A Detection Method for Fusion Offset Based on the Spot of Ring-Core Fiber

Abstract

The present invention belongs to the field of fiber optic sensing technology, and more specifically, relates to a detection method for the fusion offset based on the optical fiber spot of a ring-core optical fiber. By using the optical fiber spot of the ring-core optical fiber with stable and high morphological structure differentiation, the spot recognition training of the high-precision convolutional neural network is carried out on the spots collected under different misaligned fusion offsets, breaking the existing limitation of measuring the fiber fusion loss only by the backward Rayleigh scattering of the higher-order modes of the few-mode optical fiber, realizing the measurement of the fiber offset during the fusion process, and improving the accuracy of the fusion quality detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber sensing, and more specifically, relates to a detection method for the splicing offset based on the spot of a ring-core optical fiber. Background Art

[0002] Currently, most of the detection methods for optical fiber splicing offset are based on optical time domain reflectometry. The most common method for detecting the quality of optical fiber splicing is to use a single-mode optical time domain reflectometer (OTDR), which detects based on the backward Rayleigh scattering of the fundamental mode in a single-mode optical fiber; this method can only support the detection of the fundamental mode in an optical fiber and is not applicable to few-mode optical fibers that support multiple modes and have large differences in the transmission loss characteristics of each mode; moreover, the detection sensitivity of the fundamental mode is low, which will affect the accuracy of judgment. Some people have also proposed a detection method for few-mode optical fiber links based on the backward Rayleigh scattering of high-order modes, which utilizes the high sensitivity of high-order modes in few-mode optical fibers to optical fiber splicing errors to improve the accuracy of judging the splicing quality. However, this method only obtains the loss caused by splicing and can only judge the relative quality of splicing from the magnitude of the loss, and cannot obtain the numerical value of the splicing offset. Summary of the Invention

[0003] In order to overcome the above defects in the prior art, the present invention provides a detection method for the splicing offset based on the spot of a ring-core optical fiber, which realizes the measurement of the optical fiber offset during the splicing process and improves the accuracy of splicing quality detection.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is: a detection method for the splicing offset based on the spot of a ring-core optical fiber, comprising the following steps:

[0005] S1. System setup: Set up a characterization system for the misaligned splicing of ring-core optical fibers. The system includes: a laser, a lens, a quarter-wave plate, a mirror, a vortex phase plate, two ring-core optical fibers to be spliced, an optical fiber fusion splicer, and a CCD camera;

[0006] S2. Data acquisition: After the light emitted by the laser is collimated by the lens, it becomes circularly polarized after passing through the quarter-wave plate; the light is transmitted to the vortex phase plate for mode modulation and then coupled into the first ring-core optical fiber to be spliced through the lens; in the optical fiber fusion splicer, first align the first and second ring-core optical fibers to be spliced, and then adjust the axial offset between the two ring-core optical fibers to be spliced according to experimental needs; at the end of the second ring-core optical fiber to be spliced, place a CCD camera to collect the optical fiber output spots under different splicing offset conditions;

[0007] S3. Processing of the spot image: Process the collected spot data on a computer. First, perform absolute value difference processing on each image collected at different offsets of the two optical fibers with the average image of the images collected under the alignment condition, and then crop the image after the absolute value difference as the input image of the convolutional neural network;

[0008] S4. Training and prediction of the neural network: The cropped image is input into the convolutional neural network. Use the convolutional neural network to learn the optical fiber spots under different splicing offsets, and establish a non-linear mapping relationship between the splicing offset and the change of the optical fiber spot. In the training stage, the offset predicted by the output layer will construct a cross-entropy loss function with the actual corresponding offset of the input image, and update the parameters of the convolutional neural network through the method of backpropagation of gradients and stochastic gradient descent; In the prediction stage, the data output by the output layer will be directly used as the offset corresponding to the image.

[0009] In the present invention, use the convolutional neural network to learn the optical fiber spots under different splicing offsets, establish a non-linear mapping relationship between the splicing offset and the change of the optical fiber spot, realize the measurement of the optical fiber splicing offset, and thus realize the evaluation of the optical fiber splicing quality.

