Optical fiber coil defect detection and type identification method, system, medium and equipment

Abstract

The invention discloses an optical fiber coil defect detection and type identification method and system based on deep learning, a medium, equipment and a program, and belongs to the technical field of deep learning target detection. The method comprises the steps that optical fiber coil data are collected and preprocessed, and an optical fiber coil image data set is obtained; and inputting the optical fiber coil data set into a pre-trained target detector based on deep learning to carry out defect detection and type identification so as to obtain a defect detection and type identification result. According to the method, the deep learning technology is applied to defect detection and type identification of the guidance optical fiber coil, high-accuracy identification of tiny defects can be achieved, the automation level of the detection process can be remarkably improved, and the labor cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of deep learning target detection technology, and specifically to a method, system, medium, equipment and program for optical fiber package defect detection and type identification based on deep learning. Background Art

[0002] Fiber optic technology, a rapidly developing technology following the development of cable technology in the 1970s, uses optical fibers to sense or transmit external signals. Due to its low loss, wide transmission bandwidth, high capacity, resistance to electromagnetic interference, excellent confidentiality, and nuclear radiation, as well as its light weight and compact size, fiber optic technology has achieved rapid development in fields such as communications. Fiber optic guidance technology has become a key approach to extending the range and improving the accuracy of wire-guided missiles. This technology involves transmitting image data from a weapon's image sensor to an operator control station via a link after vertically launching a missile a certain distance from a launch platform that cannot see the target. Control data is then transmitted over the link for manual looping. Due to its advantages such as high transmission capacity, low latency, and resistance to electromagnetic interference, fiber optic guidance technology has a wide range of applications and has become a key means of remote weapon control. Fiber optic guided missiles currently in service include Japan's Type 96 Heavy Matt missile and Israel's Spike series of missiles.

[0003] Guidance fiber serves as the bidirectional data transmission channel between the launch platform and the missile. Its reliable release is crucial to the missile's ability to achieve its intended strategic objective. Due to its specialized application, the guidance fiber package must be compact, capable of smooth, high-speed release during missile flight without loosening or tangling, adaptable to diverse environments, and within acceptable deformation limits. These objectives place high demands on the winding and molding of the guidance fiber package. During the winding and molding process, fluctuations in fiber diameter, winding tension, and hysteresis angle, uneven adhesive coating, and varying thermal expansion coefficients of various materials can lead to various structural defects within the package. These defects can significantly reduce the reliability of the guidance fiber release. Therefore, the winding quality of the guidance fiber package must be inspected and evaluated.

[0004] Non-destructive testing (NDT) is a technique that uses physical or chemical methods to inspect and test the internal and surface structure, condition, or defects of an object without compromising its integrity or performance. Currently, NDT methods used on structures similar to those studied in this method include machine vision, distributed strain measurement, optical coherence tomography (OCT), and computed tomography (CT). Machine vision can only detect surface defects during the winding process, while most structural defects appear after the fiber optic package is wound. Distributed strain measurement can identify obvious collapse defects in the fiber optic package, but its resolution is limited to 5 cm, making it difficult to accurately locate the defect. OCT can perform tomographic testing on fiber coils, but its scanning depth is limited to 2 mm, and the images are not clear or intuitive, requiring expert experience to infer the type and location of the defect.

[0005] CT is a nondestructive three-dimensional imaging technique that uses an X-ray beam to penetrate an object, measuring its attenuation and distribution. Computerized information processing and image reconstruction techniques then produce a three-dimensional visualization of the object's internal structure. Compared to other nondestructive testing methods, CT can rapidly produce high-resolution, three-dimensional visualizations of optical fiber bundles, providing clear, intuitive, and easy-to-identify images. This method, through the application of CT technology, can precisely reveal the internal structure of optical fiber bundles, providing a solid foundation for in-depth analysis and assessment of potential defects.

[0006] While existing inspection methods can identify certain structural defects in optical fiber bundles, they still require trained experts to observe and examine the resulting data to infer the defect type and location. However, expert inspection is labor-intensive, requiring inspectors to possess a high degree of focus and discernment to identify even tiny defects in optical fiber bundles. Furthermore, this method is limited by human factors, including fatigue and subjective judgment, which can lead to inconsistent and unstable test results, making it difficult to meet the ever-increasing demand for high efficiency and precision in modern industrial production. With the expansion of production scale and the increasing quality requirements for optical fiber bundles, there is an urgent need for more advanced inspection technologies to detect and identify defects in guided optical fiber bundles. Summary of the Invention

[0007] Existing defect detection technologies for guided optical fiber packages suffer from limited detection ranges and reliance on subjective expert judgment, making them difficult to meet the high efficiency and high precision requirements of modern industrial production. This invention provides a method for optical fiber package defect detection and type recognition based on deep learning. By applying deep learning technology to defect detection and type recognition for guided optical fiber packages, this method not only achieves highly accurate identification of minor defects, but also significantly improves the automation level of the detection process and reduces labor costs.

[0008] In order to achieve the above objectives, the present invention provides the following technical solutions.

