set;

[0072] After training the training set, the minimum value of the loss function is less than 10 -4 When , the output is used as a model file;

[0073] Step S5, use the verification set to verify the trained model file, if the accuracy is greater than the first threshold range, it is qualified, otherwise use a new random seed function to repeat steps S3 to S4 for iterative model training and verification until the accuracy is greater than the first threshold range.

[0074] It should be noted that, in this embodiment, the first threshold range in step S5 is 98.6%.

[0075] It should be noted that after iterative training using the test set, 100,000 iterations were performed as a cycle, and the batch size was 1000. The learning rate parameter was set to 0.001, and the learning rate was multiplied by 0.8 every 10,000 iterations; the Adam (Adaptive Moment Estimation) optimizer was used, and the beta parameter was set to 0.9 and 0.999.

[0076] In some embodiments of the present application, in step S1, the reference image is discretized into a P×Q pixel matrix, defined as ,

[0077] Target light field information It is expressed as follows:

[0078] ,

[0079] Among them, x and y are the corresponding coordinates of the light spot occupied by the light reflected or transmitted by the object through the optical component and converged on the photosensitive film of the single-pixel detector; u1, u2, ..., u M are different bias voltage values; V1, V2, V3, ..., V N is the signal strength detected by different electrodes on a single pixel sensor.

[0080] Furthermore, the acquired optical signal is converted into a voltage signal, which can be expressed as:

[0081] ,

[0082] Among them, x and y are the corresponding coordinates of the area where the light reflected or transmitted by the object is concentrated on the photosensitive film of the single-pixel detector through the optical component, and the area where the light spot occupies is ; and is the electrode position; is the Seebeck coefficient of the hole; γ is the wavelength of the incident light The divergence relationship correlation factor of; u is the bias voltage; is the voltage temperature drift rate; T is the film temperature; is the bias voltage at room temperature 25°C; is the average incident light intensity; is the reflectivity of the film; d is the thickness of the film; is the film absorption rate; is the beam spot radius.

[0083] In this embodiment, the photocurrent or resistance change generated by the photosensitive film can be converted into a voltage signal.

[0084] It should be noted that in step S2, the neural light field model shows the light spot distribution characteristics according to each voltage intensity distribution and the relationship with the focal length f of the optical component to establish

[0085] ,

[0086] Where n is the encoding focal length encoding dimension, Decoded values ​​for different positions;

[0087] The encoded focal length of the spot distribution feature is then input into a small multi-layer perceptron with only 3 hidden layers and 256 neurons in each hidden layer.

[0088] It should be noted that in step S2, the feature extraction network adopts a multilayer perceptron, which includes 12 hidden layers, 1024 neurons activated by a rectified linear unit, and the feature dimension is set to 302.

[0089] In this embodiment, 12 hidden layers give the network enough depth to capture deep features and complex patterns in the data. 1024 neurons per layer provide sufficient width, increase the representation power and capacity of the network, improve the ability of feature extraction, and show high accuracy on the training set and test set.

[0090] It should be noted that the single-pixel imaging system proposed in a possible embodiment of the present application includes:

[0091] A light source, used for emitting a broadband light source;

[0092] An optical component for converging and splitting the reflected light or transmitted light irradiating the target object;

[0093] An acquisition module, used to acquire light signals of multiple targets after being illuminated by a light source, and obtain target light field information according to the light signals;

[0094] A reference unit, configured to obtain a reference image according to the optical signal;

[0095] The module is used to establish a neural light field model and a feature extraction network based on the reference image and the target light field information, and to predict the input target light field information to obtain a predicted image, wherein the predicted image is:

[0096] ,

[0097] in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image;

[0098] The sorting module is used to randomly sample the corresponding data set into training set, test set and validation set in a ratio of 4:3:3 according to the same random seed function;

[0099] The training module is used to train the neural light field model. The loss function is as follows:

[0100] ,

[0101] in, is the number of training sets, Z is the total number of training sets, is the target light field information of the i-th training set, is the pixel matrix of the reference image of the i-th training set;

[0102] After training the training set, the minimum value of the loss function is less than 10 -4 When , the output is used as a model file;

[0103] A verification module is used to verify the trained model file. If the accuracy is greater than a first threshold range, it is qualified. Otherwise, a new random seed function is used to repeat steps S3 to S4 to iteratively train and verify the model until the accuracy is greater than the first threshold range;

[0104] The image generating unit generates an image of a target object.

