Surface Structure Quality Detection Method and System for DFB Laser Chip

Abstract

The present invention provides a method and system for detecting the surface structure quality of a DFB laser chip, relating to the technical field of image processing, including: comprehensively scanning the DFB laser chip according to an industrial camera to obtain a panoramic image of the chip; performing enhancement processing according to a chip image enhancement function to obtain a DFB laser chip image; performing watershed feature segmentation to obtain multiple chip region images; performing expected deviation detection learning according to a chip surface structure quality detection factor to build a chip surface quality deviation detection channel; respectively performing surface structure quality analysis on multiple chip region images to obtain multiple chip region surface quality coefficients; and drawing a surface detection cloud map of the DFB laser chip. The present invention solves the technical problem that the existing chip quality detection methods usually rely on simple image processing algorithms, can only analyze from a macroscopic perspective, have limited ability to identify chip surface defects, and result in low detection accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method and system for detecting the surface structure quality of a DFB laser chip. Background Art

[0002] A DFB (Distributed Feedback) laser is a semiconductor laser widely used in the fields of optical communication, sensors, spectrometers, etc. Its performance highly depends on the surface structure quality of the chip. How to efficiently and accurately detect the surface structure quality of the chip and ensure that the chip production meets the high-standard requirements is a crucial link in the production process. However, the existing automated detection methods usually rely on simple image processing algorithms, which can usually only detect macroscopic geometric shape deviations and cannot accurately process the microscopic features on the chip surface, resulting in limited ability to identify chip surface defects, and thus low detection accuracy, making it difficult to meet the production needs of optical communication devices with high-precision requirements. Summary of the Invention

[0003] The present application provides a method and system for detecting the surface structure quality of a DFB laser chip, aiming to solve the technical problem that the existing chip quality detection methods usually rely on simple image processing algorithms, can only analyze from a macroscopic perspective, have limited ability to identify chip surface defects, and result in low detection accuracy.

[0004] In the first aspect disclosed by the present application, a method for detecting the surface structure quality of a DFB laser chip is provided. The method includes: comprehensively scanning the DFB laser chip according to an industrial camera to obtain a panoramic image of the chip; performing enhancement processing on the panoramic image of the chip according to a chip image enhancement function to obtain an image of the DFB laser chip; performing watershed feature segmentation on the image of the DFB laser chip to obtain multiple chip region images; performing expected deviation detection learning according to a chip surface structure quality detection factor to build a chip surface quality deviation detection channel, where the chip surface structure quality detection factor includes surface texture features, geometric features, and microstructural features; based on the chip surface quality deviation detection channel, respectively performing surface structure quality analysis on the multiple chip region images according to a chip surface quality sorting function to obtain multiple chip region surface quality coefficients; and drawing a surface detection cloud map of the DFB laser chip according to the multiple chip region surface quality coefficients.

[0005] The second aspect disclosed in this application provides a surface structure quality detection system for a DFB laser chip. The system is used for the above-mentioned surface structure quality detection method for a DFB laser chip. The system includes: a chip full-scan module, which is used to perform a full scan of the DFB laser chip according to an industrial camera to obtain a chip panoramic image; an image enhancement processing module, which is used to perform enhancement processing on the chip panoramic image according to a chip image enhancement function to obtain a DFB laser chip image; a feature segmentation module, which is used to perform watershed feature segmentation on the DFB laser chip image to obtain multiple chip area images; a detection learning module, which is used to perform expected deviation detection learning according to chip surface structure quality detection factors and build a chip surface quality deviation detection channel, where the chip surface structure quality detection factors include surface texture features, geometric features, and microstructural features; a quality analysis module, which is used to perform surface structure quality analysis on the multiple chip area images respectively based on the chip surface quality deviation detection channel according to a chip surface quality sorting function to obtain multiple chip area surface quality coefficients; a detection cloud map drawing module, which is used to draw a surface detection cloud map of the DFB laser chip according to the multiple chip area surface quality coefficients.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] By using an industrial camera to comprehensively scan the DFB laser chip, a panoramic image of the chip is obtained, providing a complete view of all areas on the chip surface and offering comprehensive and accurate basic data for subsequent analysis. The panoramic image is processed using an image enhancement function to obtain the DFB laser chip image, enhancing the contrast and detail presentation of the image, improving the detectability of complex surface structures, and reducing analysis errors caused by blurred images or insufficient contrast. The enhanced chip image is segmented using the watershed algorithm to obtain multiple independent chip region images. The watershed algorithm is based on image gradients and local minima, ensuring precise segmentation of the chip surface area and avoiding problems such as over-segmentation or unclear region boundaries, providing an efficient basis for feature analysis of each subsequent independent region and further improving the accuracy and locality of detection. According to the chip surface structure quality detection factors, including texture features, geometric features, and micro-structure features, expected deviation detection learning is carried out to establish a chip surface quality deviation detection channel, which can automatically identify deviations in surface texture, geometry, and micro-structure, improving the automation of detection and reducing human errors. Based on the chip surface quality deviation detection channel, combined with the surface quality sorting function, the surface structures of multiple chip regions are analyzed for quality to generate the surface quality coefficient for each region, which can generate specific and quantifiable quality indicators for each chip region. This quantified result helps to identify problem areas and high-quality areas, facilitating subsequent quality control and production optimization. According to the surface quality coefficients of multiple chip regions, a surface detection cloud map of the DFB laser chip is drawn. The cloud map can comprehensively display the quality distribution of the entire chip surface, not only identifying problems in individual regions but also showing the quality trend of the entire chip. This visualization tool provides an efficient quality monitoring means for production management, helping to improve the quality control level of the production line.

[0008] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented in accordance with the content of the specification. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specifically illustrates the specific embodiments of this application. Brief Description of the Drawings

[0009] Figure 1 It is a schematic flowchart of the method for detecting the surface structure quality of a DFB laser chip provided by an embodiment of this application;

[0010] Figure 2 It is a schematic structural diagram of the system for detecting the surface structure quality of a DFB laser chip provided by an embodiment of this application.

