A gallium oxide transistor defect detection method and system based on image processing

Abstract

The application discloses a gallium oxide transistor defect detection method and system based on image processing, and relates to the technical field of image processing.The method comprises the following steps: acquiring a defect image of a gallium oxide transistor; performing image processing on the defect image; performing feature extraction on the defect image after image processing; calculating the fuzzy membership degree of the defect image features; taking the distribution center of the defect image features as the origin and the distribution range of the defect image features as the radius to construct a hypersphere; constructing a target function and constraint conditions related to the hypersphere; under the constraint of the constraint conditions, determining optimal hypersphere parameters by minimizing the target function; constructing a decision function according to the optimal hypersphere parameters; acquiring image features of a gallium oxide transistor to be detected; and performing defect detection on the gallium oxide transistor by the decision function to obtain a defect detection result.The application can stably work in a complex and noisy environment, and ensures the stability and reliability of defect detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and particularly relates to a gallium oxide transistor defect detection method and system based on image processing. BACKGROUND

[0002] The gallium oxide transistor defect detection method based on image processing is a technology that uses computer vision and image processing technology, combines machine learning algorithms, and automatically identifies and analyzes surface defects of gallium oxide transistors, which can accurately determine whether the transistor has defects, thereby improving the quality control and production efficiency of the transistor.

[0003] In the manufacturing process of gallium oxide transistors, any minor surface defect can seriously affect its performance and reliability. The traditional manual detection method is not only inefficient, but also easily affected by subjective factors, and is difficult to meet the requirements of modern production for high precision and high efficiency. Therefore, it is crucial to use an automatic detection method based on image processing.

[0004] However, traditional gallium oxide transistor defect detection is usually checked by hand, which requires a lot of time and effort, and is easily affected by human factors, resulting in inconsistent detection results and missed detection. Traditional automatic detection methods rely on electrical characteristics and surface scanning technologies, which can detect some obvious defects, but have limitations in detecting complex or minor defects, cannot handle complex backgrounds or minor surface flaws, and are easily disturbed in a noisy environment, cannot effectively remove noise, and have insufficient detection accuracy and robustness. SUMMARY

[0005] The purpose of the present application is to provide a gallium oxide transistor defect detection method and system based on image processing. It can solve the technical problems of traditional gallium oxide transistor defect detection, which is usually checked by hand, requires a lot of time and effort, and is easily affected by human factors, resulting in inconsistent detection results and missed detection. Traditional automatic detection methods rely on electrical characteristics and surface scanning technologies, which can detect some obvious defects, but have limitations in detecting complex or minor defects, cannot handle complex backgrounds or minor surface flaws, and are easily disturbed in a noisy environment, cannot effectively remove noise, and have insufficient detection accuracy and robustness.

[0006] To achieve the above purpose, the present application is implemented according to the following technical scheme:

[0007] The gallium oxide transistor defect detection method based on image processing comprises the following steps:

[0008] S1: Obtain a defect image of a gallium oxide transistor;

[0009] S2: performing diffusion processing, image enhancement processing and edge detection processing on the defect image;

[0010] S3: performing feature extraction on the defect image after image processing to determine defect image features;

[0011] S4: calculating fuzzy membership degrees of the defect image features;

[0012] S5: constructing a hypersphere with the distribution center of the defect image features as the origin and the distribution range of the defect image features as the radius;

[0013] S6: constructing a target function and constraint conditions related to the hypersphere based on the fuzzy membership degrees;

[0014] S7: under the constraint of the constraint conditions, adopting an optimization algorithm to determine optimal hypersphere parameters by minimizing the target function;

[0015] S8: constructing a decision function according to the optimal hypersphere parameters;

[0016] S9: obtaining image features of a gallium oxide transistor to be detected;

[0017] S10: performing defect detection on the gallium oxide transistor to be detected through the decision function according to the image features to obtain a defect detection result.

[0018] The gallium oxide transistor defect detection system based on image processing of the application comprises a processor and a memory;

[0019] The memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the gallium oxide transistor defect detection method based on image processing as described in the first aspect.

[0020] The gallium oxide transistor defect detection system based on image processing of the application has the following advantages:

[0021] By automatically acquiring the defect image of the gallium oxide transistor and performing image processing, detailed information of the surface of the transistor can be quickly acquired, time and energy consumption of manual inspection is eliminated, human factors are avoided, through feature extraction, detailed features can be extracted, the detection accuracy is greatly improved, the fuzzy membership degree calculation based on image features can process fuzzy and uncertain situations, the construction of hypersphere utilizes the distribution center and distribution range of image features, so that the boundary of the defect category is accurately constructed in the high-dimensional feature space, the normal and defective transistors are effectively distinguished, and the classification accuracy is improved, by establishing the objective function and constraint condition, combining the fuzzy membership degree, the parameters of the hypersphere are optimized, so that the construction of the hypersphere is not only more accurate, but also has stronger robustness in the noise or uncertain environment, the occurrence of missed detection and false detection is reduced, the decision function constructed based on the optimal hypersphere parameter can accurately judge whether the transistor has defects according to the image features to be detected, the misjudgment and missed judgment problems are avoided, the defective transistor and the normal transistor can be accurately classified, and the detection accuracy and efficiency are improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a flowchart of a gallium oxide transistor defect detection method based on image processing provided by an embodiment of the application.