[0010] Further, the ring-core optical fiber is a radially first-order restricted ring-core optical fiber.

[0011] Further, the absolute value difference processing includes: subtracting the gray value of the pixel point of each image from the gray value of the corresponding pixel point of the average image of the images collected under the alignment condition.

[0012] Further, the specific steps of S3 include:

[0013] S31. Calculate the average image of the images collected for the two optical fibers under the alignment condition, expressed as:

[0014]

[0015] where A mean (i,j) represents the gray value of the obtained average image at the pixel point (i,j), and A k (i,j) represents the gray value of the image collected under the alignment condition at the pixel point (i,j);

[0016] S32. Perform absolute value difference processing on the image collected when the optical fiber is offset with the obtained average image, expressed as:

[0017] A′ p (i,j) = A p (i,j) - A mean (i,j)

[0018] where A′p (i, j) represents the gray value of the image after absolute difference at the pixel point (i, j), A p (i, j) represents the gray value of the image collected under the offset condition at the pixel point (i, j), A mean (i, j) represents the gray level of the average image at the pixel point (i, j);

[0019] S33. Crop the image after absolute difference, and crop it from the size of 960 * 1280 to 412 * 412 as the input image of the convolutional neural network.

[0020] Furthermore, the convolutional neural network includes an input layer, two convolutional layers, two downsampling layers, two fully connected layers and an output layer.

[0021] Furthermore, the step S4 specifically includes:

[0022] S41. The input image X1 is compressed to the size of 256 * 256 in the input layer and becomes the image X2 after zero-mean normalization, and then enters the first convolutional layer C1;

[0023] S42. After the image X2 is convolved by 8 convolutional kernels with a size of 5 * 5 * 1, a stride of 1, and a padding of 2, 8 feature maps X3_i {i = 1,..., 8} are generated. After the 8 feature maps are batch-normalized, they are activated by the ReLU function to become X3_i' {i = 1,..., 8}, and then enter the first pooling layer S1;

[0024] S43. The 8 feature maps X3_i' {i = 1,..., 8} become X4_i {i = 1,..., 8} after average pooling with a kernel size of 2 * 2 and a stride of 2, and then enter the second convolutional layer C2;

[0025] S44. After the 8 feature maps X4_i {i = 1,..., 8} are convolved by 16 convolutional kernels with a size of 5 * 5 * 8, a stride of 1, and a padding of 2, 16 feature maps X5_i {i = 1,..., 16} are generated. After the 16 feature maps are batch-normalized, they are activated by the ReLU function to become X5_i' {i = 1,..., 16}, and then enter the second pooling layer S2;

[0026] S45. The 16 feature maps X5_i' {i = 1,..., 16} become X6_i {i = 1,..., 16} after average pooling with a kernel size of 2 * 2 and a stride of 2. Then the 16 feature maps are unfolded and concatenated into a one-dimensional vector, which is input to the fully connected layer FC1. The fully connected layer FC1 has 256 neurons;

[0027] After the output of the fully connected layer FC1 is activated by the ReLU function, it is input into the fully connected layer FC2, and the fully connected layer FC2 has 9 neurons;

[0028] S47. The outputs of the 9 neurons of FC2 enter the output layer, and the output layer uses the softmax function to judge the fiber splicing offset corresponding to the input image; in the training stage, the offset predicted by the output layer will construct a cross-entropy loss function with the actual corresponding offset of the input image, and the parameters of the convolutional neural network are updated by the method of backpropagation of gradients and stochastic gradient descent; in the prediction stage, the data output by the output layer will be directly used as the offset corresponding to the image.