[0011] Description of reference numerals: Chip full-scan module 10, image enhancement processing module 20, feature segmentation module 30, detection and learning module 40, quality analysis module 50, detection cloud map drawing module 60. Detailed implementation manners

[0012] In an embodiment of the present application, by providing a method and a system for detecting the surface structure quality of a DFB laser chip, the technical problem in the prior art that the chip quality detection method usually relies on simple image processing algorithms, can only analyze from a macroscopic perspective, has limited ability to identify surface defects of the chip, and results in low detection accuracy is solved.

[0013] After introducing the basic principle of the present application, various non-limiting implementation manners of the present application will be specifically introduced below in conjunction with the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0014] Embodiment 1, as Figure 1 shown, an embodiment of the present application provides a method for detecting the surface structure quality of a DFB laser chip, and the method includes:

[0015] Perform a full scan on the DFB laser chip according to an industrial camera to obtain a panoramic chip image.

[0016] A DFB laser chip is a semiconductor laser chip that uses a diffraction grating to generate feedback. Its surface structure is very precise and is usually used in high-precision fields such as optical communication. The structural characteristics of the DFB chip are very critical for its optical performance. Therefore, surface quality detection is particularly important. An industrial camera is a high-precision and high-resolution camera device dedicated to image acquisition in industrial applications. Compared with ordinary cameras, industrial cameras have better image quality, reliability and durability, and are suitable for precision detection environments. Perform a full scan on the DFB laser chip through an industrial camera. Specifically, first place the DFB laser chip in the field of view of the industrial camera, adjust the focusing system of the camera to make the surface of the chip clearly imaged. The industrial camera gradually scans the entire surface of the chip by moving the field of view. If a single lens cannot cover the entire surface of the chip, the industrial camera will take multiple images and combine these images into a complete panoramic chip image through a stitching algorithm. This panoramic chip image contains details of the entire surface of the DFB laser chip and provides basic data for subsequent image enhancement, feature extraction, and quality analysis.

[0017] Perform enhancement processing on the panoramic chip image according to a chip image enhancement function to obtain a DFB laser chip image.

[0018] Due to the complex microstructure of the DFB laser chip surface, the original image may have low contrast, making it difficult to distinguish surface details; uneven illumination affects image quality; and detail information may be covered by noise. By enhancing the image through the chip image enhancement function, the contrast of the image can be improved, thereby highlighting important surface features and improving the overall quality of the image.

[0019] Furthermore, the chip image enhancement function is: ;

[0020] Among them, g(x,y) represents the grayscale value at the position (x,y) in the DFB laser chip image, round represents the rounding process, f(x,y) represents the grayscale value at the position (x,y) in the chip panoramic image, mean represents the average grayscale value of the chip panoramic image, and Factor represents the image enhancement coefficient.

[0021] Specifically, the chip image enhancement function is: ;

[0022] The core idea of ​​this function is to adjust the grayscale value of each pixel based on the difference between the grayscale value of each pixel and the average grayscale value of the image, so as to enhance the contrast of the image. Specifically, first calculate the difference between the original grayscale value and the average grayscale value of the image. This difference reflects the brightness deviation of the pixel relative to the overall image. If the difference is positive, it means that the pixel is brighter than the average. If the difference is negative, it means that the pixel is darker than the average. Multiply the above difference by the enhancement coefficient Factor to amplify or reduce the difference in grayscale values, so that the pixels in the image that deviate from the average brightness are enhanced more obviously. Finally, round the enhanced difference and add it back to the original grayscale value to obtain a new grayscale value. This new grayscale value is the pixel value after enhancement, which makes the details of the image clearer and helps with subsequent detection and analysis.

[0023] Watershed feature segmentation is performed according to the DFB laser chip image to obtain a plurality of chip area images.

[0024] The watershed algorithm is based on the gradient information of the image. The image is regarded as a terrain, where high gradient values represent ridges and low gradient values represent basins. By simulating the process of water filling, the algorithm assigns each pixel in the image to different basins, gradually separating different regions. Specifically, based on the enhanced DFB laser chip image, the gradient value of each pixel is calculated to generate the gradient image of the chip. The highlighted parts of the gradient image correspond to the edges and protrusions of the chip surface features. There may be noise and over-segmentation in the gradient image, so it needs to be smoothed, such as using methods like Gaussian filtering. Smoothing helps reduce the noise in the image while retaining the main edge information. In the smoothed gradient image, local minima are found as the seed points for segmentation. These seed points represent the initial boundary positions of different chip regions. Based on the above seed points, the watershed algorithm is applied to divide the chip image into regions, generating a preliminary watershed segmentation result. At this time, the image is divided into multiple non-overlapping regions, and each region corresponds to a surface feature of the chip. By refining the segmentation boundaries, finally, more accurate images of multiple chip regions are obtained.

[0025] Perform expected deviation detection learning according to the chip surface structure quality detection factors, and build a chip surface quality deviation detection channel. Among them, the chip surface structure quality detection factors include surface texture features, geometric features, and microstructural features.

[0026] The chip surface structure quality detection factors include surface texture features, geometric features, and microstructural features. Among them, surface texture features reflect the microscopic surface structure of the chip material, such as roughness, repeated patterns, etc. Texture features can be measured by indicators such as the gray distribution pattern and texture complexity of the image; geometric features include the shape, edges, dimensions, and curvature of different regions on the chip surface. These features can reflect the macroscopic morphology of the chip surface, such as whether the edge shape of the chip is regular and whether the surface is flat, etc.; microstructural features involve the microscopic structural details of the chip surface at a smaller scale, such as nanoscale surface structures and the arrangement of micro-components. These features can be detected through microscopic imaging technology, and then the surface accuracy can be evaluated.

[0027] Perform expected deviation detection learning based on these chip surface structure quality detection factors. Expected deviation detection learning refers to establishing an expected model based on normal chip surface structure samples, and then comparing and analyzing the actually detected chip surface structure with this expected model to calculate the deviation degree. If the actual structure deviates too much from the expected model, it indicates that there is a problem with the chip surface quality. Based on the quality detection data of a large number of chips, through supervised learning, a model capable of detecting quality deviation is trained. For example, by analyzing the surface texture, geometry, and micro-structure features of a large number of normal samples and abnormal samples, the deviation of the chip surface quality can be automatically identified, and finally an automated chip surface quality deviation detection channel is established, which can detect whether there are quality problems by analyzing the surface features of the chip.