[0009] In a first aspect, the present invention provides a method for optical fiber package defect detection and type identification based on deep learning, comprising: Collect optical fiber package data and perform preprocessing to obtain an optical fiber package image dataset; Input the optical fiber package dataset into the pre-trained deep learning-based object detector for defect detection and type recognition, and obtain the defect detection and type recognition results; The training method of the pre-trained target detector comprises: Collecting and preprocessing the optical fiber package data to be tested to obtain an image dataset of the optical fiber package to be tested; The training set of the optical fiber package dataset to be detected is input into the target detector, and hyperparameter optimization is performed to obtain the optimal hyperparameter configuration; Train the target detector based on the optimal hyperparameter configuration to obtain a trained target detector; The test set of the optical fiber package data set to be detected is input into the trained target detector for defect detection and type recognition to obtain the defect detection and type recognition results.

[0010] As a further improvement of the present invention, the collecting of the optical fiber package data to be detected and preprocessing to obtain the optical fiber package image data set to be detected includes: The optical fiber package to be inspected is photographed and imaged by the industrial CT system to construct the internal features of the optical fiber package to be inspected and obtain a three-dimensional visualization model of the optical fiber package to be inspected; The three-dimensional visualization model of the optical fiber package to be inspected is uniformly discretized into a two-dimensional image stack to obtain two-dimensional cross-sectional image data of the optical fiber package to be inspected, thereby forming an image dataset of the optical fiber package to be inspected; The optical fiber package image dataset to be detected is divided into a training set and a test set.

[0011] As a further improvement of the present invention, the training set of the optical fiber package data set to be detected is input into the target detector, and hyperparameter optimization is performed to obtain the optimal hyperparameter configuration, including: Transforming the training set of the optical fiber package data set to be detected to obtain an expanded training set; Label the defects of the expanded training set to obtain the label data of the training set; The labeled data of the training set is input into the target detector, and the hyperparameters are optimized according to the learning rate and batch size to obtain the optimal hyperparameter configuration.

[0012] As a further improvement of the present invention, the training set of the optical fiber package data set to be detected is transformed to obtain an expanded training set, including: The image data in the training set of the optical fiber package dataset to be detected is rotated, scaled, flipped and sharpened to obtain an expanded training set.

[0013] As a further improvement of the present invention, the object detector is trained based on the optimal hyperparameter configuration to obtain a trained object detector, including: The detection performance and training speed of the target detector are trained based on the optimal hyperparameter configuration to obtain a trained target detector.

[0014] As a further improvement of the present invention, the test set of the optical fiber package data set to be detected is input into the target detector to perform defect detection and type identification to obtain defect detection and type identification results; Input the test set of the optical fiber package data set to be detected into the trained target detector to perform defect detection and type recognition, and obtain the defect detection and type recognition results; If the accuracy of the defect detection and type recognition results meets the set state, the recognition results are output; If the accuracy of the defect detection and type recognition results does not meet the set state, the target detector will be retrained until the accuracy of the recognition result meets the set state and the recognition result is output.

[0015] In a second aspect, the present invention provides a deep learning-based optical fiber package defect detection and type recognition system, comprising: Dataset acquisition module: used to collect optical fiber package data and perform preprocessing to obtain optical fiber package image dataset; Identification result module: used to input the optical fiber package data set into the pre-trained target detector for defect detection and type recognition, and obtain defect detection and type recognition results; The training system of the pre-trained target detector comprises: The data set module to be tested is used to collect the optical fiber package data to be tested and pre-process it to obtain the image data set of the optical fiber package to be tested; Obtaining the optimal hyperparameter module: This module is used to input the training set of the optical fiber package dataset to be detected into the target detector and perform hyperparameter optimization to obtain the optimal hyperparameter configuration; Training target detector module: used to train the target detector based on the optimal hyperparameter configuration to obtain the trained target detector; Obtaining identification result module: used to input the test set of the optical fiber package data set to be detected into the trained target detector for defect detection and type recognition, and obtain defect detection and type recognition results.

[0016] In a third aspect, the present invention provides an electronic device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and runnable on the processor, and when the processor executes the computer program, it implements the steps of the optical fiber cable package defect detection and type identification method based on deep learning.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of the optical fiber cable package defect detection and type identification method based on deep learning.

[0018] In a fifth aspect, the present invention provides a computer program product, characterized in that it includes computer instructions, which, when executed by a processor, implement the steps of the optical fiber cable package defect detection and type identification method based on deep learning.

[0019] Compared with the prior art, the present invention has the following beneficial effects: This method collects and preprocesses fiber optic cable data to construct a high-quality image dataset of fiber optic cable structural defects, providing a solid foundation for subsequent deep learning model training. Leveraging a pretrained deep learning target detector, this method can automatically and efficiently identify various defect types in fiber optic cable packages, including but not limited to cracks, voids, and end-face extrusion. Its detection accuracy far exceeds that of traditional methods, effectively avoiding misjudgments and missed detections caused by human factors, thereby ensuring the stability and reliability of product quality. Furthermore, the generalization capabilities of deep learning models enable this method to address the inspection needs of fiber optic cable packages of different batches and specifications, greatly expanding the detection range and enhancing the applicability and flexibility of the technology.