[0029] The present invention also provides a detection system for the splicing offset of a ring-core fiber spot, including:

[0030] Data acquisition module: including a laser, a lens, a quarter-wave plate, a mirror, a vortex phase plate, two ring-core fibers to be spliced, a fiber splicer, and a CCD camera; the light emitted by the laser is collimated by the lens and then becomes circularly polarized after passing through the quarter-wave plate; the light is transmitted to the vortex phase plate for mode modulation and then coupled into the first ring-core fiber to be spliced through the lens; in the fiber splicer, the first and the second ring-core fibers to be spliced are first aligned, and then the axial offset between the two ring-core fibers to be spliced is adjusted according to experimental needs; at the end of the second ring-core fiber to be spliced, a CCD camera is placed to collect the fiber output spots under different splicing offsets;

[0031] Spot image processing module: used to process the collected spot data on a computer, first perform absolute difference processing on each image collected by the two fibers at different offsets and the average image of the images collected in the aligned case, and then crop the image after the absolute difference processing as the input image of the convolutional neural network;

[0032] Convolutional neural network training and prediction module: used to input the cropped image into the convolutional neural network, and use the convolutional neural network to train and predict the image. In the training stage, the offset predicted by the output layer will construct a cross-entropy loss function with the actual corresponding offset of the input image, and the parameters of the convolutional neural network are updated by the method of backpropagation of gradients and stochastic gradient descent; in the prediction stage, the data output by the output layer will be directly used as the offset corresponding to the image.

[0033] Further, the spot image processing module includes:

[0034] Image averaging unit: used to calculate the average image of the images collected by the two fibers in the aligned case, expressed as:

[0035]

[0036] where A mean (i, j) represents the gray value of the obtained average image at the pixel point (i, j), and A k (i, j) represents the gray value of the image collected under the alignment condition at the pixel point (i, j);

[0037] Image difference unit: used to perform absolute value difference processing on the image collected under the condition of fiber offset and the obtained average image, expressed as:

[0038] A' p (i, j) = A p (i, j) - A mean (i, j)

[0039] where A' p (i, j) represents the gray value of the image after absolute value difference at the pixel point (i, j), A p (i, j) represents the gray value of the image collected under the offset condition at the pixel point (i, j), A mean (i, j) represents the gray level of the average image at the pixel point (i, j);

[0040] Image clipping unit: used to perform clipping processing on the image after absolute value difference, and clip it from the size of 960 * 1280 to the size of 412 * 412 as the input image of the convolutional neural network.

[0041] The present invention also provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the above-mentioned detection method for the fusion offset amount based on the ring-core fiber spot.

[0042] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned detection method for the fusion offset amount based on the ring-core fiber spot is implemented.

[0043] Compared with the prior art, the beneficial effects are as follows: The detection method for the fusion offset amount based on the ring-core fiber spot provided by the present invention utilizes the fiber spot of the ring-core fiber with stable and high morphological structure differentiation to perform high-precision convolutional neural network spot recognition training on the spots collected under different misaligned fusion offset amounts, breaking the limitation of the existing measurement of fiber fusion loss only using the backward Rayleigh scattering of the higher-order modes of few-mode fibers, realizing the measurement of the fiber offset amount during the fusion process, and improving the accuracy of fusion quality detection. Description of the Drawings

[0044] Figure 1It is the distribution diagram of the refractive index of the ring-core optical fiber selected in Embodiment 1 of the present invention.

[0045] Figure 2 It is the schematic diagram of the system structure in Embodiment 2 of the present invention.

[0046] Figure 3 It is the schematic diagram of the spot image processing flow in Embodiment 1 of the present invention.

[0047] Figure 4 It is the schematic diagram of the structure of the convolutional neural network in Embodiment 1 of the present invention. Detailed implementation manners

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The present invention will be described below in one of the embodiments in conjunction with the specific implementation manners. Among them, the accompanying drawings are only for illustrative purposes and represent only schematic diagrams, rather than physical diagrams, and should not be construed as a limitation to this patent; in order to better illustrate the embodiments of the present invention, some components in the accompanying drawings will be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0049] In the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the accompanying drawings are only for illustrative purposes and should not be construed as a limitation to this patent. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances. In addition, if there is a description involving "first", "second", etc. in the embodiments of the present invention, the description of "first", "second", etc. is only for descriptive purposes and should not be construed as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the meaning of "and / or" appearing throughout the text is that it includes three parallel solutions. Taking "A and / or B" as an example, it includes the A solution, or the B solution, or the solution where A and B are satisfied simultaneously.

