A method for detecting volume defects in a crystal

Abstract

A method for detecting internal body defects of a crystal comprises the following steps: S1, taking a crystal with internal body defects after detection; S2, irradiating the crystal from the side with a line light source to obtain a crystal tomography image of scattered light containing the internal body defects of the crystal; performing size correction on the crystal tomography image; S3, performing time series analysis and processing on the corrected crystal tomography image to highlight the characteristics of the defect point; S4, combining the internal defect position of the known crystal and the detection time series signal of the position, training an RNN model to obtain a crystal internal defect detection model; S5, repeating the training to obtain a final crystal internal defect detection model; S6, inputting the time series signal of the crystal to be tested obtained by the S2-S3 processing into the final model, i.e. obtaining the internal defect position information; S7, filtering out false defects according to the representation of the defect on the plane. The present invention solves the problems of complex crystal internal defect recognition process and low recognition efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of internal defects detection of uniform light-transmitting materials, and in particular relates to a method for detecting internal defects of a crystal. Background Art

[0002] During the crystal growth process, changes in temperature, pressure, and medium component concentration may cause defects inside the crystal, which will affect the mechanical properties of the crystal. When crystals with body defects are used to produce optical devices, the stray light generated will reduce the imaging quality. When they are used to produce parts and equipment, the product performance will be reduced and even the product function will be affected. Therefore, inspecting, identifying, and locating body defects inside the crystal is an important part of crystal quality inspection and a necessary measure to improve the performance of crystal products.

[0003] Existing methods for detecting crystal defects include digital holographic imaging, traditional laser scattering tomography, and parallel confocal microscopy, each with its own advantages and disadvantages, among which:

[0004] Digital holographic imaging uses the interference and diffraction effects of lasers to achieve lossless recording and reproduction of the three-dimensional shape of an object. In this technology, the laser is divided into two: one beam illuminates the crystal sample (object light) and the other serves as a reference light. The reflected and transmitted light waves generated by the interaction between the object light and the sample intersect with the reference light on the detector to generate an interference pattern containing the phase, amplitude and internal defect information of the light wave. This pattern is captured by a highly sensitive camera and converted into a digital hologram, which is then processed by a computer algorithm to accurately reconstruct the three-dimensional structural image of the sample. This technology can not only intuitively display the internal morphology of the sample, but also effectively evaluate its volume defects. Although this method can provide high-resolution three-dimensional images and identify tiny defects, the reconstruction and analysis of digital holographic images require complex calculations, resulting in slow detection speeds in industrial production and cannot meet the online detection needs of crystal production.

[0005] Traditional laser scattering tomography uses the principle of laser scattering to detect internal defects in crystals. This technology uses a laser beam to irradiate a crystal sample, and the defective area scatters light to form a specific light wave field. The detector collects scattered light data, and uses image processing algorithms to parse the three-dimensional structural information of the sample, and then constructs a three-dimensional image. This image accurately displays the internal defects of the crystal, providing key support for material performance evaluation and application. Laser scattering tomography is a non-contact detection technology that ensures that the sample is protected from physical damage during the detection process, and is particularly suitable for non-destructive testing of precious or fragile materials. However, this traditional technology mainly relies on image processing, and the laser irradiation process is prone to introduce significant noise, which limits its accuracy in identifying tiny body defects.

[0006] The parallel confocal microscopy method uses a spectroscopic element to subdivide the light source into an array of point light sources. Each beam of light is focused on the object to be inspected to form a light spot. These light spot images are directly mapped to the detector to achieve precise point-to-point matching and efficiently capture the two-dimensional information of the object. By accumulating data from multiple two-dimensional layers, a three-dimensional image of the object can be constructed, which is suitable for the fine detection of internal defects in optical crystals. However, this technology faces a significant contradiction between numerical aperture (NA) and field of view (FOV): although high NA improves resolution, it severely limits FOV to millimeter level, hindering the rapid and comprehensive detection of internal defects in crystals, and it is difficult to meet the efficient detection needs of crystal production lines.