[0020] Furthermore, the training method of the present invention ensures the scientific nature and effectiveness of model training by systematically dividing and utilizing the optical fiber cable package data set to be detected, that is, dividing it into a training set and a test set. During the training process, by continuously adjusting and optimizing the model parameters, the model achieves deep learning and accurate capture of the defect characteristics of the optical fiber cable package. This process not only improves the recognition accuracy of the model, but also enhances its robustness and stability. The excellent performance of the target detector obtained based on the optimal weight training on the test set verifies the reliability and accuracy of this method in practical applications. This series of training and optimization steps not only improves the detection efficiency, but also ensures the consistency and repeatability of the detection results, providing strong technical support for quality control in industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are for illustration purposes only and are not intended to limit the scope of the present disclosure in any way. In the drawings: Figure 1 This is a flow chart of a method for optical fiber package defect detection and type identification based on deep learning in the present invention; Figure 2 A flowchart of a training method for a pre-trained object detector according to the present invention is provided; Figure 3 Schematic diagram of the semi-polar coordinate system of the optical fiber package of the present invention; Figure 4 A two-dimensional cross-sectional image of the optical fiber package of the present invention; Figure 5 Typical crack patterns of two optical fiber packages shown in the present invention; Figure 6 Two typical end face extrusion patterns of optical fiber packages shown in this invention, Figure 7 Two typical cavity patterns of optical fiber packages shown in the present invention; Figure 8 Schematic diagram of the effects obtained by different image transformation methods for optical fiber package defect detection and type identification adopted in the expansion of the present invention; wherein (a) is the effect obtained by rotating 90° counterclockwise, (b) is the effect obtained by rotating 180° counterclockwise, (c) is the effect obtained by rotating 270° counterclockwise, (d) is the effect obtained by scaling by 0.5 times, (e) is the effect obtained by scaling by 2 times, (f) is the effect obtained by horizontal flipping, (g) is the effect obtained by vertical flipping, and (h) is the effect obtained by sharpening by 1.5 times; Figure 9 Training results for different target detection algorithms; Figure 10 This is a structural diagram of a deep learning-based optical fiber package defect detection and type recognition system of the present invention; Figure 11A schematic diagram of the structure of a training system for a pre-trained object detector of the present invention; Figure 12 Schematic diagram of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. The embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] Aiming at the problems that the existing guided optical fiber package defect detection technology has limited detection range and the detection results need to rely on the subjective judgment of experts, which is difficult to meet the requirements of high efficiency and high precision in modern industrial production. The present invention provides a method for optical fiber package defect detection and type identification based on deep learning, such as Figure 1 Shown, including: S100: Collecting optical fiber package data and performing preprocessing to obtain an optical fiber package image dataset; S200: Inputting the optical fiber package dataset into a pre-trained deep learning-based target detector to perform defect detection and type recognition, thereby obtaining defect detection and type recognition results; like Figure 2 As shown, the present invention provides a training method for a pre-trained target detector, comprising: S201: collecting and preprocessing optical fiber package data to be inspected to obtain an image dataset of the optical fiber package to be inspected; S202: Inputting the training set of the optical fiber package data set to be detected into the target detector, and performing hyperparameter optimization to obtain the optimal hyperparameter configuration; S203: Training the target detector based on the optimal hyperparameter configuration to obtain the optimal weight of the trained target detector; S204: Inputting the test set of the optical fiber package data set to be detected into the trained target detector to perform defect detection and type recognition, and obtaining defect detection and type recognition results.

[0025] The present invention applies deep learning technology to defect detection and type identification of guided optical fiber packages, which can not only achieve high-accuracy identification of tiny defects, but also significantly improve the automation level of the detection process and reduce labor costs.

[0026] The present invention will be further explained below with reference to the accompanying drawings.

[0027] The purpose of this invention is to provide an automated, high-precision method for optical fiber package defect detection and type identification. To achieve this, the present invention proposes an optical fiber package defect detection and type identification method based on deep learning and industrial CT technology. This method, for the first time, accurately and automatically detects various structural defects within large, complex optical fiber packages.

[0028] The present invention provides a method for optical fiber package defect detection and type identification based on deep learning, which specifically includes the following steps: S1: Scan the sample optical fiber package using industrial CT equipment, reconstruct the internal structural features of the sample optical fiber package, obtain its 3D visualization model, and establish a semi-polar coordinate system; S11: Use industrial CT equipment with a micro-focus radiation source to photograph the sample optical fiber package to improve the spatial resolution of the imaging; S12: Due to the large size of the sample fiber package, the offset scanning method is used to scan it multiple times from top to bottom; S13: The image formed by the geometric projection is optically amplified by coupling the objective lens to the detector to further improve the imaging resolution; S14: Reconstruct the internal structural features of the sample optical fiber package using the FDK algorithm to obtain its 3D visualization model; S2: The obtained 3D visualization model is evenly discretized into 2D image stacks I1~I n ; S21: If Figure 3 As shown, the center of the bottom end face of the optical fiber package is the origin of the semi-polar coordinate system, the XOY plane is on the bottom end face, the X axis and the Y axis are perpendicular to each other, and the Z axis is perpendicular to the XOY plane; the θ coordinate takes the X axis as the starting point and is along the circumferential direction, with a range of [0, 2π]; the relationship between ρ and the x and y coordinates is shown in formula (1); (1) Where ρ represents the distance from any point on the surface of the optical fiber package to the Z axis, as Figure 3 shown.

