Microscope adjustment structure, optical module detection device and optical module detection method

Abstract

The present invention discloses a microscope adjustment structure, an optical module detection device and an optical module detection method. The structure includes an inner frame, a middle frame and an outer frame. The inner frame has a detection through hole penetrating along the Z-axis direction; the middle frame is arranged around the outer periphery of the inner frame, and the inner frame can be displaced relative to the middle frame along the X-axis direction; the outer frame includes an outer frame body arranged around the outer periphery of the middle frame, a front stopper and a rear stopper respectively fixed at opposite ends of the outer frame body, and the middle frame can be displaced relative to the outer frame body along the Y-axis direction; the middle frame and the inner frame are mutually limited in the Y-axis direction, and the middle frame and the front stopper form a limit match for limiting the displacement of the inner frame along the Z-axis direction, and at least one of the outer frame body and the front stopper and the middle frame are mutually limited in the X-axis direction, and the front stopper and the rear stopper form a limit match for limiting the displacement of the middle frame along the Z-axis direction. The technical solution of the present invention can improve the end face detection efficiency of the optical module and ensure the reliability of the detection result.

Description

Technical Field

[0001] The present invention relates to the technical field of optical module detection, and in particular to a microscope adjustment structure, an optical module detection device and an optical module detection method. Background Art

[0002] Optical modules are key components in optical communication equipment, and their core components are optical fiber connectors. During the production and use of optical modules, the end face quality of the optical fiber connector directly affects the transmission performance of the optical module. Defects such as scratches, contamination, and pits on the end face will cause optical signal attenuation, increased reflection loss, and even cause the performance of the entire communication system to deteriorate. Therefore, during the production and maintenance of optical modules, it is necessary to use an optical module detection device to detect the end face of the optical fiber connector to ensure that its quality meets the use requirements.

[0003] Existing optical module detection devices usually include a clamping mechanism for fixing the optical module, a microscopic imaging camera for collecting end face images, and a light source assembly for providing lighting. During detection, the operator needs to adjust the position of the optical module through the clamping mechanism so that its end face enters the field of view of the microscope and achieves focus. During this adjustment process, since the relative position of the optical module relative to the microscope objective lens needs to be repeatedly adjusted to obtain a clear end face image, shaking is prone to occur. This shaking can cause blurred end face imaging, especially under high magnification observation, even a slight shake can cause a large image offset, making it difficult to accurately identify end face defects, reducing detection efficiency and the reliability of detection results. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problems that in the process of inspecting the end face of an optical module by an existing optical module inspection device, image deviation occurs, resulting in difficulty in accurately identifying end face defects, low inspection efficiency and unreliable inspection results.

[0005] A first aspect of the present invention provides a microscope adjustment structure, the microscope adjustment structure comprising an inner frame, a middle frame and an outer frame;

[0006] The inner frame has a detection through hole penetrating along the Z-axis direction;

[0007] The middle frame is arranged around the outer periphery of the inner frame, and the inner frame can be displaced along the X-axis direction relative to the middle frame;

[0008] The outer frame includes an outer frame body arranged around the outer periphery of the middle frame, and a front stopper and a rear stopper respectively fixed at opposite ends of the outer frame body, and the middle frame can be displaced along the Y-axis direction relative to the outer frame body;

[0009] The middle frame and the inner frame are mutually limited in the Y-axis direction, and the middle frame and the front limit member form a limit fit for limiting the displacement of the inner frame along the Z-axis direction. At least one of the outer frame body and the front limit member and the middle frame are mutually limited in the X-axis direction, and the front limit member and the rear limit member form a limit fit for limiting the displacement of the middle frame along the Z-axis direction.

[0010] Optionally, the inner circumferential wall of the middle frame is provided with a first limiting boss extending along the X-axis direction, the front limiting piece, the inner circumferential wall of the middle frame and the first limiting boss are arranged to form a first guide groove extending along the X-axis direction, and a first guide portion is protrudingly provided on the outer circumferential wall of the inner frame, the first guide portion is slidably matched with the first guide groove, and the first guide portion is respectively abutted against two groove walls opposite to the first guide groove along the Y-axis direction.

[0011] Optionally, the microscope adjustment structure includes an X-axis adjustment member and an X-axis elastic member, one end of the X-axis adjustment member passes through the outer frame body and the middle frame in sequence along the X-axis direction from the outside of the outer frame body and then abuts against the inner frame;

[0012] The X-axis elastic member is arranged on a side of the inner frame away from the X-axis adjusting member, and two ends of the X-axis elastic member are respectively connected to the inner frame and the middle frame.

[0013] Optionally, the front limit member is a plate-shaped structure, and one of the front limit member and the middle frame is provided with a second guide groove extending along the Y-axis direction, and the other is provided with a second guide portion adapted to the second guide groove, the second guide portion slidingly cooperates with the second guide groove, and the second guide portion respectively abuts against two groove walls opposite to the second guide groove along the X-axis direction.

[0014] Optionally, the microscope adjustment structure includes a Y-axis adjustment member and a Y-axis elastic member, one end of the Y-axis adjustment member passes through the outer frame body along the Y-axis direction from the outside of the outer frame body and then abuts against the inner frame;

[0015] The Y-axis elastic member is arranged on a side of the middle frame away from the Y-axis adjusting member, and two ends of the Y-axis elastic member are respectively connected to the middle frame and the outer frame body.

[0016] A second aspect of the present invention provides an optical module detection device, the optical module detection device includes a microscope adjustment structure, the microscope adjustment structure includes an inner frame, a middle frame and an outer frame;

[0017] The inner frame has a detection through hole penetrating along the Z-axis direction;

[0018] The middle frame is arranged around the outer periphery of the inner frame, and the inner frame can be displaced along the X-axis direction relative to the middle frame;

[0019] The outer frame includes an outer frame body arranged around the outer periphery of the middle frame, and a front stopper and a rear stopper respectively fixed at opposite ends of the outer frame body, and the middle frame can be displaced along the Y-axis direction relative to the outer frame body;

[0020] The middle frame and the inner frame are mutually limited in the Y-axis direction, and the middle frame and the front limit member form a limit fit for limiting the displacement of the inner frame along the Z-axis direction. At least one of the outer frame body and the front limit member and the middle frame are mutually limited in the X-axis direction, and the front limit member and the rear limit member form a limit fit for limiting the displacement of the middle frame along the Z-axis direction.

[0021] A third aspect of the present invention provides an optical module detection method, comprising:

[0022] The acquired end face image of the optical module is used to extract geometric features, surface features and optical features through a deep separable convolutional network, and the extracted features are adaptively fused to obtain end face feature data;

[0023] According to the end face feature data, the displacement change between adjacent image frames is obtained to determine the degree of shaking, and the compensation parameters are calculated using a double-stream network to perform position correction and clarity enhancement on the end face image to obtain a compensated end face image;

[0024] Performing regional positioning on the compensated end face image to obtain a to-be-detected area, and performing feature analysis on the to-be-detected area to obtain defect feature data;

[0025] Classifying and processing the defect feature data to obtain preliminary defect detection results;

[0026] The reliability of the preliminary defect detection results and the defect degree assessment are performed to generate final detection results.