[0007] Currently, there is an urgent need for an efficient and accurate detection technology to quickly locate and measure the size and position of defects inside crystals, providing effective guarantees for the high-quality preparation of optical components. Summary of the invention

[0008] The purpose of the present invention is to provide a method for detecting body defects inside a crystal, so as to solve the problem mentioned in the background art that several currently used crystal defect detection methods cannot meet the efficient and accurate detection requirements of a crystal production line.

[0009] To achieve the above object, the present invention provides a method for detecting volume defects inside a crystal, comprising the following steps:

[0010] S1. Take a crystal with internal body defects that has been tested;

[0011] S2, irradiating the crystal from the side with a line light source to obtain crystal tomography image information including scattered light of volume defects inside the crystal;

[0012] Correct the crystal tomography images and adjust them to a uniform standard size to ensure the consistency of analysis;

[0013] S3, performing time series analysis on the grayscale value time series of each pixel point of the corrected crystal tomography image, using signal processing technology to highlight the defect information in the time series data and the characteristics of the defect points;

[0014] S4, combining the known internal defect position of the crystal and the detection timing signal of the position, training the RNN model to obtain a crystal internal defect detection model;

[0015] S5. Set a threshold and repeat the training until the threshold is reached; that is, the final crystal internal defect detection model is obtained;

[0016] S6, processing the crystal to be tested by using steps S2 to S3, inputting the obtained timing signal into the trained final model, and obtaining the internal defect position;

[0017] S7. Based on the representation of defects on the plane, the convolution kernel is used to filter out false defects caused by discrete noise.

[0018] In a specific embodiment, the method for detecting body defects inside a crystal uses a crystal image acquisition device, which includes a CCD camera, a camera bracket, a motor module, a laser line structure light source, a rotating platform and a controller;

[0019] The CCD camera is arranged on the camera bracket, and the rotating platform is located directly below the CCD camera. The rotating platform is used to carry the crystal to be detected, and the crystal to be detected placed thereon can be driven to rotate by controlling the rotation of the rotating platform; the laser line structure light source is arranged on the side of the crystal to be detected; the CCD camera is used to obtain crystal tomography image information of the crystal to be detected;

[0020] The camera bracket and the laser line structure light source are both arranged on the motor module. By controlling the motor module, the camera bracket and the laser line structure light source can be moved upward or downward synchronously; the CCD camera, the motor module, the laser line structure light source and the rotating platform are all connected to the controller, and the controller is used to control the shooting of the CCD camera, the lifting and lowering of the motor module, the switching of the laser line structure light source and the rotation of the rotating platform.

[0021] In a specific implementation, the laser line structure light source is a laser line structure light source with a wavelength of 532 nm or 650 nm.

[0022] In a specific implementation, the laser line structure light source generates sheet laser light for exciting the region to be detected of the crystal, and makes the action surface of the sheet laser light transmitted into the crystal a horizontal surface.

[0023] In a specific embodiment, in step S2, the process of obtaining the crystal tomography image information containing scattered light of the internal volume defect of the crystal is:

[0024] The controller uses electrical signals to drive the motor module, so that the CCD camera and the laser line structured light source move synchronously and uniformly along the optical axis from bottom to top or from top to bottom. During the movement, the CCD camera continues to shoot video and record the structured light image formed by the laser on the crystal detection surface. After completing a detection, the controller sends an electrical signal to control the rotating platform to rotate the crystal by a certain angle, repeating the detection process to obtain detection videos at different angles.

[0025] In a specific embodiment, in step S2, correcting the crystal tomography image includes shearing the detection videos at different angles;

[0026] Cut off the partial images occupying the laser light outlet in different videos; seamlessly connect and splice the cut multiple videos; and splice to obtain a complete crystal detection image.

