Fusion splicer, fusion splicing system, and method for fusion splicing optical fibers

Abstract

This fusion splicer comprises an imaging unit, a determination unit, and a splicing unit. The imaging unit captures images of a pair of optical fibers to generate imaging data. The determination unit determines the type of each of the pair of optical fibers on the basis of a plurality of feature values obtained from imaging data from the imaging unit. The determination unit employs determination results from either of a first and a second determination algorithm. The first determination algorithm is predefined using a method other than machine learning on the basis of the correlation between a plurality of feature values obtained for optical fibers from imaging data and the types of said optical fibers. The second determination algorithm includes a determination model. The determination model is created by machine learning that uses sample data that indicates the correspondence between a plurality of feature values obtained for optical fibers from imaging data and the types of said optical fibers. On the basis of determination results from the determination unit, the splicing unit fusion splices the pair of optical fibers per splicing conditions that correspond to the combination of types of the pair of optical fibers.

Description

[Technical field]

[0001] This disclosure relates to a fusion splicer, a fusion splicing system, and a method for fusion splicing optical fibers. This application claims priority to International Application No. PCT / JP2020 / 016859, filed on April 17, 2020, and incorporates by reference all of the contents of said international application. [Background technology]

[0002] Patent Documents 1 and 2 disclose a fusion splicer, a fusion splicing system, and a method for fusion splicing optical fibers. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2002-169050 A [Patent Document 2] JP 2020-20997 A Summary of the Invention

[0004] The fusion splicer of the present disclosure includes an imaging unit, a discrimination unit, and a connection unit. The imaging unit captures an image of a pair of optical fibers to generate imaging data. The discrimination unit discriminates the type of each of the pair of optical fibers based on a plurality of feature quantities obtained from the imaging data provided by the imaging unit. The discrimination unit has first and second discrimination algorithms for discriminating the type of optical fibers, and adopts a discrimination result by either the first or second discrimination algorithm. The first discrimination algorithm is predefined by a method other than machine learning based on a correlation between a plurality of feature quantities obtained from the imaging data of the optical fibers and the type of the optical fiber from which the feature quantities are obtained. The second discrimination algorithm includes a discrimination model for discriminating the type of optical fibers to be connected based on the imaging data of the optical fibers to be connected. The discrimination model is created by machine learning using sample data showing a correspondence relationship between a plurality of feature quantities obtained from the imaging data of the optical fibers and the type of the optical fiber from which the feature quantities are obtained. The connection unit fusion-splices the pair of optical fibers to each other under a connection condition according to the combination of the types of the pair of optical fibers based on the discrimination result by the discrimination unit.

[0005] A fusion splicing system according to the present disclosure includes a plurality of fusion splicers, which are the above-mentioned fusion splicers, and a model creation device. The model creation device collects sample data from the plurality of fusion splicers, performs machine learning to create a discrimination model, and provides the discrimination model to the plurality of fusion splicers.

[0006] The method for fusion splicing optical fibers of the present disclosure includes a step of generating imaging data, a step of determining, and a step of fusion splicing. In the step of generating imaging data, imaging data is generated by imaging a pair of optical fibers. In the step of determining, the type of each of the pair of optical fibers is determined based on a plurality of feature amounts obtained from the imaging data acquired in the step of generating imaging data. In the step of determining, a determination result by either a first or a second determination algorithm for determining the type of optical fiber is adopted. The first determination algorithm is predefined by a method other than machine learning based on a correlation between a plurality of feature amounts obtained from the imaging data of the optical fibers and the type of the optical fiber from which the feature amounts are obtained. The second determination algorithm includes a determination model for determining the type of optical fibers to be spliced ​​based on the imaging data of the optical fibers to be spliced. The determination model is created by machine learning using sample data showing a correspondence relationship between a plurality of feature amounts obtained from the imaging data of the optical fibers and the type of the optical fiber from which the feature amounts are obtained. In the step of fusion splicing, the pair of optical fibers are fusion spliced ​​to each other under a connection condition according to the combination of the types of the pair of optical fibers based on the determination result in the step of determining. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram illustrating a schematic configuration of an optical fiber fusion splicing system according to an embodiment of the present disclosure. [Diagram 2] FIG. 2 is a perspective view showing the appearance of the fusion splicer with the windshield cover closed. [Diagram 3] FIG. 3 is a perspective view showing the appearance of the fusion splicer with the windshield cover open to allow the internal structure of the fusion splicer to be seen. [Figure 4] FIG. 4 is a block diagram showing the functional configuration of the fusion splicer. [Diagram 5] FIG. 5 is a block diagram showing the hardware configuration of the fusion splicer. [Figure 6] FIG. 6 is a diagram illustrating the operation of the connection portion. [Figure 7] FIG. 7 is a diagram illustrating the operation of the connection portion. [Figure 8] FIG. 8 is a diagram illustrating the operation of the connection portion. [Figure 9] FIG. 9 is a front view of the end face of one of the optical fibers. [Figure 10] FIG. 10 is a diagram showing a schematic diagram of imaging data obtained in the imaging section. [Figure 11] FIG. 11 is a block diagram showing the functional configuration of the model generating device. [Figure 12] FIG. 12 is a block diagram showing the hardware configuration of the model creating device. [Figure 13] FIG. 13 is a flow chart illustrating a method according to one embodiment. [Figure 14] FIG. 14 is a flowchart showing a method according to a modified example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] [Problem that this disclosure aims to solve] There are various types of optical fibers. The types of optical fibers are distinguished by, for example, features related to applications and optical properties, and structural features. Features related to applications and optical properties include single mode fiber (SMF: Single Mode Fiber), multimode fiber (MMF: Multi Mode Fiber), general-purpose single mode fiber, dispersion shifted single mode fiber (DSF: Dispersion Shifted SMF), and non-zero dispersion shifted single mode fiber (NZDSF: Non-Zero DSF). Structural features include the diameter of the optical fiber, the core diameter, the material of the core and cladding, and the radial refractive index distribution. The optimal fusion conditions for fusion splicing a pair of optical fibers, such as the discharge time and the relative positions of the optical fibers, vary depending on the combination of the types of the pair of optical fibers. However, the types of optical fibers that have already been installed are often unknown. Therefore, it is important for the fusion splicer to accurately determine the combination of the types of the pair of optical fibers to be spliced.

[0009] For example, in the system described in Patent Document 2, a discrimination model capable of discriminating the type of optical fiber from the brightness distribution data in the radial direction of the optical fiber is created by machine learning. However, even if only a discrimination model based on machine learning is used, there is a limit to the discrimination accuracy.

[0010] [Effects of this disclosure] According to the present disclosure, it is possible to provide a fusion splicer, a fusion splicing system, and a method for fusion splicing optical fibers that can improve the accuracy of identifying the type of optical fiber.

[0011] [Description of the embodiments of the present disclosure] First, the embodiments of the present disclosure will be listed and described. A fusion splicer according to one embodiment includes an imaging unit, a discrimination unit, and a connection unit. The imaging unit captures an image of a pair of optical fibers to generate imaging data. The discrimination unit discriminates the type of each of the pair of optical fibers based on a plurality of feature amounts obtained from the imaging data provided by the imaging unit. The discrimination unit has first and second discrimination algorithms for discriminating the type of optical fibers, and adopts a discrimination result by either the first or second discrimination algorithm. The first discrimination algorithm is predefined by a method other than machine learning based on a correlation between a plurality of feature amounts obtained from the imaging data of the optical fibers and the type of the optical fibers from which the feature amounts are obtained. The second discrimination algorithm includes a discrimination model for discriminating the type of optical fibers to be connected based on the imaging data of the optical fibers to be connected. The discrimination model is created by machine learning using sample data showing a correspondence relationship between a plurality of feature amounts obtained from the imaging data of the optical fibers and the type of the optical fibers from which the feature amounts are obtained. The connection unit fusion-splices the pair of optical fibers to each other under a connection condition according to the combination of the types of the pair of optical fibers based on the discrimination result in the discrimination unit.

