End face quality detection device and method of high-speed optical module
Through the combination of multi-spectral imaging technology and intelligent processors, the problem of difficulty in detecting small defects and distinguishing defects of different properties of single spectral detection is solved, and high accuracy and comprehensive detection of the end surface of the optical module is achieved, which enhances the reliability of the detection and comprehensive.
Patent Information
- Application Number
- CN202411563263.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-11-05
AI Technical Summary
The existing optical module end-face detection technology relies on single spectral imaging, making it difficult to detect small defects, distinguish defects of different properties, be affected by ambient light, and is difficult to provide detailed material information, resulting in low detection accuracy and comprehensiveness.
Multispectral imaging technology is used to combine rotatable transportation and image acquisition components, and intelligent processors are used to pre-process, feature extraction and fusion of multispectral images, identify end-face defects, and generate defect density maps.
It significantly improves the accuracy and comprehensiveness of the detection, and can capture defect characteristics in different bands, reduce the impact of environmental factors, provide richer material information, and ensure the reliability and comprehensiveness of the detection.
Smart Images

Figure CN119064375A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of defect detection technology for optical module end faces, and more particularly to an end face quality detection device and method for high-speed optical modules. Background Art
[0002] Optical modules are key components in fiber optic communication systems, converting electrical signals into optical signals or vice versa, thereby enabling high-speed, long-distance data transmission. The performance of optical modules directly affects the quality and reliability of the entire communication system, and the quality of the module's end face is one of the most important factors determining its performance. Contaminants, scratches, or other defects on the end face can cause signal attenuation, reflection, or scattering, severely impacting the transmission efficiency and lifespan of the optical module. Therefore, rigorous quality inspection of the end face is crucial during the production and use of optical modules.
[0003] Existing optical module end-face inspection technologies primarily rely on single-spectrum imaging, typically using images from the visible light band for analysis. This approach has several significant drawbacks: First, certain types of defects are difficult to detect in the visible spectrum. For example, minute surface cracks may be almost invisible in visible light but produce significant scattering effects in the near-infrared band. Second, a single spectrum struggles to distinguish defects of different natures. For instance, surface contamination and material defects may present similar characteristics in visible light images, leading to misjudgments. Furthermore, single-spectrum detection is sensitive to changes in ambient light and is easily affected by external lighting conditions, reducing the stability and repeatability of the detection. Additionally, some optical coatings or treated end faces may appear uniform in visible light, masking potential quality issues that might be more easily detected in other spectra. Finally, single-spectrum detection cannot provide detailed information on the end-face material composition, limiting in-depth analysis of potential problems. In contrast, the high-speed optical module end-face quality inspection device of this invention employs multispectral imaging technology, combining a rotatable transport component, an image acquisition component, and an intelligent processor, enabling comprehensive and accurate identification of various types of end-face defects. This method can capture defect features in different bands, distinguish problems of different natures, reduce the influence of environmental factors, detect hidden defects, and provide richer material information, thereby greatly improving the accuracy, reliability and comprehensiveness of detection. Summary of the Invention
[0004] The main objective of this invention is to solve the technical problems of low accuracy, reliability, and comprehensiveness of existing optical fiber endfaces that use single-spectrum detection optical modules.
[0005] The first aspect of the present invention provides an end-face quality inspection device for a high-speed optical module, the end-face quality inspection device for the high-speed optical module comprising a body, a transport component, an image acquisition component, and a processor; The transport component is rotatably mounted on the body, and the transport component has a fixing position for fixing the optical module to be tested; An image acquisition component is rotatably mounted on the body, and the image acquisition component is used to acquire a multispectral image of the end face of the optical module to be tested; The processor is electrically connected to both the transport component and the image acquisition component. The processor is used to adjust the relative position of the end face of the optical module under test and the image acquisition component so that the end face of the optical module under test is perpendicular to the imaging center line of the image acquisition component. The processor is able to receive and process the multispectral image to identify the defect type and degree of the end face of the optical module under test.
[0006] Optionally, the transport assembly includes a rotating base, a lifting member, and a fixing member. The rotating base has a mounting surface, the lifting member is fixedly mounted on the mounting surface, and the end of the lifting member away from the mounting surface is used to support the optical module to be tested. The fixing member is movably mounted on the mounting surface, and the end of the fixing member away from the mounting surface is used to be fixedly connected to the optical module to be tested. The fixing member can move in a direction away from or towards the supporting member, so that the interval between the fixing member and the supporting member can be adjusted to accommodate optical modules of different lengths to be tested.
[0007] Optionally, the supporting member includes a frame fixed to the mounting surface and a bracket reciprocally disposed on the frame, wherein the end of the bracket away from the mounting surface is formed as a supporting part, and the bracket can reciprocate in a direction perpendicular to the mounting surface.
[0008] Optionally, a first bolt hole is formed on the fixing member, and a plurality of second bolt holes are formed on the mounting surface. The plurality of second bolt holes are arranged at intervals along the line connecting the fixing member and the supporting member. The end face quality detection device of the high-speed optical module includes bolts, and the bolts are threadedly connected to the first bolt hole and any of the second bolt holes respectively, so as to detachably fix the fixing member to the mounting surface.
[0009] Optionally, the transport assembly includes an X-axis slide, a Y-axis slide, a Z-axis slide, and a rotary drive. The X-axis slide is movably mounted on the machine body along the X-axis direction, the Y-axis slide is movably mounted on the X-axis slide along the Y-axis direction, the Z-axis slide is movably mounted on the Y-axis slide along the Z-axis direction, and the rotary drive is mounted on the Z-axis slide. The output shaft of the rotary drive is connected to the rotary seat.
[0010] A second aspect of the present invention provides a method for detecting the end-face quality of a high-speed optical module, comprising: Adjust the optical module to be tested so that its end face is perpendicular to the imaging center line of the image acquisition component; Multispectral imaging is performed on the end face of the optical module to be tested to obtain a multispectral image set including visible light band, near-infrared band and ultraviolet band; Based on the multispectral image set, the image of each band is preprocessed; A multi-scale feature extraction network is used to extract features from the preprocessed multispectral image set to obtain feature maps at multiple scales. The multi-scale feature maps are fused to obtain a fused feature map; The fused feature map is analyzed using a defect detection network to identify the location, type, and severity of defects; Based on the identified defect information, the end face integrity rate is calculated and a defect density map is generated.
