Fiber detection and alignment system

12223,673

18/737846

June 07, 2024

A system includes an imaging system and a fiber detection and alignment system. The fiber detection and alignment system includes one or more neural networks trained to detect end faces of one or more fibers in an image and to predict rotation angle and direction of rotation for each fiber that will align an axis of the fiber to associated reference keys. In operation, the imaging system generates an image of an end face of one or more polarization-maintaining (PM) fibers to be aligned. The fiber detection and alignment system generates fiber detection information, such as bounding boxes around end faces of each fiber in the image, and fiber alignment information, such as an alignment instruction indicating a rotation angle and direction of rotation that will align fast or slow axes of each fiber to associated reference keys. The novel system provides a scalable technique to automatically align PM fiber.

Neptec OS, Inc. (Fremont, CA, US)

Neptec OS, Inc. (Fremont, CA, US)

1. A method comprising: detecting an end face of a polarization-maintaining (PM) optical fiber in an image; and generating a rotation angle and direction from the image, wherein the rotation angle and direction aligns a fast or slow axis of the PM optical fiber with a reference key, wherein a neural network system generates bounding box coordinates that identify the end face of the PM optical fiber in the image, wherein the neural network system generates the rotation angle and direction that align the fast or slow axis of the PM optical fiber with the reference key, and wherein the neural network system is trained with labeled images having bounding box information that identifies fiber end faces and is also trained with labeled images having rotation angle and direction information that indicates how to align fiber end faces.

2. The method of claim 1 , wherein the PM optical fiber is part of a connector or a multi-fiber polarization-maintaining ribbon.

3. The method of claim 1 , further comprising: using a fiber rotator to rotate the PM optical fiber detected in the image by the rotation angle generated from the image.

4. The method of claim 1 , further comprising: receiving the image from a microscope of a fiber alignment station; and presenting the image on a display of the fiber alignment station.

5. The method of claim 1 , further comprising: displaying a bounding box overlaid above the end face of the PM optical fiber in the image; and displaying the rotation angle and direction.

6. The method of claim 1 , further comprising: generating a confidence score associated with the bounding box information.

7. A system comprising: an imaging system configured to generate an image having an end face of a polarization-maintaining (PM) optical fiber; and a fiber detection and alignment system, wherein the fiber detection and alignment system generates fiber detection information and fiber alignment information from the image that aligns a fast or slow axis of the PM optical fiber to a reference key, and wherein the fiber detection and alignment system comprises: a fiber detection neural network system that generates bounding box coordinates that identify the end face of the PM optical fiber in the image, wherein the fiber detection neural network system is trained with labeled images having bounding box information identifying fiber; and a fiber alignment neural network system that predicts the fiber alignment information, wherein the fiber alignment neural network system is trained with labeled images having rotation angle and direction information that aligns PM optical fiber.

8. The system of claim 7 , wherein the fiber detection information includes a bounding box around the end face of the PM optical fiber, and wherein the fiber alignment information includes a rotation angle and direction indicating how to align the fast or slow axis of the PM optical fiber to the reference key.

9. The system of claim 8 , wherein the fiber alignment information further comprises: a confidence score associated with the bounding box.

10. The system of claim 7 , further comprising: a fiber rotator, wherein the fiber rotator rotates the PM optical fiber in accordance with the fiber detection information and fiber alignment information generated by the fiber detection and alignment system.

11. The system of claim 7 , wherein parts of the fiber detection and alignment system are distributed over a network and run one or more processors.

12. The system of claim 7 , further comprising: a display, wherein the display is configured to display the image and the fiber detection information and fiber alignment information overlaid above the image; a processor; and a memory.

13. A fiber detection and alignment system comprising: an imaging system configured to generate an image having an end face of a polarization-maintaining (PM) optical fiber; and means for generating fiber detection information from the image and fiber alignment information from the image that aligns a fast or slow axis of the PM optical fiber to a reference key, wherein the means includes a neural network system that generates bounding box coordinates that identify the end face of the PM optical fiber in the image and predicts the rotation angle and direction that align the fast or slow axis of the PM optical fiber with the reference key, and wherein the neural network system is trained with labeled images having bounding box information that identifies fiber end faces and is also trained with labeled images having rotation angle and direction information that indicates how to align fiber end faces.

14. The fiber detection and alignment system of claim 13 , wherein the means comprises a fiber detection and alignment system, a memory, and a processor.

15. The method of claim 1 , further comprising: detecting a height and a width of the image; and transforming the image to a square shape by cropping a larger of the height or width of the image to match a shorter of the height or width of the image.

16. The method of claim 6 , further comprising: comparing the confidence score to a configurable threshold.

17. The method of claim 1 , wherein in training the neural network system, parameters of the neural network system are updated using sine and cosine values of a doubled rotation angle.

18. The method of claim 1 , further comprising: detecting a second end face of a second PM optical fiber in the image; generating a second rotation angle and direction from the image, wherein the second rotation angle and direction aligns the second fast or slow axis of the PM optical fiber with a second reference key; overlaying a first bounding box over the first PM optical fiber in the image along with the rotation angle and direction generated for the first PM optical fiber; and overlaying a second bounding box over the second PM optical fiber in the image along with the second rotation angle and direction generated for the second PM optical fiber.

19. The method of claim 1 , wherein the neural network system includes one or more neural networks realized as a residual network (ResNet), a YOLO (You Only Look Once) network, a convolutional neural network, or a pretrained neural network.

20. The system of claim 9 , wherein the confidence score is compared to a configurable threshold.

2003/0165283 September 2003 Huang
2005/0254754 November 2005 Huang
2012/0172700 July 2012 Krishnan
2014/0363133 December 2014 Schwarzenbach
2015/0131881 May 2015 Gnanamani
2015/0238148 August 2015 Georgescu
2021/0056293 February 2021 Yin
2021/0150230 May 2021 Smolyanskiy
2022/0301258 September 2022 Song

edspgr.12223673