[0012] A fusion splicing system according to one embodiment includes a plurality of fusion splicers, which are the above-mentioned fusion splicers, and a model creation device. The model creation device collects sample data from the plurality of fusion splicers, performs machine learning to create a discrimination model, and provides the discrimination model to the plurality of fusion splicers.

[0013] A method for fusion splicing optical fibers according to an embodiment includes a step of generating imaging data, a step of determining, and a step of fusion splicing. The step of generating imaging data involves generating imaging data by imaging a pair of optical fibers. In the step of determining, the type of each of the pair of optical fibers is determined based on a plurality of feature amounts obtained from the imaging data acquired in the step of generating. In the step of determining, a determination result by either a first or a second determination algorithm for determining the type of optical fiber is adopted. The first determination algorithm is predefined by a method other than machine learning based on a correlation between a plurality of feature amounts obtained from the imaging data of the optical fibers and the type of the optical fiber from which the feature amounts are obtained. The second determination algorithm includes a determination model for determining the type of optical fibers to be spliced ​​based on the imaging data of the optical fibers to be spliced. The determination model is created by machine learning using sample data showing a correspondence relationship between a plurality of feature amounts obtained from the imaging data of the optical fibers and the type of the optical fiber from which the feature amounts are obtained. In the step of fusion splicing, the pair of optical fibers are fusion spliced ​​to each other under a connection condition according to the combination of the types of the pair of optical fibers based on the determination result in the step of determining.

[0014] In the above fusion splicer, fusion splicing system, and fusion splicing method, the type of optical fiber is identified using first and second discrimination algorithms. The first discrimination algorithm is predefined by a method other than machine learning based on the correlation between a plurality of feature amounts obtained from imaging data of the optical fiber and the type of the optical fiber, and discrimination accuracy equivalent to that of the conventional method can be expected. The second discrimination algorithm includes a discrimination model created by machine learning using sample data showing the correspondence between a plurality of feature amounts and the type of optical fiber. Therefore, it is possible to expect high-precision discrimination based on machine learning for optical fiber types that are difficult to discriminate or tend to be misdiscriminated by the first discrimination algorithm. Therefore, according to the above fusion splicer, fusion splicing system, and fusion splicing method, the discrimination accuracy of the optical fiber type can be improved compared to the conventional method by adopting the discrimination result by either the first or second discrimination algorithm.

[0015] In the above-mentioned fusion splicer, fusion splicing system, and fusion splicing method, the machine learning may be deep learning, in which case the accuracy of identifying the optical fiber type can be further improved.

[0016] In the above fusion splicer, fusion splicing system, and fusion splicing method, the discrimination unit (in the discrimination step) may adopt a discrimination result by one of the first and second discrimination algorithms when a predetermined feature included in the plurality of feature amounts is greater than a threshold, and may adopt a discrimination result by the other of the first and second discrimination algorithms when the predetermined feature amount is smaller than the threshold. For example, such a method makes it easy to select which discrimination result of the first and second discrimination algorithms to adopt. In this case, the threshold value may be a value determined based on a comparison between the discrimination accuracy by the first discrimination algorithm and the discrimination accuracy by the second discrimination algorithm when the predetermined feature amount changes. This can further improve the discrimination accuracy of the optical fiber type.

[0017] In the above fusion splicer, fusion splicing system, and fusion splicing method, the discrimination unit (in the discrimination step) may adopt the discrimination result when the type of each of the pair of optical fibers can be discriminated by the first discrimination algorithm, and adopt the discrimination result by the second discrimination algorithm when the type of each of the pair of optical fibers cannot be discriminated by the first discrimination algorithm. For example, such a method can improve the discrimination accuracy of the optical fiber type. In this case, the discrimination unit (in the discrimination step) may first execute the first discrimination algorithm, and execute the second discrimination algorithm when the type of each of the pair of optical fibers cannot be discriminated by the first discrimination algorithm. This can reduce the amount of calculations (in the discrimination step) of the discrimination unit. Alternatively, the discrimination unit (in the discrimination step) may execute the first discrimination algorithm and the second discrimination algorithm in parallel. This can reduce the time required to obtain the final discrimination result.

[0018] In the above fusion splicer, fusion splicing system, and fusion splicing method, the imaging unit (in the step of generating imaging data) may image the pair of optical fibers at least twice to generate at least two sets of imaging data. The discrimination unit (in the step of discriminating) may adopt a discrimination result by one of the first and second discrimination algorithms when a variation in a predetermined feature amount between at least two feature amount groups consisting of a plurality of feature amounts, obtained from at least two sets of imaging data, is greater than a threshold value. The discrimination unit (in the step of discriminating) may adopt a discrimination result by one of the first and second discrimination algorithms when a variation in the predetermined feature amount is smaller than a threshold value. This can further improve the accuracy of discriminating the optical fiber type.

[0019] In the above fusion splicer, fusion splicing system, and fusion splicing method, the imaging unit (in the step of generating imaging data) may capture images of the pair of optical fibers at least twice to generate at least two sets of imaging data. The discrimination unit (in the step of discriminating) may execute first and second discrimination algorithms based on at least two feature groups consisting of a plurality of feature amounts obtained from the at least two sets of imaging data. The discrimination unit (in the step of discriminating) may adopt one of the at least two discrimination results obtained by the first discrimination algorithm and the at least two discrimination results obtained by the second discrimination algorithm, whichever has the smaller variation in discrimination result. This can further improve the accuracy of discriminating the optical fiber type.

[0020] In the above fusion splicer, fusion splicing system, and fusion splicing method, the imaging positions of the at least two sets of imaging data in the optical axis direction of the pair of optical fibers may be the same as or different from each other.

[0021] In the above fusion splicing system, the model creation device may classify a plurality of fusion splicers into two or more groups that are estimated to have similar tendencies in the imaging data, and create a discrimination model for each group. Then, the second discrimination algorithm of the discrimination unit of each fusion splicer may obtain a discrimination model corresponding to the group to which the fusion splicer belongs from the model creation device. This allows machine learning to be performed limitedly within groups with similar tendencies in the imaging data, for example, within groups with less mechanical and structural variations in the fusion splicers and within groups with less mechanical and structural variations in the imaging units. This further improves the accuracy of discrimination of optical fiber types by the second discrimination algorithm.

[0022] In the above fusion splicing system, the sample data provided to the machine learning of the model creation device may include both sample data when the types of the pair of optical fibers can be discriminated by the first discrimination algorithm, and sample data when the types of the pair of optical fibers cannot be discriminated or are misdiscriminated by the first discrimination algorithm. In this case, optical fiber types that the first discrimination algorithm has difficulty discriminating can be included in the machine learning of the model creation device, thereby improving the overall accuracy of discriminating optical fiber types.

[0023] In the above fusion splicing system, the sample data provided for machine learning by the model creation device includes only sample data when the types of the pair of optical fibers can be discriminated by the first discrimination algorithm, and the discrimination unit of each fusion splicer may improve the discrimination model by performing machine learning using its own sample data when the types of the pair of optical fibers cannot be discriminated or are misdiscriminated by the first discrimination algorithm. In this case, the discrimination accuracy of the second discrimination algorithm can be improved for each fusion splicer for optical fiber types that cannot be discriminated by the first discrimination algorithm due to mechanical and structural variations of each fusion splicer, for example mechanical and structural variations of the imaging unit.