[0011] Optionally, the step of using a multi-scale feature extraction network to extract features from the preprocessed multispectral image set to obtain multiple scale feature maps includes: Construct a multi-scale feature extraction network comprising an encoder and a decoder, wherein the encoder comprises multiple downsampling stages and the decoder comprises multiple upsampling stages corresponding to the downsampling stages; A residual global perception module is introduced in each stage of the encoder and decoder. The residual global perception module divides the input feature map into multiple groups and applies spatial specific convolution and channel specific convolution within and between groups, respectively. A skip connection is added between corresponding stages of the encoder and decoder to connect the feature map of the encoder stage with the feature map of the corresponding stage of the decoder. The multi-scale feature extraction network processes the preprocessed multispectral image set and outputs feature maps from each stage of the decoder to obtain feature maps at multiple scales.
[0012] Optionally, the step of fusing features from the multi-scale feature maps to obtain a fused feature map includes: The multi-scale feature map is subjected to channel compression and excitation processing to obtain a channel weight map; A two-dimensional spatial attention map is generated based on the multi-scale feature map; The channel weight map is multiplied element-wise with the two-dimensional spatial attention map to obtain an attention-weighted feature map; The attention-weighted feature map and the multi-scale feature map are fused to obtain an intermediate feature map; The intermediate feature map is reconstructed using a feature pyramid network at multiple scales to obtain a fused feature map.
[0013] Optionally, the step of analyzing the fused feature map using a defect detection network to identify the location, type, and severity of defects includes: A region proposal network is applied to the fused feature map to obtain candidate defect regions, wherein the region proposal network uses circular anchor boxes instead of rectangular anchor boxes; Based on the candidate defect regions, the fused feature map is used to extract regional features to obtain a regional feature map; The region feature map is input into a dual-branch network, where one branch performs defect classification and the other branch performs bounding box regression to obtain the defect type probability distribution and refined bounding box coordinates. Based on the probability distribution of the defect type and the refined bounding box coordinates, a non-maximum suppression algorithm is used for post-processing to obtain the final defect detection result; Based on the final defect detection results, each detected defect is quantitatively scored according to a preset evaluation standard to obtain information on the location, type, and severity of the defect.
[0014] Optionally, the step of calculating the end-face integrity rate and generating a defect density map based on the identified defect information includes: Based on the location, type, and severity of the defects, the end face is divided into regions to obtain defective and non-defective regions. The ratio of the area of the non-defective region to the total area of the end face is calculated to obtain the end face integrity rate. Based on the spatial distribution and severity of the defective regions, a two-dimensional heat map is generated to obtain a defect density map.
[0015] The high-speed optical module end-face quality inspection device of this invention effectively solves many shortcomings of existing single-spectrum detection methods through an innovative combination of multispectral imaging technology, rotatable transport and image acquisition components, and an intelligent processor. Multispectral imaging covers the visible, near-infrared, and ultraviolet bands, enabling comprehensive capture of various defect features. For example, minute cracks that are difficult to detect under visible light can be clearly detected in the near-infrared band, and certain contaminants are more apparent under ultraviolet light, thus significantly improving the accuracy and comprehensiveness of the inspection. The rotatable transport component allows for precise adjustment of the angle of the optical module under inspection, making it suitable for different types of optical modules, such as straight and angled end faces, ensuring that the end face is perpendicular to the imaging center line for optimal imaging results. Simultaneously, the rotatable image acquisition component further enhances the system's flexibility, enabling image acquisition from multiple angles, suitable for complex shapes or specially processed end faces, and reducing information loss. Attached Figure Description
[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.
[0017] Figure 1 This is a schematic diagram of an embodiment of the end-face quality inspection device for the high-speed optical module of the present invention; Figure 2 for Figure 1 A magnified view of a section at point A in the middle; Figure 3 for Figure 1 A structural diagram with some parts omitted; Figure 4 for Figure 3 A partial enlarged view of point B in the middle; Figure 5 for Figure 3 A structural diagram from another perspective; Figure 6 for Figure 5 A magnified view of a section at point C; Figure 7 This is a flowchart of the end-face quality inspection method for the high-speed optical module of the present invention.
[0018] Description of Figure Numbers: 1. Body; 2. Transport assembly; 21. Rotary seat; 211. Mounting surface; 212. Second bolt hole; 22. Lifting component; 221. Frame; 222. Bracket; 222a. Supporting part; 23. Fixing component; 231. First bolt hole; 24. X-axis slide; 25. Y-axis slide; 26. Z-axis slide; 27. Rotation drive component; 3. Image acquisition assembly; 100. Optical module to be tested.
[0019] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. Detailed Implementation
[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0021] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.
[0022] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the term "and / or" throughout the text includes three solutions; taking A and / or B as an example, it includes technical solution A, technical solution B, and a technical solution that simultaneously satisfies A and B. Furthermore, the technical solutions of various embodiments can be combined with each other, provided that they are feasible for those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.
[0023] One embodiment of this application provides a device for inspecting the end-face quality of a high-speed optical module. Please refer to [link / reference]. Figures 1 to 6 The high-speed optical module end-face quality inspection device includes a main body 1, a transport component 2, an image acquisition component 3, and a processor. The transport component 2 is rotatably mounted on the body 1, and the transport component 2 has a fixing position for fixing the optical module 100 to be tested; The image acquisition component 3 is rotatably mounted on the body 1, and the image acquisition component 3 is used to acquire the multispectral image of the end face of the optical module 100 to be tested; The processor is electrically connected to the transport component 2 and the image acquisition component 3 respectively. The processor is used to adjust the relative position of the end face of the optical module 100 to be tested and the image acquisition component 3 so that the end face of the optical module 100 to be tested is perpendicular to the imaging center line of the image acquisition component 3. The processor can receive and process the multispectral image to identify the defect type and degree of the end face of the optical module 100 to be tested.
[0024] Specifically, in this embodiment, the body 1 can be a frame structure made of metal or engineering plastics, used to support and fix other components. The fixing position can be formed by adjustable clamps or robotic arms to secure the optical module 100 to be tested. The image acquisition component 3 can include a high-resolution camera, spectrometer, or multi-band filter group for capturing multispectral images. The processor can be an industrial-grade computer or embedded system for data processing and equipment control.
