A spatial light angle information determination method and system based on optical fiber image recognition

Abstract

The application provides a spatial light angle information determination method and system based on optical fiber image recognition, comprising: constructing a training set; wherein the training set comprises spatial light angle information and optical fiber exit images; constructing a convolutional neural network model, training the convolutional neural network model by using the training set, inputting the collected optical fiber exit images into the trained convolutional neural network model, and obtaining spatial light angle information. The time resolution of the system is greatly improved, the error caused by system movement is avoided, and complex calculation is not required.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of light field information acquisition, and particularly relates to a spatial light angle information determination method and system based on optical fiber image recognition. BACKGROUND

[0002] In the field of 3D imaging, the reconstruction of target 3D information can be realized by collecting and processing light field information. From a strict physical model, light field information generally includes seven variables describing the space-time characteristics and wave characteristics. However, for applications targeting 3D imaging and detection, the amplitude, phase and angle information of the light field will have a key impact on the high-precision reconstruction of 3D information.

[0003] For the acquisition of phase information, the interference or structured light illumination method is generally used, which has high requirements for the environment and system stability and poor anti-interference ability. For the measurement of angle information, traditional methods include 3D imaging methods based on camera arrays or microlens arrays, but such methods cannot quantify the angle information. In computational imaging, the light field moment imaging technology can extract the angle information of the light field, which requires obtaining two scene images at different focal lengths, so that the system needs to be moved and the time resolution of the system is affected to some extent. SUMMARY

[0004] To solve the above technical problems, the application provides a spatial light angle information determination method and system based on optical fiber image recognition, which establishes a mapping relationship between the optical fiber exit image and the spatial light incident angle by building a neural network to identify the features of the exit image under different incident angles, and directly obtains the spatial light angle information from the exit image. This method greatly improves the time resolution of the system and avoids errors caused by system movement.

[0005] To achieve the above-mentioned purposes, the application provides a spatial light angle information determination method based on optical fiber image recognition, which includes:

[0006] A training set is constructed, wherein the training set includes spatial light angle information and optical fiber exit images;

[0007] A convolutional neural network model is constructed, the training set is used to train the convolutional neural network model, the collected optical fiber exit images are input into the trained convolutional neural network model, and the spatial light angle information is obtained.

[0008] Optionally, the construction of the training set includes:

[0009] An optical fiber transmission model is constructed, and the optical fiber exit light field is obtained based on the optical fiber transmission model;

[0010] construct a representation model between a spatial light incidence angle and a fiber exit image based on the fiber exit light field;

[0011] obtain the training set based on the representation model.

[0012] Optionally, obtaining the training set based on the representation model comprises:

[0013] input different spatial light incidence angles into the representation model to obtain corresponding fiber exit images, and construct the training set based on the different spatial light incidence angles and the corresponding fiber exit images.

[0014] Optionally, the representation model is:

[0015]

[0016] wherein c m(θ)，in is an eigenmode coefficient, is a fiber exit end light field, x2 is a fiber exit end face horizontal coordinate, y2 is a fiber exit end face vertical coordinate, θ is an incidence angle of spatial light, E m is a complex amplitude distribution of an electric field of an mth normalized eigenmode excited by spatial light at a fiber end face, x1 is a fiber incidence end face horizontal coordinate, y1 is a fiber incidence end face vertical coordinate, j is an imaginary unit, β m is a propagation constant of an mth eigenmode excited by spatial light at a fiber end face, z f is a transmission distance of light in a fiber.

[0017] Optionally, the fiber transmission model is:

[0018]

[0019] wherein, is a fiber incidence end light field, E m (x1, y1) is an electric field of an mth normalized eigenmode, M is a total number of eigenmodes supported by the fiber, c m is an eigenmode coefficient.

[0020] Further, the fiber exit light field is:

[0021]

[0022] wherein, is a fiber exit light field, E m (x1, y1) is an electric field of an mth normalized eigenmode, M is a total number of eigenmodes supported by the fiber, c m is an eigenmode coefficient.

[0023] Optionally, training the convolutional neural network model using the training set comprises:

[0024] training the convolutional neural network model by taking the fiber exit image as input and the spatial light angle information as output.