[0024] In the above fusion splicing system, the sample data provided for the machine learning of the model creation device may include sample data when the type of each of the pair of optical fibers can be discriminated by the first discrimination algorithm, and sample data when the type of each of the pair of optical fibers cannot be discriminated or is misdiscriminated by the first discrimination algorithm. The discrimination unit of each fusion splicer may perform machine learning using its own sample data when the type of each of the pair of optical fibers cannot be discriminated or is misdiscriminated by the first discrimination algorithm to improve the discrimination model. However, the sample data provided to the model creation device is excluded. In this case, the optical fiber type that the first discrimination algorithm is not good at can be included in the machine learning of the model creation device. In addition, for optical fiber types that cannot be discriminated by the first discrimination algorithm due to mechanical and structural variations of each fusion splicer, for example, mechanical and structural variations of the imaging unit, the discrimination accuracy of the second discrimination algorithm can be improved for each fusion splicer. Therefore, the overall discrimination accuracy of the optical fiber type can be further improved.

[0025] In the above fusion splicing method, two or more optical fibers of known types may be imaged to generate imaging data, and the types of the two or more optical fibers may be identified by first and second discrimination algorithms based on a plurality of feature quantities obtained from the imaging data. Then, of the first and second discrimination algorithms, one having higher discrimination accuracy may be adopted in the discrimination step. This can further increase the accuracy of identifying the optical fiber type.

[0026] [Details of the embodiment of the present disclosure] Specific examples of the fusion splicer, fusion splicing system, and method for fusion splicing optical fibers according to the present disclosure will be described below with reference to the drawings. Note that the present invention is not limited to these examples, but is defined by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims. In the following description, the same elements in the description of the drawings will be given the same reference numerals, and duplicated descriptions will be omitted.

[0027] FIG. 1 is a diagram illustrating a schematic configuration of a fusion splicing system 1A according to an embodiment of the present disclosure. The fusion splicing system 1A includes a plurality of fusion splicers 10 and a model creation device 20. The fusion splicers 10 are devices that perform fusion splicing of optical fibers. The model creation device 20 is a device that creates a discrimination model for discriminating the type of optical fiber. The model creation device 20 is a computer that can communicate with the plurality of fusion splicers 10 via an information and communication network 30. The information and communication network 30 is, for example, the Internet. The location area of ​​the model creation device 20 is separate from the location area of ​​the fusion splicers 10.

[0028] 2 and 3 are perspective views showing the appearance of the fusion splicer 10. FIG. 2 shows the appearance with the windshield cover closed, and FIG. 3 shows the appearance with the windshield cover open to show the internal structure of the fusion splicer 10. As shown in FIGS. 2 and 3, the fusion splicer 10 has a box-shaped housing 2. A splicing section 3 for fusion splicing optical fibers and a heater 4 are provided on the upper part of the housing 2. The heater 4 is a section for heat-shrinking a fiber reinforcement sleeve that is placed over the spliced ​​portion of the optical fibers fusion-spliced ​​in the splicing section 3. The fusion splicer 10 has a monitor 5 that displays the fusion splicing status of the optical fibers captured by an imaging section (described later) disposed inside the housing 2. Furthermore, the fusion splicer 10 has a windshield cover 6 for preventing wind from entering the splicing section 3.

[0029] The splice section 3 has a holder placement section on which a pair of optical fiber holders 3a can be placed, a pair of fiber positioning sections 3b, and a pair of discharge electrodes 3c. The optical fibers to be fused are held and fixed in the optical fiber holders 3a, and the optical fiber holders 3a are placed and fixed on the holder placement sections. The fiber positioning sections 3b are disposed between the pair of optical fiber holders 3a, and position the tips of the optical fibers held in each of the optical fiber holders 3a. The discharge electrode 3c is an electrode for fusing the tips of the optical fibers together by arc discharge, and is disposed between the pair of fiber positioning sections 3b.

[0030] The windshield cover 6 is connected to the housing 2 so as to be openable and closable to cover the connection portion 3. Each of the side surfaces 6a of the windshield cover 6 is formed with an introduction port 6b for introducing an optical fiber into the connection portion 3, i.e., into each of the optical fiber holders 3a.

[0031] FIG. 4 is a block diagram showing a functional configuration of the fusion splicer 10. FIG. 5 is a block diagram showing a hardware configuration of the fusion splicer 10. As shown in FIG. 4, the fusion splicer 10 functionally includes a splicing unit 3, a communication unit 11, an imaging unit 12, a feature extraction unit 13, a discrimination unit 14, and a fusion control unit 15. The imaging unit 12 includes an imaging element and an observation optical unit that outputs an enlarged image of an imaged object to the imaging element. The observation optical unit includes, for example, one or more lenses. As shown in FIG. 5, the fusion splicer 10 includes a computer having hardware such as a CPU 10a, a RAM 10b, a ROM 10c, an input device 10d, an auxiliary storage device 10e, and an output device 10f as a control unit. Each function of the fusion splicer 10 is realized by these components operating according to a program or the like. These elements in the control unit are electrically connected to the splicing unit 3, the monitor 5, the wireless communication module as the communication unit 11, and the imaging unit 12 described above. The input device 10d may include a touch panel that is integral with the monitor 5.

[0032] The communication unit 11 is, for example, configured with a wireless LAN module. The communication unit 11 transmits and receives various data to and from the model creation device 20 via an information communication network 30 such as the Internet. The imaging unit 12 captures an image of a pair of optical fibers to be connected from the radial direction of the optical fibers through an observation optical unit (lens) while the pair of optical fibers face each other, and generates imaging data. The feature amount extraction unit 13 extracts a plurality of feature amounts for identifying the type of optical fiber from the imaging data obtained from the imaging unit 12. The feature amounts include brightness information in the radial direction of the optical fiber. The brightness information in the radial direction of the optical fiber includes, for example, at least one of the brightness distribution in the radial direction of the optical fiber, the outer diameter of the optical fiber, the outer diameter of the core, the ratio of the outer diameter of the core to the outer diameter of the optical fiber, the ratio of the area of ​​the core and the clad of the optical fiber, the sum of the brightness of the optical fiber, the position and number of inflection points of the brightness distribution in the cross section of the optical fiber, the brightness difference between the core part and the clad part of the optical fiber, and the width of the core part that has a specific brightness or more. The imaging data used to extract the feature amount may include data acquired while discharging the pair of optical fibers to be connected in a state where the optical fibers face each other. In this case, the feature amount may include, for example, at least one of the light intensity at a specific position and the time change of the light intensity at the specific position.

[0033] The discrimination unit 14 discriminates the type of each of the pair of optical fibers to be connected based on a plurality of feature quantities provided from the feature quantity extraction unit 13. To this end, the discrimination unit 14 stores and holds a first discrimination algorithm 14a and a second discrimination algorithm 14b for discriminating the type of optical fiber. The first discrimination algorithm 14a is predefined by a method other than machine learning based on the correlation between a plurality of feature quantities and the type of optical fiber. For example, the first discrimination algorithm 14a empirically or by testing determines a typical feature quantity threshold value according to the type of optical fiber, and discriminates the type of optical fiber based on the magnitude relationship between the feature quantity and the threshold value. As an example, in order to discriminate between a single mode fiber and a multimode fiber, a core outer diameter is used as a feature quantity. In that case, if the core outer diameter as a feature quantity is smaller than a predetermined threshold value, the fiber is determined to be a single mode fiber, and if the core outer diameter is larger than the predetermined threshold value, the fiber is determined to be a multimode fiber.

[0034] The second discrimination algorithm 14b includes a discrimination model Md for discriminating the type of optical fiber to be connected based on the imaging data of the optical fiber to be connected. The discrimination model Md is created by machine learning by the model creation device 20 using sample data showing the correspondence between a plurality of feature quantities and the type of optical fiber. The discrimination model Md discriminates the type of each of the pair of optical fibers by inputting the feature quantities obtained from the feature quantity extraction unit 13. These discrimination algorithms 14a and 14b are stored in, for example, the ROM 10c or the auxiliary storage device 10e. The discrimination unit 14 selects and adopts a discrimination result by one of the discrimination algorithms 14a and 14b by one of the following methods A, B, C, and D.