[0027] Optionally, the step of obtaining the displacement change between adjacent image frames based on the end face feature data, determining the degree of shaking, calculating compensation parameters using a dual-stream network, performing position correction and clarity enhancement on the end face image, and obtaining a compensated end face image includes:

[0028] Acquire end surface feature data of two adjacent image frames, perform optical flow calculation on the end surface feature data, and obtain a displacement vector field;

[0029] According to the displacement vector field, the displacement component and the acceleration component in the X-axis and Y-axis directions are calculated, and the shaking level is determined by combining the displacement threshold and the acceleration threshold;

[0030] Based on the shake level, generating a position compensation coefficient and a sharpening compensation coefficient by using a compensation prediction network, wherein the position compensation coefficient is used to correct the offset in the X-axis and Y-axis directions, and the sharpening compensation coefficient is used to correct different degrees of motion blur;

[0031] The position compensation coefficient and the sharpening compensation coefficient are respectively input into the spatial domain compensation network and the frequency domain compensation network to obtain the spatial domain compensation result and the frequency domain compensation result, and the spatial domain compensation result and the frequency domain compensation result are adaptively fused to obtain the compensated end face image.

[0032] Optionally, the inputting the position compensation coefficient and the sharpening compensation coefficient into a spatial domain compensation network and a frequency domain compensation network respectively to obtain a spatial domain compensation result and a frequency domain compensation result, and adaptively fusing the spatial domain compensation result and the frequency domain compensation result to obtain a compensated end face image includes:

[0033] The position compensation coefficient is used to modulate the convolution kernel parameters in the spatial domain compensation network to generate an adaptive convolution kernel, and the end face image is subjected to spatial domain transformation by the adaptive convolution kernel to obtain a spatial domain compensation result;

[0034] Using the sharpening compensation coefficient to modulate the Fourier transform coefficient in the frequency domain compensation network to generate a frequency domain compensation matrix, and multiplying the frequency domain compensation matrix with the frequency spectrum of the end face image to obtain a frequency domain compensation result;

[0035] The local clarity index of the spatial domain compensation result and the frequency domain compensation result is calculated, a fusion weight map is generated according to the local clarity index, and the spatial domain compensation result and the frequency domain compensation result are weightedly fused using the fusion weight map to obtain a compensated end face image.

[0036] Optionally, performing regional positioning on the compensated end face image to obtain a to-be-detected area, and performing feature analysis on the to-be-detected area to obtain defect feature data, includes:

[0037] Constructing a multi-scale pyramid structure for the compensated end face image, extracting local contrast features and edge gradient features at each scale level, and generating a regional saliency map;

[0038] Based on the peak distribution of the regional saliency map, an adaptive watershed algorithm is used to segment the salient region, and the segmented region is screened in combination with the prior information of the fiber core position to obtain the region to be detected;

[0039] Constructing an eight-directional Gabor filter group with an angle interval of 15°, performing filtering processing on each direction of the area to be detected to obtain a texture response map, and constructing a defect feature vector by combining the regional grayscale mean, standard deviation and gradient amplitude;

[0040] The defect feature vector is subjected to dimensionality reduction processing to obtain defect feature data.

[0041] The microscope adjustment structure provided in the present application forms a stable adjustment system through a three-layer nested structure of an inner frame, a middle frame and an outer frame, and a limit matching mechanism between the frames. Among them, the limit matching between the inner frame and the middle frame enables the inner frame to move only along the X-axis direction, and the limit matching between the middle frame and the outer frame enables the middle frame to move only along the Y-axis direction. At the same time, the limit matching between the front limiter, the rear limiter and each frame suppresses the displacement in the Z-axis direction. This structural design allows the operator to decompose the adjustment action in the three-dimensional space into two independent planar movements of the X-axis and the Y-axis when adjusting the position of the optical module for focusing, and each frame is limited to the specified direction of movement during the adjustment process, which effectively avoids shaking during the adjustment process, ensures the stability and clarity of the end face image, and improves the detection efficiency and the reliability of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 creative work.

[0043] Figure 1 It is a structural schematic diagram of an embodiment of an optical module detection device of the present invention;

[0044] Figure 2 for Figure 1 The schematic diagram of the structure after omitting the optical module and other structures;

[0045] Figure 3 It is a structural schematic diagram of the microscope adjustment structure of the present invention;

[0046] Figure 4 for Figure 3 A structural diagram from another perspective;

[0047] Figure 5 for Figure 4 The schematic diagram of the structure after the moving cylinder is omitted;

[0048] Figure 6 for Figure 5 Schematic diagram of the decomposition of

[0049] Figure 7 for Figure 6 A structural diagram from another perspective;

[0050] Figure 8 The figure is a flow chart of the optical module detection method of the present invention.

[0051] Description of Figure Numbers:

[0052] 100. objective lens; 200. microscope adjustment structure; 1. inner frame; 11. detection through hole; 12. first guide part; 2. middle frame; 21. first limit boss; 22. second guide groove; 3. outer frame; 31. outer frame body; 32. front limit piece; 321. second guide part; 322. positioning screw hole; 33. rear limit piece; 4. first guide groove; 5. X-axis adjustment piece; 6. X-axis elastic piece; 7. Y-axis adjustment piece; 8. Y-axis elastic piece; 9. moving cylinder; 300. optical module; 400. imaging camera; 500. light source assembly.

[0053] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0055] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0056] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, "and / or" in the full text includes three solutions. Taking A and / or B as an example, it includes technical solution A, technical solution B, and technical solution that satisfies both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, which must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0057] The present invention provides a microscope adjustment structure 200 .

[0058] In the embodiment of the present invention, Figures 1 to 7 As shown, the microscope adjustment structure 200 includes an inner frame 1, a middle frame 2 and an outer frame 3;

[0059] The inner frame 1 has a detection through hole 11 penetrating along the Z-axis direction;

[0060] The middle frame 2 is arranged around the outer periphery of the inner frame 1, and the inner frame 1 can be displaced along the X-axis direction relative to the middle frame 2;

[0061] The outer frame 3 includes an outer frame body 31 arranged around the outer periphery of the middle frame 2, and a front stopper 32 and a rear stopper 33 respectively fixed at opposite ends of the outer frame body 31. The middle frame 2 can be displaced along the Y-axis direction relative to the outer frame body 31; wherein,

[0062] The middle frame 2 and the inner frame 1 are mutually limited in the Y-axis direction, and the middle frame 2 and the front limit member 32 form a limit cooperation for limiting the displacement of the inner frame 1 along the Z-axis direction. At least one of the outer frame body 31 and the front limit member 32 and the middle frame 2 are mutually limited in the X-axis direction, and the front limit member 32 and the rear limit member 33 form a limit cooperation for limiting the displacement of the middle frame 2 along the Z-axis direction.