[0029] S22: With a step length of g, uniformly intercept n two-dimensional cross-sectional images of the optical fiber package along the θ direction, such as Figure 4 As shown, n is obtained by formula (2); (2) S3: Sets detection standards for three types of structural defects commonly found in optical fiber packages that have a significant impact on information transmission stability and optical fiber release reliability; S31: Cracks are gaps that continuously penetrate multiple layers inside the cable package and have a width greater than half the fiber diameter in the Z-axis direction. The detection standard is set as a penetration depth of more than two fiber layers in the ρ direction and a width greater than half the fiber diameter at the widest point in the Z-axis direction. In this method, the fiber diameter including the coating layer is 0.5 mm. The standard crack defect is as follows: Figure 5 As shown; S32: End face extrusion refers to the phenomenon that multiple turns of optical fiber that should be arranged on the end face of the cable package slip onto the limit buckle of the cable package core tube. The standard is set as the ρ direction range of the extruded optical fiber exceeds two layers, or the number of turns of the extruded optical fiber exceeds 3 turns. Typical end face basic defects are as follows: Figure 6 As shown; S33: A cavity is a cavity that exists inside the cable package and has a width in the Z-axis direction greater than one fiber diameter and exists mainly in a single layer of fiber. The standard is set as a cavity that has a width in the Z-axis direction greater than one fiber diameter and exists only in a single layer of fiber in the ρ direction. Typical cavity defects are Figure 7 As shown; S4: Select representative patterns from the 2D image stack as the original database for optical fiber package structure defect detection and type identification; S41: This method performed high-resolution CT imaging on seven different types of optical fiber bundle samples, ultimately obtaining a total of 15,120 two-dimensional cross-sectional images of the optical fiber bundle; S42: After analyzing each of the 15,120 images, a series of representative patterns were selected, resulting in an original database of 433 images. S5: A series of image processing technologies are used to transform the original database, ultimately obtaining a database for optical fiber package structure defect detection and type recognition; S51: rotating the original image at different angles to simulate observation angles in different directions; S52: scaling the image size to simulate the visual effects when observing from far and close distances; S53: The image is flipped upside down and left to right to simulate the characteristics of symmetrical and asymmetrical defects; S54: Adjust the sharpness factor of the original image to enhance the edges and details of the image, so that the model can more clearly identify tiny defect features; S55: After the above image processing means, the expanded optical fiber package defect detection and type recognition database is obtained as follows Figure 8 As shown; S6: Use labelimg tool to label defects in the expanded database; S7: Use four object detectors to train based on the above database, and select the object detector with the best training results; S71: Four deep learning-based object detectors are tested: Single Shot MultiBox Detector (SSD), Faster R-CNN, CenterNet, and Fully Convolutional One-Stage Object Detection (FCOS). S72: We compared the training performance of the four object detectors mentioned above in terms of mean average precision (mAP), average recall (AR), detection time, and memory usage. mAP is a key metric in object detection algorithms used to evaluate the performance of a model across all categories. It is calculated by calculating the average precision (AP) for each defect category and then averaging it. The AP value is calculated based on the Precision-Recall (PR) curve, which is composed of precision values ​​at different recall rates.

[0030] The mAP value is obtained by formula (3): (3) in, represents the AP value of the i-th category, which is calculated by formula (4), and n represents the total number of categories.

[0031] (4) in, represents the precision at the kth recall point, calculated by formula (5), is the difference between the kth and k+1th recall points.

[0032] (5) Among them, TP (True Positive) represents the number of true positive examples; FP (False Positive) is the number of false positive examples.

[0033] The average recall rate (AR) measures the algorithm's ability to find all positive examples, that is, the proportion of samples that are correctly detected among all samples that are actually positive. Recall is also called the true positive rate. The recall value is calculated using Equation (6).

[0034] (6) Detection time refers to the time it takes from inputting an image to outputting the final result. In optical fiber defect detection, detection speed is directly related to production efficiency and real-time requirements. Memory usage refers to the computer memory resources occupied by the algorithm during execution.

[0035] S7.3: Take the learning rate (lr) = 0.0005 and the batch size (batch size) = 16, and train the following eight object detectors 50 times under the same configuration. The training results are as follows Figure 8 , thus the target detector that is most suitable for this method is the CenterNet detector; S8: Optimize the hyperparameters of the target detector to obtain the optimal weight model; S81: Select learning rate (lr) and batch size (batch-size) as hyperparameters for experiment; S82: Obtain the optimal training configuration parameters through 3 groups of twelve hyperparameter tests.

[0036] S9: Apply the obtained optimal weight model to a new set of test data sets to obtain the final test results.