[0035] (Method A) The discrimination unit 14 adopts the discrimination result of the first discrimination algorithm 14a when the possibility of misclassification by the first discrimination algorithm 14a is low, and adopts the discrimination result of the second discrimination algorithm 14b when the possibility of misclassification by the first discrimination algorithm 14a is high. The degree of the possibility of misclassification can be determined, for example, by the magnitude of a predetermined feature included in the plurality of feature amounts and a threshold. That is, when the predetermined feature included in the plurality of feature amounts is larger than a predetermined threshold, the discrimination result by one of the discrimination algorithms 14a and 14b is adopted, and when the predetermined feature amount is smaller than the predetermined threshold, the discrimination result by the other of the discrimination algorithms 14a and 14b is adopted. In other words, when the predetermined feature amount is smaller (or larger) than a predetermined threshold, the first discrimination algorithm 14a is adopted, and when the predetermined feature amount is larger (or smaller) than the predetermined threshold, the second discrimination algorithm 14b is adopted.

[0036] The predetermined threshold is a value determined based on a comparison between the discrimination accuracy by the first discrimination algorithm 14a and the discrimination accuracy by the second discrimination algorithm 14b when the predetermined feature quantity changes. In other words, the predetermined threshold is a value of the predetermined feature quantity when the discrimination accuracy is reversed between the first discrimination algorithm 14a and the second discrimination algorithm 14b. Therefore, in a range in which the predetermined feature quantity is smaller than the predetermined threshold, the discrimination accuracy by the first discrimination algorithm 14a is higher (or lower) than the discrimination accuracy by the second discrimination algorithm 14b. In a range in which the predetermined feature quantity is larger than the predetermined threshold, the discrimination accuracy by the second discrimination algorithm 14b is higher (or lower) than the discrimination accuracy by the first discrimination algorithm 14a. Then, the discrimination unit 14 adopts the discrimination result of the discrimination algorithm 14a or 14b, which has the higher discrimination accuracy, based on the magnitude relationship between the predetermined feature quantity and the predetermined threshold. Note that the discrimination accuracy of the discrimination algorithms 14a and 14b changes over time as the fusion splicer 10 is operated. The above-mentioned predetermined threshold value is determined, for example, by sequentially calculating the discrimination accuracy of the discrimination algorithms 14a and 14b while the fusion splicer 10 is in operation.

[0037] (Method B) When the type of each of the pair of optical fibers can be discriminated by the first discrimination algorithm 14a, the discrimination unit 14 adopts the discrimination result by the second discrimination algorithm 14b. When the type of each of the pair of optical fibers cannot be discriminated by the first discrimination algorithm 14a, the discrimination unit 14 adopts the discrimination result by the second discrimination algorithm 14b. Here, "the optical fiber type can be discriminated" means that the optical fiber type corresponding to the multiple feature quantities extracted by the feature quantity extracting unit 13 exists in the first discrimination algorithm 14a. "The optical fiber type cannot be discriminated" means that the optical fiber type corresponding to the multiple feature quantities extracted by the feature quantity extracting unit 13 does not exist in the first discrimination algorithm 14a. In this method B, the discrimination unit 14 may first execute the first discrimination algorithm 14a, and when the type of each of the pair of optical fibers cannot be discriminated by the first discrimination algorithm 14a, execute the second discrimination algorithm 14b. Alternatively, the discrimination unit 14 may execute the first discrimination algorithm 14a and the second discrimination algorithm 14b in parallel.

[0038] (Method C) In this method, first, the imaging unit 12 captures the pair of optical fibers F1, F2 at least twice to generate at least two pieces of imaging data PX, PY. The feature extraction unit 13 extracts at least two feature groups consisting of a plurality of feature amounts from the at least two pieces of imaging data PX, PY. When the variation of the predetermined feature between the at least two feature groups is greater than a threshold, the discrimination unit 14 adopts the discrimination result by one of the discrimination algorithms 14a, 14b, that is, the algorithm with the smaller decrease in discrimination accuracy caused by the variation of the predetermined feature. When the variation of the predetermined feature is smaller than a threshold, the discrimination unit 14 adopts the discrimination result by one of the discrimination algorithms 14a, 14b. The predetermined feature is, for example, the outer diameter of the core. In this method, the imaging positions of the at least two pieces of imaging data PX, PY in the optical axis direction of the pair of optical fibers F1, F2 may be the same as each other or may be different from each other. The imaging data PX and PY having different imaging positions are obtained, for example, by moving the imaging unit 12 in the optical axis direction of the pair of optical fibers F1 and F2 for each imaging.

[0039] (Method D) In this method as well, first, the imaging unit 12 captures the pair of optical fibers F1, F2 at least twice to generate at least two pieces of imaging data PX, PY. The feature extraction unit 13 extracts at least two feature groups consisting of a plurality of feature amounts from the at least two pieces of imaging data PX, PY. The discrimination unit 14 executes both discrimination algorithms 14a, 14b based on the at least two feature groups. The discrimination unit 14 adopts one of the at least two discrimination results obtained by the first discrimination algorithm 14a and the at least two discrimination results obtained by the second discrimination algorithm 14b, which has a smaller variation in the discrimination result. In this method as well, the imaging positions of the at least two pieces of imaging data PX, PY in the optical axis direction of the pair of optical fibers F1, F2 may be the same as each other or may be different from each other.

[0040] The discrimination result adopted by the discrimination unit 14 is displayed on the monitor 5. If the types of the pair of optical fibers displayed on the monitor 5 are incorrect, the user inputs the correct types via the input device 10d to correct the discrimination result. In this case, the discrimination unit 14 makes an erroneous discrimination, and this correction is fed back to the discrimination accuracy of each of the discrimination algorithms 14a and 14b described above. Alternatively, the user may input the types of the pair of optical fibers via the input device 10d regardless of the discrimination result by the discrimination unit 14. In this case, the input by the user is given priority and the types of the pair of optical fibers are specified. Alternatively, one of the manufacturing conditions preset for each type of optical fiber may be selected instead of inputting the type of the corresponding optical fiber itself. In this case, the correctness of the discrimination results of each of the discrimination algorithms 14a and 14b is also fed back to the discrimination accuracy.

[0041] The fusion control unit 15 controls the operation of the splicing unit 3. That is, the fusion control unit 15 controls the contact operation of the tips of the pair of optical fibers and the arc discharge in the splicing unit 3 in response to the operation of a switch by the user. The contact operation of the tips of the pair of optical fibers includes the positioning process of the optical fibers by the fiber positioning unit 3b, that is, the control of the tip positions of each optical fiber. The control of the arc discharge includes the control of the discharge power, the discharge start timing, and the discharge end timing. Various connection conditions such as the tip position of the optical fiber and the discharge power are set in advance for each combination of the types of the pair of optical fibers, and are stored in, for example, the ROM 10c. The fusion control unit 15 selects the connection conditions according to the combination of the types of the pair of optical fibers determined by the discrimination unit 14 or input by the user. That is, the splicing unit 3 recognizes the combination of the types of the pair of optical fibers based on the discrimination result in the discrimination unit 14 or the input result by the user, and fusion splices the pair of optical fibers to each other under the connection conditions according to the combination.

[0042] The operation of the splicing unit 3 is as follows. First, as shown in FIG. 6, a user holds a pair of optical fibers F1 and F2 to be spliced ​​in the optical fiber holder 3a. At this time, the end face F1a of the optical fiber F1 and the end face F2a of the optical fiber F2 are arranged to face each other. Next, the user instructs the fusion splicer 10 to start fusion splicing. This instruction is given, for example, via a switch input. In response to this instruction, as shown in FIG. 7, the fusion control unit 15 positions the optical fibers F1 and F2 based on the positions of the end faces F1a and F2a set as the splicing condition. Thereafter, as shown in FIG. 8, the fusion control unit 15 starts an arc discharge between the pair of discharge electrodes 3c.