[0063] Specifically, the microscope adjustment structure 200 is applied to an optical module detection device. A positioning screw hole 322 for installing the optical module 300 is provided on the front limiter 32, and an observation hole is provided at the center of the front limiter 32, so that the end face of the optical module 300 installed at the positioning screw hole 322 is optically coaxially arranged with the detection through hole 11 of the inner frame 1 through the observation hole. The rear end of the inner frame 1 is fixedly connected to the moving cylinder 9, and the moving cylinder 9 is used to fix the microscope objective 100. A light source assembly 500 for lighting and an imaging camera 400 for image acquisition are also provided in the housing of the optical module detection device, and the optical axis of the imaging camera 400 is spatially aligned with the detection through hole 11. During the detection process, the operator can drive the inner frame 1 to move in the X-axis direction through the X-axis adjustment member 5. Since the moving cylinder 9 is fixedly connected to the inner frame 1, the objective lens 100 moves accordingly, thereby adjusting the relative position between the end face of the optical module 300 and the objective lens 100 in the X-axis direction; similarly, when the middle frame 2 is driven to move in the Y-axis direction through the Y-axis adjustment structure, since the inner frame 1 is installed in the middle frame 2, the inner frame 1 and the connected moving cylinder 9 and the objective lens 100 also move in the Y-axis direction, thereby adjusting the relative position between the end face of the optical module 300 and the objective lens 100 in the Y-axis direction. This separate adjustment mechanism enables the operator to independently and accurately control the position adjustment in the X-axis and Y-axis directions.

[0064] In this embodiment, the frame refers to a supporting structure made of rigid materials such as metal or engineering plastics, wherein the inner frame 1, the middle frame 2 and the outer frame body 31 can all be made of stainless steel, aluminum alloy or reinforced engineering plastics; the front limit member 32 and the rear limit member 33 can be a plate-like structure or a ring-like structure, also made of metal or engineering plastics.

[0065] In the above, the "mutual limiting fit" mentioned refers to the mechanical structure between two components to limit movement in a certain direction while allowing movement in other directions. For example, the middle frame 2 and the inner frame 1 are mutually limited in the Y-axis direction, which means that the inner frame 1 cannot move relative to the middle frame 2 in the Y-axis direction, but can move in the X-axis direction; similarly, the outer frame body 31 and the front limiting member 32 are mutually limited in the X-axis direction with the middle frame 2, which means that the middle frame 2 cannot move relative to the outer frame 3 in the X-axis direction, but can move in the Y-axis direction.

[0066] It can be understood that the microscope adjustment structure 200 provided in the present application forms a stable adjustment system through the three-layer nested structure of the inner frame 1, the middle frame 2 and the outer frame 3, and the limit matching mechanism between the frames. Among them, the limit matching between the inner frame 1 and the middle frame 2 enables the inner frame 1 to move only along the X-axis direction, and the limit matching between the middle frame 2 and the outer frame 3 enables the middle frame 2 to move only along the Y-axis direction. At the same time, the limit matching between the front limit member 32, the rear limit member 33 and each frame suppresses the displacement in the Z-axis direction. This structural design allows the operator to decompose the adjustment action in the three-dimensional space into two independent plane movements of the X-axis and the Y-axis when adjusting the position of the optical module 300 for focusing, and each frame is limited to the specified movement direction during the adjustment process, which effectively avoids shaking during the adjustment process, ensures the stability and clarity of the end face image, and improves the detection efficiency and the reliability of the detection results.

[0067] Optionally, the inner circumferential wall of the middle frame 2 is provided with a first limiting boss 21 extending along the X-axis direction, the front limiting member 32, the inner circumferential wall of the middle frame 2 and the first limiting boss 21 are arranged to form a first guide groove 4 extending along the X-axis direction, and a first guide portion 12 is protrudingly provided on the outer circumferential wall of the inner frame 1, the first guide portion 12 is slidably matched with the first guide groove 4, and the first guide portion 12 is respectively abutted against two opposite groove walls of the first guide groove 4 along the Y-axis direction.

[0068] In this embodiment, the first limiting boss 21 can be a raised structure integrally formed of metal or engineering plastic or welded and fixed on the inner circumferential wall of the middle frame 2; the first guide groove 4 is a groove structure extending along the X-axis direction and is jointly surrounded by the inner circumferential wall of the middle frame 2, the front limiting member 32 and the first limiting boss 21; the first guide portion 12 can be a raised structure integrally formed of metal or engineering plastic or welded and fixed on the outer circumferential wall of the inner frame 1, and the shape of the first guide portion 12 is adapted to the shape of the first guide groove 4.

[0069] "Enclosed formation" means that the first limiting boss 21, the inner peripheral wall of the middle frame 2 and the front limiting member 32 together form a first guide groove 4 in spatial position, and the first guide groove 4 extends along the X-axis direction. "Aggressive" means that there is physical contact between the first guide portion 12 and the front limiting member 32, and between the first guide portion 12 and the first limiting boss 21, and the limiting in the Z-axis direction is achieved through such contact.

[0070] Optionally, the microscope adjustment structure 200 includes an X-axis adjustment member 5 and an X-axis elastic member 6, one end of the X-axis adjustment member 5 passes through the outer frame body 31 and the middle frame 2 in sequence along the X-axis direction from the outside of the outer frame body 31 and then abuts against the inner frame 1;

[0071] The X-axis elastic member 6 is disposed on a side of the inner frame 1 away from the X-axis adjusting member 5 , and two ends of the X-axis elastic member 6 are respectively connected to the inner frame 1 and the middle frame 2 .

[0072] The X-axis adjustment member 5 can be a lead screw, a screw or other mechanical structure with adjustment function, generally a precision-machined metal screw; the X-axis elastic member 6 can be a compression spring, an elastic rubber member or other structural member with elastic deformation function, and a compression spring is preferably used in this embodiment.

[0073] In terms of connection, the X-axis adjustment member 5 passes through a pre-processed through hole from the outside of the outer frame body 31, and the through hole can be a smooth hole or a threaded hole. After the X-axis adjustment member 5 passes through the outer frame body 31 and the middle frame 2, its end is against the end face of the inner frame 1, and point contact can be achieved through a ball head structure, or surface contact can be achieved through a plane structure. The two ends of the X-axis elastic member 6 are fixedly connected to the inner frame 1 and the middle frame 2 respectively, and the fixing method can be a slot fixing, an end seat fixing or other mechanical fixing methods. The position where one end is fixed to the inner frame 1 is on the side away from the X-axis adjustment member 5, so that the X-axis elastic member 6 can provide a reverse thrust for the X-axis adjustment member 5.