[0027] In a specific embodiment, in step S2, after the shearing process, the correction of the crystal tomography image also includes: calculating the actual size of each detection surface according to the camera focal length, crystal diameter, height and refractive index, and uniformly adjusting the detection surface size to the first frame detection surface size through image scaling and cropping.

[0028] In a specific implementation, the RNN model includes a hidden layer, an output layer and an input layer, and the number of neurons in the output layer is the same as the number of neurons in the input layer.

[0029] In a specific implementation, all preprocessed grayscale value time series are grouped into a training set and a validation set; the training of the RNN model specifically includes: using the training set to train the RNN model to obtain the training set accuracy, using the validation set to verify the trained crystal internal defect detection model to obtain the validation set accuracy, and verifying whether the crystal internal defect detection model is qualified according to the training set accuracy and the validation set accuracy: if the training set accuracy or the validation set accuracy does not reach the set threshold, then adjusting the number of hidden layers, the number of neurons in each hidden layer, regularization and the amount of training data, and retraining the crystal internal defect detection model.

[0030] In a specific implementation, in step S3, the signal processing method includes one or more of coefficient fitting, absolute difference comparison, Fourier transform, wavelet transform, first-order differential, second-order differential, and principal component analysis.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. Automatic detection: Only by placing the crystal on the detection table, the present invention can automatically drive the CCD camera to capture images, and import the captured images into a professional detection program, so as to easily realize efficient detection of internal defects of the crystal.

[0033] 2. High consistency: By adopting advanced detection algorithms, the present invention ensures the strict unification of detection standards and processes, greatly reducing errors and deviations caused by human factors, thereby ensuring the accuracy and reliability of the detection results.

[0034] 3. High precision: This method cleverly combines two-dimensional image representation with one-dimensional time series signal analysis and processing, thereby achieving more accurate identification of defects in transparent crystals.

[0035] 4. Wide application scope: The invention has a wide application scope and is not limited to crystals. It is also applicable to the internal defect detection of glass and other types of light-transmitting materials, demonstrating its strong versatility and practicality.

[0036] The present invention decomposes the image sequence collected by the CCD camera to obtain the gray value time series of each pixel collection point on the crystal surface, that is, the brightness change curve. The internal defect position of the crystal corresponds to the peak position of the time series of the pixel point. Therefore, the gray time series of the pixel collection point with the known internal defect position after image correction and signal processing is used as sample data, and the crystal internal defect detection model obtained by training can automatically identify the internal defect position of each pixel collection point of the crystal, solving the problem of complex crystal internal defect recognition process and low recognition efficiency.

[0037] The present invention uses optical principles and combines the refractive index characteristics of the measured crystal and the refraction law of light in the crystal to establish a mathematical model, corrects the crystal in the picture, eliminates the image deformation caused by refraction and perspective distortion, ensures the accuracy of the crystal tomography information in the picture, and enables it to truly reflect the actual appearance proportions of the crystal.

[0038] In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention is further described in detail below. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0040] Figure 1 It is a schematic diagram of the axial scanning method inside the crystal according to the present invention;

[0041] Among them, 1. CCD camera; 2. Camera bracket; 3. Motor module; 4. Crystal to be tested; 5. Laser line structure light source; 6. Rotating platform; 7. Controller. DETAILED DESCRIPTION

[0042] The embodiments of the present invention are described in detail below. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0043] A method for detecting body defects inside a crystal of the present invention, wherein the method for detecting body defects inside a crystal adopts a crystal image acquisition device, wherein the crystal image acquisition device comprises a CCD camera 1, a camera bracket 2, a motor module 3, a laser line structure light source 5, a rotating platform 6, and a controller 7;