[0050] Embodiment 1:

[0051] A detection method for the fusion offset based on the spot of a ring-core optical fiber includes the following steps:

[0052] Step 1. Selection of optical fiber: Select a radially first-order confinement ring-core optical fiber with a large propagation constant difference between different mode groups and weak coupling between mode groups. Such a design maintains the clustering characteristics between modes while ensuring stable spot and high morphological structure distinguishability. For misaligned fusion splicing with different offsets, the spots of the ring-core optical fiber show different changes.

[0053] As Figure 1 shown, it is the refractive index distribution of the used optical fiber. A refractive index notch is introduced at the top of the ring-core optical fiber to modulate the refractive index distribution of the ring-core optical fiber and reduce the mode coupling caused by perturbation. This ring-core optical fiber supports the transmission of four mode groups, and the effective refractive index difference Δ eff is greater than 2×10 -3 , which ensures weak coupling between each mode group, keeps the clustering characteristics between modes, and at the same time ensures stable fiber speckle and obvious morphological structure distinguishability.

[0054] Step 2. As Figure 2 shown, construction of the system: Construct a characterization system for misaligned fusion splicing of ring-core optical fibers. The system includes: a laser, a lens, a quarter-wave plate, a mirror, a vortex phase plate, two ring-core optical fibers to be fused, an optical fiber fusion splicer, and a CCD camera;

[0055] Step 3. Data acquisition: The light emitted by the laser is collimated by the lens and then becomes circularly polarized after passing through the quarter-wave plate; the light is transmitted to the vortex phase plate for mode modulation and then coupled into the first ring-core optical fiber to be fused through the lens; in the optical fiber fusion splicer, first align the first and the second ring-core optical fibers to be fused, and then adjust the axial offset between the two ring-core optical fibers to be fused according to experimental needs; at the end of the second ring-core optical fiber to be fused, place a CCD camera to collect the fiber output spots under different fusion offsets.

[0056] Step 4. Processing of spot images: Process the collected spot data on the computer. First, perform absolute value difference processing on each image collected from the two optical fibers at different offsets and the average image of the images collected under the alignment condition, and then crop the image after the absolute value difference as the input image of the convolutional neural network.

[0057] As Figure 3 shown, it specifically includes the following steps:

[0058] S31. Calculate the average image of the images collected from the two optical fibers under the alignment condition, expressed as:

[0059]

[0060] where A mean(i, j) represents the gray value of the obtained average image at the pixel point (i, j), A k (i, j) represents the gray value of the image collected under the alignment condition at the pixel point (i, j);

[0061] S32. Perform absolute value difference processing on the image collected under the condition of fiber offset and the obtained average image, expressed as:

[0062] A′ p (i, j) = A p (i, j) - A mean (i, j)

[0063] where A′ p (i, j) represents the gray value of the image after absolute value difference at the pixel point (i, j), A p (i, j) represents the gray value of the image collected under the offset condition at the pixel point (i, j), A mean (i, j) represents the gray level of the average image at the pixel point (i, j);

[0064] S33. Perform clipping processing on the image after absolute value difference, and clip it from the size of 960 * 1280 to 412 * 412 as the input image of the convolutional neural network.

[0065] Step 5. Training and prediction of the neural network. The convolutional neural network used in this patent includes an input layer, two convolutional layers, two downsampling layers, two fully connected layers, and an output layer. The specific network structure is as Figure 4 shown.