[0105] In some embodiments of the present application, the acquisition module includes a single-pixel detector for detecting converged light and obtaining information of the target light field, wherein the single-pixel detector is a single photosensitive film with a thermoelectric effect, has N electrodes for receiving electrical signals, N>3, and is modulated by a bias voltage with a certain step size, and the modulation amount is M.

[0106] It should be noted that the certain step length in this embodiment can be understood as a preset step length value. For example, if the adjustable range of the bias voltage is -20V to 20V, and it starts from -20V and is adjusted in steps of 5V, then the modulation amount M is 9.

[0107] In some embodiments of the present application, the target light field information includes light field information generated by the target object based on converged light, and the light field information is an M×N matrix composed of M modulation quantities and N electrode received electrical signals.

[0108] In some embodiments of the present application, the optical component includes one or more optical elements for converging and splitting the reflected light or transmitted light irradiating the target object. The field of view angle and the aperture of the lens group can be designed according to the illumination mode of the object and the pixel size of the single-pixel detector.

[0109] In some embodiments of the present application, the image generation unit includes:

[0110] The multi-electrode signal amplification circuit is used to detect the signals of multiple electrodes at the same time and amplify them uniformly with a certain amplification resistor;

[0111] The signal acquisition and processing circuit is used to acquire the signal after amplification of each electrode, and the circuit processes the acquired signal to obtain an image of the object.

[0112] Compared with the prior art, the present invention has the following beneficial effects:

[0113] The present invention does not require complex optomechanical structures and spatial modulation methods, and can use a single-pixel sensor to achieve high-precision imaging of objects. At the same time, it uses a semi-supervised deep learning algorithm, which is beneficial to improving training speed and saving hardware resources.

[0114] This embodiment provides a transmissive single-pixel imaging system for performing transmissive object imaging, including:

[0115] A light source, used for emitting a broadband light source;

[0116] An optical assembly, the optical assembly comprising a lens L1, a lens L2 and a beam splitter, the lens L1 focusing a light beam onto an object, using the beam splitter to disperse a transmitted light beam passing through the object, wherein a group of transmitted split light beams passes through the lens L2;

[0117] A single-pixel detector, wherein the light source, the optical component and the single-pixel detector are all located on the same optical axis, and the single-pixel detector receives the split light from the lens L2;

[0118] A signal acquisition unit, connected to the single-pixel detector, for acquiring a light signal after the object is illuminated by a light source, and obtaining target light field information according to the light signal;

[0119] A reference unit, including a lens L3 and an image sensor, wherein another group of transmitted split beams passes through the lens L3 and is acquired by the image sensor, wherein the image sensor includes a CCD (Charge Coupled Device) and a CMOS (Complementary Metal-Oxide Semiconductor);

[0120] The image generating unit is used to generate an object image.

[0121] It should be noted that CCD converts light signals into electric charges and further into digital signals through a series of charge-coupled capacitors and amplifiers. Its core part is a two-dimensional array, each unit of which contains a photosensitive element and a storage unit. When light shines on the photosensitive element, the generated charge is accumulated and stored in the corresponding storage unit, and then transmitted to the output end line by line through charge coupling technology. CMOS has an independent sensor and amplifier for each pixel, which directly converts the light signal into a voltage signal for output. Each pixel of CMOS is independent of each other and can be controlled individually. This makes CMOS sensors have higher integration and lower power consumption.

[0122] like Figure 2 As shown, the object indicated in the figure is the target object, the single pixel detector is a 4-electrode system, the bias voltage is modulated with a certain step size of 0.1V, and the modulation range is from -2V to 2V, then M is 42, N is 4; the reference image P×Q=64×64, then the single pixel imaging method includes the following steps:

[0123] Step 111, using a single pixel imaging system with a thermoelectric effect to acquire the signal after the light irradiation of the digital character, and obtain a reference image and target light field information respectively, wherein:

[0124] The reference image is discretized into a 64×64 pixel matrix, defined as ;

[0125] Time-correlated target light field information It is expressed as follows:

[0126] ,

[0127] Step S112: The reference image is used as a label set and the target light field information is used as a signal set, a neural light field model and a feature extraction network are established, and a prediction is performed according to the input target light field information to obtain a predicted image, wherein the predicted image is:

[0128] ,

[0129] in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image; both the gain factor and the characteristic factor are the parameter adjustment factors in the multilayer perceptron.