[0044] The CCD camera 1 is arranged on the camera bracket 2, and the rotating platform 6 is located directly below the CCD camera 1. The rotating platform 6 is used to carry the crystal 4 to be detected, and the rotation of the rotating platform 6 can drive the crystal 4 to be detected placed thereon to rotate; the laser line structure light source 5 is arranged on the side of the crystal 4 to be detected; the CCD camera 1 is used to obtain the crystal tomography image information of the crystal 4 to be detected;

[0045] The camera bracket 2 and the laser line structure light source 5 are both arranged on the motor module 3. By controlling the motor module 3, the camera bracket 2 and the laser line structure light source 5 can be moved upward or downward synchronously; the CCD camera 1, the motor module 3, the laser line structure light source 5 and the rotating platform 6 are all connected to the controller 7, and the controller 7 is used to control the shooting of the CCD camera 1, the lifting and lowering of the motor module 3, the switching of the laser line structure light source 5 and the rotation of the rotating platform 6.

[0046] The detection method specifically includes the following steps:

[0047] S1. Take a crystal known to have internal body defects that has been tested.

[0048] S2, using a laser line structure light source with a wavelength of 532nm or 650nm to illuminate the crystal from the side to obtain crystal tomography image information of scattered light including internal volume defects of the crystal;

[0049] The crystal tomography images were corrected and adjusted to a uniform standard size to ensure the consistency of the analysis.

[0050] The laser line structure light source generates sheet laser light for exciting the to-be-detected area of ​​the crystal, and makes the action surface of the sheet laser light transmitted into the crystal a horizontal surface.

[0051] In step S2, the process of obtaining the crystal tomography image information containing the scattered light of the internal volume defect of the crystal is as follows:

[0052] The controller 7 drives the motor module 3 with an electrical signal, so that the CCD camera 1 and the laser line structure light source 5 move synchronously and uniformly from bottom to top or from top to bottom along the optical axis; during the movement, the CCD camera 1 continuously shoots video to record the structured light image formed by the laser on the crystal detection surface; after completing one detection, the controller 7 sends an electrical signal to control the rotating platform 6 to rotate the crystal by a certain angle, repeats the detection process, and obtains detection videos at different angles.

[0053] Cut the detection videos at different angles;

[0054] Cut off the partial images occupying the laser light outlet in different videos; seamlessly connect and splice the cut multiple videos; and splice to obtain a complete crystal detection image.

[0055] After the shearing process, the correction of the crystal tomography image further includes: uniformly adjusting the size of the detection surface to the size of the detection surface of the first frame, which is achieved by image scaling and cropping.

[0056] The refraction of light inside the crystal will cause image distortion, and fixed-focus cameras will produce perspective distortion when shooting. To solve the above problems, optical principles are used to establish a mathematical model based on the refractive index of the measured crystal, and the crystal in the picture is corrected so that the crystal tomography information in the corrected time-series picture conforms to the appearance ratio of the actual crystal.

[0057] S3. Perform time series analysis on the grayscale value time series of each pixel point of the corrected crystal tomography image, and use signal processing technology to highlight the defect information in the time series data and the characteristics of the defect points.

[0058] S4. Based on the known internal defect position of the crystal and the detection timing signal of the position, the RNN model is trained to obtain a crystal internal defect detection model.

[0059] In the corrected image sequence, each pixel accurately maps the same position of the crystal cross section, and different frames sequentially display the detailed internal structure information of the crystal along the depth direction. By performing signal processing on the time series signal of each pixel, the signal processing method includes one or more of coefficient fitting, absolute difference comparison, Fourier transform, wavelet transform, first-order differential, second-order differential, and principal component analysis, based on the unique characteristics exhibited when the light passes through the body defect area and the non-body defect area, the defect signal is enhanced. Subsequently, the processed time series signal is extended to the length required by the neurons in the input layer of the adaptation model. The present invention extracts the time series signal containing defect information as the input of the model, and is equipped with an equal-length label signal (where 0 represents a non-defective area and 1 represents a defective area) as the output of the model, thereby training a detection model that can identify and accurately locate internal defects in the crystal.