[0025] In some embodiments, the transport component 2 and the image acquisition component 3 can be rotatable via bearings or slide rails. The processor can establish an electrical connection with the transport component 2 and the image acquisition component 3 via a data cable or wireless module to transmit data and control commands.
[0026] "Vertical imaging centerline" means that the optical axis of the image acquisition component 3 forms a 90-degree angle with the end face of the optical module 100 under test. This configuration ensures image clarity and accuracy. The processor's "adjusting relative position" means that it can control the movement of the transport component 2 and the image acquisition component 3 to achieve the optimal imaging position.
[0027] It is understood that the high-speed optical module end-face quality inspection device of the present invention effectively solves many shortcomings of existing single-spectrum detection methods through the innovative combination of multispectral imaging technology, a rotatable transport and image acquisition component 3, and an intelligent processor. Multispectral imaging covers visible light, near-infrared, and ultraviolet bands, enabling comprehensive capture of various defect features. For example, tiny cracks that are difficult to detect under visible light can be clearly detected in the near-infrared band, and certain contaminants are more obvious under ultraviolet light, thus significantly improving the accuracy and comprehensiveness of the detection. The rotatable transport component 2 allows for precise adjustment of the angle of the optical module 100 under inspection, making it suitable for different types of optical modules such as straight and angled end faces, ensuring that the end face is perpendicular to the imaging center line for optimal imaging results. Simultaneously, the rotatable image acquisition component 3 further enhances the system's flexibility, enabling image acquisition from multiple angles, suitable for complex shapes or specially processed end faces, and reducing information loss.
[0028] Optionally, the transport assembly 2 includes a rotating base 21, a lifting member 22, and a fixing member 23. The rotating base 21 has a mounting surface 211. The lifting member 22 is fixedly mounted on the mounting surface 211, and the end of the lifting member 22 away from the mounting surface 211 is used to support the optical module 100 to be tested. The fixing member 23 is movably mounted on the mounting surface 211, and the end of the fixing member 23 away from the mounting surface 211 is used to be fixedly connected to the optical module 100 to be tested. The fixing member 23 can move away from or towards the supporting member 22, so that the interval between the fixing member 23 and the supporting member 22 can be adjusted and adapted to optical modules 100 of different lengths to be tested.
[0029] In this embodiment, the rotating base 21 can be a disc, polygonal platform, or other irregularly shaped platform made of metal or high-strength engineering plastic, which is used to support and rotate the optical module 100 to be tested. The mounting surface 211 is a flat surface on the rotating base 21 used for mounting other components.
[0030] The supporting component 22 can be fixed to the mounting surface 211 by bolts, welding, or embedded installation. The fixing component 23 can be connected to the mounting surface 211 by a slide, guide rail, bolt, or gear and rack mechanism to achieve a movable characteristic. The fixing component 23 can be fixedly connected to the optical module 100 under test by bolts, mechanical clips, or vacuum adsorption to ensure that the optical module remains stable during the test.
[0031] The phrase "the fixing member 23 can move in a direction away from or towards the supporting member 22" means that the fixing member 23 can reciprocate in a straight line to accommodate optical modules of different lengths. This design allows the interval between the fixing member 23 and the supporting member 22 to be adjusted as needed. This adjustable design greatly improves the applicability and flexibility of the device. It can adapt to various optical modules of different lengths on the market, and can be quickly adjusted without replacing parts, improving detection efficiency. At the same time, this design also ensures that optical modules of different sizes can be stably and reliably supported and fixed, thereby guaranteeing the accuracy and stability of the optical module position during the detection process and improving the accuracy and reliability of the detection results.
[0032] Optionally, the lifting member 22 includes a frame 221 fixed to the mounting surface 211 and a bracket 222 reciprocally disposed on the frame 221. The end of the bracket 222 away from the mounting surface 211 is formed as a supporting portion 222a, and the bracket 222 can reciprocate in a direction perpendicular to the mounting surface 211.
[0033] The frame 221 can be fixed to the mounting surface 211 by bolt connection, welding or integral molding. The bracket 222 can be connected to the frame 221 by slide rail, guide column or screw nut mechanism to realize reciprocating movement in the vertical direction (perpendicular to the mounting surface 211).
[0034] The phrase "the bracket 222 can reciprocate in a direction perpendicular to the mounting surface 211" means that the bracket 222 can move up and down in a plane perpendicular to the mounting surface 211. This height-adjustable support 22 design significantly enhances the adaptability and accuracy of the device. It can precisely adjust the height position of the optical module, ensuring that the end face of the optical module maintains the optimal relative position with the image acquisition component 3, thereby obtaining the clearest image. At the same time, this design also allows the device to adapt to optical modules of different sizes and shapes, improving the versatility and efficiency of the inspection process. The vertical adjustment function also helps to reduce stress during the installation of the optical module, reducing the risk of damage to the optical module, and further ensuring the reliability of the inspection and the integrity of the optical module.
[0035] Optionally, a first bolt hole 231 is formed on the fixing member 23, and a plurality of second bolt holes 212 are formed on the mounting surface 211. The plurality of second bolt holes 212 are arranged at intervals in the line connecting the fixing member 23 and the supporting member 22. The end face quality detection device of the high-speed optical module includes bolts, and the bolts are threadedly connected to the first bolt hole 231 and any of the second bolt holes 212 respectively, so as to detachably fix the fixing member 23 to the mounting surface 211.
[0036] The first bolt hole 231 can be a circular or elliptical through-hole penetrating the fastener 23 to accommodate a bolt. The second bolt hole 212 can be a threaded hole or a smooth through-hole provided on the mounting surface 211 for use with a nut. The bolt can be a standard hex bolt, an internal hex bolt, or a quick-release bolt with a hand-tightening head, used to detachably connect the fastener 23 and the mounting surface 211 together.
[0037] It's easy to understand that this adjustable fixing method greatly improves the flexibility and adaptability of the device. By selecting different second bolt holes 212, the operator can quickly and accurately adjust the position of the fixing member 23 to accommodate optical modules of different lengths. This design not only simplifies the adjustment process and reduces adjustment time, but also ensures the accuracy and repeatability of the adjustment. At the same time, the detachable design facilitates maintenance and replacement, improving the maintainability and service life of the device.