[0028] Based on the chip surface quality deviation detection channel, according to the chip surface quality sorting function, perform surface structure quality analysis on the multiple chip region images respectively to obtain multiple chip region surface quality coefficients.

[0029] The chip surface quality sorting function is a function used to synthesize multiple quality detection factors. This function considers the deviation coefficients of different features, assigns weights to each feature, and then calculates a comprehensive chip region surface quality coefficient.

[0030] Based on the chip surface quality detection factors, including surface texture features, geometric features, and micro-structure features, perform feature extraction on the images of each region respectively, and obtain the texture feature deviation coefficient, geometric feature deviation coefficient, and micro-structure feature deviation coefficient of each region through the chip surface quality deviation detection channel. Substitute the three feature deviation coefficients of each region into the chip surface quality sorting function, and combine the weight conditions to calculate the surface quality coefficient of each chip region, obtaining multiple chip region surface quality coefficients. Each coefficient reflects the surface quality of the corresponding region. The closer the value is to the expected standard, the better the quality of the region; the greater the deviation, the more serious the quality problem.

[0031] Draw a surface detection cloud map of the DFB laser chip according to the multiple chip region surface quality coefficients.

[0032] A cloud map is a visualization tool used to display the quality distribution in different regions. The cloud map can help identify areas with severe quality deviations, which are usually marked with special colors for easy location by operators and taking repair measures. Specifically, according to the surface quality coefficients of each region, these coefficients are mapped to a color range, usually represented by different colors from low to high. For example, red represents poor quality and green represents good quality. According to the layout of the chip surface area, the surface quality coefficients of each region are presented in the cloud map in a color-coded manner, and finally, a cloud map for detecting the surface of the DFB laser chip is obtained. Different colors in the cloud map represent the quality levels of different regions of the chip. Through color comparison, areas with quality problems can be quickly identified, which provides an intuitive visualization result for chip quality assessment and helps improve the quality control efficiency in the chip manufacturing process.

[0033] Furthermore, perform watershed feature segmentation on the DFB laser chip image to obtain multiple chip region images, including:

[0034] Perform gradient detection on the DFB laser chip image to obtain a chip gradient image; perform smoothing processing on the chip gradient image to obtain a smoothed chip gradient image; perform local minimum detection on the smoothed chip gradient image to determine multiple region seed points; based on the multiple region seed points, perform watershed segmentation on the smoothed chip gradient image to obtain an image watershed segmentation result; perform boundary optimization based on the image watershed segmentation result to obtain the multiple chip region images.

[0035] Visualize the gradient value of each pixel in the DFB laser chip image. Exemplarily, use an operator, such as the Sobel operator, to calculate the gradients of the image in the horizontal and vertical directions respectively. By combining the horizontal and vertical gradients, calculate the gradient magnitude of each pixel, also known as the gradient size, and map the gradient magnitude to a grayscale value to generate a chip gradient image. Regions with high gradient values in the image are usually parts with drastic gray-scale changes in the image, such as edges and texture mutations.

[0036] In the high-frequency regions of the gradient image, such as fine textures and noise regions, there may be a large number of subtle changes, resulting in instability in the subsequent segmentation process. Through smoothing processing, these unnecessary details can be eliminated while maintaining the main edge information, thus simplifying the subsequent image processing steps. Exemplarily, adopt the Gaussian filtering method. Through convolution operation, use a Gaussian kernel to smooth the image and reduce noise. The size and standard deviation of the filter can be adjusted according to the actual noise level of the image. Through smoothing processing, eliminate the high-frequency noise in the gradient image and generate a more stable smoothed chip gradient image.

[0037] A local minimum is a point where, within a certain local range, the pixel value is smaller than all the pixel values in its neighborhood. For a gradient image, local minima usually lie in flat regions or deep basins, and these points are the starting points for region growing in image segmentation. For each pixel in a gradient-smoothed image, its surrounding neighborhood, such as a 3x3 or 5x5 neighborhood, is examined. If the value of this pixel is lower than all the pixel values in its neighborhood, it is considered a local minimum. To avoid detecting too many local minima, a threshold can be set to retain only the local minima with a gradient value less than a certain specific threshold, which can further reduce noise and unimportant points. By detecting local minima, multiple region seed points are determined, and these points serve as the starting positions for the subsequent watershed algorithm. Each seed point represents a region, and the watershed algorithm will start from these points and gradually perform region growing and segmentation.

[0038] The watershed algorithm is based on the gradient image and regards the image as a terrain model. Regions with large gradient values represent ridges, and regions with small gradient values represent valleys. Water starts flowing from the valleys. When the water flows from different valleys (i.e., seed points) meet, a segmentation boundary is formed. Specifically, the detected local minima are used as the initial seed points for the watershed algorithm. Starting from each seed point, it expands to the surrounding pixels with smaller gradients, gradually covering the entire image. When the regions from different seed points meet, a segmentation boundary is formed, and finally the image is segmented into multiple non-overlapping regions, obtaining the multiple chip region images, where each region represents an independent chip surface area.

[0039] The watershed algorithm sometimes over-segments the image, resulting in too many regions, especially in areas with noise or small features. Sometimes it may also be inaccurate and unable to accurately reflect the true structure of the chip. Therefore, it is necessary to optimize the boundaries of the initially segmented regions to make the final segmentation result more in line with the actual requirements. Methods for boundary optimization include merging small regions, edge smoothing, region growing method, contour extraction, etc. Exemplarily, for the small regions generated by over-segmentation, they can be merged based on similar features, such as color, texture, or geometric shape. By merging adjacent regions, unnecessary small region segmentation is reduced. After boundary optimization, more accurate multiple chip region images are obtained.