[0130] Among them, the neural light field model is established. The neural light field model shows the distribution characteristics of the light spot according to each voltage intensity distribution and the relationship with the focal length f of the optical component.

[0131] ,

[0132] Where n is the encoding focal length encoding dimension, which is the same as the number of thin film electrodes. Decoded values ​​for different positions;

[0133] The encoded focal length of the spot distribution feature is then input into a small multi-layer perceptron with only 3 hidden layers and 256 neurons in each hidden layer.

[0134] The feature extraction network established uses a multi-layer perceptron, which contains 12 hidden layers, 1024 neurons activated by rectified linear units, and the feature dimension is set to 302.

[0135] Step S113, according to the same random seed function, the corresponding data set consisting of the label set and the signal set is randomly sampled in a ratio of 4:3:3 and divided into a training set, a test set and a validation set;

[0136] Step S114: training with the neural light field model, the loss function is as follows:

[0137] ,

[0138] in, is the number of training sets, Z is the total number of training sets, is the target light field information of the i-th training set, is the pixel matrix of the reference image of the i-th training set.

[0139] After training the training set, the minimum value of the loss function is less than 10 -4 When the model file is output,

[0140] After iterative training with the test set, 100,000 iterations were performed as a cycle, with a batch size of 1000. The learning rate parameter was set to 0.001, and the learning rate was multiplied by 0.8 every 10,000 iterations; the Adam optimizer was used, and the beta parameter was set to 0.9 and 0.999.

[0141] Step S115: Use the validation set to verify the trained model file. If the accuracy rate is greater than 98.6%, it is qualified. Otherwise, use a new random seed function to repeat steps S113 to S114 for iterative model training and verification until the accuracy rate is greater than 98.6%.

[0142] The final training results are visible Figure 3 The reconstruction results show that based on the optomechanical structure and spatial modulation method of the present invention, a single-pixel sensor can be used to achieve high-precision imaging of objects, thereby improving the training speed and saving hardware resources.

[0143] This embodiment provides a reflective single-pixel imaging system for performing reflective object imaging, including:

[0144] A light source, used for emitting a broadband light source;

[0145] An optical component, the optical component includes a lens L1, a lens L2 and a beam splitter, the lens L1 gathers a reflected light beam reflected by an object to the beam splitter, and the reflected light beam is dispersed by the beam splitter, wherein a group of reflected split light beams passes through the lens L2;

[0146] A single-pixel detector, wherein the light source, the optical component and the single-pixel detector are all located on the same optical axis, and the single-pixel detector receives the split light from the lens L2;

[0147] A signal acquisition unit, connected to the single-pixel detector, for acquiring a light signal after the object is illuminated by a light source, and obtaining target light field information according to the light signal;

[0148] A reference unit, including a lens L3 and an image sensor, wherein another group of reflected split beams passes through the lens L3 and is acquired by the image sensor, wherein the image sensor includes a CCD (Charge Coupled Device) and a CMOS (Complementary Metal-Oxide Semiconductor);

[0149] The image generating unit is used to generate an object image.

[0150] It should be noted that CCD converts light signals into electric charges and further into digital signals through a series of charge-coupled capacitors and amplifiers. Its core part is a two-dimensional array, each unit of which contains a photosensitive element and a storage unit. When light shines on the photosensitive element, the generated charge is accumulated and stored in the corresponding storage unit, and then transmitted to the output end line by line through charge coupling technology. CMOS has an independent sensor and amplifier for each pixel, which directly converts the light signal into a voltage signal for output. Each pixel of CMOS is independent of each other and can be controlled individually. This makes CMOS sensors have higher integration and lower power consumption.

[0151] like Figure 4 As shown, the object indicated in the figure is the target object, the single pixel detector is a 10-electrode system, the bias voltage is modulated with a certain step size of 0.2V, and the modulation range is from -2V to 2V, then M is 21, N is 10; the reference image P×Q=128×128, then the single pixel imaging method includes the following steps:

[0152] Step S121, using a single pixel imaging system with a thermoelectric effect to acquire the signal after the light irradiation of the digital character, and obtain a reference image and target light field information respectively, wherein:

[0153] The reference image is discretized into a 128×128 pixel matrix, defined as ,

[0154] Target light field information It is expressed as follows:

[0155] ,

[0156] Among them, x and y are the corresponding coordinates of the light spot occupied by the light reflected or transmitted by the object through the optical component and converged on the photosensitive film of the single-pixel detector; u1, u2, ..., u M are different bias voltage values; V1, V2, V3, ..., V N is the signal strength detected by different electrodes on a single pixel sensor.