[0074] This adjustment structure forms a balanced adjustment mechanism through the precise adjustment of the X-axis adjustment member 5 and the elastic force of the X-axis elastic member 6. The X-axis elastic member 6 provides a constant damping force, which can avoid a large displacement during the adjustment process, making the position adjustment in the X-axis direction more delicate and stable. At the same time, when the X-axis adjustment member 5 rotates in the opposite direction, the elastic force of the X-axis elastic member 6 can also drive the inner frame 1 to move in the direction of the X-axis adjustment member 5, playing a role in assisting reset. This bidirectional adjustment feature makes the position adjustment in the X-axis direction more flexible and controllable.

[0075] Optionally, the front limit member 32 is a plate-shaped structure, and one of the front limit member 32 and the middle frame 2 is provided with a second guide groove 22 extending along the Y-axis direction, and the other is provided with a second guide portion 321 adapted to the second guide groove 22, the second guide portion 321 is slidably matched with the second guide groove 22, and the second guide portion 321 is respectively abutted against two groove walls opposite to the second guide groove 22 along the X-axis direction.

[0076] The plate-shaped structure refers to a flat component with a certain thickness of the front limit member 32, which can be made of metal plates or engineering plastic plates; the second guide groove 22 refers to a groove structure extending along the Y-axis direction, which can be directly processed on the front limit member 32 or the middle frame 2 by mechanical processing; the second guide portion 321 refers to a protruding structure that cooperates with the second guide groove 22, and its shape and size are compatible with the second guide groove 22, and can also be directly processed on the front limit member 32 or the middle frame 2 by mechanical processing.

[0077] The design of this guide structure makes the middle frame 2 more stable when moving along the Y-axis direction. The cooperation between the second guide groove 22 and the second guide portion 321 can not only guide the middle frame 2 to move along the Y-axis direction, but also effectively prevent the middle frame 2 from shaking in the X-axis direction through the abutment between the second guide portion 321 and the guide groove wall.

[0078] Optionally, the microscope adjustment structure 200 includes a Y-axis adjustment member 7 and a Y-axis elastic member 8, one end of the Y-axis adjustment member 7 passes through the outer frame body 31 along the Y-axis direction from the outside of the outer frame body 31 and then abuts against the inner frame 1;

[0079] The Y-axis elastic member 8 is disposed on a side of the middle frame 2 away from the Y-axis adjusting member 7 , and two ends of the Y-axis elastic member 8 are respectively connected to the middle frame 2 and the outer frame body 31 .

[0080] The Y-axis adjustment member 7 can be a lead screw, a screw or other mechanical structure with adjustment function, generally a precision-machined metal screw; the Y-axis elastic member 8 can be a compression spring, an elastic rubber member or other structural member with elastic deformation function, preferably a compression spring in this embodiment. The combination of the two can achieve accurate adjustment of the position of the middle frame 2.

[0081] In terms of connection, the Y-axis adjustment member 7 passes through a pre-processed through hole from the outside of the outer frame body 31, and the through hole can be a smooth hole or a threaded hole. After the Y-axis adjustment member 7 passes through the outer frame body 31, its end is against the end face of the middle frame 2, and point contact can be achieved through a ball head structure, or surface contact can be achieved through a plane structure. The two ends of the Y-axis elastic member 8 are fixedly connected to the middle frame 2 and the outer frame body 31 respectively, and the fixing method can be a slot fixing, an end seat fixing or other mechanical fixing methods. The position where one end is fixed to the middle frame 2 is on the side away from the Y-axis adjustment member 7, so that the Y-axis elastic member 8 can provide a reverse thrust for the Y-axis adjustment member 7.

[0082] This adjustment structure forms a balanced adjustment mechanism through the precise adjustment of the Y-axis adjustment member 7 and the elastic force of the Y-axis elastic member 8. The Y-axis elastic member 8 provides a constant damping force, which can avoid large displacements during the adjustment process, making the position adjustment in the Y-axis direction more delicate and stable. At the same time, when the Y-axis adjustment member 7 rotates in the opposite direction, the elastic force of the Y-axis elastic member 8 can also drive the middle frame 2 to move in the direction of the Y-axis adjustment member 7, playing a role in assisting resetting. This bidirectional adjustment feature makes the position adjustment in the Y-axis direction more flexible and controllable.

[0083] It should be noted that, in this embodiment, the rear limiter 33 includes a limiter plate and a fastening screw, wherein the limiter plate may be a plate-shaped structure made of a metal plate or an engineering plastic plate, and the fastening screw may be a machine screw, a self-tapping screw or other standard parts with a fastening function. The number of the rear limiters 33 may be set according to actual needs, and a plurality of rear limiters 33 may be evenly arranged along the circumferential direction at the rear end of the outer frame body 31, thereby providing a more uniform limiter effect.

[0084] In terms of connection, the limit plate is fixedly connected to the outer frame body 31 by means of a fastening screw. Specifically, a plurality of threaded holes are provided at the rear end of the outer frame body 31, and a through hole is provided at a corresponding position on the limit plate. After the fastening screw passes through the through hole on the limit plate, it is threadedly connected with the threaded hole on the outer frame body 31, thereby fixing the limit plate to the rear end of the outer frame body 31. The end face of the limit plate abuts against the rear end face of the middle frame 2, which may be surface contact, point contact, or line contact.

[0085] The present invention also proposes an optical module detection device, which includes a microscope adjustment structure 200. The specific structure of the microscope adjustment structure 200 refers to the above embodiment. Since the optical module detection device adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought by the technical solutions of the above embodiments, which will not be repeated here one by one.

[0086] The microscope adjustment structure provided in this solution effectively reduces the shaking during the adjustment process through mechanical constraints, but in order to further improve the detection accuracy, it is also necessary to compensate for the residual slight shaking from the perspective of image processing. Therefore, the present invention also proposes an optical module detection method, which ensures that accurate detection results can be obtained even in the presence of slight shaking through technical means such as real-time shaking analysis, dynamic compensation and adaptive detection.

[0087] Specifically, see Figure 8 , the optical module detection method includes:

[0088] The acquired end face image of the optical module is used to extract geometric features, surface features, and optical features through a deep separable convolutional network, and the extracted features are adaptively fused to obtain end face feature data;

[0089] Specifically, the feature extraction of the acquired end-face image of the optical module adopts a deep separable convolutional network structure. The network preprocesses the input end-face image, including adjusting the image size to 256×256 pixels and normalizing the pixel values. The deep separable convolutional network improves the operation efficiency by splitting the traditional single convolution operation into two steps: first, the deep convolution performs feature extraction on each input channel separately. For example, for the three RGB channels of the image, a 3×3 convolution kernel is used for feature extraction respectively; then, the point-by-point convolution combines the features of each channel through a 1×1 convolution kernel to achieve cross-channel information fusion. This structural design reduces the amount of calculation to 1 / 8 to 1 / 10 of the traditional convolutional network while maintaining the feature extraction capability. In the processing of the end-face image of the optical module, based on this efficient network structure, the system extracts different types of features respectively through three parallel branches. The first branch extracts geometric features, focusing on the core arrangement features on the end face, such as the roundness of the core, positional relationship and other spatial information; through the spatial pyramid pooling structure, this branch can simultaneously obtain geometric information at different scales, such as extracting the shape features of a single core and the arrangement features between multiple cores. The second branch extracts surface features, mainly analyzing the surface state of the end face, such as scratches, stains and other defect features; this branch uses multi-scale convolution kernels to detect surface defects of different sizes, from subtle scratches to large areas of contamination. The third branch extracts optical features, mainly analyzing the reflection and scattering characteristics of the end face, which are crucial for evaluating the optical performance of the end face; for example, by analyzing the reflection intensity distribution of the end face, the polishing quality of the end face can be judged.