[0037] In summary, with the rapid development of deep learning technology, deep learning-based object detection methods, with their powerful feature learning capabilities and excellent generalization performance, offer new possibilities for automated and intelligent optical fiber cable defect detection. Deep learning object detection technologies, such as You Only Look Once (YOLO) and Faster R-CNN, have demonstrated excellent detection performance in multiple fields. Through end-to-end training, these methods automatically learn the mapping relationship from raw data to target detection results, greatly improving detection accuracy and efficiency. Applying deep learning technology to defect detection and type recognition in guided optical fiber cable not only achieves highly accurate identification of tiny defects, but also significantly improves the automation level of the detection process and reduces labor costs.

[0038] The present invention is further explained below with reference to specific embodiments.

[0039] Example The present invention provides a method for optical fiber package defect detection and type identification based on deep learning, comprising: S1: An industrial CT system with a micro-focus radiation source is used to photograph the sample optical fiber package. The system mainly consists of subsystems such as the micro-focus radiation source, collimator system, detector system, multi-degree-of-freedom mechanical turntable system, data acquisition and transmission system, image reconstruction workstation, and radiation safety protection system. Due to the large size of the sample fiber package, multiple scans are performed from top to bottom using an offset scanning method. The system utilizes a microfocus radiation source with a focal spot size of only 6 microns, resulting in a negligible penumbra. This minimizes the distance from the sample to the focal spot, reducing the imaging range and improving imaging resolution. Finally, the geometrically projected image is optically amplified by a detector coupled to the objective lens, achieving high-resolution imaging of large samples.

[0040] Since X-rays attenuate when penetrating an object, the detector receives the attenuated photons and uses a mathematical 3D reconstruction algorithm to reconstruct the internal structural features of the object being measured, resulting in a 3D image of the object. Finally, the FDK algorithm is used to reconstruct the internal structural features of the sample fiber optic cable package, creating a 3D visualization model.

[0041] S2: If Figure 3 As shown, the center of the bottom end face of the optical fiber package is the origin of the semi-polar coordinate system, the XOY plane is on the bottom end face, the X axis and the Y axis are perpendicular to each other, and the Z axis is perpendicular to the XOY plane; the θ coordinate takes the X axis as the starting point and is along the circumferential direction, with a range of [0, 2π]; the relationship between ρ and the x and y coordinates is shown in formula (1); (1) With g as the step length, n two-dimensional cross-sectional images of the optical fiber package are uniformly intercepted along the θ annular direction, as shown in Figure 4 As shown, n is obtained by formula (2); (2) S3: Cracks refer to gaps that continuously penetrate multiple layers in the Z-axis direction and are larger than half the fiber diameter. The main reason for this defect is that the various materials that make up the optical fiber have different thermal expansion coefficients, which leads to uneven internal stress distribution after the wire package experiences temperature changes, resulting in anisotropic expansion, which causes the adhesive to crack and eventually form cracks. The detection standard is set as a penetration depth of more than 2 layers of optical fiber in the ρ direction, and the widest part in the Z-axis direction should be larger than half the fiber diameter. In this method, the diameter of the optical fiber including the coating layer is taken as 0.5mm. The standard crack defect is as follows: Figure 5 As shown; End face extrusion refers to the phenomenon that multiple turns of optical fiber that should be arranged at the end face of the wire package slip onto the limiting buckle of the wire package core tube. This is because after the wire package experiences temperature changes, the thermal expansion coefficients of the various components are different, resulting in uneven stress distribution within the wire package, and concentratedly reflected on the end face of the wire package. In addition, the two innermost layers of optical fiber will also be squeezed by the limiting buckle of the core tube at the end face, which causes the two innermost layers of optical fiber in the wire package to easily bulge at the end face, causing the outer layer of optical fiber to bulge and slip onto the limiting buckle of the wire package core tube. The standard is set as the ρ direction range of the extruded optical fiber exceeds two layers, or the number of turns of the extruded optical fiber exceeds 3 turns. Typical end face basic defects are as follows: Figure 6 As shown; A void is a cavity that exists inside the cable package and has a width greater than one fiber diameter in the Z-axis direction and is primarily found in a single layer of fiber. This cavity is primarily caused by different thermal expansion coefficients of the various materials that make up the fiber, a small fiber winding hysteresis angle, and uneven adhesive coating. The standard is a cavity that has a width greater than one fiber diameter in the Z-axis direction and exists only in a single layer of fiber in the ρ direction. Typical void defects include: Figure 7 As shown; S4: This method uses a Metro Voxel industrial CT system to perform high-resolution CT imaging on seven different fiber optic cable samples, followed by high-precision stitching and imaging processing. The resulting images total 15,120 two-dimensional cross-sectional images of the fiber optic cable, providing a comprehensive and detailed visual database for studying the internal structure and potential defects of the cable.

[0042] After analyzing each of the 15,120 images, a representative set of patterns was selected. These patterns not only exhibited the typical characteristics of each defect type, but also included other notable features such as standard winding patterns, uneven brightness distribution, areas with artifacts, and easily confused details. After rigorous screening, a raw database of 433 images was ultimately obtained.