[0023] Figure 2 is a structural schematic diagram of a gallium oxide transistor defect detection system based on image processing provided by an embodiment of the application. DETAILED DESCRIPTION

[0024] The application will be further described below in combination with the drawings and specific embodiments, the illustrative embodiments and the description of the application are used to explain the application, but not as a limitation of the application.

[0025] Referring to the drawings accompanying the specification Figure 1 , a flowchart of a gallium oxide transistor defect detection method based on image processing provided by an embodiment of the application is shown.

[0026] An embodiment of the application provides a gallium oxide transistor defect detection method based on image processing, which can include the following steps:

[0027] S1: Acquire the defect image of the gallium oxide transistor.

[0028] In a possible implementation, S1 is specifically:

[0029] The defect image of the gallium oxide transistor is acquired by a scanning electron microscope.

[0030] Scanning electron microscopy (SEM) is a device that uses an electron beam to scan the surface of a sample and analyzes the reflected electron signals to form a high-resolution image. SEM is widely used to observe material surfaces and microstructures, providing resolution up to the nanometer level. Gallium oxide (Ga2O3) transistors are transistors made of a wide-bandgap semiconductor material, possessing high breakdown voltage, excellent thermal stability, and high-temperature resistance, and are commonly used in high-power and high-frequency electronic devices.

[0031] It should be noted that SEM provides extremely high image resolution, which can clearly show tiny defects and structural problems on the surface of transistors, typically reaching nanometer-level resolution. This helps detect different types of defects, such as cracks, corrosion, or contamination. Through high-precision images, the defect features of gallium oxide transistors can be accurately captured, thus providing reliable data support for subsequent defect detection and analysis.

[0032] In this invention, transistor defects primarily refer to structural defects on or inside the gallium oxide transistor, such as cracks, bubbles, scratches, surface contamination, and other minor defects caused during the manufacturing process. These defects affect the transistor's performance and reliability, leading to abnormal electrical performance and even transistor failure. Therefore, accurate detection of these defects is crucial to ensuring the quality and performance of gallium oxide transistors.

[0033] S2: Perform image processing on the defect image.

[0034] It should be noted that image processing of defect images can effectively improve image quality, remove noise, and enhance key features. Through processing methods such as diffusion, image enhancement, and edge detection, the details of the defect area can be highlighted, the interference of background clutter on the detection results can be reduced, the defect detection can be made more accurate and stable, and the robustness of the entire detection system can be improved.

[0035] In one possible implementation, image processing specifically includes diffusion processing, image enhancement processing, and edge detection processing.

[0036] Diffusion processing is an image processing technique used to smooth noise in an image while preserving its edge information. In diffusion processing, the value of each pixel in an image is influenced by the values ​​of its surrounding pixels. Image enhancement is the process of improving the visual quality of an image, making it more suitable for further processing. Edge detection is a common image analysis method designed to identify regions in an image where brightness changes significantly. These regions often correspond to the boundaries or structural features of objects. Edge detection helps extract edge information from an image, enabling the identification of object contours or key features. It is particularly important in defect detection because defects often manifest as changes in surface or structure.

[0037] In one possible implementation, S2 specifically includes:

[0038] S201: Construct a PM model and use the PM model to perform diffusion processing on the defect image to obtain a diffused image.

[0039] The PM model is a nonlinear diffusion model used in image processing, particularly for denoising and preserving edge features. This model adjusts the diffusion coefficient based on local grayscale changes in the image, allowing smooth areas in the image to diffuse more noticeably while edge areas are preserved. This effectively removes noise and retains important image details. The core idea of ​​the PM model is to locally and adaptively adjust the diffusion intensity, suppressing diffusion at edges and enhancing diffusion in smooth areas.

[0040] Specifically, by using the PM model to perform diffusion processing on defect images, noise can be effectively removed and the image can be smoothed, while preserving edge features in the image. This can reduce false positives caused by noise and improve the accuracy of subsequent image processing and feature extraction, thereby improving the overall accuracy and robustness of defect detection.

[0041] S202: Perform image enhancement processing on the diffused image using Fourier transform and a high-pass filter to determine the enhanced image.

[0042] The Fourier transform is a mathematical method that converts a signal from the time domain (spatial domain) to the frequency domain. High-frequency components in the frequency domain typically correspond to details, edges, and noise in an image. A high-pass filter is a filter that allows high-frequency signals to pass through while suppressing low-frequency signals. High-frequency components in an image usually contain details, edges, and noise, while low-frequency components contain smooth areas and large areas of background. By using a high-pass filter, low-frequency components can be removed from the image, highlighting details and edge information, thereby enhancing image clarity.

[0043] Specifically, image enhancement through Fourier transform and high-pass filter can effectively extract high-frequency information from the image in the frequency domain and remove low-frequency noise in the background. After Fourier transform converts the image to the frequency domain, it can accurately identify details and edges in the image. Using a high-pass filter to enhance these high-frequency information makes details such as defects and edges more obvious, improves the contrast and detail of the image, and makes subsequent defect detection more accurate.