[0043] Immediately after the arc discharge starts, the end faces F1a, F2a are separated from each other. This arc discharge corresponds to a preliminary discharge for softening the end faces F1a, F2a before fusing. When the arc discharge starts, the fusion control unit 15 controls the position of the fiber positioning unit 3b to bring the end faces F1a, F2a closer to each other and bring them into contact with each other. The fusion control unit 15 then continues the arc discharge to perform the main discharge. This causes the end faces F1a, F2a to further soften and be fused to each other.

[0044] In this embodiment, the connection conditions include at least one of the positions of the end faces F1a, F2a before the start of discharge, the distance between the end faces F1a, F2a before the start of discharge, the preliminary discharge time, the main discharge time, the amount of pushing after the end faces F1a, F2a come into contact with each other, the amount of pulling back after the end faces F1a, F2a are pushed back, the preliminary discharge power, the main discharge power, and the discharge power during pulling back.

[0045] The position of each end face F1a, F2a before the start of discharge refers to the position of each end face F1a, F2a based on the line connecting the central axes of the pair of discharge electrodes 3c, i.e., the discharge central axis, at the start of the preliminary discharge. The distance between the discharge central axis and each end face F1a, F2a changes depending on these end face positions. This increases or decreases the amount of heat, i.e., the amount of melting. In addition, the time required for the end faces F1a, F2a to move until they come into contact with each other changes. The distance between each end face F1a, F2a before the start of discharge refers to the distance between the end faces F1a, F2a in the state shown in FIG. 7, i.e., at the start of the preliminary discharge. The time required for the end faces F1a, F2a to move until they come into contact with each other changes depending on this distance. The preliminary discharge time refers to the time from the start of arc discharge in the state shown in FIG. 7 to the start of relative movement of the optical fibers F1, F2 to bring the end faces F1a, F2a into contact with each other. The main discharge time refers to the time from when the end faces F1a, F2a come into contact with each other until the arc discharge is terminated, in other words, the time until the application of voltage to the pair of discharge electrodes 3c is stopped. The preliminary discharge and the main discharge are performed successively in time. The pushing amount after the end faces F1a, F2a come into contact with each other refers to the movement distance of the optical fiber holder 3a when the optical fibers F1, F2 are moved relatively to move the optical fibers F1, F2 so that the end faces F1a, F2a come into contact with each other, and then the optical fibers F1, F2 are further moved relatively in the same direction during discharge. The pulling back amount after the end faces F1a, F2a are pushed into each other refers to the movement distance of each optical fiber holder 3a when the optical fibers F1, F2 are moved relatively in the opposite direction during discharge, that is, in the direction in which the end faces F1a, F2a move away from each other, after the end faces F1a, F2a come into contact with each other and then the end faces F1a, F2a are further pushed in. The preliminary discharge power refers to the arc discharge power during the period from when the arc discharge starts in the state shown in FIG. 7 until when the optical fibers F1, F2 start to move relative to each other to bring the end faces F1a, F2a into contact with each other.

[0046] 9 is a view of the end face F2a of one optical fiber F2 as viewed from the front, i.e., from the optical axis direction. Arrows MSX and MSY in the figure indicate the imaging directions of the imaging units 12. That is, in this example, at least two imaging units 12 are installed, and the two imaging units 12 capture images of the end faces F1a, F2a from radial directions of the optical fibers F1, F2 that are perpendicular to each other. A light source for illuminating the optical fibers F1, F2 is disposed at a position facing the imaging units 12 across the optical fibers F1, F2. The light source is, for example, a light emitting diode.

[0047] Fig. 10 is a diagram showing a schematic diagram of imaging data PX obtained by the imaging unit 12 imaging from the direction MSX and imaging data PY obtained by the imaging unit 12 imaging from the direction MSY. As shown in Fig. 10, in these imaging data PX and PY, the positions and shapes of the optical fibers F1 and F2 are confirmed by the outlines of the core CR and the cladding CL. The core CR is brightened by the illumination light from the light source. The cladding CL is darkened by the refraction of the illumination light from the light source.

[0048] Fig. 11 is a block diagram showing a functional configuration of the model creating device 20. Fig. 12 is a block diagram showing a hardware configuration of the model creating device 20. As shown in Fig. 11, the model creating device 20 functionally includes a communication unit 21 and a discrimination model creating unit 22. As shown in Fig. 12, the model creating device 20 includes a computer including hardware such as a CPU 20a, a RAM 20b, a ROM 20c, an input device 20d, a communication module 20e, an auxiliary storage device 20f, and an output device 20g. Each function of the model creating device 20 is realized by operating these components according to a program or the like.

[0049] The communication unit 21 shown in FIG. 11 communicates with a plurality of fusion splicers 10 via an information and communication network 30 (see FIG. 1) such as the Internet. The communication unit 21 receives, from a plurality of fusion splicers 10, the feature values ​​extracted from the imaging data PX, PY and information on the types of the optical fibers F1, F2 via the information and communication network 30. The communication unit 21 may receive the imaging data PX, PY itself instead of the feature values ​​extracted from the imaging data PX, PY. In that case, the model creating device 20 extracts the feature values ​​from the imaging data PX, PY. The information on the types of the optical fibers F1, F2 may be only information input by the user. In other words, the communication unit 21 receives, from each fusion splicer 10, the information on the types of the optical fibers F1, F2 input by the user and the feature values ​​extracted from the imaging data PX, PY of the optical fibers F1, F2 or the imaging data itself. The information on the types of the optical fibers F1, F2 input by the user includes a case where one of the manufacturing conditions preset for each type of the optical fibers F1, F2 is selected instead of inputting the types of the optical fibers F1, F2 themselves. The communication unit 21 provides the received information to the discrimination model creation unit 22 as sample data Da indicating the correspondence between the feature amounts obtained from the imaging data PX, PY of the optical fibers F1, F2 and the types of the optical fibers F1, F2.

[0050] The discrimination model creation unit 22 performs machine learning using the sample data Da provided by the communication unit 21. Then, the discrimination model creation unit 22 creates a discrimination model Md for discriminating the types of the optical fibers F1, F2 based on the image data PX, PY. The machine learning is preferably deep learning. As the machine learning technique, various techniques included in so-called supervised learning, such as neural networks and support vector machines, can be applied. The discrimination model creation unit 22 continuously performs machine learning using a huge amount of sample data Da obtained from a large number of fusion splicers 10 in operation, and improves the accuracy of the discrimination model Md. The discrimination model creation unit 22 of this embodiment classifies a plurality of fusion splicers 10 into two or more groups that are estimated to have similar tendencies in the image data PX, PY. Then, the discrimination model creation unit 22 collects sample data Da for each group and creates a discrimination model Md for each group. Creating a discrimination model Md for each group means performing machine learning using only sample data Da obtained from multiple fusion splicers 10 belonging to a certain group, and providing the created discrimination model Md only to the fusion splicers 10 belonging to that group.

[0051] The two or more groups estimated to have similar tendencies of the imaging data PX and PY are classified based on, for example, similarity of the inspection results of the fusion splicer 10, similarity of the inspection conditions of the fusion splicer 10, similarity of the manufacturer and manufacturing date and time of the fusion splicer 10, similarity of the manufacturer and manufacturing date and time of the imaging unit 12, similarity of the environmental conditions at the place where the fusion splicer 10 is used, similarity of the deterioration state of the fusion splicer 10, and similarity of the type of optical fiber to be spliced. The similarity of the inspection results of the fusion splicer 10 is, for example, similarity of the luminance distribution in the imaging data PX and PY. The similarity of the inspection conditions of the fusion splicer 10 is similarity of the environmental conditions when an image of a reference optical fiber is captured during the inspection of the fusion splicer 10, for example, similarity of at least one of the temperature (air temperature), humidity, and air pressure when an image of the reference optical fiber is captured. The similarity of the environmental conditions at the location where the fusion splicer 10 is used is, for example, similarity in at least one of the temperature, humidity, and atmospheric pressure at the location where the fusion splicer 10 is used. The similarity of the deterioration state of the fusion splicer 10 is, for example, similarity in at least one of the number of discharges of the fusion splicer 10, the connection frequency, the degree of dirt on the discharge electrode 3c, the dimming state of the light source that illuminates the optical fiber from the opposite side of the imaging unit 12, the degree of dirt on the observation optical unit of the imaging unit 12, and the equipment diagnosis results.