[0043] S5: To further increase the size and diversity of the database, this method employs a series of image processing techniques to transform the original database images. These transformations, including rotation, scaling, flipping, and sharpening, are designed to simulate different observation conditions and imaging effects, thereby enhancing the model's generalization and adaptability to various image transformations.

[0044] The original image was rotated 90, 180, and 270 degrees to simulate observation angles in different directions. This step helps the model learn the characteristics of defects in different directions.

[0045] The images are scaled by 0.5x and 2.0x to simulate the visual effects of observing from far and close distances. This not only increases the diversity of the images but also enables the model to adapt to inputs of different resolutions.

[0046] The images were flipped upside down and left to right to simulate the characteristics of symmetrical and asymmetrical defects. This transformation helps improve the model's ability to recognize changes in defect location.

[0047] By adjusting the image sharpening factor to 1.5, the edges and details of the image are enhanced, enabling the model to more clearly identify tiny defect features.

[0048] By using the above image processing technology, this method successfully expanded the original database (containing 433 images) to 9 times its original size, and finally obtained a fiber optic cable CT cross-section database containing 3897 images; S6: Use labelimg tool to label defects in the expanded database; S7: Given that structural defects in optical fiber bundles are relatively small relative to the overall bundle size, and the need for real-time inspection of optical fiber bundle winding quality, high requirements are placed on the small target detection capability and speed of target detectors. Therefore, this method selects the following four detectors that are both suitable for small target detection and have high detection speeds. We explore their principles and characteristics, and comprehensively evaluate and compare their detection results using actual image data to select the most suitable model for this method's application scenario.

[0049] SSD (Single Shot MultiBox Detector) is a single-stage detector that detects objects of varying sizes by performing detection on CNN feature maps at different levels. Shallow feature maps have a larger receptive field, making them suitable for detecting small objects. SSD also incorporates PriorBoxes as strong prior knowledge of candidate regions. By adjusting the size and aspect ratio of PriorBoxes, it can predict object boxes at various scales and locations, helping to improve the detection accuracy of small objects. Another feature of SSD is its efficient computational performance, making it suitable for applications requiring real-time response.

[0050] Faster R-CNN is a two-stage detector that introduces a region proposal network (RPN) to replace the traditional selective search algorithm for generating candidate regions. This improvement significantly improves object detection accuracy. By performing detection on feature maps of different scales, Faster R-CNN can also detect objects of different sizes, making it particularly suitable for small objects. Furthermore, although Faster R-CNN is a two-stage detector, its RPN and subsequent detection network share feature maps extracted by the same convolutional neural network (CNN). This means that the entire image is convolved only once, reducing repeated computations and significantly improving detection speed.

[0051] CenterNet is an innovative anchor-free object detection algorithm. Based on high-resolution feature maps, it detects objects by predicting their center point, width, height, and category. This approach offers significant advantages when dealing with small and densely packed objects because it directly regresses the object's size and position without relying on pre-set anchor boxes or performing non-maximum suppression (NMS). Furthermore, since NMS is not required, CenterNet also achieves faster inference speeds.

[0052] FCOS (Fully Convolutional One-Stage Object Detection) also adopts an anchor-free design. Its core concept is to predict the object's category, location (using the distance between the center point and the boundary), and center-ness directly on the feature map. By introducing center-ness to measure the proximity of the predicted point to the object center, detection quality is significantly improved. The use of a Feature Pyramid Network (FPN) enables it to handle objects of varying sizes, enhancing its detection capabilities for small objects.

[0053] The performance and practicality of object detection algorithms are crucial in the research of optical fiber defect detection and type identification. To comprehensively compare and evaluate the performance of the aforementioned object detection algorithms in this field, this method considers several aspects, including mean average precision (mAP), average recall (AR), detection time, and memory usage.

[0054] mAP is an important metric used in object detection algorithms to evaluate the performance of a model across all categories. It is calculated by calculating the average precision (AP) for each defect category and then taking the average. The AP value is calculated based on the Precision-Recall (PR) curve, which is composed of precision values ​​at different recall rates. The mAP value is calculated using formula (3): (3) Where APi represents the AP value of the i-th category, which is calculated by formula (4), and n represents the total number of categories.

[0055] (4) where pk is the precision at the kth recall point, calculated by (8), and Δrk is the difference between the kth and k+1th recall points.

[0056] Formula (5) Where TP (True Positive) is the number of true positive examples, and FP (False Positive) is the number of false positive examples.

[0057] The average recall rate (AR) measures the algorithm's ability to find all positive examples, that is, the proportion of samples that are correctly detected among all samples that are actually positive. Recall rate is also called the true positive rate. The recall value is calculated by formula (6).

[0058] (6) Detection time refers to the time it takes from inputting an image to outputting the final result. In optical fiber defect detection, detection speed is directly related to production efficiency and real-time requirements. Memory usage refers to the computer memory resources occupied by the algorithm during execution. In optical fiber defect detection, due to the large amount of inspection object data, the algorithm's memory usage directly affects the stability and scalability of the detection system.