[0157] The voltage signal received by each electrode of the single-pixel detector can be expressed as:

[0158] ,

[0159] Among them, x and y are the corresponding coordinates of the area where the light reflected or transmitted by the object is concentrated on the photosensitive film of the single-pixel detector through the optical component, and the area where the light spot occupies is ; and is the electrode position; is the Seebeck coefficient of the hole; γ is the wavelength of the incident light The divergence relationship correlation factor of; u is the bias voltage; is the voltage temperature drift rate; T is the film temperature; is the bias voltage at room temperature 25°C; is the average incident light intensity; is the reflectivity of the film; d is the thickness of the film; is the film absorption rate; is the beam spot radius.

[0160] Step S122: The reference image is used as a label set and the target light field information is used as a signal set, a neural light field model and a feature extraction network are established, and a prediction is performed according to the input target light field information to obtain a predicted image, wherein the predicted image is:

[0161] ,

[0162] in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image; both the gain factor and the characteristic factor are the parameter adjustment factors in the multilayer perceptron.

[0163] Among them, the neural light field model is established. The neural light field model shows the distribution characteristics of the light spot according to each voltage intensity distribution and the relationship with the focal length f of the optical component.

[0164] ,

[0165] Where n is the encoding focal length encoding dimension, which is the same as the number of thin film electrodes; Decoded values ​​for different positions.

[0166] The spot distribution feature encoding the focal length is then input into a small multilayer perceptron with only 3 hidden layers and 256 neurons in each hidden layer.

[0167] The feature extraction network established uses a deep understanding of deep learning segmentation network (U-net), which contains 20 layers of convolution;

[0168] Step S123, according to the same random seed function, the corresponding data set consisting of the label set and the signal set is randomly sampled in a ratio of 4:3:3 and divided into a training set, a test set and a validation set;

[0169] Step S124: training with the neural light field model, the loss function is as follows:

[0170] ,

[0171] in, is the number of training sets, and Z is the total number of training sets;

[0172] After training the training set, the minimum value of the loss function is less than 10 -4 When , the output is used as a model file;

[0173] After iterative training with the test set, 100,000 iterations were performed as a cycle, with a batch size of 1000. The learning rate parameter was set to 0.001, and the learning rate was multiplied by 0.8 every 10,000 iterations; the Adam optimizer was used, and the beta parameter was set to 0.9 and 0.999.

[0174] Step S125: Use the validation set to verify the trained model file. If the accuracy rate is greater than 98.6%, it is qualified. Otherwise, use a new random seed function to repeat steps S123 to S124 for iterative model training and verification until the accuracy rate is greater than 98.6%.

[0175] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “clockwise”, “counterclockwise”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation to the invention.

[0176] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example.

[0177] Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Mentioning "embodiment" in this article means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present embodiment application. The appearance of this phrase in various positions in the specification is not necessarily the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It can be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0178] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A single pixel imaging method, characterized in that: include: Step S1, obtaining light signals of multiple targets after being illuminated by a light source, and obtaining a reference image and target light field information according to the light signals, wherein the obtained light signals are converted into voltage signal intensity, and the voltage signal intensity detected by the nth electrode is expressed as: ， Among them, x and y are the corresponding coordinates of the area where the light reflected or transmitted by the object is concentrated on the photosensitive film of the single-pixel detector through the optical component, and the area where the light spot occupies is ; and is the electrode position; is the Seebeck coefficient of the hole; γ is the wavelength of the incident light The divergence relationship correlation factor of; u is the bias voltage; is the voltage temperature drift rate; T is the film temperature; is the bias voltage at room temperature 25°C; is the average incident light intensity; is the reflectivity of the film; d is the thickness of the film; is the film absorption rate; is the beam spot radius; The reference image is discretized into a P×Q pixel matrix, defined as , Target light field information It is expressed as follows: ， Among them, x and y are the corresponding coordinates of the light spot occupied by the light reflected or transmitted by the object through the optical component and converged on the photosensitive film of the single-pixel detector; u1, u2, ..., u M are different bias voltage values; V1, V2, V3, ..., V N is the signal intensity detected by different electrodes on a single-pixel sensor; Step S2, taking the reference image as a label set and the target light field information as a signal set, establishing a neural light field model and a feature extraction network according to the reference image and the target light field information, and performing prediction according to the input target light field information to obtain a predicted image, wherein the predicted image is: ， in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image, is the target light field information; Step S3, according to the same random seed function, randomly sampling the corresponding data set consisting of the label set and the signal set in a first range ratio into a training set, a test set and a validation set; Step S4, training the neural light field model, the loss function is as follows: ， in, is the number of training sets, Z is the total number of training sets, is the target light field information of the i-th training set, is the pixel matrix of the reference image of the i-th training set, When the minimum value of the loss function obtained after training the training set is less than the set threshold, the output is used as the model file; Step S5, use the verification set to verify the trained model file, if the accuracy is greater than the first threshold range, it is qualified, otherwise use a new random seed function to repeat steps S3 to S4 for iterative model training and verification until the accuracy is greater than the first threshold range.