[0038] Optionally, the transport assembly 2 includes an X-axis slide 24, a Y-axis slide 25, a Z-axis slide 26, and a rotary drive 27. The X-axis slide 24 is movably mounted on the body 1 along the X-axis direction, the Y-axis slide 25 is movably mounted on the X-axis slide 24 along the Y-axis direction, the Z-axis slide 26 is movably mounted on the Y-axis slide 25 along the Z-axis direction, and the rotary drive 27 is mounted on the Z-axis slide 26. The output shaft of the rotary drive 27 is connected to the rotary seat 21 in a transmission connection.
[0039] Specifically, the X-axis slide 24 can be connected to the machine body 1 via a linear guide or a slider to achieve movement in the X direction. The Y-axis slide 25 can be mounted on the X-axis slide 24 in a similar manner, and the Z-axis slide 26 is mounted on the Y-axis slide 25, forming a three-dimensional motion system. The rotary drive component 27 can be fixed to the Z-axis slide 26 via a flange or bracket. The output shaft of the rotary drive component 27 can be connected to the rotary seat 21 via a coupling, gear drive, or belt drive to achieve power transmission.
[0040] This three-axis slide with rotary drive design enhances the flexibility and precision of the transport assembly 2. It can precisely position the optical module 100 to be inspected in three-dimensional space and adjust the angle of the optical module to accommodate various types of end faces, including straight and angled end faces. This omnidirectional adjustment capability ensures that the end face of the optical module and the image acquisition assembly 3 always maintain the optimal relative position, thereby obtaining the clearest and most accurate images. Simultaneously, this design also improves the automation and efficiency of the device, reduces manual intervention, and enhances the consistency and repeatability of the inspection.
[0041] Another embodiment of this application provides a method for inspecting the end-face quality of a high-speed optical module, which is applied to the end-face quality inspection device for the high-speed optical module provided in the above embodiments. Figure 7 This is a flowchart of the end-face quality inspection method for high-speed optical modules.
[0042] See also Figure 7 The end-face quality inspection methods for high-speed optical modules include: Adjust the optical module to be tested so that its end face is perpendicular to the imaging center line of the image acquisition component; Specifically, firstly, the transport component uses its three-axis slide system to roughly move the optical module under test to a preset position. The X-axis slide controls the optical module's forward and backward movement relative to the image acquisition component, the Y-axis slide controls its left and right movement, and the Z-axis slide adjusts its height. Subsequently, the rotation drive mechanism comes into play, adjusting the orientation of the optical module's end face by precisely controlling the angle of the rotation seat. For example, for an optical module with an 8-degree angled end face, the rotation drive mechanism needs to drive the rotation seat to rotate precisely 8 degrees so that the end face is perpendicular to the imaging center line.
[0043] During this process, the image acquisition component continuously captures real-time images of the optical module's end face. The processor analyzes these images to determine if the end face has reached a vertical position. The processor can use edge detection algorithms to analyze the symmetry of the end face contour, or use laser ranging technology to accurately measure the distance from each point on the end face to the image acquisition component. If a deviation is detected, the processor calculates the angle and position that need to be adjusted in real time and sends instructions to the transport component for fine-tuning.
[0044] The entire adjustment process is automated, greatly reducing human error and improving the efficiency and accuracy of detection. This precise adjustment capability allows the device to adapt to various types of optical modules, including those with straight end faces, angled end faces, and even complex curved surfaces. By ensuring that the end face is strictly perpendicular to the imaging center line, the device lays a solid foundation for subsequent multispectral imaging and defect detection, greatly improving the reliability and consistency of the detection results.
[0045] Please continue reading Figure 7Multispectral imaging is performed on the end face of the optical module to be tested to obtain a multispectral image set including visible light band, near-infrared band and ultraviolet band; Specifically, firstly, the image acquisition component is equipped with a high-performance multispectral camera capable of imaging in the visible, near-infrared, and ultraviolet bands. During imaging, the processor controls the image acquisition component to sequentially switch filters or light sources in different bands to acquire images across various spectral ranges. For example, for a standard FC / PC fiber optic connector, imaging is first performed in the 400-700nm visible light band, which helps capture the overall morphology of the end face and obvious surface defects. Subsequently, the system switches to the 750-1400nm near-infrared band, which is particularly effective for detecting microcracks and internal stress, as certain defects produce significant scattering or reflection in this band. Finally, the system images in the 200-400nm ultraviolet band, which is very helpful for detecting surface contaminants and the integrity of certain special coatings.
[0046] The entire multispectral imaging process is precisely controlled by the processor, ensuring that images in each band are acquired under optimal exposure and focal length conditions. The processor also adjusts parameters of the image acquisition components in real time, such as exposure time, gain, and contrast, to adapt to the optical characteristics of different bands. For example, when imaging in the ultraviolet band, since most materials have low reflectivity in this band, the system may automatically increase the exposure time or increase the sensor gain. Furthermore, to ensure precise alignment between images from different bands, the system performs rapid image registration after each band switch, typically achieved by identifying feature points on the end face.
[0047] The advantage of this multispectral imaging method lies in its ability to comprehensively capture various features and potential defects on the end face of the optical module. Visible light images provide intuitive surface information, near-infrared images can reveal deep structural problems, while ultraviolet images are particularly suitable for detecting surface contamination and the integrity of special coatings. By comprehensively analyzing images from these three bands, defects that are difficult to detect using traditional single-spectral detection can be detected, greatly improving the comprehensiveness and accuracy of the detection. In addition, multispectral images provide a rich data foundation for subsequent deep learning analysis, enabling the system to more accurately identify and classify various types of defects, thus providing a more reliable basis for the quality control of optical modules.
[0048] Please continue reading Figure 7 Based on the multispectral image set, the image of each band is preprocessed; Specifically, the preprocessing first involves dark-field correction, which eliminates the effects of inherent sensor noise and thermal noise by subtracting the dark-field image acquired under no-light conditions. Dark-field correction is particularly important for near-infrared images because thermal noise is more pronounced in this band. Next, flat-field correction is performed, using a flat-field image acquired under a uniform light source to compensate for non-uniformity in the optical system and differences in pixel response. This step is crucial for ensuring consistent image brightness, especially when detecting defects in the edge regions of fiber optic end faces.