[0059] Take the learning rate (lr) = 0.0005, batch size (batch size) = 16, and train the following eight target detectors 50 times under the same configuration. The training results are as follows Figure 9 As shown in the figure, mAP_50:95 refers to the average mAP value for IoU thresholds between 0.5 and 0.95, mAP_50 refers to the mAP value when the IoU threshold is 0.5, mAP_s refers to the mAP value for small-scale object detection, AR_50:95 refers to the average AR value for IoU thresholds between 0.5 and 0.95, Time refers to the time required for different algorithms to train 50 times, and Memory refers to the memory occupied during training. As can be seen from the table, the SSD algorithm and CenterNet algorithm perform the best, reaching mAP of 0.908 and 0.895 respectively when the IoU is 0.5. These two algorithms also perform the best in terms of average recall. Among them, the SSD algorithm takes almost three times as long as CenterNet, but the achieved results are close to CenterNet, so only CenterNet is used for subsequent experiments and optimization.

[0060] S8: In object detection algorithms, the learning rate (LR) and batch size are two crucial hyperparameters that significantly impact model training and performance. The learning rate determines the step size for updating model weights and is a key factor in controlling model training speed and convergence stability. The batch size, on the other hand, determines the number of samples processed during each training step.

[0061] Therefore, we chose learning rate and batch size as hyperparameters for experimentation because they have a direct impact on the model's training speed and convergence stability. The optimal learning rate and batch size often vary depending on the dataset and hardware resources. Proper hyperparameter tuning can significantly improve the detection performance and training speed of the object detection algorithm.

[0062] The hyperparameter training results of the CenterNet algorithm are shown in Table 1: Table 1 CenterNet algorithm hyperparameter training results

[0063] S9: The optimal hyperparameters are used to train the CenterNet detector, resulting in the optimal weight model. This model is then used on a new, unseen test dataset, yielding the following test results, as shown in Table 2: Table 2 Test performance of weighted models before and after data augmentation

[0064] It can be seen that the performance of the weight model obtained after data enhancement training has been significantly improved in the test dataset compared with the weight model obtained before data enhancement training.

[0065] Experimental results demonstrate that the proposed method can achieve high-precision and high-efficiency detection and type identification of various structural defects within optical fiber bundles. This research not only provides strong technical support for the quality control of optical fiber bundles, but also provides new ideas and methods for non-destructive testing of other complex structures.

[0066] The second purpose of the present invention is to propose a fiber optic cable defect detection and type recognition system based on deep learning, such as Figure 10 Shown, including: The data set acquisition module 100 is used to collect optical fiber package data and perform preprocessing to obtain an optical fiber package image data set; Identification result module 200: used to input the optical fiber package data set into the pre-trained target detector to perform defect detection and type identification, and obtain defect detection and type identification results; like Figure 11 As shown, a training system for a pre-trained target detector includes: The data set module 201 is used to collect the optical fiber package data to be detected and pre-process it to obtain the image data set of the optical fiber package to be detected; Obtaining optimal hyperparameter module 202: used to input the training set of the optical fiber package data set to be detected into the target detector, and perform hyperparameter optimization to obtain the optimal hyperparameter configuration; The training target detector module 203 is used to train the target detector based on the optimal hyperparameter configuration to obtain a trained target detector; Obtaining identification result module 204: is used to input the test set of the optical fiber package data set to be detected into the trained target detector to perform defect detection and type identification, and obtain defect detection and type identification results.

[0067] like Figure 12 As shown, a third object of the present invention is to provide an electronic device, comprising: a processor 301, a memory 302, and a display screen 303. The memory 302 and the display screen 303 are both connected to the processor 301, for example, via a bus 304. Optionally, the electronic device may further include a transceiver 305. It should be noted that in actual applications, the number of transceivers 305 is not limited to one, and the structure of the electronic device does not constitute a limitation on the embodiments of the present application.

[0068] Processor 301 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0069] Bus 304 may include a path for transmitting information between the aforementioned components. Bus 304 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, for example. Bus 304 may be divided into an address bus, a data bus, a control bus, and other components.

[0070] The memory 302 may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0071] The memory 302 is used to store application code for executing the solution of the present application, and the execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 302 to implement the content shown in the above method embodiment.

[0072] The electronic device shown in the figure is only an example and should not limit the functions and scope of use of the embodiments of the present application.

[0073] The fourth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is stored thereon, and when the program is executed by a processor, the computer program is realized as described above. Figures 1 to 2 The various processes of the illustrated method embodiment include, for example, a memory including instructions, and the instructions can be executed by a processor of an electronic device to perform the above method.

[0074] A computer-readable storage medium may be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium may be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, a computer-readable storage medium may be a portable computer disk, a hard drive, a USB flash drive, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a rostrum random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, an optical disc, a magnetic disk, a mechanical encoding device, or any combination thereof.

[0075] The fifth object of the present invention is to provide a computer program product comprising computer instructions, which, when executed by a processor, implement the above Figures 1 to 2 The various processes of the method embodiment shown can achieve the same technical effect, and to avoid repetition, they will not be described here.