[0060] The RNN model includes a hidden layer, an output layer and an input layer, and the number of neurons in the output layer is the same as the number of neurons in the input layer. The number of neurons is set to the maximum length of the crystal detection timing signal with a length of 280 mm. When the detection accuracy is 1 mm7 frames, the number of neurons is set to 280×7=1960. The deviation and weight of the crystal internal defect detection model are determined by training the RNN model.

[0061] S5. Set a threshold and repeat the training until the threshold is reached; thus, the final crystal internal defect detection model is obtained.

[0062] All pre-processed grayscale value time series are grouped into a training set and a validation set; the training RNN model specifically includes: using the training set to train the RNN model to obtain the training set accuracy, using the validation set to verify the trained crystal internal defect detection model to obtain the validation set accuracy, and verifying whether the crystal internal defect detection model is qualified according to the training set accuracy and the validation set accuracy: if the training set accuracy or the validation set accuracy does not reach the set threshold, then adjust the number of hidden layers, the number of neurons in each hidden layer, the regularization and the number of training data, and retrain the crystal internal defect detection model. If both the training set accuracy and the validation set accuracy reach the set threshold, the final crystal internal defect detection model is obtained.

[0063] S6. The unknown crystal to be tested is processed using steps S2 to S3, and the obtained timing signal is input into the trained final model to obtain the internal defect position.

[0064] S7. According to the representation of defects on the plane, the convolution kernel is used to filter out discretely distributed false defects. Defects have a certain size, so the clustered defect points are judged as true defect points, and the discretely distributed defect points are false defect points.

[0065] Finally, the three-dimensional coordinates of the volume defect point are accurately located, and the obtained three-dimensional coordinate information is used to reconstruct the volume defect in the crystal in three dimensions to generate a visual model.

[0066] Example 1

[0067] The present invention provides a method for detecting body defects inside a crystal, wherein the method for detecting body defects inside a crystal adopts a crystal image acquisition device, wherein the crystal image acquisition device comprises a CCD camera 1, a camera bracket 2, a motor module 3, a 532nm or 650nm laser line structure light source 5, a rotating platform 6, and a controller 7;

[0068] The CCD camera 1 is arranged on the camera bracket 2, and the rotating platform 6 is located directly below the CCD camera 1. The rotating platform 6 is used to carry the crystal 4 to be detected, and the crystal 4 to be detected placed thereon can be driven to rotate by controlling the rotation of the rotating platform 6; a 532nm or 650nm laser line structure light source 5 is arranged on the side of the crystal 4 to be detected;

[0069] The camera bracket 2 and the 532nm or 650nm laser line structure light source 5 are both arranged on the motor module 3. By controlling the motor module 3, the camera bracket 2 and the 532nm or 650nm laser line structure light source 5 can be moved upward or downward synchronously; the CCD camera 1, the motor module 3, the 532nm or 650nm laser line structure light source 5 and the rotating platform 6 are all connected to the controller 7, and the controller 7 is used to control the shooting of the CCD camera 1, the lifting and lowering of the motor module 3, the switching of the 532nm or 650nm laser line structure light source 5 and the rotation of the rotating platform 6.

[0070] The method for detecting body defects inside a crystal comprises the following steps when used specifically:

[0071] S1. First obtain a crystal with internal body defects that has been tested.

[0072] S2. In a dark field environment, a laser line structure light source with a wavelength of 532nm or 650nm is used to irradiate the side of the crystal to be tested obtained in step S1. The CCD camera 1 collects the test image from just above the crystal 4 to be tested, and the field of view of the CCD camera is accurately aligned with the upper surface of the crystal to be tested, and the optical axis direction of the CCD camera 1 is perpendicular to the crystal 4 to be tested and the rotating platform 6, and the centers of the three circles are coaxial. The line structure laser light source 5 generates a sheet laser, which is used to excite the crystal to be tested area, and makes the action surface of the sheet laser transmitted into the laser crystal to be tested a horizontal plane.