[0040] Furthermore, perform expected deviation detection learning according to the chip surface structure quality detection factor, and build a chip surface quality deviation detection channel, including:

[0041] Based on the chip surface structure quality detection factors, perform backtracking of chip detection records to obtain chip texture feature deviation detection records, chip geometric feature deviation detection records, and chip microstructure feature deviation detection records; perform expected deviation detection learning based on the chip texture feature deviation detection records to build a chip texture feature deviation detection channel; perform expected deviation detection learning based on the chip geometric feature deviation detection records to build a chip geometric feature deviation detection channel; perform expected deviation detection learning based on the chip microstructure feature deviation detection records to build a chip microstructure feature deviation detection channel; connect the chip texture feature deviation detection channel, the chip geometric feature deviation detection channel, and the chip microstructure feature deviation detection channel as parallel independent nodes to generate the chip surface quality deviation detection channel.

[0042] The chip surface structure quality detection factors include surface texture features, geometric features, and microstructure features. Performing backtracking of chip detection records means retrieving the previous detection data on the chip surface quality and extracting valuable features and deviation information. Specifically, perform backtracking of chip detection records based on surface texture features to obtain chip texture feature deviation detection records, which contain the deviation between the expected value and the actual detected value of the chip surface texture. Texture features include surface roughness, distribution of repeating patterns, etc. The degree of surface texture deviation is related to the uniformity or structural integrity of the chip surface; obtain chip geometric feature deviation detection records in the same way, which involve the deviation of the chip surface geometric shape, such as surface curvature, edge profile, dimensions, etc. These records can reflect the differences in the chip geometric structure, such as whether it meets the design requirements; obtain chip microstructure feature deviation detection records in the same way, which involve the deviation of the chip surface microstructure. By detecting the deviation between the actual morphology of the microstructure and the expected model, the chip surface quality is judged.

[0043] Expected deviation detection learning refers to training a model with historical data so that it can predict the deviation of the chip surface texture and determine whether the chip surface meets the expected standard. In this way, automated detection can be carried out on new samples to identify areas with large deviations in texture features. Specifically, the chip texture feature deviation detection records are used as the training data set. Each piece of data contains the actual texture features of the chip (input data) and the deviation value from the expected features (output label). A suitable supervised learning model is selected for expected deviation detection, such as neural networks, decision trees, etc. The existing chip texture feature deviation detection records are used to train the model so that it can predict the deviation of the chip surface texture from the expected value according to the input texture features. Through cross-validation, the accuracy of the model is verified and the model is optimized to ensure that it can accurately predict new data. After training is completed, automated detection can be achieved through the model, and a texture feature deviation detection channel is established. This channel can evaluate the texture quality of the chip surface in real time and automatically mark the areas with large deviations.

[0044] The construction processes of the chip geometric feature deviation detection channel and the chip microstructure feature deviation detection channel are exactly the same as that of the chip texture feature deviation detection channel. For the sake of brevity of the specification, they are not described here.

[0045] The texture feature, geometric feature, and microstructure feature deviation detection channels are connected in parallel, and each channel serves as an independent analysis node. In this way, quality evaluations of different features of the chip surface can be independently obtained from each node, and finally a chip surface quality deviation detection channel is generated. This channel can comprehensively consider all surface features of the chip and provide a comprehensive quality evaluation.

[0046] Furthermore, expected deviation detection learning is carried out based on the chip texture feature deviation detection records, and a chip texture feature deviation detection channel is built, including:

[0047] The chip texture feature deviation detection record includes the expected record of the chip area texture features, the detected record of the chip area texture features, and the deviation coefficient record of the chip area texture features; based on the expected record of the chip area texture features, the detected record of the chip area texture features, and the deviation coefficient record of the chip area texture features, supervised learning is performed on K expected deviation detection learning models to generate K chip texture feature deviation detection models, where K is a positive integer greater than 1; using the output data sets of the K chip texture feature deviation detection models as input information and the deviation coefficient record of the chip area texture features as output information, a chip texture feature deviation detection fusion model is trained; the K chip texture feature deviation detection models are merged as parallel independent nodes to generate a chip texture feature deviation detection processing layer; the chip texture feature deviation detection processing layer is merged with the input layer of the chip texture feature deviation detection fusion model to generate the chip texture feature deviation detection channel.

[0048] The expected record of the chip area texture features refers to the texture feature information that the chip surface should present under ideal conditions. This record is based on design specifications or quality standards and represents the expected texture features of different regions on the chip surface; the detected record of the chip area texture features refers to the texture feature information of the chip surface actually detected, and these data are obtained through detection devices, reflecting the actual state of the chip surface; the deviation coefficient record of the chip area texture features is a quantitative value calculated by comparing the difference between the actually detected texture features and the expected features, and the deviation coefficient records the deviation degree of each chip area texture feature.

[0049] Supervised learning is used to train the model through sample data with known inputs and outputs, enabling the model to predict the corresponding output based on new input data. Specifically, K expected deviation detection learning models are constructed based on neural networks, decision trees, etc. Here, K represents multiple different models, and K is a positive integer greater than 1. Each model focuses on different texture feature detections to improve the overall detection effect. For example, some models can be used to detect relatively regular surface textures, while others can be specifically used to handle complex or random texture patterns. The setting of multiple models can improve the accuracy and robustness of the detection. Using the expected record, the detected record, and the deviation coefficient record as the training set, K models are trained through supervised learning. Each model learns how to predict the deviation coefficient of the chip area based on the input texture feature parameters. Through cross-validation and tuning, it is ensured that each model can also maintain good prediction ability on new data. Each model can finally predict the texture feature deviation situation of the chip area based on the new chip texture feature detection record, and finally obtain K chip texture feature deviation detection models.

[0050] Apply the trained K chip texture feature deviation detection models to the test set or a new data set, collect their prediction outputs. Each model outputs a predicted deviation coefficient. Combine these prediction results into a new data set as the input data for the fusion model. Each record in this input data set contains the prediction values of the K models. Record the actual chip area texture feature deviation coefficient as the label for the fusion model. Select a fusion model algorithm, such as linear regression, random forest, etc., to establish a fusion model. The fusion model learns how to combine these predictions based on the prediction values of the K models and outputs a more accurate deviation coefficient. Finally, train a chip texture feature deviation detection fusion model. By combining the prediction results of multiple deviation detection models, this fusion model can reduce the error of a single model and provide a more robust prediction.