[0076] Many embodiments and applications beyond the examples provided will be apparent to those skilled in the art upon reading the foregoing description. Therefore, the scope of the present teachings should be determined not with reference to the foregoing description, but rather with reference to the preceding claims, along with the full scope of equivalents to which such claims are entitled. For the purpose of completeness, all articles and references, including the disclosures of patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein from the preceding claims is not a disclaimer of such subject matter, nor should it be interpreted that the applicants did not consider such subject matter to be part of the disclosed inventive subject matter.

[0077] The above content is a further detailed description of the present invention, and it cannot be considered that the specific implementation methods of the present invention are limited to these. For ordinary technicians in the technical field to which the present invention belongs, they can make several simple deductions or substitutions without departing from the concept of the present invention, which should be regarded as belonging to the scope of protection of the present invention determined by the submitted claims.

Claims

1. A method for optical fiber package defect detection and type identification based on deep learning, characterized in that: include: Collecting optical fiber line package data and preprocessing it to obtain an optical fiber line package image data set; The optical fiber package data set is input into the pre-trained deep learning-based target detector for defect detection and type recognition, and the defect detection and type recognition results are obtained; The training method of the pre-trained target detector comprises: Collecting and preprocessing the optical fiber package data to be detected to obtain an image data set of the optical fiber package to be detected; The training set of the optical fiber package data set to be detected is input into the target detector, and the hyperparameter optimization is performed to obtain the optimal hyperparameter configuration; The target detector is trained based on the optimal hyperparameter configuration to obtain a trained target detector; The test set of the optical fiber package data set to be detected is input into the trained target detector for defect detection and type recognition to obtain the defect detection and type recognition results.

2. The optical fiber package defect detection and type identification method based on deep learning according to claim 1 is characterized in that: The collecting and preprocessing of the optical fiber package data to be detected to obtain the optical fiber package image data set to be detected includes: The optical fiber package to be inspected is photographed and imaged by the industrial CT system, the internal features of the optical fiber package to be inspected are constructed, and a three-dimensional visualization model of the optical fiber package to be inspected is obtained; The three-dimensional visualization model of the optical fiber package to be detected is evenly discretized into a two-dimensional image stack to obtain two-dimensional cross-sectional image data of the optical fiber package to be detected, thereby forming an image data set of the optical fiber package to be detected; The optical fiber package image dataset to be detected is divided into a training set and a test set.

3. The optical fiber package defect detection and type identification method based on deep learning according to claim 1, characterized in that: The step of inputting the training set of the optical fiber package data set to be detected into the target detector and performing hyperparameter optimization to obtain the optimal hyperparameter configuration includes: Transforming the training set of the optical fiber line package data set to be detected to obtain an expanded training set; Annotate the defects of the expanded training set to obtain label data of the training set; The labeled data of the training set is input into the target detector, and the hyperparameters are optimized according to the learning rate and batch size to obtain the optimal hyperparameter configuration.

4. The optical fiber package defect detection and type identification method based on deep learning according to claim 3 is characterized in that: The training set of the optical fiber line package data set to be detected is transformed to obtain an expanded training set, including: The image data in the training set of the optical fiber package data set to be detected is rotated, scaled, flipped and sharpened to obtain an expanded training set.

5. The optical fiber package defect detection and type identification method based on deep learning according to claim 1, characterized in that: The target detector is trained based on the optimal hyperparameter configuration to obtain a trained target detector, including: The detection performance and training speed of the target detector are trained based on the optimal hyperparameter configuration to obtain a trained target detector.

6. The optical fiber package defect detection and type identification method based on deep learning according to claim 1, characterized in that: The test set of the optical fiber package data set to be detected is input into the target detector for defect detection and type recognition to obtain defect detection and type recognition results; Input the test set of the optical fiber package data set to be detected into the trained target detector to perform defect detection and type recognition, and obtain defect detection and type recognition results; If the accuracy of the defect detection and type recognition results meets the set state, the recognition results are output; If the accuracy of the defect detection and type recognition results does not meet the set state, the target detector is retrained until the accuracy of the recognition result meets the set state, and the recognition result is output.

7. A fiber optic cable package defect detection and type recognition system based on deep learning, characterized in that: include: Dataset acquisition module: used to collect optical fiber package data and perform preprocessing to obtain optical fiber package image data set; Identification result module: used to input the optical fiber package data set into the pre-trained target detector for defect detection and type recognition, and obtain defect detection and type recognition results; The training system of the pre-trained target detector comprises: The module for the data set to be detected is used to collect the optical fiber package data to be detected and perform preprocessing to obtain the image data set of the optical fiber package to be detected; Obtaining the optimal hyperparameter module: used to input the training set of the optical fiber cable package data set to be detected into the target detector, and perform hyperparameter optimization to obtain the optimal hyperparameter configuration; Training target detector module: used to train the target detector based on the optimal hyperparameter configuration to obtain the trained target detector; Obtaining identification result module: used to input the test set of the optical fiber line package data set to be detected into the trained target detector for defect detection and type recognition, and obtain defect detection and type recognition results.

8. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method implements the steps of a method for optical fiber package defect detection and type identification based on deep learning as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the optical fiber line package defect detection and type identification method based on deep learning as described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of a method for optical fiber line package defect detection and type identification based on deep learning as described in any one of claims 1 to 6.

Images