[0073] The controller 7 uses electrical signals to drive the stepper motor in the motor module 3 to drive the camera bracket 2, the CCD camera 1 and the laser line structure light source 5 with a wavelength of 532nm or 650nm to move synchronously and uniformly along the optical axis in the vertical direction to scan the depths of the crystal under test.

[0074] During the movement, the CCD camera 1 continuously shoots video and records the structured light image formed by the laser on the crystal detection surface. After completing one detection, the controller 7 sends an electrical signal to control the rotating platform 6 to rotate the crystal to be detected by a certain angle, repeating the detection process to obtain detection videos at different angles.

[0075] First, the detection videos at different angles are cut. Since the laser light outlet will occupy a part of the video screen, and this part usually contains a lot of noise, it needs to be cut off. In specific operations, the part of the screen occupying the light outlet in the two videos can be subtracted respectively. The multiple videos after cutting are spliced. Since different videos shoot different sides of the crystal respectively, a complete crystal detection image can be obtained after splicing. When splicing, it is necessary to ensure that the videos are seamlessly connected at the splicing point to eliminate the influence of the laser light outlet on the detection results.

[0076] According to the focal length f of the camera and the refractive index n of the air1 , crystal length D, crystal radius R, crystal refractive index n 2 And the number of pixels occupied by the crystal cross-section radius in the first and last frames of the video 0 、r n Calculate the flange distance d of the camera.

[0077] d=f(n 1 ,n 2 ,f,D,R,r 0 ,r n )

[0078] After the flange distance d is calculated, according to the air refractive index n 1 , the depth d of the detection surface relative to the crystal surface 1 , the distance d between the CCD camera and the crystal surface 2 , crystal refractive index n 2 , crystal radius R, flange distance d, calculate the number of pixels r occupied by the crystal cross-section radius of the detection surface in each frame i .

[0079] r i =f(n 1 ,n 2 ,d 1 ,d 2 ,R,d)

[0080] The above operation obtains the pixel size of each detection surface. The camera imaging has the characteristics of large near and small far, so the detection section sequence is an increasing sequence. In order to realize the subsequent timing analysis of the signal, the detection surface size is uniformly adjusted to the detection surface size r0 of the first frame. This step is achieved by image scaling and cropping. The crystal tomography image with perspective distortion and image distortion problems eliminated is obtained.

[0081] Perform time series analysis on each pixel of the corrected crystal tomography image, and enhance the identifiability of defects by highlighting the brightness changes of pixels between different frames, that is, the gray value differences, which is used as a key means to enhance defect recognition features. Signal processing methods are used to highlight the defect information hidden in the time series data. Defect information is expressed as a spike signal, and non-defect information is expressed as a flat signal. Through signal processing methods, defect features expressed as spike signals can be identified and emphasized, while non-defect features expressed as flat signals can be weakened or ignored, thereby improving the accuracy and reliability of defect detection. Signal processing methods include coefficient fitting, absolute difference comparison, Fourier transform, wavelet transform, first-order differential, second-order differential, and principal component analysis.

[0082] In order to ensure that the constructed model has wide compatibility, that is, it can detect defect information in time series signals of any length, an effective padding strategy is adopted. Specifically, the last value of the time series sequence is used as the padding value, and the value is repeated until the length of the time series signal reaches the value required by the neurons in the model input layer. This padding method ensures that all the padding parts in the padded time series signal have the same value, so it does not contain spike signals, thereby avoiding potential interference with defect detection. At the same time, this strategy meets the specific requirements of the recurrent neural network (RNN) model for the input data size.