[0049] Subsequently, the system employs an adaptive median filter for denoising, effectively removing impulse noise from the image while preserving edge and detail information. This method is particularly suitable for processing fiber optic end-face images that may contain minor scratches or blemishes. For example, for an end-face image of an FC / PC connector, the adaptive median filter can effectively remove random noise caused by dust without obscuring the fine structure of the end-face. Following this, the system applies a non-local means (NLM) algorithm for further denoising. This algorithm suppresses noise by finding similar pixel blocks in different regions of the image, making it particularly suitable for preserving image texture details.
[0050] To enhance image contrast and detail visibility, the system applies a Contrast-Limited Adaptive Histogram Equalization (CLAHE) algorithm to each band of the image. This method effectively improves local image contrast while avoiding excessive noise enhancement. The CLAHE algorithm is particularly effective for ultraviolet (UV) band images, as images in this band typically have low contrast. CLAHE processing makes minute contaminants or coating defects on the end face more visible, facilitating subsequent defect detection.
[0051] In the final stage of preprocessing, the system performs precise registration of the multispectral images. Using the visible light image as a reference, the near-infrared and ultraviolet images are spatially aligned. This step employs feature-point-based registration methods, such as the SIFT (Scale Invariant Feature Transform) algorithm, to extract feature points, and then uses the RANSAC (Random Sample Consensus) algorithm to estimate the transformation matrix. Finally, image resampling is performed through bicubic interpolation to ensure precise spatial correspondence between images of different bands. This precise registration is crucial for subsequent multispectral analysis because it allows the system to compare information from different bands at the pixel level.
[0052] The advantages of this series of preprocessing steps are that they significantly improve image quality and data reliability. By eliminating various noises and systematic errors, enhancing image contrast, and ensuring precise alignment of images across different bands, a solid foundation is laid for subsequent feature extraction and defect detection. This comprehensive and meticulous preprocessing not only improves detection accuracy but also enhances the system's sensitivity to minute defects. For example, a tiny crack that is almost invisible in the original image may become clearly discernible after these preprocessing steps. Furthermore, this standardized preprocessing workflow ensures consistency between images acquired in different batches and at different times, which is particularly important for long-term quality monitoring and trend analysis. In summary, this preprocessing process greatly improves the reliability, accuracy, and sensitivity of the entire detection system, providing strong support for high-quality optical module end-face inspection.
[0053] Please continue reading Figure 7 A multi-scale feature extraction network is used to extract features from the preprocessed multispectral image set to obtain feature maps at multiple scales. In one embodiment of the present invention, the step of using a multi-scale feature extraction network to extract features from a preprocessed multispectral image set to obtain feature maps at multiple scales includes: constructing a multi-scale feature extraction network comprising an encoder and a decoder, wherein the encoder comprises multiple downsampling stages and the decoder comprises multiple upsampling stages corresponding to the downsampling stages; introducing a residual global perception module in each stage of the encoder and decoder, wherein the residual global perception module divides the input feature map into multiple groups and applies spatially specific convolution and channel-specific convolution within and between groups, respectively; adding skip connections between corresponding stages of the encoder and decoder to connect the feature maps of the encoder stage with the feature maps of the corresponding stages of the decoder; processing the preprocessed multispectral image set through the multi-scale feature extraction network, and outputting feature maps from each stage of the decoder to obtain feature maps at multiple scales.
[0054] Specifically, the encoder contains multiple downsampling stages, each consisting of two 3x3 convolutional layers and a max-pooling layer. For example, for an input image of 640×480 pixels, the first downsampling stage halves its size to 320×240 while increasing the number of feature channels. This progressively decreasing spatial resolution design allows the network to capture features at different scales, from detail to the global. Conversely, the decoder contains multiple upsampling stages, each using transposed convolutions to enlarge the spatial size of the feature maps, restoring them to their original resolution.
[0055] In each stage of the encoder and decoder, this invention introduces an innovative Residual Global Awareness Module (RGPM). RGPM first divides the input feature map into multiple groups, for example, dividing a 64-channel feature map into 4 groups of 16 channels each. Within each group, RGPM applies spatially specific convolutions, which share parameters in the spatial dimension but are independent between channels. This effectively captures spatial correlations while maintaining channel independence. Next, RGPM applies channel-specific convolutions between groups, which share parameters in the channel dimension but are independent between spatial locations. This design allows the network to learn the inter-channel relationships while maintaining the independence of spatial information. Through this alternating convolution within and between groups, RGPM can simultaneously capture both local and global contextual information, significantly enhancing the network's feature extraction capabilities.
[0056] To better preserve and utilize information at different scales, this invention adds skip connections between corresponding stages of the encoder and decoder. These connections directly link the feature map of a certain stage of the encoder to the feature map of the corresponding stage of the decoder. For example, the feature map of the second stage of the encoder is connected to the feature map of the penultimate stage of the decoder. This design effectively solves the gradient vanishing problem in deep networks while also allowing high-resolution detail information to be directly passed to later stages of the decoder. This is particularly important for fiber optic end-face inspection, as it ensures that details such as tiny scratches or cracks are not lost in the deeper layers of the network.
[0057] When processing the preprocessed multispectral image set, the multi-scale feature extraction network first uses images from different spectral bands as input channels. For example, visible, near-infrared, and ultraviolet images can form an input tensor with multiple channels. The encoder part of the network progressively extracts and compresses features, while the decoder part reconstructs and refines these features. Notably, each stage of the decoder outputs a feature map, which represents image features at different scales. For example, for a network with five downsampling / upsampling stages, five feature maps at different scales will be obtained, corresponding to 1 / 32, 1 / 16, 1 / 8, 1 / 4, and 1 / 2 the size of the original image, respectively.
[0058] The advantage of this multi-scale feature extraction method lies in its ability to comprehensively capture various features of the fiber endface. Larger-scale feature maps contain global information about the endface, such as its overall shape and large-area defects, while smaller-scale feature maps retain fine local details, such as tiny scratches or blemishes. By combining information from different scales, the network can more accurately identify and locate defects of various types and sizes. Furthermore, the introduction of the residual global perception module greatly enhances the network's ability to capture complex features, enabling it to better understand the relationships between images in different spectral bands, thereby extracting richer and more meaningful features.