[0052] The discrimination model Md thus created by collecting the sample data Da for each group is transmitted to and provided to the fusion splicers 10 belonging to the corresponding group via the communication unit 21. The second discrimination algorithm 14b of the discrimination unit 14 of each fusion splicer 10 obtains the discrimination model Md corresponding to the group to which the fusion splicer 10 belongs from the model creation device 20, and discriminates the type of each of the pair of optical fibers F1, F2.

[0053] The sample data Da provided for the machine learning of the discrimination model creation unit 22 includes both sample data when the types of the pair of optical fibers F1, F2 can be discriminated by the first discrimination algorithm 14a, and sample data when the types of the pair of optical fibers F1, F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a. Alternatively, the sample data Da provided for the machine learning of the discrimination model creation unit 22 may include only sample data when the types of the pair of optical fibers F1, F2 can be discriminated by the first discrimination algorithm 14a. In this case, the discrimination unit 14 of each fusion splicer 10 performs additional machine learning using its own sample data when the types of the pair of optical fibers F1, F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a, and improves the discrimination model Md held by itself.

[0054] The sample data Da provided for the machine learning of the discrimination model creation unit 22 may include both sample data when the types of the pair of optical fibers F1, F2 can be discriminated by the first discrimination algorithm 14a, and sample data when the types of the pair of optical fibers F1, F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a. In this case, the discrimination unit 14 of each fusion splicer 10 may perform additional machine learning using its own sample data when the types of the pair of optical fibers F1, F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a, and improve the discrimination model Md held by itself. However, the sample data used for the additional machine learning does not include the sample data Da provided to the model creation device 20. In this case, the sample data that cannot be discriminated or is misdiscriminated for a certain period or number from the shipment of the fusion splicer 10 may be provided for the machine learning of the discrimination model creation unit 22. Then, the sample data that cannot be discriminated or is misdiscriminated after that may be provided for the additional machine learning in the discrimination unit 14 of each fusion splicer 10.

[0055] FIG. 13 is a flowchart showing a method for fusion splicing optical fibers according to this embodiment. This method can be preferably realized by using the above-mentioned fusion splicing system 1A. First, in a model creation step ST1, machine learning is performed using sample data Da showing a correspondence between a plurality of feature quantities obtained from imaging data of the optical fiber and the type of the optical fiber. Then, a discrimination model Md is created for discriminating the type of the optical fibers F1, F2 to be spliced ​​based on the imaging data PX, PY of the optical fibers F1, F2. In this model creation step ST1, a plurality of fusion splicers 10 are classified into two or more groups that are estimated to have similar tendencies in the imaging data PX, PY. Then, sample data Da is collected for each group to create a discrimination model Md.

[0056] Next, in the imaging step ST2, the pair of optical fibers F1, F2 are imaged to generate imaging data PX, PY. Then, in the discrimination step ST3, the type of each of the pair of optical fibers F1, F2 is discriminated based on a plurality of feature amounts obtained from the imaging data PX, PY generated in the imaging step ST2. In this discrimination step ST3, a discrimination result by one of the discrimination algorithms 14a, 14b for discriminating the type of the optical fibers F1, F2 is adopted. As described above, the first discrimination algorithm 14a is predefined by a method other than machine learning based on the correlation between a plurality of feature amounts obtained from the imaging data PX, PY of the optical fibers F1, F2 and the type of the optical fibers F1, F2. The second discrimination algorithm 14b includes a discrimination model Md created in the model creation step ST1. The discrimination model Md corresponds to a group to which the fusion splicer 10 performing the discrimination step ST3 belongs. Next, in a splicing step ST4, the pair of optical fibers F1, F2 are fusion-spliced ​​to each other under splicing conditions according to the combination of the types of the pair of optical fibers F1, F2 based on the results of the discrimination in the discrimination step ST3.

[0057] As shown in FIG. 14, a step ST5 may be added to the above method. In the step ST5, the discrimination accuracy is measured for each of the discrimination algorithms 14a and 14b. This step ST5 is performed before the imaging step ST2 or the discrimination step ST3. Specifically, first, two or more optical fibers whose types are known are imaged by the imaging unit 12 to generate image data PX and PY. Next, the feature extraction unit 13 extracts a plurality of feature amounts from the image data PX and PY. Then, based on the plurality of feature amounts, the discrimination unit 14 discriminates the types of the two or more optical fibers by both the discrimination algorithms 14a and 14b, and the discrimination results are compared with the known types to obtain the discrimination accuracy of each of the discrimination algorithms 14a and 14b. In the discrimination step ST3, the discrimination algorithm 14a or 14b that has a higher discrimination accuracy in step ST5 is adopted.

[0058] The effects obtained by the fusion splicing system 1A, the fusion splicer 10, and the fusion splicing method of the present embodiment described above will be described. In the present embodiment, the types of the optical fibers F1, F2 are discriminated using the discrimination algorithms 14a and 14b. The first discrimination algorithm 14a is predefined by a method other than machine learning based on the correlation between a plurality of feature amounts obtained from the imaging data of the optical fibers F1, F2 and the types of the optical fibers F1, F2, and can be expected to have discrimination accuracy equivalent to that of the conventional method. The second discrimination algorithm 14b includes a discrimination model Md created by machine learning using sample data Da showing the correspondence between a plurality of feature amounts and the types of the optical fibers F1, F2. Therefore, it is possible to expect high-precision discrimination based on machine learning for the types of the optical fibers F1, F2 that are difficult to discriminate or tend to be erroneously discriminated by the first discrimination algorithm 14a. Therefore, according to the present embodiment, by adopting the discrimination result by either the discrimination algorithm 14a or 14b, it is possible to improve the discrimination accuracy of the types of the optical fibers F1, F2 compared to the conventional method.

[0059] As described above, the machine learning may be deep learning, in which case the accuracy of distinguishing the types of the optical fibers F1 and F2 can be further improved.

[0060] As described above, the discrimination unit 14 (in the discrimination step ST3) may adopt the discrimination result by one of the discrimination algorithms 14a and 14b when a predetermined feature included in the plurality of feature amounts is greater than a threshold value, and may adopt the discrimination result by the other of the discrimination algorithms 14a and 14b when the predetermined feature amount is smaller than the threshold value. For example, such a method may be used to easily select which discrimination result of the discrimination algorithms 14a and 14b to adopt. In this case, the threshold value may be a value determined based on a comparison between the discrimination accuracy by the first discrimination algorithm 14a and the discrimination accuracy by the second discrimination algorithm 14b when the predetermined feature amount changes. This may further improve the discrimination accuracy of the types of the optical fibers F1 and F2.

[0061] As described above, the discrimination unit 14 (in the discrimination step ST3) may adopt the discrimination result when the types of the optical fibers F1 and F2 can be discriminated by the first discrimination algorithm 14a, and may adopt the discrimination result by the second discrimination algorithm 14b when the types of the optical fibers F1 and F2 cannot be discriminated by the first discrimination algorithm 14a. For example, by using such a method, the discrimination accuracy of the types of the optical fibers F1 and F2 can be improved. In this case, the discrimination unit 14 (in the discrimination step ST3) may first execute the first discrimination algorithm 14a, and execute the second discrimination algorithm 14b when the types of the optical fibers F1 and F2 cannot be discriminated by the first discrimination algorithm 14a. This can reduce the amount of calculations (in the discrimination step ST3) of the discrimination unit 14. Alternatively, the discrimination unit 14 (in the discrimination step ST3) may execute the first discrimination algorithm 14a and the second discrimination algorithm 14b in parallel. This can shorten the time required to obtain the final discrimination result.