[0083] The filled signal is used as the input of the RNN model, and the label signal of the same length (where 0 is a non-defective area and 1 is a defective area) is used as the output of the RNN model. The model is trained by using the collected time series signal of known defect depth information and the label signal of the same length as the training set. The crystal internal defect detection model is obtained, and the training accuracy of the crystal internal defect detection model can be automatically obtained. The time series signal of the known defect depth information is obtained by collecting crystals with internal body defects that have been manually detected. The RNN model specifically adopts the LSTM model.

[0084] Repeat the above-mentioned collection and processing of the crystals with internal body defects that have been detected, and then input them into the crystal internal defect detection model for training until the training accuracy of the crystal internal defect detection model reaches the set threshold, thus obtaining the final crystal internal defect detection model.

[0085] The unknown crystal to be tested is placed in a crystal image acquisition device for acquisition to obtain a crystal tomography image of the unknown crystal to be tested.

[0086] After the crystal tomography image of the unknown crystal to be tested has gone through the aforementioned series of preprocessing steps for the crystal tomography image, and after filling the timing signal, the collected and preprocessed crystal data of the unknown crystal to be tested is input into the trained RNN model to obtain an output sequence that matches the length of the input data. Subsequently, the data portion increased by the filling is removed, and only the output of the original length is retained, and it is reconstructed into a three-dimensional matrix according to the timing order of the input model. The matrix only contains two values ​​0 and 1, where 0 represents a non-defective area and 1 represents a defective area. This three-dimensional matrix intuitively shows the defect distribution inside the crystal to be tested.

[0087] Based on the three-dimensional distribution characteristics of defects, the algorithm is used to determine the closely clustered and interconnected defect points as actual defects, and the discretely distributed isolated points are regarded as noise points and removed. According to the actual application requirements, the image is convolved using convolution kernels of different sizes to extract and highlight volume defects of different sizes. Through three-dimensional reconstruction technology, the specific distribution of these volume defects inside the crystal is intuitively displayed.

[0088] In order to analyze the defect characteristics more deeply, a two-dimensional convolution operation is performed on the matrix. The matrix is ​​convolved with a convolution kernel, which will filter out defects with a volume smaller than the convolution kernel size and extract defects greater than or equal to the convolution kernel. Convolution kernels of different sizes are used as screening tools to extract and highlight volume defects of different sizes. Finally, with the help of three-dimensional reconstruction technology, the specific distribution of the extracted volume defects inside the crystal is presented in an intuitive way.

[0089] This solution decomposes the image sequence collected by the CCD camera to obtain the grayscale value time series of each pixel acquisition point on the crystal surface, that is, the brightness change curve. The internal defect position of the crystal corresponds to the peak position of the pixel time series. Therefore, the grayscale time series of pixel acquisition points with known internal defect positions is used as sample data, and the crystal internal defect detection model trained can automatically identify the internal defect position of each pixel acquisition point of the crystal, solving the current problems of complex crystal internal defect recognition process and low recognition efficiency.

[0090] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions and substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for detecting volume defects inside a crystal, characterized in that: The following steps are involved: S1. Take a crystal with internal body defects that has been tested; S2. Using a line light source to illuminate the crystal from the side to obtain crystal tomography image information containing scattered light from internal defects of the crystal; correcting the crystal tomography image to adjust the crystal tomography image to a uniform standard size to ensure consistency of analysis; In the step S2, the correction of the crystal tomography image includes cutting the detection videos at different angles; Cut off the part of the screen that occupies the laser light outlet in different videos; perform seamless splicing on the cut multiple videos; Stitching to get a complete crystal detection image; After the shearing process, the correction of the crystal tomography image also includes: calculating the actual size of each detection surface according to the camera focal length, crystal diameter, height and refractive index, and uniformly adjusting the detection surface size to the first frame detection surface size through image scaling and cropping; S3, performing time series analysis on the grayscale value time series of each pixel point of the corrected crystal tomography image, using signal processing technology to highlight the defect information in the time series data and the characteristics of the defect points; S4, combining the known internal defect position of the crystal and the detection timing signal of the position, training the RNN model to obtain a crystal internal defect detection model; S5. Set a threshold and repeat the training until the threshold is reached; that is, the final crystal internal defect detection model is obtained; S6, processing the crystal to be tested by using steps S2 to S3, inputting the obtained timing signal into the trained final model, and obtaining the internal defect position; S7. Based on the representation of defects on the plane, the convolution kernel is used to filter out false defects caused by discrete noise.