[0051] Set the trained K deviation detection models as parallel computing nodes. Each model independently processes the input chip texture feature data and makes independent predictions on the same data set, outputting its own prediction results. By running the K models in parallel, a chip texture feature deviation detection processing layer is generated. This processing layer can simultaneously perform multiple more refined texture feature analyses on the input data, significantly improving the detection efficiency and accuracy.

[0052] Integrate the processing layer with the fusion model to generate a complete chip texture feature deviation detection channel. This channel can start from the input chip texture feature data, go through multi-model processing and fusion analysis, automate and efficiently perform quality detection on the surface texture of the chip, and finally provide the deviation coefficient as the detection result.

[0053] Furthermore, based on the chip surface quality deviation detection channel, according to the chip surface quality sorting function, perform surface structure quality analysis on the multiple chip area images respectively to obtain multiple chip area surface quality coefficients, including:

[0054] Extract the first chip area image according to the multiple chip area images; perform multi-feature detection on the first chip area image based on the chip surface structure quality detection factor to obtain the first chip area texture detection information, the first chip area geometric detection information, and the first chip area microstructure detection information; based on the first chip area texture detection information, the first chip area geometric detection information, and the first chip area microstructure detection information, obtain the first chip area texture feature deviation coefficient, the first chip area geometric feature deviation coefficient, and the first chip area microstructure feature deviation coefficient according to the chip surface quality deviation detection channel; input the first chip area texture feature deviation coefficient, the first chip area geometric feature deviation coefficient, and the first chip area microstructure feature deviation coefficient into the chip surface quality sorting function to obtain the first chip area surface quality coefficient; add the first chip area surface quality coefficient to the multiple chip area surface quality coefficients, and continue to perform surface structure quality analysis on the multiple chip area images according to the chip surface quality sorting function and the chip surface quality deviation detection channel to obtain the multiple chip area surface quality coefficients.

[0055] Randomly extract an analysis object from the multiple chip area images as the first chip area image.

[0056] The chip surface structure quality detection factor includes surface texture features, geometric features, and microstructure features. Perform multi-feature detection on the first chip area image based on surface texture features, geometric features, and microstructure features. Specifically, use image processing techniques to extract texture feature information. Common methods include gray-level co-occurrence matrix analysis, Fourier transform, local binary pattern, etc. Through these methods, the complexity, directionality, and roughness of the texture can be quantified to obtain the first chip area texture detection information; use geometric analysis techniques to detect the geometric shape of the chip area. Common geometric features include area, boundary length, curvature, etc. Usually, geometric information can be obtained through edge detection and morphological processing to obtain the first chip area geometric detection information; microstructure detection usually requires higher-resolution image processing means, such as microscopic imaging technology, which can detect the arrangement pattern and feature size of the fine structure on the chip surface to obtain the first chip area microstructure detection information.

[0057] Through the chip texture feature deviation detection channel, the chip geometric feature deviation detection channel, and the chip microstructure feature deviation detection channel of the chip surface quality deviation detection channel, the first chip area texture detection information, the first chip area geometric detection information, and the first chip area microstructure detection information are respectively analyzed. Each channel analyzes according to the patterns and features learned during the training process and outputs the first chip area texture feature deviation coefficient, the first chip area geometric feature deviation coefficient, and the first chip area microstructure feature deviation coefficient. These deviation coefficients can help determine the deviation degree of the chip area in different quality dimensions, and then decide whether this area meets the quality standards.

[0058] The chip surface quality combing function is used to combine the deviation situations of different features, including texture, geometry, and microstructure features, to form a comprehensive surface quality evaluation index. Taking the first chip area texture feature deviation coefficient, the first chip area geometric feature deviation coefficient, and the first chip area microstructure feature deviation coefficient as input parameters and substituting them into the chip surface quality combing function, this function calculates the first chip area surface quality coefficient according to the weights of each feature, and this coefficient can quantitatively represent the comprehensive surface quality of this area.

[0059] Add the calculated first chip area surface quality coefficient to the surface quality coefficients of the analyzed areas. According to the chip surface quality combing function and the chip surface quality deviation detection channel, perform the same analysis process on the remaining chip area images. When all areas have been analyzed, finally obtain the quality coefficients of different areas on the entire chip surface, that is, multiple chip area surface quality coefficients. Through these coefficients, a chip surface quality distribution map can be generated to identify areas that may have problems.

[0060] Furthermore, based on the first chip area texture detection information, the first chip area geometric detection information, and the first chip area microstructure detection information, according to the chip surface quality deviation detection channel, obtaining the first chip area texture feature deviation coefficient, the first chip area geometric feature deviation coefficient, and the first chip area microstructure feature deviation coefficient includes:

[0061] Based on the chip surface structure quality detection factors, perform multi-feature expectations on the chip area of the first chip area image to obtain the first chip area texture expectation information, the first chip area geometric expectation information, and the first chip area microstructure expectation information; input the first chip area texture expectation information and the first chip area texture detection information into the chip texture feature deviation detection channel to obtain the first chip area texture feature deviation coefficient; based on the first chip area geometric expectation information and the first chip area geometric detection information, obtain the first chip area geometric feature deviation coefficient according to the chip geometric feature deviation detection channel; based on the first chip area microstructure expectation information and the first chip area microstructure detection information, output the first chip area microstructure feature deviation coefficient according to the chip microstructure feature deviation detection channel.

[0062] The expectation information is the surface structure characteristics of the chip area in the design or ideal situation. These expectation information can be generated according to design specifications, process standards, or historical data of high-quality samples. The chip surface structure quality detection factors include surface texture characteristics, geometric characteristics, and microstructure characteristics. Each characteristic has its expected state, which is used to detect whether the actual situation of the chip surface deviates from the design standard. The first chip area texture expectation information is the standard information of the ideal texture characteristics of this area. The expected value is usually the optimal texture characteristics obtained based on process parameters or historical data; the first chip area geometric expectation information is the expectation information of the geometric shape of this area, and the expectation information can come from the chip design drawing or the standard manufacturing specification; the first chip area microstructure expectation information is the expected structure at the microscopic level of this area. The microstructure expectation information is usually extracted based on microscopic imaging technology and reflects the standards that the microscopic structure characteristics of the chip surface should meet.