[0066] The clipped image is input into the convolutional neural network, and the convolutional neural network is used to train and predict the image, specifically including:

[0067] S41. The input image X1 is compressed to the size of 256 * 256 in the input layer and becomes the image X2 after zero-mean normalization, and then enters the first convolutional layer C1;

[0068] S42. After the image X2 is convolved by 8 convolutional kernels with a size of 5 * 5 * 1, a step size of 1, and a padding of 2, 8 feature maps X3_i {i = 1,..., 8} are generated. After the 8 feature maps are batch-normalized, they are activated by the ReLU function to become X3_i' {i = 1,..., 8}, and then enter the first pooling layer S1;

[0069] S43. The 8 feature maps X3_i' {i = 1,..., 8} become X4_i {i = 1,..., 8} after average pooling with a kernel size of 2 * 2 and a step size of 2, and then enter the second convolutional layer C2;

[0070] After the 8 feature maps X4_i {i = 1, ..., 8} are convolved by 16 convolutional kernels with a kernel size of 5*5*8, a stride of 1, and a padding of 2, 16 feature maps X5_i {i = 1, ..., 16} are generated. After the 16 feature maps are batch-normalized, they are activated by the ReLU function to become X5_i’ {i = 1, ..., 16} and enter the second pooling layer S2;

[0071] The 16 feature maps X5_i’ {i = 1, ..., 16} become X6_i {i = 1, ..., 16} after average pooling with a kernel size of 2*2 and a stride of 2. The 16 feature maps are then unfolded and concatenated into a one-dimensional vector and input into the fully connected layer FC1, which has 256 neurons;

[0072] The output of the fully connected layer FC1 is activated by the ReLU function and then input into the fully connected layer FC2, which has 9 neurons;

[0073] The output of the 9 neurons in FC2 enters the output layer, and the output layer uses the softmax function to determine the fusion offset of the optical fiber corresponding to the input image; in the training phase, the offset predicted by the output layer will construct a cross-entropy loss function with the actual offset corresponding to the input image, and the parameters of the convolutional neural network are updated by the method of backpropagation of gradients and stochastic gradient descent; in the prediction phase, the data output by the output layer will be directly used as the offset corresponding to the image.

[0074] Embodiment 2

[0075] This embodiment provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the method for detecting the fusion offset of the ring-core optical fiber spot described in Embodiment 1.

[0076] Embodiment 3

[0077] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for detecting the fusion offset of the ring-core optical fiber spot described in Embodiment 1.