[0025] Optionally, the convolutional neural network model comprises a plurality of convolutional blocks.

[0026] The convolutional block comprises a convolutional layer-BN layer-ReLU layer-convolutional layer-BN layer-ReLU layer-convolutional layer-BN layer-ReLU layer-pooling layer and Sigmoid activation function layer.

[0027] In another aspect to achieve the above object, the present application also provides a spatial light angle information determination system based on fiber image recognition, comprising a first construction module, a second construction module and an output module.

[0028] The first construction module is configured to construct a training set, wherein the training set comprises spatial light angle information and fiber exit images.

[0029] The second construction module is configured to construct a convolutional neural network model and train the convolutional neural network model using the training set.

[0030] The output module is configured to input the collected fiber exit images into the trained convolutional neural network model to obtain spatial light angle information.

[0031] Constructing the training set comprises:

[0032] Constructing a fiber transmission model and obtaining a fiber exit light field based on the fiber transmission model.

[0033] Constructing a representation model between spatial light incident angle and fiber exit image based on the fiber exit light field.

[0034] Inputting different spatial light incident angles into the representation model to obtain corresponding fiber exit images, and constructing the training set based on different spatial light incident angles and corresponding fiber exit images.

[0035] The representation model is:

[0036]

[0037] wherein, c m(θ)，in is an eigenmode coefficient, is a fiber exit end light field, x2 is a fiber exit end face horizontal coordinate, y2 is a fiber exit end face vertical coordinate, θ is a spatial light incident angle, E mE m(x 1, y 1) is a complex amplitude distribution of the mth normalized eigenmode excited by spatial light at the end face of the optical fiber, x 1 is a transverse coordinate of the incident end face of the optical fiber, y 1 is a longitudinal coordinate of the incident end face of the optical fiber, j is an imaginary unit, and β m is a propagation constant of the mth eigenmode excited by spatial light at the end face of the optical fiber m E m(x 1, y 1) is a complex amplitude distribution of the mth normalized eigenmode excited by spatial light at the end face of the optical fiber, x 1 is a transverse coordinate of the incident end face of the optical fiber, y 1 is a longitudinal coordinate of the incident end face of the optical fiber, j is an imaginary unit, and β m is a propagation constant of the mth eigenmode excited by spatial light at the end face of the optical fiber f E m(x 1, y 1) is a complex amplitude distribution of the mth normalized eigenmode excited by spatial light at the end face of the optical fiber, x 1 is a transverse coordinate of the incident end face of the optical fiber, y 1 is a longitudinal coordinate of the incident end face of the optical fiber, j is an imaginary unit, and β m is a propagation constant of the mth eigenmode excited by spatial light at the end face of the optical fiber

[0038] Compared with the prior art, the present application has the following advantages and technical effects:

[0039] The present application establishes the mapping relationship between the optical fiber exit image and the spatial light incident angle by building a convolutional neural network to perform feature recognition on the exit image under different incident angles, and directly obtains the spatial light angle information from the exit image. This method greatly improves the time resolution of the system, avoids errors caused by system movement, and does not require complex calculations. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and the illustrative embodiments of the present application and their description serve the purpose of explaining the present application. The accompanying drawings should not be construed as an inappropriate limitation on the present application. In the drawings:

[0041] Figure 1 FIG. 1 is a flowchart of a spatial light angle information determination method according to an embodiment of the present application.

[0042] Figure 2 FIG. 2 is a schematic diagram of a spatial light angle information determination system according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0044] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0045] The present application proposes a spatial light angle information determination method based on optical fiber image recognition, comprising:

[0046] A training set is constructed, wherein the training set comprises spatial light angle information and optical fiber exit images;

[0047] A convolutional neural network model is constructed, the training set is used to train the convolutional neural network model, the collected fiber exit image is input into the trained convolutional neural network model, and spatial light angle information is obtained.

[0048] Further, constructing the training set comprises:

[0049] A fiber transmission model is constructed, and the fiber exit light field is obtained based on the fiber transmission model;

[0050] Based on the fiber exit light field, a representation model between the spatial light incidence angle and the fiber exit image is constructed;

[0051] Based on the representation model, the training set is obtained.