[0059] In summary, this meticulously designed multi-scale feature extraction network provides a strong foundation for subsequent defect detection and classification tasks. It not only processes image information across different spectral bands but also extracts and integrates features at multiple scales, significantly improving the system's ability to detect various types of defects. The flexibility and powerful feature representation capabilities of this method enable it to perform exceptionally well in handling various complex fiber optic end-face inspection scenarios, providing a reliable quality control method for high-quality optical module production.
[0060] Please continue reading Figure 7 The multi-scale feature map is fused to obtain a fused feature map. In one embodiment of the present invention, the step of fusing features from the multi-scale feature map to obtain a fused feature map includes: performing channel compression and excitation processing on the multi-scale feature map to obtain a channel weight map; generating a two-dimensional spatial attention map based on the multi-scale feature map; performing element-wise multiplication of the channel weight map and the two-dimensional spatial attention map to obtain an attention-weighted feature map; fusing features from the attention-weighted feature map and the multi-scale feature map to obtain an intermediate feature map; and using a feature pyramid network to reconstruct multi-scale features from the intermediate feature map to obtain the fused feature map.
[0061] Specifically, firstly, channel compression and activation processing is performed on the multi-scale feature maps. This step employs an improved squeeze-and-excitation module. Specifically, global average pooling is first performed on the feature map at each scale, compressing the spatial dimension into a vector representing the global response of each channel. Subsequently, this vector is processed through two fully connected layers and a non-linear activation function (such as ReLU) to obtain a weight vector with the same number of input channels. This weight vector is the channel weight map, reflecting the importance of different channels. For example, for a feature map containing 64 channels, channel compression and activation processing will generate a 64-dimensional weight vector, where each element corresponds to the importance score of a channel.
[0062] Simultaneously, a two-dimensional spatial attention map is generated based on the multi-scale feature maps. This step utilizes a spatial attention module, which first performs max pooling and average pooling on the feature maps along the channel dimension, resulting in two two-dimensional feature maps. These two feature maps are concatenated and then passed through a convolutional layer and a sigmoid activation function to generate a two-dimensional spatial attention map. This attention map reflects the importance of different spatial locations in the feature maps. This step is particularly meaningful for fiber end-face inspection because it can highlight areas on the end face that may contain defects, such as the end-face edges or central regions.
[0063] Next, the channel weight map and the two-dimensional spatial attention map are multiplied element-wise to obtain an attention-weighted feature map. This step essentially combines channel attention and spatial attention to generate a comprehensive attention mechanism. For example, if a channel scores highly in the channel weight map and a spatial location also scores highly in the spatial attention map, then the feature of that location in that channel will be significantly enhanced. This combination allows the network to simultaneously focus on important feature channels and important spatial locations, greatly improving the expressive power of the features.
[0064] Then, the attention-weighted feature map is fused with the original multi-scale feature map to obtain an intermediate feature map. This step is usually achieved through element-wise addition or concatenation operations. This fusion preserves the comprehensive information of the original feature map while highlighting the important features selected by the attention mechanism, thereby improving the discriminative power of the features while maintaining their integrity.
[0065] Finally, the intermediate feature maps are reconstructed at multiple scales using a Feature Pyramid Network (FPN) to obtain the final fused feature map. The core idea of FPN is to construct a top-down feature pyramid while using lateral connections to fuse features at different resolutions. Specifically, FPN starts with the highest-level (lowest resolution) feature map and gradually constructs high-resolution feature maps through upsampling and 1×1 convolutions. At each layer, the upsampled feature map is fused with the intermediate feature map of the corresponding resolution. This process continues until the original resolution is reached, ultimately resulting in a series of feature maps with different resolutions but rich semantic information—this is the fused feature map.
[0066] This complex feature fusion and reconstruction process offers several advantages. First, channel compression and excitation processing adaptively adjust the importance of different channels, which is particularly useful for processing multispectral images, as information from different bands may have varying degrees of importance. Second, spatial attention mechanisms help the network focus on key regions where defects may exist, improving detection accuracy. Combining these two attention mechanisms further enhances the network's expressive power and specificity. The use of a feature pyramid network ensures that the detection system obtains semantically rich feature representations at different scales, which is crucial for detecting defects of varying sizes.
[0067] In summary, this feature fusion and reconstruction process significantly enhances the network's ability to detect fiber end-face defects. It not only effectively integrates multi-scale, multi-spectral information but also intelligently focuses on important features and regions while maintaining sensitivity to defects at different scales. The flexibility and powerful feature representation capabilities of this method enable it to perform exceptionally well in handling various complex fiber end-face inspection scenarios, providing a reliable and precise quality control method for high-quality optical module production.
[0068] Please continue reading Figure 7 The fused feature map is analyzed using a defect detection network to identify the location, type, and severity of defects. In one embodiment of the present invention, the step of analyzing the fused feature map using a defect detection network to identify the location, type, and severity of defects includes: applying a region proposal network to the fused feature map to obtain candidate defect regions, wherein the region proposal network uses circular anchor boxes instead of rectangular anchor boxes; extracting region features from the fused feature map based on the candidate defect regions to obtain a region feature map; inputting the region feature map into a dual-branch network, where one branch performs defect classification and the other branch performs bounding box regression to obtain a defect type probability distribution and refined bounding box coordinates; performing post-processing using a non-maximum suppression algorithm based on the defect type probability distribution and the refined bounding box coordinates to obtain the final defect detection result; and, based on the final defect detection result, quantifying and scoring each detected defect according to a preset evaluation standard to obtain information on the location, type, and severity of the defects.
[0069] Specifically, firstly, a region proposal network is applied to the fused feature map. This network is unique in that it uses circular anchor frames instead of traditional rectangular ones. This design is well-suited to the circular characteristics of fiber end faces, as they are typically circular, allowing for more precise localization of potential defect regions. For example, for a standard single-mode fiber end face with a diameter of 125 micrometers, the region proposal network might generate a series of circular anchor frames of varying sizes, ranging from 5 to 50 micrometers, to cover a wide range of possible defect sizes. The use of these circular anchor frames significantly improves the accuracy and efficiency of candidate defect regions while reducing unnecessary background information interference.
[0070] Next, based on the obtained candidate defect regions, region features are extracted from the fused feature map to obtain a region feature map. This step typically uses the RoIAlign (Region of Interest Align) technique, which can accurately map irregularly shaped candidate regions onto a fixed-size feature map, maintaining the continuity of spatial information. For example, each candidate defect region may be mapped onto a 7×7 feature map, regardless of the size of the original candidate region. This method ensures consistency in subsequent analysis while preserving detailed information.