[0078] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limiting the embodiments of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A detection method for the fusion offset based on the spot of ring-core fiber, characterized in that, it includes the following steps: S1. System construction: Construct a characterization system for the misaligned fusion of ring-core fibers. The system includes: a laser, a lens, a quarter-wave plate, a mirror, a vortex phase plate, two ring-core fibers to be fused, a fiber fusion splicer, and a CCD camera; S2. Data acquisition: The light emitted by the laser is collimated by the lens and then becomes circularly polarized after passing through the quarter-wave plate; the light is transmitted to the vortex phase plate for mode modulation and then coupled into the first ring-core fiber to be fused through the lens; in the fiber fusion splicer, first align the first and the second ring-core fibers to be fused, and then adjust the axial offset between the two ring-core fibers to be fused according to experimental needs; at the end of the second ring-core fiber to be fused, place a CCD camera to collect the fiber output spots under different fusion offsets; S3. Processing of the spot image: Process the collected spot data on a computer. First, perform absolute value difference processing on each image collected for the two fibers at different offsets and the average image of the images collected under the aligned condition, and then crop the image after the absolute value difference processing as the input image of the convolutional neural network; the specific steps of step S3 include: S31. Calculate the average image of the images collected for the two fibers under the aligned condition, expressed as: where A mean (i, j) represents the gray value of the obtained average image at the pixel point (i, j), A k (i, j) represents the gray value of the image collected under the alignment condition at the pixel point (i, j); S32. Perform absolute value difference processing on the image collected when the fiber is offset and the obtained average image, expressed as: A ′ p (i,j) = A p (i,j) - A mean (i,j) Where A ′ p (i, j) represents the gray value of the image at the pixel point (i, j) after absolute value difference, and A p (i, j) represents the gray value of the image acquired under the offset condition at the pixel point (i, j); S33. Perform cropping processing on the image after the absolute value difference processing, crop it from the size of 960*1280 to the size of 412*412 as the input image of the convolutional neural network; S4. Training and prediction of the neural network: The cropped image is input into the convolutional neural network, and the convolutional neural network is used to train and predict the image. In the training stage, the offset predicted by the output layer will construct a cross-entropy loss function with the actual corresponding offset of the input image, and update the parameters of the convolutional neural network through the method of backpropagation of gradients and stochastic gradient descent; in the prediction stage, the data output by the output layer will be directly used as the offset corresponding to the image; the specific steps of step S4 include: S41. Input image X 1 It is compressed to a size of 256*256 in the input layer and becomes image X after zero-centering 2 and enters the first convolutional layer C1; S42. Image X 2 After convolution with 8 convolutional kernels of size 5*5*1, stride 1, and padding 2, 8 feature maps X are generated 3_i {i = 1,...,8}, after the 8 feature maps are batch-normalized, they are activated by the ReLU function to become X 3_i ’{i = 1,...,8}, enter the first pooling layer S1; 8 activated Figure X 3_i ’{i = 1,...,8} becomes X after average pooling with a kernel size of 2*2 and a stride of 2 4_i {i = 1,...,8} enters the second convolutional layer C2; Graph X after average pooling 4_i After convolution with 16 convolutional kernels of size 5*5*8, stride 1, and padding 2 for {i = 1,..., 8}, 16 feature maps X are generated 5_i For {i = 1,..., 16}, after batch normalization of the 16 feature maps, they are activated by the ReLU function to become X 5_i For {i = 1,..., 16}, they enter the second pooling layer S2; 16 activated feature maps X 5_i ’{i = 1,..., 16} becomes X after average pooling with a kernel size of 2*2 and a stride of 2 6_i {i = 1,..., 16}, the 16 feature maps after average pooling are then unfolded and concatenated into a one-dimensional vector and input into the fully connected layer FC1, and the fully connected layer FC1 has 256 neurons; S46. The output of the fully connected layer FC1 is activated by the ReLU function and then input into the fully connected layer FC2, and the fully connected layer FC2 has 9 neurons; S47. The outputs of the 9 neurons of FC2 enter the output layer, and the output layer judges the fiber fusion offset corresponding to the input image through the softmax function; in the training stage, the offset predicted by the output layer will construct a cross-entropy loss function with the actual corresponding offset of the input image, and update the parameters of the convolutional neural network through the method of backpropagation of gradients and stochastic gradient descent; in the prediction stage, the data output by the output layer will be directly used as the offset corresponding to the image.

2. The detection method for the fusion offset based on the spot of ring-core fiber according to claim 1, characterized in that, the ring-core fiber is a radially first-order restricted ring-core fiber.

3. The detection method for the fusion offset based on the spot of ring-core fiber according to claim 1, It is characterized in that The absolute difference processing includes: subtracting the gray value of each pixel point of the image from the gray value of the corresponding pixel point of the average image of the image collected under the aligned condition.

4. The detection method of the fusion offset based on the coreless fiber spot according to claim 1, It is characterized in that The convolutional neural network includes an input layer, two convolutional layers, two downsampling layers, two fully connected layers and an output layer.