2. The single pixel imaging method according to claim 1, characterized in that: In step S2, the neural light field model displays the light spot distribution characteristics according to each voltage intensity distribution and the relationship with the focal length f of the optical component to establish decoding at different positions: ， Where n is the encoding focal length encoding dimension, Decoded values ​​for different positions; The encoded focal length of the spot distribution feature is then input into a small multi-layer perceptron with only 3 hidden layers and 256 neurons in each hidden layer.

3. The single pixel imaging method according to claim 1, characterized in that: In step S2, the feature extraction network uses a multilayer perceptron, which includes 12 hidden layers, 1024 neurons activated by a rectified linear unit, and the feature dimension is set to 302.

4. A single pixel imaging system, used to implement the single pixel imaging method according to any one of claims 1 to 3, characterized in that: include: A light source, used for emitting a broadband light source; An optical component for converging and splitting reflected light or transmitted light irradiated on a target object; An acquisition module, used to acquire light signals of multiple targets after being illuminated by a light source, and obtain target light field information according to the light signals; A reference unit, used for obtaining a reference image according to the optical signal; The module is used to establish a neural light field model and a feature extraction network based on the reference image and the target light field information, and to predict the input target light field information to obtain a predicted image, wherein the predicted image is: ， in, is the neural light field model, is the gain factor of the object image in the light field, is the feature extraction network, is the characteristic factor of the image, is the target light field information; A sorting module is used to randomly sample the corresponding data set according to the same random seed function and divide it into a training set, a test set and a validation set in a first range ratio; The training module is used to train the neural light field model. The loss function is as follows: ， in, is the number of training sets, Z is the total number of training sets, is the target light field information of the i-th training set, is the pixel matrix of the reference image of the i-th training set; When the minimum value of the loss function obtained after training the training set is less than the set threshold, the output is used as the model file; A verification module is used to verify the trained model file. If the accuracy rate is greater than the first threshold range, the model file is qualified. Otherwise, a new random seed function is used to repeat steps S3 to S4 to iteratively train and verify the model until the accuracy rate is greater than the first threshold range. The image generating unit generates an image of a target object.

5. The single pixel imaging system according to claim 4, characterized in that: The acquisition module includes a single-pixel detector for detecting converged light and obtaining information of the target light field, wherein the single-pixel detector is a single photosensitive film with a thermoelectric effect, has N electrodes for receiving electrical signals, N>3, and is modulated by a bias voltage with a preset step value, and the modulation amount is M.

6. The single pixel imaging system according to claim 4, characterized in that: The target light field information includes light field information generated by the target object based on the converged light, and the light field information is an M×N matrix composed of M modulation quantities and N electrode receiving electrical signals.

7. The single pixel imaging system according to claim 4, characterized in that: The optical assembly includes one or more optical elements for converging and splitting the reflected light or transmitted light irradiating the target object.

8. The single pixel imaging system according to claim 4, characterized in that: The image generation unit includes: The multi-electrode signal amplification circuit is used to detect the signals of multiple electrodes at the same time and amplify them uniformly with a certain amplification resistor; The signal acquisition and processing circuit is used to acquire the signal after amplification of each electrode, and the circuit processes the acquired signal to obtain an image of the object.

Images