[0052] Further, based on the representation model, the training set is obtained, comprising:

[0053] Different spatial light incidence angles are input into the representation model to obtain corresponding fiber exit images, and the training set is constructed based on different spatial light incidence angles and corresponding fiber exit images.

[0054] Further, the training set is used to train the convolutional neural network model, comprising:

[0055] The fiber exit image is used as input, and the spatial light angle information is used as output, and the convolutional neural network model is trained.

[0056] Further, the convolutional neural network model comprises a plurality of convolutional blocks.

[0057] The convolutional block comprises a convolutional layer-BN layer-ReLU layer-convolutional layer-BN layer-ReLU layer-convolutional layer-BN layer-ReLU layer-pooling layer and Sigmoid activation function layer.

[0058] Embodiment 1

[0059] At present, the acquisition of angle information in the light field requires shooting scene images at different focal lengths, which affects the time resolution of the system. The purpose of the present embodiment is to provide a spatial light angle information determination method based on fiber exit image recognition. By building a neural network, the exit image under different incidence angles is identified, the mapping relationship between the fiber exit image and the spatial light incidence angle is established, and the spatial light angle information is directly obtained from the exit image. This method greatly improves the time resolution of the system and avoids errors caused by system movement.

[0060] As Figure 1As shown, the embodiment provides a spatial light angle information determination method based on optical fiber image recognition, comprising:

[0061] (1) Establish an optical fiber transmission model to calculate the optical fiber exit light field. All eigenmodes in the optical fiber form a complete orthonormal basis, so the optical fiber near field in a single polarization direction can be represented as a linear superposition of each eigenmode:

[0062]

[0063] where E m (x1,y1) is the electric field of the mth normalized eigenmode, M is the total number of eigenmodes supported by the optical fiber, c m is the mode coefficient of the eigenmode, determined by the integral of the product of the incident light field and the complex conjugate of the eigenmode electric field, is the optical fiber incident end light field.

[0064]

[0065] where, is the conjugate complex electric field of the mth eigenmode in the optical fiber. After a transmission distance z f in the optical fiber, the exit light field of the optical fiber is:

[0066]

[0067] where β m is the propagation constant of the mth eigenmode.

[0068] (2) Establish a representation model between the spatial light incident angle and the exit image. Different angles of spatial light can be described by a complex amplitude distribution:

[0069]

[0070] For spatial light with a certain tilt angle, the mode field distribution excited at the end face of the optical fiber can be represented as a superposition of the optical fiber eigenmode excited at this incident angle:

[0071]

[0072] The corresponding eigenmode coefficient is:

[0073]

[0074] Different angles of spatial light incident to the end face of the optical fiber have different coefficients of excited eigenmodes, so there is a one-to-one correspondence between the incident angle and the eigenmode coefficient. After a certain distance of transmission in the optical fiber, the corresponding exit light field can be written as:​

[0075]

[0076] wherein c m(θ)，in is the eigenmode coefficient, is the fiber exit light field, x2 is the transverse coordinate of the fiber exit end face, y2 is the longitudinal coordinate of the fiber exit end face, θ is the incident angle of the spatial light, E m is the complex amplitude distribution of the electric field of the mth normalized eigenmode excited by the spatial light at the fiber end face, x1 is the transverse coordinate of the fiber incident end face, y1 is the longitudinal coordinate of the fiber incident end face, j is the imaginary unit, β m is the propagation constant of the mth eigenmode excited by the spatial light at the fiber end face, z f is the transmission distance of the light in the fiber.

[0077] Since the incident angle of the spatial light changes, the coefficient of the eigenmode supported by the fiber will also change, so the fiber exit light field obtained finally also changes. Therefore, the incident angle of the spatial light and the fiber exit light field are in one-to-one correspondence, that is, the fiber exit image will change with the change of the angle of the light incident to the fiber end face.