2. The method for detecting volume defects inside a crystal according to claim 1, characterized in that: The method for detecting body defects inside a crystal adopts a crystal image acquisition device, which comprises a CCD camera (1), a camera bracket (2), a motor module (3), a laser line structure light source (5), a rotating platform (6) and a controller (7); The CCD camera (1) is arranged on a camera support (2); the rotating platform (6) is located directly below the CCD camera (1); the rotating platform (6) is used to carry the crystal to be detected (4); and the crystal to be detected (4) placed thereon can be driven to rotate by controlling the rotating platform (6) to rotate; the laser line structure light source (5) is arranged on the side of the crystal to be detected (4); and the CCD camera (1) is used to obtain crystal tomography image information of the crystal to be detected (4); The camera support (2) and the laser line structure light source (5) are both arranged on the motor module (3); by controlling the motor module (3), the camera support (2) and the laser line structure light source (5) can be synchronously moved upward or downward; the CCD camera (1), the motor module (3), the laser line structure light source (5) and the rotating platform (6) are all connected to the controller (7); the controller (7) is used to control the shooting of the CCD camera (1), the lifting and lowering of the motor module (3), the switching of the laser line structure light source (5) and the rotation of the rotating platform (6).

3. The method for detecting volume defects inside a crystal according to claim 2, characterized in that: The laser line structure light source (5) is a laser line structure light source with a wavelength of 532nm or 650nm.

4. The method for detecting volume defects inside a crystal according to claim 2, characterized in that: The laser line structure light source (5) generates sheet laser light for exciting the to-be-detected area of ​​the crystal, and makes the action surface of the sheet laser light transmitted into the crystal a horizontal surface.

5. The method for detecting volume defects inside a crystal according to claim 2, characterized in that: In step S2, the process of obtaining the crystal tomography image information containing the scattered light of the internal volume defect of the crystal is as follows: The controller (7) drives the motor module (3) using an electrical signal, so that the CCD camera (1) and the laser line structure light source (5) move synchronously and uniformly along the optical axis from bottom to top or from top to bottom; during the movement, the CCD camera (1) continuously shoots video and records the structured light image formed by the laser on the crystal detection surface; after completing one detection, the controller (7) sends an electrical signal to control the rotating platform (6) to rotate the crystal by a certain angle, and repeats the detection process to obtain detection videos at different angles.

6. The method for detecting volume defects inside a crystal according to claim 1, characterized in that: The RNN model includes a hidden layer, an output layer and an input layer, and the number of neurons in the output layer is the same as the number of neurons in the input layer.

7. The method for detecting volume defects in a crystal according to claim 6, characterized in that: Group all preprocessed grayscale value time series into training set and validation set; The training RNN model specifically includes: using the training set to train the RNN model to obtain the training set accuracy, using the verification set to verify the trained crystal internal defect detection model to obtain the verification set accuracy, and verifying whether the crystal internal defect detection model is qualified according to the training set accuracy and the verification set accuracy: if the training set accuracy or the verification set accuracy does not reach the set threshold, adjusting the number of hidden layers, the number of neurons in each hidden layer, regularization and the amount of training data, and retraining the crystal internal defect detection model.

8. The method for detecting volume defects inside a crystal according to claim 1, characterized in that: In step S3, the signal processing method includes one or more of coefficient fitting, absolute difference comparison, Fourier transform, wavelet transform, first-order differential, second-order differential, and principal component analysis.

Images