[0062] As described above, the imaging unit 12 (in the imaging step ST2) may capture images of the pair of optical fibers F1, F2 at least twice to generate at least two sets of imaging data PX, PY. Then, the discrimination unit 14 (in the discrimination step ST3) may adopt a discrimination result by one of the discrimination algorithms 14a, 14b when the variation of a predetermined feature amount between at least two feature amount groups consisting of a plurality of feature amounts obtained from at least two sets of imaging data PX, PY is greater than a threshold value. The discrimination unit 14 (in the discrimination step ST3) may adopt a discrimination result by one of the discrimination algorithms 14a, 14b when the variation of the predetermined feature amount is smaller than a threshold value. This can further improve the discrimination accuracy of the types of the optical fibers F1, F2.

[0063] As described above, the imaging unit (in the imaging step ST2) may capture images of the pair of optical fibers F1, F2 at least twice to generate at least two sets of image data PX, PY. Then, the discrimination unit 14 (in the discrimination step ST3) may execute the discrimination algorithms 14a, 14b based on at least two feature groups consisting of a plurality of feature amounts obtained from the at least two sets of image data PX, PY. The discrimination unit 14 (in the discrimination step ST3) may adopt the discrimination result with the smaller variation between the at least two discrimination results by the first discrimination algorithm 14a and the at least two discrimination results by the second discrimination algorithm 14b. This can further improve the discrimination accuracy of the types of the optical fibers F1, F2.

[0064] As described above, the model creation device 20 may classify the multiple fusion splicers 10 into two or more groups that are estimated to have similar tendencies of the image data PX, PY, and create a discrimination model Md for each group. Then, the second discrimination algorithm 14b of the discrimination unit 14 of each fusion splicer 10 may obtain a discrimination model Md corresponding to the group to which each fusion splicer 10 belongs from the model creation device 20. This allows machine learning to be limited to groups in which the tendencies of the image data PX, PY are similar, for example, groups in which the mechanical and structural variations of the fusion splicers 10 are small, or groups in which the mechanical and structural variations of the imaging unit 12 are small. Therefore, the accuracy of discrimination of the types of the optical fibers F1, F2 by the second discrimination algorithm 14b can be further improved.

[0065] As described above, the sample data Da provided for the machine learning of the model creation device 20 may include both sample data when the types of the optical fibers F1 and F2 can be discriminated by the first discrimination algorithm 14a, and sample data when the types of the optical fibers F1 and F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a. In this case, the types of the optical fibers F1 and F2 that the first discrimination algorithm 14a has difficulty discriminating can be included in the machine learning of the model creation device 20, and the overall accuracy of discriminating the types of the optical fibers F1 and F2 can be improved.

[0066] As described above, the sample data Da provided for machine learning by the model creating device 20 includes only sample data when the types of the optical fibers F1, F2 can be discriminated by the first discrimination algorithm 14a, and the discrimination unit 14 of each fusion splicer 10 may perform machine learning using its own sample data when the types of the optical fibers F1, F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a, thereby improving the discrimination model Md. In this case, the discrimination accuracy of the second discrimination algorithm 14b can be improved for each fusion splicer 10 for the types of the optical fibers F1, F2 that cannot be discriminated by the first discrimination algorithm 14a due to mechanical and structural variations of each fusion splicer 10, for example, mechanical and structural variations of the imaging unit 12.

[0067] As described above, the sample data Da provided for the machine learning of the model creation device 20 may include sample data when the types of the optical fibers F1 and F2 can be discriminated by the first discrimination algorithm 14a, and sample data when the types of the optical fibers F1 and F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a. The discrimination unit 14 of each fusion splicer 10 may perform machine learning using its own sample data when the types of the optical fibers F1 and F2 cannot be discriminated or are misdiscriminated by the first discrimination algorithm 14a, and improve the discrimination model Md. However, the sample data provided to the model creation device 20 is excluded. In this case, the types of the optical fibers F1 and F2 that the first discrimination algorithm 14a has difficulty discriminating can be included in the machine learning of the model creation device 20. In addition, for the types of the optical fibers F1, F2 that cannot be identified by the first identification algorithm 14a due to mechanical and structural variations of each fusion splicer 10, such as mechanical and structural variations of the imaging unit 12, the identification accuracy of the second identification algorithm 14b can be improved for each fusion splicer 10. Therefore, the overall identification accuracy of the types of the optical fibers F1, F2 can be further improved.

[0068] As described above, two or more optical fibers whose types are known may be imaged to generate image data PX, PY, and the types of the two or more optical fibers may be discriminated by the discrimination algorithms 14a, 14b based on a plurality of feature quantities obtained from the image data PX, PY. Then, the discrimination algorithm 14a, 14b that has a higher discrimination accuracy may be adopted in the discrimination step ST3. This can further improve the discrimination accuracy of the types of the optical fibers F1, F2.

[0069] The fusion splicer, the fusion splicing system, and the method for fusion splicing optical fibers according to the present disclosure are not limited to the above-mentioned embodiment, and various other modifications are possible. For example, in the fusion splicer 10 of the above-mentioned embodiment, the criteria for adopting the discrimination result of the second discrimination algorithm 14b are exemplified as a case where discrimination is not possible using the first discrimination algorithm 14a and a case where the discrimination result using the first discrimination algorithm 14a is highly likely to be erroneous. Other criteria may be used as long as they can improve the overall discrimination accuracy.

[0070] In the fusion splicer 10 of the above embodiment, either of the discrimination algorithms 14a and 14b is used to fusion splice a pair of optical fibers to each other under connection conditions according to the combination of the types of the pair of optical fibers. In addition, either of the discrimination algorithms 14a and 14b may be used, for example, to recognize the position of the core when aligning the pair of optical fibers after discriminating the types of the pair of optical fibers. [Explanation of symbols]

[0071] 1A…Fusion splicing system 2. Housing 3…Connection 3a...Optical fiber holder 3b…Fiber positioning part 3c…discharge electrode 4…heater 5. Monitor 6…Windshield cover 6a…side 6b... Entry point 10…Fusion splicer 10a…CPU 10b…RAM 10c…ROM 10d…Input device 10e…Auxiliary storage device 10f…Output device 11…Communications Department 12...Imaging unit 13...Feature extraction unit 14...Discrimination section 14a…First discrimination algorithm 14b…Second discrimination algorithm 15...Fusing control section 20...Model making equipment 20a…CPU 20b…RAM 20c…ROM 20d…Input device 20e…Communication module 20f…Auxiliary storage device 20g…Output device 21…Communications Department 22…Discrimination model creation section 30…Information and communication network CL: Clad CR…Core Da…Sample data F1, F2: Optical fiber F1a,F2a…end face Md…Discrimination model MSX,MSY…direction PX, PY: Imaging data ST1: Model creation process ST2: Imaging process ST3…Discrimination process ST4…Connection process