[0071] The extracted region feature maps are then input into an innovative dual-branch network. One branch of this network is responsible for defect classification, while the other branch performs bounding box regression. The classification branch, through a series of convolutional and fully connected layers, ultimately outputs the probability distribution for each possible defect type. For example, for fiber optic end faces, defect types include scratches, cracks, contamination, and bubbles. Simultaneously, the regression branch refines the bounding box coordinates, outputting the center coordinates, radius (for circular bounding boxes), or adjusted width and height (for rectangular bounding boxes). This dual-branch structure allows the network to simultaneously locate and identify defects, improving detection efficiency and accuracy.
[0072] To handle potentially overlapping detection results, this invention employs a non-maximum suppression (NMS) algorithm for post-processing. The NMS algorithm compares the detection confidence levels of overlapping regions, retaining the detection result with the highest confidence while suppressing other overlapping low-confidence detections. This step effectively eliminates redundant detection results and improves the accuracy of the final detection result. For example, if multiple detection boxes overlap and mark the same scratch, NMS will retain the detection box with the highest confidence and delete the other overlapping detection boxes.
[0073] Finally, based on the final defect detection results, this invention quantifies and scores each detected defect according to preset evaluation criteria. This evaluation process considers multiple factors, including the size, location, shape, and performance under different spectra. For example, for a detected scratch, the system comprehensively considers its length, depth (inferred from multispectral information), and distance from the fiber core. Each factor has a weight, ultimately yielding a comprehensive severity score. This quantitative scoring method not only provides information on the location and type of the defect but also indicates its severity, providing an important basis for subsequent quality control decisions.
[0074] The advantages of this entire defect detection and assessment process lie in its high accuracy, efficiency, and adaptability. The use of circular anchor frames is particularly suitable for the geometry of fiber endfaces, improving the accuracy of initial detection. The dual-branch network structure allows for simultaneous defect localization and classification, improving processing efficiency. Non-maximum suppression ensures the accuracy and non-redundancy of the detection results. Finally, the quantitative scoring system provides an objective and comprehensive defect assessment, which helps in developing precise quality control standards. This method not only accurately detects various types of defects but also assesses their severity, providing a powerful and flexible tool for quality control in the optical module manufacturing process, contributing to improved product quality and production efficiency.
[0075] Please continue reading Figure 7 Based on the identified defect information, the end face integrity rate is calculated and a defect density map is generated.
[0076] In one embodiment of the present invention, the step of calculating the end face integrity rate and generating a defect density map based on the identified defect information includes: dividing the end face into regions based on the location, type, and severity information of the defects to obtain defect regions and non-defect regions; calculating the ratio of the area of the non-defect regions to the total area of the end face to obtain the end face integrity rate; and generating a two-dimensional heat map based on the spatial distribution and severity of the defect regions to obtain the defect density map.
[0077] Specifically, firstly, the end face is precisely divided into defective and non-defective regions based on this information. This process utilizes previously obtained defect boundary box information, combined with defect type and severity, to accurately delineate the extent of each defect. For example, for a standard single-mode fiber end face with a diameter of 125 micrometers, if a scratch with a length of 10 micrometers is detected, the system will accurately mark this defective region on the end face image based on the scratch's location and direction.
[0078] Subsequently, this invention calculates the end-face integrity rate by dividing the area of the non-defective region by the total area of the end face. This ratio directly reflects the overall quality of the end face. For example, if the total area of an optical fiber end face is 12,271 square micrometers (corresponding to a circle with a diameter of 125 micrometers), and the total area of all detected defective regions is 100 square micrometers, then the end-face integrity rate will be 99.19%. This indicator provides a simple and clear reference standard for quality control.
[0079] Next, this invention generates a two-dimensional heatmap, also known as a defect density map, based on the spatial distribution and severity of the defect areas. In this heatmap, different colors represent different defect densities or severity levels. For example, red may represent highly concentrated or severe defect areas, while blue indicates minor or scattered defects. This visualization method not only intuitively shows the distribution of defects but also reflects their severity. For fiber optic end faces, this heatmap can clearly show whether defects are concentrated in the core region of the fiber, which has a significant impact on the transmission performance of the fiber.
[0080] The advantage of this comprehensive evaluation method lies in its multi-dimensional quality assessment. The end-face integrity rate provides a quantitative indicator of overall quality, facilitating rapid judgment and comparison of the quality of different fiber end-faces. Meanwhile, the defect density map provides more detailed and intuitive information on quality distribution, helping to identify potential problem areas and locations of defect concentrations. The combination of these two methods not only helps quality control personnel quickly assess the overall quality of fiber end-faces but also guides them to focus on specific problem areas, thereby developing more targeted improvement measures. Furthermore, this method provides a powerful tool for quality trend analysis in mass production, helping to identify systemic problems in the production process, thereby continuously optimizing production processes and improving product quality.
[0081] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the inventive concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.
Claims
1. A high-speed optical module end face quality detection device, characterized in that: include: Body (1); A transport component (2) is rotatably disposed on the body (1), and the transport component (2) is formed with a fixing position for fixing the optical module (100) to be detected; An image acquisition component (3) is rotatably disposed on the body (1), and the image acquisition component (3) is used to acquire a multi-spectral image of the end face of the optical module (100) to be detected; A processor is electrically connected to the transport component (2) and the image acquisition component (3), respectively, and the processor is used to adjust the relative position of the end face of the optical module (100) to be detected and the image acquisition component (3) so that the end face of the optical module (100) to be detected is perpendicular to the imaging center line of the image acquisition component (3), and the processor is capable of receiving and processing the multispectral image to identify the type and degree of defects on the end face of the optical module (100) to be detected.
2. The end face quality detection device for a high-speed optical module according to claim 1, characterized in that: The transport component (2) comprises a rotating seat (21), a supporting member (22) and a fixing member (23); the rotating seat (21) is formed with a mounting surface (211); the supporting member (22) is fixedly mounted on the mounting surface (211); an end of the supporting member (22) away from the mounting surface (211) is used to support the optical module (100) to be detected; the fixing member (23) is movably mounted on the mounting surface (211); an end of the fixing member (23) away from the mounting surface (211) is used to be fixedly connected to the optical module (100) to be detected; wherein: The fixing member (23) can move in a direction away from or close to the supporting member (22), so that the interval between the fixing member (23) and the supporting member (22) can be adjusted and adapted to optical modules (100) to be detected of different lengths.