[0063] Input the texture expectation information and the detection information into the texture feature deviation detection channel. The channel automatically analyzes the difference between the two based on the features and patterns learned during the model training process and quantifies this difference into a deviation coefficient to obtain the first chip area texture feature deviation coefficient. This coefficient can quantitatively represent the gap between the surface texture of this area and the ideal state.

[0064] Similarly, input the geometric expectation information and the detection information into the geometric feature deviation detection channel, and obtain the first chip area geometric feature deviation coefficient by analyzing and calculating the deviation degree. This coefficient is used to measure the deviation degree between the actual geometric shape and the design expectation.

[0065] Similarly, by inputting the microstructure expectation information and the detection information, the microstructure feature deviation detection channel can calculate the deviation degree of the microstructure and obtain the first chip area microstructure feature deviation coefficient. This coefficient quantifies the difference between the actual microstructure and the design standard.

[0066] Furthermore, the chip surface quality combing function is as follows: ;

[0067] wherein, XQY represents the surface quality coefficient of the chip area, XAL represents the texture feature deviation coefficient of the chip area, XBL represents the geometric feature deviation coefficient of the chip area, XCL represents the micro-structure feature deviation coefficient of the chip area, α, β, and γ represent the chip surface quality combing weight conditions, muy represents the chip surface quality combing factor, and 0 < muy < 1.

[0068] Specifically, the chip surface quality combing function is as follows: ;

[0069] This chip surface quality combing function multiplies the deviation coefficients XAL, XBL, and XCL of each feature by their respective weight coefficients α, β, and γ, and then sums them up, representing the contributions of different features to the quality assessment. muy represents the chip surface quality combing factor, and its value range is 0 < muy < 1. It affects the calculation of the surface quality coefficient by taking the logarithm of the result of the entire function. Due to the characteristics of the logarithmic function, when the value of muy is smaller, the result after taking the logarithm is more negative, making the value of the entire XQY smaller, which will regulate the result of the entire surface quality assessment, and finally obtain a comprehensive surface quality coefficient of the chip area. This coefficient quantifies the overall quality of the chip area surface.

[0070] In summary, the surface structure quality detection method for DFB laser chips provided by the embodiments of the present application has the following technical effects:

[0071] By using an industrial camera to comprehensively scan the DFB laser chip, a panoramic image of the chip is obtained, providing a complete view of all areas on the chip surface and offering comprehensive and accurate basic data for subsequent analysis. The panoramic image is processed using an image enhancement function to obtain the DFB laser chip image, enhancing the contrast and detail presentation of the image, improving the detectability of complex surface structures, and reducing analysis errors caused by image blurring or insufficient contrast. The enhanced chip image is segmented using the watershed algorithm to obtain multiple independent chip region images. The watershed algorithm, based on image gradients and local minima, ensures the precise segmentation of the chip surface area, avoiding over-segmentation or unclear region boundaries and providing an efficient basis for the feature analysis of each subsequent independent region, further improving the accuracy and locality of the detection. According to the chip surface structure quality detection factors, including texture features, geometric features, and micro-structure features, expectation deviation detection learning is carried out to establish a chip surface quality deviation detection channel, which can automatically identify deviations in surface texture, geometry, and micro-structure, enhancing the automation of the detection and reducing human errors. Based on the chip surface quality deviation detection channel, combined with the surface quality sorting function, the surface structures of multiple chip regions are analyzed for quality, generating a surface quality coefficient for each region, which can generate specific and quantifiable quality indicators for each chip region. This quantified result helps to identify problem areas and high-quality areas, facilitating subsequent quality control and production optimization. According to the surface quality coefficients of multiple chip regions, a surface detection cloud map of the DFB laser chip is drawn. The cloud map can comprehensively display the quality distribution of the entire chip surface, not only identifying problems in individual regions but also showing the quality trend of the entire chip. This visualization tool provides an efficient quality monitoring means for production management, contributing to improving the quality control level of the production line.

[0072] Embodiment 2, based on the same inventive concept as the surface structure quality detection method for the DFB laser chip in the foregoing embodiment, as Figure 2 shown, the embodiment of the present application provides a surface structure quality detection system for the DFB laser chip, and the system includes:

[0073] Chip comprehensive scanning module 10, which is used to comprehensively scan the DFB laser chip according to an industrial camera to obtain a panoramic image of the chip; Image enhancement processing module 20, which is used to enhance the panoramic image of the chip according to the chip image enhancement function to obtain a DFB laser chip image; Feature segmentation module 30, which is used to perform watershed feature segmentation on the DFB laser chip image to obtain multiple chip area images; Detection learning module 40, which is used to perform expected deviation detection learning according to the chip surface structure quality detection factor and build a chip surface quality deviation detection channel, where the chip surface structure quality detection factor includes surface texture features, geometric features, and microstructural features; Quality analysis module 50, which is used to perform surface structure quality analysis on the multiple chip area images respectively according to the chip surface quality sorting function based on the chip surface quality deviation detection channel to obtain multiple chip area surface quality coefficients; Detection cloud map drawing module 60, which is used to draw a surface detection cloud map of the DFB laser chip according to the multiple chip area surface quality coefficients.

[0074] Furthermore, the chip image enhancement function is: ;

[0075] Among them, g(x,y) represents the gray value at the position (x,y) in the DFB laser chip image, round represents rounding processing, f(x,y) represents the gray value at the position (x,y) in the panoramic image of the chip, mean represents the average gray value of the panoramic image of the chip, and Factor represents the image enhancement coefficient.

[0076] Furthermore, the system further includes a plurality of chip area image acquisition modules to perform the following operation steps:

[0077] Perform gradient detection on the DFB laser chip image to obtain a chip gradient image; perform smoothing processing on the chip gradient image to obtain a smoothed chip gradient image; perform local minimum detection on the smoothed chip gradient image to determine multiple regional seed points; based on the multiple regional seed points, perform watershed segmentation on the smoothed chip gradient image to obtain an image watershed segmentation result; perform boundary optimization based on the image watershed segmentation result to obtain the multiple chip area images.