[0078] (3) Generation of the neural network training set. With the central normal line of the fiber end face as the optical axis, the angle range of the spatial light that can be coupled into the fiber and excite the transmission mode is calculated through the numerical aperture of the fiber. By changing the angle information (θ1, θ2, θ3, …, θ n of the spatial light incident to the fiber, the corresponding fiber exit image is obtained, which is used as the neural network training set. The exit image is used as the input, and the angle information of the spatial light is used as the output.

[0079] (4) Building of the convolutional neural network (CNN) to establish the mapping relationship between the incident angle of the spatial light and the exit image. The CNN is composed of 3 convolutional blocks, each of which contains 2 convolutional layers, 2 BN layers, 2 LeakyReLU and 1 pooling layer, and finally outputs the angle information of the spatial light through the Sigmoid activation function.

[0080] (5) Spatial light angle information determination. The collected fiber exit image is input into the trained CNN neural network, and the corresponding spatial light angle information can be directly obtained.

[0081] The embodiment has the advantages that a spatial light angle information determination method based on optical fiber exit image recognition is provided. By building a convolutional neural network, the exit images under different incident angles are recognized, the mapping relationship between the optical fiber exit image and the spatial light incident angle is established, and the spatial light angle information is directly obtained from the exit image. This method greatly improves the time resolution of the system, avoids the error caused by system movement, and does not require complex calculation.

[0082] Embodiment 2

[0083] As shown in Figure 2 , the embodiment provides a spatial light angle information determination system based on optical fiber image recognition, comprising: a first construction module, a second construction module and an output module;

[0084] The first construction module is used for constructing a training set; wherein the training set comprises spatial light angle information and optical fiber exit images;

[0085] The second construction module is used for constructing a convolutional neural network model, and training the convolutional neural network model by using the training set;

[0086] The output module is used for inputting the collected optical fiber exit images into the trained convolutional neural network model to obtain the spatial light angle information.

[0087] Further, the training set comprises:

[0088] An optical fiber transmission model is constructed, and the optical fiber exit light field is obtained based on the optical fiber transmission model;

[0089] A representation model between the spatial light incident angle and the optical fiber exit image is constructed based on the optical fiber exit light field;

[0090] The training set is obtained based on the representation model.

[0091] The training set obtained based on the representation model comprises:

[0092] Different spatial light incident angles are input into the representation model to obtain corresponding optical fiber exit images, and the training set is constructed based on different spatial light incident angles and corresponding optical fiber exit images.

[0093] The representation model is:

[0094]

[0095] Wherein, c m(θ)，in is an eigenmode coefficient, is an optical fiber exit end light field, x2 is an optical fiber exit end face horizontal coordinate, y2 is an optical fiber exit end face vertical coordinate, θ is an incident angle of spatial light, E mLet x1 be the abscissa of the m-th normalized eigenmode excited by spatial light at the fiber end face, y1 be the ordinate of the fiber incident end face, j be the imaginary unit, and β be the ordinate of the abscissa of the fiber incident end face. m Let z be the propagation constant of the m-th eigenmode excited by spatial light at the fiber end face. f This represents the distance light travels in the optical fiber.

[0096] The fiber optic transmission model is as follows:

[0097]

[0098] in, E represents the optical field at the incident end of the optical fiber. m (x1, y1) represents the electric field of the m-th normalized eigenmode, M is the total number of eigenmodes supported by the optical fiber, and c m These are the mode coefficients of the intrinsic modes.

[0099] The optical field emitted from the fiber is:

[0100]

[0101] in, E represents the optical field emitted from the fiber. m (x1, y1) represents the electric field of the m-th normalized eigenmode, M is the total number of eigenmodes supported by the optical fiber, and c m These are the mode coefficients of the intrinsic modes.

[0102] Training a convolutional neural network model using a training set includes:

[0103] The optical fiber output image is used as input, and the spatial light angle information is used as output to train the convolutional neural network model.

[0104] A convolutional neural network model includes: several convolutional blocks;

[0105] The convolutional block consists of: convolutional layer - BN layer - ReLU layer - convolutional layer - BN layer - ReLU layer - convolutional layer - BN layer - ReLU layer - pooling layer and Sigmoid activation function layer.