[0090] The feature fusion module first dynamically evaluates the importance of various features in the current detection task: when detecting surface defects, such as scratches and stains, the weight of the surface feature will automatically increase to above 0.7; when detecting the core arrangement, the weight of the geometric feature will automatically increase to above 0.8, ensuring that the system can focus on the most important feature information at the moment. Then, different feature channels are weighted and combined according to these score values ​​to ensure that the final feature expression can highlight the most critical information at the moment. For example, when obvious surface scratches appear in the image, the weight of the surface feature branch will automatically increase; when it is necessary to determine the core position, the weight of the geometric feature branch will increase accordingly. The core considerations for adopting this feature extraction and fusion method are: deep separable convolution can reduce the amount of calculation parameters and improve the efficiency of feature extraction; the multi-branch parallel structure design enables various features to be independently and fully extracted; the adaptive fusion based on the attention mechanism can flexibly adjust the feature weight according to the actual detection needs and improve the adaptability of feature expression.

[0091] Please continue reading Figure 8, according to the end face feature data, the displacement change between adjacent image frames is obtained, the degree of shaking is determined, the compensation parameters are calculated using a dual-stream network, the position of the end face image is corrected and the clarity is enhanced to obtain a compensated end face image;

[0092] In one embodiment of the present invention, the displacement change between adjacent image frames is obtained according to the end face feature data, the degree of shaking is determined, compensation parameters are calculated using a dual-stream network, position correction and clarity enhancement are performed on the end face image, and the compensated end face image is obtained, including:

[0093] Acquire end surface feature data of two adjacent image frames, perform optical flow calculation on the end surface feature data, and obtain a displacement vector field;

[0094] According to the displacement vector field, the displacement component and the acceleration component in the X-axis and Y-axis directions are calculated, and the shaking level is determined by combining the displacement threshold and the acceleration threshold;

[0095] Based on the shake level, generating a position compensation coefficient and a sharpening compensation coefficient by using a compensation prediction network, wherein the position compensation coefficient is used to correct the offset in the X-axis and Y-axis directions, and the sharpening compensation coefficient is used to correct different degrees of motion blur;

[0096] The position compensation coefficient and the sharpening compensation coefficient are respectively input into the spatial domain compensation network and the frequency domain compensation network to obtain the spatial domain compensation result and the frequency domain compensation result, and the spatial domain compensation result and the frequency domain compensation result are adaptively fused to obtain the compensated end face image.

[0097] Specifically, during the end face detection of the optical module, the adjustment of the mechanical structure will cause shaking, and it is necessary to analyze and compensate for the shaking of the continuously collected image frames. First, the end face feature data of two adjacent frames of images are obtained, and the optical flow calculation is performed on these feature data. In order to accurately measure the impact of shaking, the system uses optical flow calculation technology to track the movement of the image. The specific process is: first, a series of fixed feature points are marked in the end face image, mainly including the edge points of the fiber core (usually 8-12 evenly distributed points on the circular edge) and surface feature points (such as obvious reference marks on the end face), and then the direction and amplitude of the movement are obtained by comparing the position changes of these feature points in two adjacent frames of images. For example, if the center point of a fiber core is offset by 5 pixels in the X-axis direction and 3 pixels in the Y-axis direction, the optical flow vector of the point is (5, 3). By analyzing the optical flow vectors of all key points, the system can accurately evaluate the direction and amplitude of shaking. In this solution, the optical flow is calculated through a dual-stream network structure. The network contains two main parts: feature extraction and motion estimation, which can accurately capture the motion information in the end face image. For example, when the optical module shakes during adjustment, the position of the fiber core in the end face image will change accordingly, and the direction and magnitude of this change can be obtained through optical flow calculation.

[0098] The system analyzes the displacement vector field obtained by optical flow calculation, and extracts the displacement magnitude in the horizontal direction (X-axis) and vertical direction (Y-axis) respectively. For example, if a feature point is detected to have moved 5 pixels in the X-axis direction and 3 pixels in the Y-axis direction, the displacement component is recorded as (5, 3). At the same time, the acceleration information is obtained by calculating the rate of change of displacement between two adjacent frames to determine the severity of the shake. The degree of shake is quantitatively analyzed by the set displacement threshold and acceleration threshold. The displacement threshold is used to determine the magnitude of the displacement, and the acceleration threshold is used to evaluate the severity of the movement. The shake level is divided into multiple levels based on the comparison results of the displacement component and the acceleration component with the corresponding thresholds. For example, the degree of shaking is divided into three levels according to the displacement and acceleration values: when the displacement is less than 0.1mm and the acceleration is less than 0.5mm / s², it is defined as slight shaking; when the displacement is in the range of 0.1-0.5mm or the acceleration is in the range of 0.5-2mm / s², it is defined as moderate shaking; when the displacement is greater than 0.5mm and the acceleration is greater than 2mm / s², it is defined as severe shaking.

[0099] The compensation prediction network generates position compensation coefficients and sharpening compensation coefficients according to the shake level. The position compensation coefficient is generated for the X-axis and Y-axis directions respectively, and is used to correct the position offset. Its value is linearly related to the displacement. For example, when the X-axis displacement is 0.2mm, the corresponding position compensation coefficient is -0.2; the sharpening compensation coefficient is used to process motion blur, and its value increases with the shake level. For example, the value for slight shake is 0.3, the value for moderate shake is 0.6, and the value for severe shake is 0.9. The compensation prediction network adopts a residual learning structure, which can accurately predict the required compensation parameters. For example, for slight shake, a smaller position compensation coefficient and sharpening compensation coefficient are generated; for severe shake, a larger compensation coefficient is generated.