[0078] Furthermore, the system further includes a chip surface quality deviation detection channel generation module to perform the following operation steps:

[0079] Perform chip detection record backtracking based on the chip surface structure quality detection factor to obtain a chip texture feature deviation detection record, a chip geometric feature deviation detection record, and a chip microstructure feature deviation detection record; perform expected deviation detection learning based on the chip texture feature deviation detection record to build a chip texture feature deviation detection channel; perform expected deviation detection learning based on the chip geometric feature deviation detection record to build a chip geometric feature deviation detection channel; perform expected deviation detection learning based on the chip microstructure feature deviation detection record to build a chip microstructure feature deviation detection channel; connect the chip texture feature deviation detection channel, the chip geometric feature deviation detection channel, and the chip microstructure feature deviation detection channel as parallel independent nodes to generate the chip surface quality deviation detection channel.

[0080] Furthermore, the system further includes a chip texture feature deviation detection channel generation module to perform the following operation steps:

[0081] The chip texture feature deviation detection record includes a chip area texture feature expectation record, a chip area texture feature detection record, and a chip area texture feature deviation coefficient record; according to the chip area texture feature expectation record, the chip area texture feature detection record, and the chip area texture feature deviation coefficient record, perform supervised learning on K expected deviation detection learning models to generate K chip texture feature deviation detection models, where K is a positive integer greater than 1; use the output data set of the K chip texture feature deviation detection models as input information and the chip area texture feature deviation coefficient record as output information to train a chip texture feature deviation detection fusion model; merge the K chip texture feature deviation detection models as parallel independent nodes to generate a chip texture feature deviation detection processing layer; merge the chip texture feature deviation detection processing layer with the input layer of the chip texture feature deviation detection fusion model to generate the chip texture feature deviation detection channel.

[0082] Furthermore, the system further includes a plurality of chip area surface quality coefficient acquisition modules to perform the following operation steps:

[0083] Extract the first chip area image according to the multiple chip area images; perform multi-feature detection on the first chip area image based on the chip surface structure quality detection factor to obtain the first chip area texture detection information, the first chip area geometric detection information, and the first chip area microstructure detection information; based on the first chip area texture detection information, the first chip area geometric detection information, and the first chip area microstructure detection information, obtain the first chip area texture feature deviation coefficient, the first chip area geometric feature deviation coefficient, and the first chip area microstructure feature deviation coefficient according to the chip surface quality deviation detection channel; input the first chip area texture feature deviation coefficient, the first chip area geometric feature deviation coefficient, and the first chip area microstructure feature deviation coefficient into the chip surface quality sorting function to obtain the first chip area surface quality coefficient; add the first chip area surface quality coefficient to the multiple chip area surface quality coefficients, and continue to perform surface structure quality analysis on the multiple chip area images according to the chip surface quality sorting function and the chip surface quality deviation detection channel to obtain the multiple chip area surface quality coefficients.

[0084] Furthermore, the system further includes a first chip area microstructure feature deviation coefficient output module to perform the following operation steps:

[0085] Perform multi-feature expectations on the chip area of the first chip area image based on the chip surface structure quality detection factor to obtain the first chip area texture expectation information, the first chip area geometric expectation information, and the first chip area microstructure expectation information; input the first chip area texture expectation information and the first chip area texture detection information into the chip texture feature deviation detection channel to obtain the first chip area texture feature deviation coefficient; based on the first chip area geometric expectation information and the first chip area geometric detection information, obtain the first chip area geometric feature deviation coefficient according to the chip geometric feature deviation detection channel; based on the first chip area microstructure expectation information and the first chip area microstructure detection information, output the first chip area microstructure feature deviation coefficient according to the chip microstructure feature deviation detection channel.

[0086] Furthermore, the chip surface quality sorting function is: ;

[0087] Wherein, XQY represents the chip area surface quality coefficient, XAL represents the chip area texture feature deviation coefficient, XBL represents the chip area geometric feature deviation coefficient, XCL represents the chip area microstructure feature deviation coefficient, α, β, γ represent the chip surface quality sorting weight conditions, muy represents the chip surface quality sorting factor, and 0 < muy < 1.

[0088] Through the foregoing detailed description of the method for detecting the surface structure quality of a DFB laser chip, those skilled in the art can clearly know the system for detecting the surface structure quality of a DFB laser chip in this embodiment. Since it corresponds to the method disclosed in the embodiment, it is described relatively simply. For related parts, reference can be made to the description in the method part.