[0106] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for determining spatial light angle information based on fiber optic image recognition, characterized in that, include: Construct the training set; The training set includes: spatial light angle information and fiber optic emission images; A convolutional neural network model is constructed, and the convolutional neural network model is trained using the training set. The collected optical fiber emission image is input into the trained convolutional neural network model to obtain spatial light angle information. Constructing the training set includes: Construct an optical fiber transmission model and obtain the optical fiber outgoing light field based on the optical fiber transmission model; Based on the optical fiber emitted light field, a characterization model is constructed between the spatial light incident angle and the optical fiber emitted image. Based on the representation model, the training set is obtained, including: Different spatial light incident angles are input into the representation model to obtain the corresponding optical fiber emission images. The training set is constructed based on the different spatial light incident angles and the corresponding optical fiber emission images. The representation model is as follows: in, These are the eigenmode coefficients. The optical field at the fiber optic output end. The x-coordinate of the fiber optic output end face. This represents the ordinate of the fiber optic output end face. Let be the incident angle of the light in space. Let be the complex amplitude distribution of the electric field of the m-th normalized eigenmode excited by spatial light at the end face of the optical fiber. The x-axis represents the incident end face of the optical fiber. The vertical coordinate of the fiber incident end face is... The imaginary unit, Let be the propagation constant of the m-th eigenmode excited by spatial light at the end face of the optical fiber. This represents the distance light travels in the optical fiber.

2. The method for determining spatial light angle information based on fiber optic image recognition according to claim 1, characterized in that, The optical fiber transmission model is as follows: in, The optical field at the fiber optic input end. For the first The electric field of a normalized eigenmode. The total number of intrinsic modes supported by the optical fiber. These are the mode coefficients of the intrinsic modes.

3. The method for determining spatial light angle information based on fiber optic image recognition according to claim 1, characterized in that, The optical field emitted from the optical fiber is: in, This refers to the light field emitted from the optical fiber. For the first The electric field of a normalized eigenmode. The total number of intrinsic modes supported by the optical fiber. These are the mode coefficients of the intrinsic modes.

4. The method for determining spatial light angle information based on fiber optic image recognition according to claim 1, characterized in that, Training the convolutional neural network model using the training set includes: The convolutional neural network model is trained by using the optical fiber output image as input and the spatial light angle information as output.

5. The method for determining spatial light angle information based on fiber optic image recognition according to claim 1, characterized in that, The convolutional neural network model includes: several convolutional blocks; The convolutional block includes: a convolutional layer, a batch null (BN) layer, a ReLU layer, a convolutional layer, a BN layer, a ReLU layer, a convolutional layer, a BN layer, a ReLU layer, a pooling layer, and a Sigmoid activation function layer.

6. A spatial light angle information determination system based on fiber optic image recognition, employing the determination method as described in any one of claims 1-5, characterized in that, include: First building module, second building module, and output module; The first construction module is used to construct the training set; The training set includes: spatial light angle information and fiber optic emission images; The second building module is used to build a convolutional neural network model and train the convolutional neural network model using the training set; The output module is used to input the acquired optical fiber emission image into the trained convolutional neural network model to obtain spatial light angle information. Constructing the training set includes: Construct an optical fiber transmission model and obtain the optical fiber outgoing light field based on the optical fiber transmission model; Based on the optical fiber emitted light field, a characterization model is constructed between the spatial light incident angle and the optical fiber emitted image. Different spatial light incident angles are input into the representation model to obtain the corresponding optical fiber emission images. The training set is constructed based on the different spatial light incident angles and the corresponding optical fiber emission images. The representation model is as follows: in, These are the eigenmode coefficients. The optical field at the fiber optic output end. The x-coordinate of the fiber optic output end face. This represents the ordinate of the fiber optic output end face. Let be the incident angle of the light in space. Let be the complex amplitude distribution of the electric field of the m-th normalized eigenmode excited by spatial light at the end face of the optical fiber. The x-axis represents the incident end face of the optical fiber. The vertical coordinate of the fiber incident end face is... The imaginary unit, Let be the propagation constant of the m-th eigenmode excited by spatial light at the end face of the optical fiber. This represents the distance light travels in the optical fiber.

Images