[0100] The compensation process is divided into two parts: spatial domain compensation and frequency domain compensation. The spatial domain compensation network processes directly in the image space and is mainly responsible for the correction of position offset; the frequency domain compensation network processes in the frequency domain of the image and is mainly used to improve clarity. The spatial domain compensation network receives the position compensation coefficient and corrects the image by means of a deformation field; the frequency domain compensation network receives the sharpening compensation coefficient and improves the image clarity by adjusting the frequency domain features. Finally, based on the characteristics of the two compensation results, a weighted fusion method is used to obtain the final compensated image. This dual-domain collaborative compensation method makes full use of the respective advantages of the spatial domain and the frequency domain: spatial domain compensation can accurately process position offsets, while frequency domain compensation is more suitable for processing blur. The combination of the two can better restore the end face image quality affected by shaking.

[0101] In one embodiment of the present invention, the position compensation coefficient and the sharpening compensation coefficient are respectively input into a spatial domain compensation network and a frequency domain compensation network to obtain a spatial domain compensation result and a frequency domain compensation result, and the spatial domain compensation result and the frequency domain compensation result are adaptively fused to obtain a compensated end face image, including:

[0102] The position compensation coefficient is used to modulate the convolution kernel parameters in the spatial domain compensation network to generate an adaptive convolution kernel, and the end face image is subjected to spatial domain transformation by the adaptive convolution kernel to obtain a spatial domain compensation result;

[0103] Using the sharpening compensation coefficient to modulate the Fourier transform coefficient in the frequency domain compensation network to generate a frequency domain compensation matrix, and multiplying the frequency domain compensation matrix with the frequency spectrum of the end face image to obtain a frequency domain compensation result;

[0104] The local clarity index of the spatial domain compensation result and the frequency domain compensation result is calculated, a fusion weight map is generated according to the local clarity index, and the spatial domain compensation result and the frequency domain compensation result are weightedly fused using the fusion weight map to obtain a compensated end face image.

[0105] Specifically, when processing the end face image of the optical module affected by shaking, compensation processing needs to be performed in the spatial domain and frequency domain respectively. The spatial domain compensation network uses an adaptive convolution kernel to perform image correction. The convolution kernel is a basic operator in image processing, which is usually used to extract image features or perform image transformation. In this scheme, the spatial domain compensation network dynamically generates convolution kernel parameters according to the position compensation coefficient, thereby realizing adaptive image transformation. Specifically, the position compensation coefficient determines the shape and weight distribution of the convolution kernel. When there is a large displacement in the image, the weight distribution of the convolution kernel will be adjusted accordingly to achieve a wider range of position correction. For example, during the detection of the end face of the optical module, if a large offset in the X-axis direction is detected, the weight distribution of the convolution kernel will be offset accordingly in the horizontal direction, thereby guiding the image to be corrected in the opposite direction. When the adaptive convolution kernel processes the end face image, it transforms point by point in the form of a local receptive field. This processing method can maintain the local structural characteristics of the image and avoid deformation or distortion.

[0106] Frequency domain processing is to enhance image quality by analyzing the frequency characteristics of the image. The system first performs Fourier transform on the end face image and decomposes the image into components of different frequencies: the low-frequency components of 0-50Hz correspond to the overall brightness and contour information of the image, the medium-frequency components of 50-150Hz contain edge and texture details, and the high-frequency components above 150Hz contain fine structure and noise information. In order to address the loss of high-frequency information caused by motion blur, the system enhances the signal in the 100-150Hz frequency band by 1.5-2 times, and appropriately attenuates the 150-200Hz frequency band that may contain noise, thereby achieving image clarity. Fourier transform is a mathematical tool that decomposes a signal into different frequency components. In image processing, high-frequency components usually correspond to image details, such as edges and textures, while low-frequency components correspond to the overall structure of the image. The sharpening compensation coefficient is used to modulate the frequency domain coefficients after Fourier transform to generate a frequency domain compensation matrix. In frequency domain processing, the image spectrum is divided into low frequency band (0-50Hz), medium frequency band (50-150Hz) and high frequency band (>150Hz). For the high frequency information loss caused by motion blur, the frequency domain compensation matrix will increase the high frequency band coefficient by 1.5-2 times; for the medium frequency band, the coefficient remains unchanged; and for the 150-200Hz frequency band that may contain noise, its coefficient is attenuated to 0.7 times the original value, thereby achieving selective enhancement of the image. By multiplying the frequency domain compensation matrix with the image spectrum, the frequency domain features of the image are adjusted, and then the processed image is converted back to the spatial domain through inverse Fourier transform to obtain the frequency domain compensation result.

[0107] In order to effectively fuse the compensation results of the spatial domain and the frequency domain, it is necessary to evaluate the local image quality of the two compensation results. The local clarity index is an important parameter to measure the clarity of the local area of ​​the image. It is evaluated by calculating the gradient features, edge features, etc. of the image. This evaluation is particularly important in the end-face image of the optical module because different areas may be affected by different degrees of shaking. For example, the fiber core area and the background area may require different compensation strategies. Based on the calculated local clarity index, a fusion weight map is generated to reflect the quality distribution of each area of ​​the image. The fusion weight map is calculated based on the local clarity index. For each pixel position (x, y), the gradient mean M(x, y) and variance V(x, y) in its 5×5 neighborhood are calculated. The weight value W(x, y) is calculated by the following formula: W(x, y)=M(x, y) / (M(x, y)+V(x, y)). When W(x, y)>0.6, the spatial domain compensation result is mainly used at this location; when W(x, y)<0.4, the frequency domain compensation result is mainly used; when 0.4≤W(x, y)≤0.6, linear fusion is performed according to the weight value. Finally, the two compensation results are combined by weighted fusion. For areas with better spatial domain compensation effects, the spatial domain compensation results are given a greater weight; for areas with better frequency domain compensation effects, more frequency domain compensation results are retained. This adaptive fusion strategy based on local quality assessment can make full use of the advantages of the two compensation methods and obtain a more ideal compensation effect.

[0108] Please continue reading Figure 8 , performing regional positioning on the compensated end face image to obtain a to-be-detected area, and performing feature analysis on the to-be-detected area to obtain defect feature data;

[0109] In one embodiment of the present invention, the performing of regional positioning on the compensated end face image to obtain the area to be detected, and performing feature analysis on the area to be detected to obtain defect feature data includes:

[0110] Constructing a multi-scale pyramid structure for the compensated end face image, extracting local contrast features and edge gradient features at each scale level, and generating a regional saliency map;

[0111] Based on the peak distribution of the regional saliency map, an adaptive watershed algorithm is used to segment the salient region, and the segmented region is screened in combination with the prior information of the fiber core position to obtain the region to be detected;

[0112] Constructing an eight-directional Gabor filter group with an angle interval of 15°, performing filtering processing on each direction of the area to be detected to obtain a texture response map, and constructing a defect feature vector by combining the regional grayscale mean, standard deviation and gradient amplitude;

[0113] The defect feature vector is subjected to dimensionality reduction processing to obtain defect feature data.