[0089] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A surface structure quality detection method for a DFB laser chip, characterized in that: The method comprises: The DFB laser chip is fully scanned using an industrial camera to obtain a panoramic image of the chip; Performing enhancement processing on the chip panoramic image according to a chip image enhancement function to obtain a DFB laser chip image; Perform watershed feature segmentation according to the DFB laser chip image to obtain multiple chip area images; Performing expected deviation detection learning according to chip surface structure quality detection factors and building a chip surface quality deviation detection channel, wherein the chip surface structure quality detection factors include surface texture features, geometric features and microstructure features; Based on the chip surface quality deviation detection channel and according to the chip surface quality combing function, performing surface structure quality analysis on the multiple chip area images respectively to obtain multiple chip area surface quality coefficients; Drawing a DFB laser chip surface detection cloud map according to the surface quality coefficients of the plurality of chip regions; Among them, the expected deviation detection learning is carried out according to the chip surface structure quality detection factor, and the chip surface quality deviation detection channel is built, including: Perform chip detection record backtracking according to the chip surface structure quality detection factor to obtain chip texture feature deviation detection records, chip geometric feature deviation detection records, and chip microstructure feature deviation detection records; Based on the chip texture feature deviation detection record, expected deviation detection learning is performed to build a chip texture feature deviation detection channel; Perform expected deviation detection learning based on the chip geometric feature deviation detection record and build a chip geometric feature deviation detection channel; Based on the chip microstructure feature deviation detection record, expected deviation detection learning is performed to build a chip microstructure feature deviation detection channel; The chip texture feature deviation detection channel, the chip geometric feature deviation detection channel and the chip microstructure feature deviation detection channel are connected as parallel independent nodes to generate the chip surface quality deviation detection channel; Wherein, based on the chip surface quality deviation detection channel and according to the chip surface quality combing function, the surface structure quality analysis is performed on the multiple chip area images respectively to obtain multiple chip area surface quality coefficients, including: extracting a first chip area image according to the plurality of chip area images; Perform multi-feature detection on the first chip region image based on the chip surface structure quality detection factor to obtain first chip region texture detection information, first chip region geometry detection information and first chip region microstructure detection information; Based on the first chip region texture detection information, the first chip region geometry detection information and the first chip region microstructure detection information, according to the chip surface quality deviation detection channel, obtaining a first chip region texture feature deviation coefficient, a first chip region geometry feature deviation coefficient and a first chip region microstructure feature deviation coefficient; Inputting the first chip region texture feature deviation coefficient, the first chip region geometric feature deviation coefficient and the first chip region microstructure feature deviation coefficient into the chip surface quality combing function to obtain the first chip region surface quality coefficient; Adding the first chip area surface quality coefficient to the plurality of chip area surface quality coefficients, and continuing to perform surface structure quality analysis on the plurality of chip area images according to the chip surface quality combing function and the chip surface quality deviation detection channel to obtain the plurality of chip area surface quality coefficients; Wherein, the chip surface quality combing function is: ； Among them, XQY represents the surface quality coefficient of the chip area, XAL represents the texture feature deviation coefficient of the chip area, XBL represents the geometric feature deviation coefficient of the chip area, XCL represents the microstructure feature deviation coefficient of the chip area, α, β, γ represent the chip surface quality combing weight conditions, muy represents the chip surface quality combing factor, 0<muy<1.

2. The method according to claim 1, characterized in that The chip image enhancement function is: ； Among them, g(x,y) represents the grayscale value at the position (x,y) in the DFB laser chip image, round represents the rounding process, f(x,y) represents the grayscale value at the position (x,y) in the chip panoramic image, mean represents the average grayscale value of the chip panoramic image, and Factor represents the image enhancement coefficient.

3. The method according to claim 1, characterized in that Watershed feature segmentation is performed according to the DFB laser chip image to obtain multiple chip area images, including: Performing gradient detection according to the DFB laser chip image to obtain a chip gradient image; Performing smoothing processing on the chip gradient image to obtain a chip gradient smoothed image; Performing local minimum detection according to the chip gradient smoothed image to determine multiple regional seed points; Based on the multiple regional seed points, performing watershed segmentation on the chip gradient smoothed image to obtain an image watershed segmentation result; Boundary optimization is performed based on the image watershed segmentation result to obtain the multiple chip area images.

4. The method according to claim 1, characterized in that Based on the chip texture feature deviation detection record, expected deviation detection learning is performed to build a chip texture feature deviation detection channel, including: The chip texture feature deviation detection record includes a chip area texture feature expectation record, a chip area texture feature detection record and a chip area texture feature deviation coefficient record; According to the chip area texture feature expectation record, the chip area texture feature detection record and the chip area texture feature deviation coefficient record, supervised learning is performed on K expected deviation detection learning models to generate K chip texture feature deviation detection models, wherein K is a positive integer greater than 1; Using the output data sets of the K chip texture feature deviation detection models as input information and the chip region texture feature deviation coefficient records as output information, training a chip texture feature deviation detection fusion model; The K chip texture feature deviation detection models are merged as parallel independent nodes to generate a chip texture feature deviation detection processing layer; The chip texture feature deviation detection processing layer is combined with the input layer of the chip texture feature deviation detection fusion model to generate the chip texture feature deviation detection channel.

5. The method according to claim 1, characterized in that Based on the first chip region texture detection information, the first chip region geometry detection information and the first chip region microstructure detection information, according to the chip surface quality deviation detection channel, obtaining a first chip region texture feature deviation coefficient, a first chip region geometry feature deviation coefficient and a first chip region microstructure feature deviation coefficient, including: Perform multi-feature expectation on the chip region of the first chip region image based on the chip surface structure quality detection factor to obtain first chip region texture expectation information, first chip region geometry expectation information and first chip region microstructure expectation information; Inputting the first chip region texture expected information and the first chip region texture detection information into a chip texture feature deviation detection channel to obtain the first chip region texture feature deviation coefficient; Based on the first chip area geometry expectation information and the first chip area geometry detection information, obtaining the first chip area geometry feature deviation coefficient according to the chip geometry feature deviation detection channel; Based on the first chip region microstructure expected information and the first chip region microstructure detection information, the first chip region microstructure feature deviation coefficient is output according to the chip microstructure feature deviation detection channel.

6. A surface structure quality detection system for DFB laser chips, characterized in that: For implementing the surface structure quality detection method for a DFB laser chip according to any one of claims 1 to 5, the system comprises: A chip comprehensive scanning module, which is used to comprehensively scan the DFB laser chip using an industrial camera to obtain a panoramic image of the chip; An image enhancement processing module, the image enhancement processing module is used to perform enhancement processing on the chip panoramic image according to a chip image enhancement function to obtain a DFB laser chip image; A feature segmentation module, the feature segmentation module is used to perform watershed feature segmentation according to the DFB laser chip image to obtain a plurality of chip area images; A detection learning module, wherein the detection learning module is used to perform expected deviation detection learning according to chip surface structure quality detection factors and build a chip surface quality deviation detection channel, wherein the chip surface structure quality detection factors include surface texture features, geometric features and microstructure features; A quality analysis module, the quality analysis module is used to perform surface structure quality analysis on the plurality of chip area images respectively based on the chip surface quality deviation detection channel and according to the chip surface quality combing function, so as to obtain a plurality of chip area surface quality coefficients; A detection cloud map drawing module is used to draw a detection cloud map of the surface of the DFB laser chip according to the surface quality coefficients of the multiple chip areas.

Images