5. A detection system for the fusion offset based on the coreless fiber spot, It is characterized in that Comprising: Data acquisition module: including a laser, a lens, a quarter-wave plate, a mirror, a vortex phase plate, two coreless fibers to be fused, a fiber fusion splicer, and a CCD camera; the light emitted by the laser is collimated by the lens and then becomes circularly polarized after passing through the quarter-wave plate; the light is transmitted to the vortex phase plate for mode modulation and then coupled into the first coreless fiber to be fused through the lens; in the fiber fusion splicer, first align the first and second coreless fibers to be fused, and then adjust the axial offset between the two coreless fibers to be fused according to experimental needs; at the end of the second coreless fiber to be fused, place a CCD camera to collect the fiber output spot under different fusion offset conditions; Spot image processing module: used to process the collected spot data on a computer. First, perform absolute difference processing on each image collected when the two fibers are at different offsets and the average image of the image collected under the aligned condition, and then crop the image after the absolute difference processing as the input image of the convolutional neural network; Convolutional neural network training and prediction module: used to input the cropped image into the convolutional neural network, use the convolutional neural network to learn the fiber spots under different fusion offsets, and establish a non-linear mapping relationship between the fusion offset and the change of the fiber spot; in the training stage, the offset predicted by the output layer will construct a cross-entropy loss function with the actual corresponding offset of the input image, and update the parameters of the convolutional neural network through the method of backpropagation of gradients and stochastic gradient descent; In the prediction stage, the data output by the output layer will be directly used as the offset corresponding to the image; Among them, the spot image processing module includes: Image averaging unit: used to calculate the average image of the images collected when the two fibers are aligned, expressed as: Where A mean (i, j) represents the gray value of the obtained average image at the pixel point (i, j), and A k (i, j) represents the gray value of the image acquired under the alignment condition at the pixel point (i, j); Image difference unit: used to perform absolute difference processing on the image collected when the fiber is offset and the obtained average image, expressed as: A ′ p (i,j) = A p (i,j) - A mean (i,j) Where A ′ p (i, j) represents the gray value of the image at the pixel point (i, j) after absolute value difference, A p (i, j) represents the gray value of the image collected under the offset condition at the pixel point (i, j); Image cropping unit: used to crop the image after the absolute difference processing, crop it from the size of 960*1280 to the size of 412*412 as the input image of the convolutional neural network; The convolutional neural network training and prediction module executes the following steps: S41. Input image X 1 It is compressed to a size of 256*256 in the input layer and becomes image X after zero-centering 2 , and enters the first convolutional layer C1; S42. Image X 2 After convolution with 8 convolutional kernels of size 5*5*1, stride 1, and padding 2, 8 feature maps X are generated 3_i {i = 1,..., 8}, after the 8 feature maps are batch-normalized, they are activated by the ReLU function to become X 3_i ’{i = 1,..., 8}, enter the first pooling layer S1; 8 activated Figure X 3_i '{i = 1,...,8} becomes X after average pooling with a kernel size of 2*2 and a stride of 2 4_i {i = 1,...,8} enters the second convolutional layer C2; Graph X after average pooling 4_i After convolution with 16 convolutional kernels of size 5*5*8, stride 1, and padding 2 for {i = 1,..., 8}, 16 feature maps X are generated 5_i For {i = 1,..., 16}, after the 16 feature maps are batch-normalized, they are activated by the ReLU function to become X 5_i For {i = 1,..., 16}, it enters the second pooling layer S2; 16 activated feature maps X 5_i ’{i = 1,..., 16} becomes X after average pooling with a kernel size of 2*2 and a stride of 2 6_i {i = 1,..., 16}, the 16 feature maps after average pooling are then unfolded and concatenated into a one-dimensional vector, which is input to the fully connected layer FC1. The fully connected layer FC1 has 256 neurons; S46. The output of the fully connected layer FC1 is activated by the ReLU function and then input to the fully connected layer FC2, and the fully connected layer FC2 has 9 neurons; The outputs of the 9 neurons of S47.FC2 enter the output layer, and the output layer uses the softmax function to determine the fusion offset of the optical fiber corresponding to the input image. During the training phase, the offset predicted by the output layer will construct a cross-entropy loss function with the actual offset corresponding to the input image, and the parameters of the convolutional neural network will be updated through the method of backpropagation of gradients and stochastic gradient descent. During the prediction phase, the data output by the output layer will be directly used as the offset corresponding to the image.

6. An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for detecting the fusion offset of the ring-core optical fiber spot according to any one of claims 1 to 4.

7. A computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the method for detecting the fusion offset of the ring-core optical fiber spot according to any one of claims 1 to 4 is implemented.

Images