[0114] Specifically, in the defect detection of the end face image of the optical module, it is first necessary to establish a multi-scale analysis structure for the compensated image. The multi-scale pyramid structure is a hierarchical representation of an image, which constructs image levels with different resolutions by downsampling layer by layer. For the end face image of the optical module with 256×256 pixels, a 4-layer pyramid structure is constructed, with the first layer being the original resolution of 256×256, the second layer being 128×128, the third layer being 64×64, and the fourth layer being 32×32. At each level, downsampling is performed by a 5×5 Gaussian filter, and the standard deviation of the filter kernel is set to 1.0. At each level, local contrast features and edge gradient features are extracted respectively. The local contrast feature reflects the brightness difference between the image area and the surrounding area, which is particularly important for detecting defects such as stains and scratches on the end face; the edge gradient feature describes the edge strength and direction information in the image, which helps to identify the core boundary and surface defects. By combining features at different levels, a regional saliency map is generated, which reflects the importance of each area in the image, where the defective area usually shows a higher saliency value.

[0115] Based on the generated regional saliency map, an adaptive watershed algorithm is used for regional segmentation. The watershed algorithm performs regional segmentation based on the gradient change of the image grayscale value, where the area with high saliency value is regarded as the local minimum point, and the segmentation boundary is determined by simulating the water injection process. In the end face detection of the optical module, these local minimum points usually correspond to the center position of the defect area. In the end face detection of the optical module, the seed point selection and segmentation threshold of the algorithm are dynamically adjusted according to the peak distribution of the saliency map. At the same time, since the core arrangement of the end face of the optical module has a certain regularity, this regularity is introduced into the segmentation process as prior information. For example, the core usually presents a regular circular arrangement, so this feature can be used to screen the segmentation results, remove the areas that do not conform to the arrangement rules, and retain the real area to be detected.

[0116] For the determined area to be inspected, a Gabor filter group is used for texture analysis. Gabor filter is a tool that can effectively capture directional texture features. A parameterized Gabor filter group is constructed, including 8 directions (0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°), 3 scales (2, 4, 8 pixels) are set for each direction, the filter kernel size is 9×9, the Gaussian envelope standard deviation is set to 2.0, and the sine wave wavelength is set to 4.0 pixels. Such parameter settings can effectively capture texture features of different directions and scales. In the end face inspection of optical modules, different types of defects will show unique texture responses in different directions. For example, radial scratches will be more obvious in the radial filter response, while annular scratches will be more prominent in the tangential filter response. The texture response map obtained by filtering is combined with the statistical characteristics of the region (such as grayscale mean, standard deviation) and gradient amplitude information to form a high-dimensional vector describing the defect characteristics.

[0117] Finally, since the feature vector contains a lot of information (including filter responses in 8 directions, multiple statistical features, etc.), its dimension is usually more than 100 dimensions, which will bring a huge computational burden. Therefore, dimensionality reduction processing is required to reduce the dimension of the feature vector to a reasonable range (usually 20-30 dimensions). The dimensionality reduction process is carried out through the principal component analysis (PCA) method, which can retain the most important feature information while removing redundancy and noise. Through dimensionality reduction processing, the most identifiable components in the feature vector are retained, and redundant and noise information is removed. For example, for scratch detection on the end face of an optical module, it may be necessary to retain only a few feature components that are most sensitive to the scratch direction. This reduced-dimensional feature data not only retains the key feature information of the defect, but also greatly reduces the computational complexity of subsequent processing.

[0118] Please continue reading Figure 8 , classify and process the defect feature data to obtain preliminary defect detection results; perform reliability verification and defect degree evaluation on the preliminary defect detection results to generate final detection results.

[0119] Specifically, during the optical module end face inspection process. Based on the defect feature data (including texture response value, grayscale statistics and gradient information) obtained in the previous steps, a two-level classification strategy is used for defect identification: the first level uses a support vector machine (SVM) to preliminarily classify defects into three categories: scratches, stains and pits, and the classification threshold is set to 0.75; the second level uses a special fine classifier for each type of defect to determine the defect level based on the quantitative indicators in the feature vector (such as length, area, depth, etc.). The classification process adopts a multi-layer classifier structure. First, the defect features are divided into different categories, such as scratches, stains, pits, etc., through a coarse classifier. For each type of defect feature, a special fine classifier is used for further analysis. For example, for scratch defects, the fine classifier will classify them into different severity levels based on the length, width, depth and other characteristics of the scratches; for stain defects, they will be subdivided according to the area, density, distribution and other characteristics of the stains.

[0120] The defect classifier classifies defects into the following levels:

[0121] Scratches: less than 10μm is mild, 10-30μm is moderate, and greater than 30μm is severe;

[0122] Stain type: area less than 20μm² is mild, 20-50μm² is moderate, and larger than 50μm² is severe;

[0123] Pit types: Depth less than 1μm is mild, 1-3μm is moderate, and greater than 3μm is severe.

[0124] Defects of different levels have different effects on optical communication performance. The core area of ​​the fiber (9μm in diameter) is the main channel for optical signal transmission. When medium-level defects (such as scratches with a length of 20-30μm or stains with an area of ​​35-50μm²) appear in this area, the insertion loss of the optical signal will increase by 0.5-1dB; when severe defects occur, the insertion loss may exceed 2dB, seriously affecting the quality of signal transmission.

[0125] In addition to the defect category information, the classification results also include specific parameters such as the defect location and size, forming a preliminary defect detection result.

[0126] A multiple verification mechanism is used to verify the reliability of the preliminary test results. First, through morphological verification, check whether the morphological characteristics of the defect conform to the typical characteristics of this type of defect. For example, scratch defects usually show linear characteristics. If the detected defect morphology deviates significantly from this characteristic, it needs to be re-evaluated. The second is contextual verification, which analyzes the characteristic distribution of the area around the defect to determine whether the test result is affected by image noise or other interference factors. Finally, optical property verification is performed. According to the optical principle of the optical module end face, it is analyzed whether the detected defect will have a substantial impact on the optical signal transmission. In the defect degree assessment stage, the system comprehensively considers multiple parameters of the defect, including the geometric size, depth, and position of the defect. Special attention is paid to the distribution of defects in the core area of ​​the fiber core, because these areas have the most significant impact on the transmission of optical signals. The evaluation results adopt a grading system to divide the defects into multiple levels according to the degree of impact. For example, for scratch defects, the hazard level is determined based on factors such as whether the scratch crosses the core area of ​​the fiber core and whether the depth of the scratch exceeds the threshold. The final test report not only contains basic information such as the location, type, and degree of the defects, but also contains a comprehensive evaluation of the impact on the performance of the optical module, providing a basis for subsequent quality control and product screening.

[0127] The present invention forms a complete solution through the collaborative design of hardware and software: at the hardware level, the microscope adjustment structure controls the shaking to the micron level through precise mechanical constraints; at the software level, the detection method processes the residual shaking through real-time image analysis and compensation algorithms. This dual protection mechanism not only ensures the detection accuracy, but also assists the operator to make precise adjustments through real-time feedback, thereby improving the detection efficiency and reliability as a whole.