Claims

1. an imaging unit that captures an image of the pair of optical fibers to generate imaging data; a discrimination unit that discriminates the type of each of the pair of optical fibers based on a plurality of feature amounts obtained from imaging data provided by the imaging unit, the discrimination unit having first and second discrimination algorithms for discriminating the type of optical fiber, and adopting a discrimination result by either the first or second discrimination algorithm, the first discrimination algorithm being defined in advance by a method other than machine learning based on a correlation between a plurality of feature amounts obtained from the imaging data of the optical fiber and the type of the optical fiber from which the feature amounts were obtained, the second discrimination algorithm including a discrimination model for discriminating the type of the optical fiber to be connected based on the imaging data of the optical fiber to be connected, the discrimination model being created by machine learning using sample data showing a correspondence relationship between a plurality of feature amounts obtained from the imaging data of the optical fiber and the type of the optical fiber from which the feature amounts were obtained; a splicing unit that fusion-splices the pair of optical fibers to each other under connection conditions corresponding to a combination of the types of the pair of optical fibers based on a result of the discrimination by the discriminating unit; Equipped with The discrimination unit is a first operation of adopting a discrimination result obtained by one of the first and second discrimination algorithms when a predetermined feature amount included in the plurality of feature amounts is greater than a threshold value, and adopting a discrimination result obtained by the other of the first and second discrimination algorithms when the predetermined feature amount is smaller than the threshold value; a second operation of adopting a discrimination result when the types of the respective optical fibers of the pair can be discriminated by the first discrimination algorithm, and adopting a discrimination result by the second discrimination algorithm when the types of the respective optical fibers of the pair cannot be discriminated by the first discrimination algorithm; a third operation of obtaining at least two feature amount groups each consisting of the plurality of feature amounts from each of at least two pieces of image data generated by the imaging unit by imaging the pair of optical fibers at least twice, and adopting a discrimination result according to one of the first and second discrimination algorithms, which has a smaller decrease in discrimination accuracy due to the variation in the predetermined feature amount, when a variation in a predetermined feature amount between the at least two feature amount groups is larger than a threshold value; and a fourth operation of obtaining at least two feature amount groups consisting of the plurality of feature amounts from each of at least two pieces of imaging data generated by the imaging unit by imaging the pair of optical fibers at least twice, executing the first and second discrimination algorithms based on the at least two feature amount groups, and adopting one of at least two discrimination results corresponding to the at least two feature amount groups, respectively, obtained by the first discrimination algorithm and at least two discrimination results corresponding to the at least two feature amount groups, respectively, obtained by the second discrimination algorithm, which has a smaller variation between the at least two discrimination results; A fusion splicer that performs at least one of the following operations.

2. The fusion splicer of claim 1 , wherein the machine learning is deep learning.

3. 3. The fusion splicer according to claim 1, wherein the threshold value is a value determined based on a comparison between a discrimination accuracy according to the first discrimination algorithm and a discrimination accuracy according to the second discrimination algorithm when the predetermined feature amount changes.

4. 4. The fusion splicer according to claim 1, wherein in the second operation, the discrimination unit first executes the first discrimination algorithm, and executes the second discrimination algorithm when the types of the pair of optical fibers cannot be discriminated by the first discrimination algorithm.

5. The fusion splicer according to claim 1 , wherein the discrimination unit executes the first discrimination algorithm and the second discrimination algorithm in parallel in the second operation.

6. A fusion splicer described in any one of claims 1 to 5, wherein, in at least one of the third operation and the fourth operation, the imaging positions of the at least two sets of imaging data in the optical axis direction of the pair of optical fibers are the same as each other.

7. A fusion splicer as described in any one of claims 1 to 5, wherein, in at least one of the third operation and the fourth operation, the imaging positions of the at least two sets of imaging data in the optical axis direction of the pair of optical fibers are different from each other.

8. A plurality of fusion splicers, each of which is the fusion splicer according to any one of claims 1 to 7; a model creation device that collects the sample data from the plurality of fusion splicers, performs the machine learning to create the discrimination model, and provides the discrimination model to the plurality of fusion splicers; A fusion splicing system comprising:

9. the model creation device classifies the plurality of fusion splicers into two or more groups that are estimated to have similar tendencies in the image data, and creates the discrimination model for each group; The fusion splicing system according to claim 8 , wherein the second discrimination algorithm of the discrimination section of each fusion splicer obtains the discrimination model corresponding to the group to which each fusion splicer belongs from the model creation device.

10. 10. The fusion splicing system according to claim 8 or 9, wherein the sample data provided to the machine learning of the model creation device includes both the sample data when the types of the respective optical fibers of the pair of optical fibers can be discriminated by the first discrimination algorithm, and the sample data when the types of the respective optical fibers of the pair of optical fibers cannot be discriminated or are erroneously discriminated by the first discrimination algorithm.

11. the sample data provided to the machine learning of the model creation device includes only the sample data when the types of the pair of optical fibers can be discriminated by the first discrimination algorithm; 10. The fusion splicing system according to claim 8 or 9, wherein the discrimination unit of each fusion splicer performs the machine learning using its own sample data when the types of the pair of optical fibers cannot be discriminated or are misdiscriminated by the first discrimination algorithm, to improve the discrimination model.

12. The sample data provided to the machine learning of the model creation device includes both the sample data when the types of the pair of optical fibers can be discriminated by the first discrimination algorithm, and the sample data when the types of the pair of optical fibers cannot be discriminated or are erroneously discriminated by the first discrimination algorithm, 10. The fusion splicing system according to claim 8 or 9, wherein the discrimination unit of each fusion splicer performs the machine learning using its own sample data (excluding the sample data provided to the model creation device) when the types of each of the pair of optical fibers cannot be discriminated by the first discrimination algorithm, to improve the discrimination model.

13. imaging the pair of optical fibers to generate imaging data; a step of discriminating the type of each of the pair of optical fibers based on a plurality of feature amounts obtained from the imaging data acquired in the generating step, employing a discrimination result by either a first or second discrimination algorithm for discriminating the type of optical fiber, the first discrimination algorithm being defined in advance by a method other than machine learning based on a correlation between a plurality of feature amounts obtained from the imaging data of the optical fiber and the type of the optical fiber from which the feature amounts were obtained, the second discrimination algorithm including a discrimination model for discriminating the type of the optical fiber to be connected based on the imaging data of the optical fiber to be connected, the discrimination model being created by machine learning using sample data showing a correspondence relationship between a plurality of feature amounts obtained from the imaging data of the optical fiber and the type of the optical fiber from which the feature amounts were obtained; a step of fusion-splicing the pair of optical fibers to each other under a splicing condition corresponding to a combination of the types of the pair of optical fibers based on a result of the discrimination in the discriminating step; Including, The determining step includes: adopting a discrimination result obtained by one of the first and second discrimination algorithms when a predetermined feature value included in the plurality of feature values ​​is greater than a threshold value, and adopting a discrimination result obtained by the other of the first and second discrimination algorithms when the predetermined feature value is smaller than the threshold value; adopting a discrimination result when the types of the respective optical fibers of the pair can be discriminated by the first discrimination algorithm, and adopting a discrimination result by the second discrimination algorithm when the types of the respective optical fibers of the pair cannot be discriminated by the first discrimination algorithm; obtaining at least two feature amount groups each including the plurality of feature amounts from each of at least two pieces of image data generated by imaging the pair of optical fibers at least twice, and adopting a discrimination result according to one of the first and second discrimination algorithms, which has a smaller decrease in discrimination accuracy due to the variation in the predetermined feature amount, when a variation in a predetermined feature amount between the at least two feature amount groups is larger than a threshold value; and obtaining at least two feature amount groups each consisting of the plurality of feature amounts from at least two sets of image data generated by imaging the pair of optical fibers at least twice, executing the first and second discrimination algorithms based on the at least two feature amount groups, and adopting one of at least two discrimination results obtained by the first discrimination algorithm and at least two discrimination results obtained by the second discrimination algorithm and corresponding to the at least two feature amount groups, the one having a smaller variation between the at least two discrimination results; 1. A method for fusion splicing optical fibers, comprising at least one of the steps of:

14. 14. The method for fusion splicing optical fibers according to claim 13, further comprising the steps of: capturing images of two or more optical fibers whose types are known to generate imaging data; discriminating the types of the two or more optical fibers using the first and second discrimination algorithms based on a plurality of feature quantities obtained from the imaging data; and adopting one of the first and second discrimination algorithms having a higher discrimination accuracy in the discrimination step.

Images