3. The end face quality detection device for a high-speed optical module according to claim 2, characterized in that: The supporting member (22) comprises a frame body (221) fixed on the mounting surface (211) and a bracket (222) reciprocatably arranged on the frame body (221), wherein the end of the bracket (222) away from the mounting surface (211) is formed as a supporting portion (222a), and the bracket (222) is reciprocatably movable in a direction perpendicular to the mounting surface (211).
4. The end face quality detection device for a high-speed optical module according to claim 2, characterized in that: A first bolt hole (231) is formed on the fixing member (23), and a plurality of second bolt holes (212) are formed on the mounting surface (211). The plurality of second bolt holes (212) are arranged at intervals in the direction of a line connecting the fixing member (23) and the supporting member (22). The end face quality detection device of the high-speed optical module comprises bolts, and the bolts are respectively threadedly connected to the first bolt hole (231) and any one of the second bolt holes (212) so as to detachably fix the fixing member (23) on the mounting surface (211).
5. The end face quality detection device for a high-speed optical module according to claim 2, characterized in that: The transport assembly (2) comprises an X-axis slide (24), a Y-axis slide (25), a Z-axis slide (26) and a rotating drive member (27); the X-axis slide (24) is movably arranged on the machine body (1) along the X-axis direction; the Y-axis slide (25) is movably arranged on the X-axis slide (24) along the Y-axis direction; the Z-axis slide (26) is movably arranged on the Y-axis slide (25) along the Z-axis direction; the rotating drive member (27) is arranged on the Z-axis slide (26); and the output shaft of the rotating drive member (27) is drivingly connected to the rotating seat (21).
6. A method for detecting the end face quality of a high-speed optical module, characterized in that: The end face quality detection method of the high-speed optical module is applied to the end face quality detection device of the high-speed optical module according to any one of claims 1 to 5, and the end face quality detection method of the high-speed optical module comprises: Adjust the optical module to be detected until its end face is perpendicular to the imaging center line of the image acquisition component; Performing multispectral imaging on the end face of the optical module to be detected to obtain a multispectral image set including a visible light band, a near infrared band, and an ultraviolet band; Preprocessing the image of each band according to the multispectral image set; A multi-scale feature extraction network is used to extract features from the preprocessed multispectral image set to obtain multiple scale feature maps; Performing feature fusion on the multi-scale feature map to obtain a fused feature map; Analyzing the fused feature map using a defect detection network to identify the location, type, and severity of the defect; Based on the identified defect information, the end face intactness rate is calculated and a defect density map is generated.
7. The end face quality detection method of a high-speed optical module according to claim 6, characterized in that: The multi-scale feature extraction network is used to extract features from the pre-processed multispectral image set to obtain multiple scale feature maps, including: Constructing a multi-scale feature extraction network including an encoder and a decoder, wherein the encoder includes a plurality of downsampling stages, and the decoder includes a plurality of upsampling stages corresponding to the downsampling stages; Introducing a residual global perception module in each stage of the encoder and decoder, the residual global perception module divides the input feature map into multiple groups and applies space-specific convolution and channel-specific convolution within and between groups respectively; Adding a skip connection between corresponding stages of the encoder and decoder to connect the feature map of the encoder stage with the feature map of the corresponding stage of the decoder; The preprocessed multispectral image set is processed by the multi-scale feature extraction network, and feature maps are output from each stage of the decoder to obtain feature maps of multiple scales.
8. The end face quality detection method of a high-speed optical module according to claim 6, characterized in that: The step of performing feature fusion on the multi-scale feature map to obtain a fused feature map includes: Performing channel compression and excitation processing on the multi-scale feature map to obtain a channel weight map; Generating a two-dimensional spatial attention map according to the multi-scale feature map; Multiplying the channel weight map by the two-dimensional spatial attention map element by element to obtain an attention weighted feature map; Performing feature fusion on the attention weighted feature map and the multi-scale feature map to obtain an intermediate feature map; A feature pyramid network is used to perform multi-scale feature reconstruction on the intermediate feature map to obtain a fused feature map.
9. The end face quality detection method of a high-speed optical module according to claim 6, characterized in that: The method of analyzing the fused feature map using a defect detection network to identify the location, type and severity of the defect includes: Applying a region proposal network to the fused feature map to obtain a candidate defect region, wherein the region proposal network uses a circular anchor box instead of a rectangular anchor box; Extracting regional features from the fused feature map according to the candidate defect region to obtain a regional feature map; The regional feature map is input into a dual-branch structure network, where one branch performs defect classification and the other branch performs bounding box regression to obtain a defect type probability distribution and refined bounding box coordinates; According to the defect type probability distribution and the refined bounding box coordinates, a non-maximum suppression algorithm is used for post-processing to obtain a final defect detection result; Based on the final defect detection result, each detected defect is quantitatively scored according to a preset evaluation standard to obtain the location, type and severity information of the defect.
10. The end face quality detection method of a high-speed optical module according to claim 6, characterized in that: The method of calculating the end face integrity rate and generating a defect density map according to the identified defect information includes: According to the location, type and severity of the defect, the end face is divided into regions to obtain defective regions and non-defective regions; Calculating the ratio of the area of the non-defective region to the total area of the end face to obtain the end face intact rate; According to the spatial distribution and severity of the defective area, a two-dimensional thermal map is generated to obtain a defect density map.
Citation Information
Patent Citations
Porcelain core end surface automatic tester
CN102759529A
Multilayer circuit board detection system
CN118858326A
Image inspecting method and device for detecting faults
CN1335496A
Method of and device for inspecting images to detect defects
US20020054293A1
Cited By
Tailless threaded sleeve deformation detection system and method based on multispectral imaging
CN119879762A
Defect detection method and device in image pair and electronic device
CN120198433A
Defect Detection Method and Device in Image Pair, and Electronic Device
CN120198433B
Defect detection method for high-density packaged chip
CN120427657A
A defect detection method for high-density packaged chips
CN120427657B