[0128] 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 changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for detecting an optical module, characterized in that: The optical module detection method is applied to an optical module detection device, and the optical module detection device includes a microscope adjustment structure, and the microscope adjustment structure includes: The inner frame (1) has a detection through hole (11) penetrating along the Z-axis direction; A middle frame (2) is arranged around the outer periphery of the inner frame (1), and the inner frame (1) can be displaced relative to the middle frame (2) along the X-axis direction; The outer frame (3) comprises an outer frame body (31) arranged around the outer periphery of the middle frame (2), a front stopper (32) and a rear stopper (33) respectively fixed at opposite ends of the outer frame body (31), and the middle frame (2) can be displaced along the Y-axis direction relative to the outer frame body (31); wherein: The middle frame (2) and the inner frame (1) are mutually limited in the Y-axis direction, the middle frame (2) and the front limit member (32) form a limit fit for limiting the displacement of the inner frame (1) along the Z-axis direction, at least one of the outer frame body (31) and the front limit member (32) and the middle frame (2) are mutually limited in the X-axis direction, and the front limit member (32) and the rear limit member (33) form a limit fit for limiting the displacement of the middle frame (2) along the Z-axis direction; The optical module detection method comprises: The acquired end face image of the optical module is used to extract geometric features, surface features and optical features through a deep separable convolutional network, and the extracted features are adaptively fused to obtain end face feature data; According to the end face feature data, the displacement change between adjacent image frames is obtained, the degree of shaking is determined, and the compensation parameters are calculated using a dual-flow network, and the position correction and clarity enhancement of the end face image are performed to obtain a compensated end face image; specifically, the method includes: obtaining the end face feature data of two adjacent image frames, performing optical flow calculation on the end face feature data, and obtaining a displacement vector field; according to the displacement vector field, the displacement component and the acceleration component in the X-axis and Y-axis directions are calculated, and the shaking level is determined by combining the displacement threshold and the acceleration threshold; based on the shaking level, a position compensation coefficient and a sharpening compensation coefficient are generated using a compensation prediction network, wherein the position compensation coefficient is used to correct the offset in the X-axis and Y-axis directions, and the sharpening compensation coefficient is used to correct different degrees of motion blur; the position compensation coefficient and the sharpening compensation coefficient are respectively input into a spatial domain compensation network and a frequency domain compensation network to obtain a spatial domain compensation result and a frequency domain compensation result, and the spatial domain compensation result and the frequency domain compensation result are adaptively fused to obtain a compensated end face image; Performing regional positioning on the compensated end face image to obtain a to-be-detected area, and performing feature analysis on the to-be-detected area to obtain defect feature data; Classifying and processing the defect feature data to obtain preliminary defect detection results; The reliability of the preliminary defect detection results and the defect degree assessment are performed to generate final detection results.

2. The optical module detection method according to claim 1, characterized in that: The inner peripheral wall of the middle frame (2) is provided with a first limiting boss (21) extending along the X-axis direction, the front limiting member (32), the inner peripheral wall of the middle frame (2) and the first limiting boss (21) are arranged to form a first guide groove (4) extending along the X-axis direction, and the outer peripheral wall of the inner frame (1) is provided with a first guide portion (12) protrudingly, the first guide portion (12) is slidably matched with the first guide groove (4), and the first guide portion (12) is respectively abutted against two opposite groove walls of the first guide groove (4) along the Y-axis direction.

3. The optical module detection method according to claim 1, characterized in that: The microscope adjustment structure comprises an X-axis adjustment member (5) and an X-axis elastic member (6); one end of the X-axis adjustment member (5) passes through the outer frame body (31) and the middle frame (2) in sequence along the X-axis direction from the outside of the outer frame body (31) and then abuts against the inner frame (1); The X-axis elastic member (6) is arranged on a side of the inner frame (1) away from the X-axis adjusting member (5), and two ends of the X-axis elastic member (6) are respectively connected to the inner frame (1) and the middle frame (2).

4. The optical module detection method according to claim 1, characterized in that: The front limit member (32) is a plate-shaped structure. One of the front limit member (32) and the middle frame (2) is provided with a second guide groove (22) extending along the Y-axis direction, and the other is provided with a second guide portion (321) adapted to the second guide groove (22). The second guide portion (321) is slidably matched with the second guide groove (22), and the second guide portion (321) is respectively abutted against two groove walls of the second guide groove (22) that are opposite to each other along the X-axis direction.

5. The optical module detection method according to claim 1, characterized in that: The microscope adjustment structure comprises a Y-axis adjustment member (7) and a Y-axis elastic member (8); one end of the Y-axis adjustment member (7) passes through the outer frame body (31) along the Y-axis direction from the outside of the outer frame body (31) and then abuts against the inner frame (1); The Y-axis elastic member (8) is arranged on a side of the middle frame (2) away from the Y-axis adjusting member (7), and two ends of the Y-axis elastic member (8) are respectively connected to the middle frame (2) and the outer frame body (31).

6. The optical module detection method according to claim 1, characterized in that: The position compensation coefficient and the sharpening compensation coefficient are respectively input into a spatial domain compensation network and a frequency domain compensation network to obtain a spatial domain compensation result and a frequency domain compensation result, and the spatial domain compensation result and the frequency domain compensation result are adaptively fused to obtain a compensated end face image, including: The position compensation coefficient is used to modulate the convolution kernel parameters in the spatial domain compensation network to generate an adaptive convolution kernel, and the end face image is subjected to spatial domain transformation by the adaptive convolution kernel to obtain a spatial domain compensation result; Using the sharpening compensation coefficient to modulate the Fourier transform coefficient in the frequency domain compensation network to generate a frequency domain compensation matrix, and multiplying the frequency domain compensation matrix with the frequency spectrum of the end face image to obtain a frequency domain compensation result; The local clarity index of the spatial domain compensation result and the frequency domain compensation result is calculated, a fusion weight map is generated according to the local clarity index, and the spatial domain compensation result and the frequency domain compensation result are weightedly fused using the fusion weight map to obtain a compensated end face image.

7. The optical module detection method according to claim 1, characterized in that: The performing regional positioning on the compensated end face image to obtain the area to be detected, and performing feature analysis on the area to be detected to obtain defect feature data, comprises: Constructing a multi-scale pyramid structure for the compensated end face image, extracting local contrast features and edge gradient features at each scale level, and generating a regional saliency map; Based on the peak distribution of the regional saliency map, an adaptive watershed algorithm is used to segment the salient region, and the segmented region is screened in combination with the prior information of the fiber core position to obtain the region to be detected; Constructing an eight-directional Gabor filter group with an angle interval of 15°, performing filtering processing on each direction of the area to be detected to obtain a texture response map, and constructing a defect feature vector by combining the regional grayscale mean, standard deviation and gradient amplitude; The defect feature vector is subjected to dimensionality reduction processing to obtain defect feature data.

Images