[0044] S203: Perform edge detection processing on the enhanced image using the Canny edge detector to determine the edge result image.

[0045] Among them, the Canny edge detector is a classic edge detection algorithm used to identify regions in an image with significant intensity changes, which usually represent the edges of objects.

[0046] Specifically, using the Canny edge detector to perform edge detection on enhanced images can effectively extract important structures and contours in the images. The Canny edge detector performs excellently in denoising and edge thinning, and can accurately locate defects and boundaries in the images, enhance the image's recognizability, and improve the accuracy of subsequent feature extraction and classification.

[0047] In one possible implementation, S201 specifically includes:

[0048] S2011: Initialize the defect image to obtain the initial image.

[0049] S2012: Calculate the gradient values ​​of the initial image in multiple directions:

[0050]

[0051] in, Indicates the initial image at position ( x , y The northward gradient at point ) Indicates the initial image at position ( x , y The southward gradient at point ) Indicates the initial image at position ( x , y The gradient in the east direction at point ) Indicates the initial image at position ( x , y The westward gradient at point ) I t ( x , y ) indicates the initial image at position ( x , y The pixel value at () It ( x , y- 1) indicates the initial image at position ( x , y- Pixel value at position 1), I t ( x , y+ 1) indicates the initial image at position ( x , y+ Pixel value at position 1), I t ( x+ 1, y ) indicates the initial image at position ( x+ 1, y The pixel value at () I t ( x- 1, y ) indicates the initial image at position ( x- 1, y The pixel value at ().

[0052] S2013: Calculate the local gray-level variance of the initial image:

[0053]

[0054] in, Indicates the first t In the next iteration, the initial image is at position ( x , y The local gray-level variance at point ) I t ( x+ i , y + j ) indicates the initial image at position ( x + i , y + j The pixel value at () This represents the average value of the surrounding 3×3 neighboring pixels.

[0055] S2014: Normalize the local grayscale variance:

[0056]

[0057] in, This represents the local gray-level variance after normalization. Indicates the process t After iteration, the minimum local gray-level variance of the initial image is... Indicates the process t The maximum local grayscale variance of the initial image after iteration.

[0058] S2015: Determine the diffusion coefficient based on the normalized local gray-level variance and gradient value:

[0059]

[0060] in, g Indicates the diffusion coefficient. This represents the gradient values ​​of the initial image in each direction. i =1,2,3,4 k 0 represents the threshold of the diffusion coefficient.

[0061] S2016: Construct a PM model based on the diffusion coefficient, and update the initial image using the PM model:

[0062]

[0063] in, I t+1 ( x , y ) indicates the first t+ After one iteration, the initial image is in ( x , y The pixel value at ().

[0064] In one possible implementation, after S2016, the following is also included:

[0065] The diffusion intensity of the updated initial image is controlled by the flux function.

[0066] The flux function is typically used to describe how information (such as pixel values) propagates in space within an image. In diffusion processing, the flux function controls the intensity and direction of information flow. In the PM model, the flux function is used to adjust the diffusion rate of each pixel, dynamically adjusting the smoothness of the image based on local features (such as edges and textures) to better preserve important structures and details. Diffusion intensity represents the speed and extent to which image pixel values ​​change during the diffusion process.

[0067] Specifically, by using a flux function to control the diffusion intensity of the updated initial image, the smoothness of different regions in the image can be adjusted more precisely. In smooth regions, the diffusion intensity is larger, which can effectively remove noise, while in edge or detail regions, the diffusion intensity is smaller, which can better preserve the structural information and important features of the image.

[0068] Optionally, the flux function is as follows:

[0069]

[0070] in, φ ( ) represents the flux function.

[0071] S2017: Repeat steps S2012 to S2016 until the maximum number of iterations is reached, and output the diffusion image.

[0072] In one possible implementation, S202 specifically includes:

[0073] S2021: The diffusion image is converted from the spatial domain to the frequency domain by Fourier transform to obtain the frequency domain image.

[0074] S2022: Enhance the frequency domain image using a high-pass filter and determine the enhanced image.

[0075] The specific formula for calculating a high-pass filter is as follows:

[0076]

[0077] in, B ( u , v ) represents the point in the frequency domain of the high-pass filter. u , v The value of ) D 0 indicates the cutoff frequency. D ( u , v ) represents a point in the frequency domain. u , v The distance to the center of the Fourier spectrum. n Indicates the order of the filter.

[0078] In one possible implementation, S203 specifically includes:

[0079] S2031: Determine the gradient of the enhanced image using a differential operator.

[0080] S2032: Calculate the edge intensity and edge direction of the enhanced image based on the gradient:

[0081]

[0082] in, M ( x , y ) indicates that the enhanced image is at location ( x , y Edge strength at ) g x and g y These represent the enhanced image at... x direction and y gradient of direction, Indicates the enhancement image at location ( x ,y The edge direction at () is given by tan, which represents the tan function. -1 This indicates the inverse operation.

[0083] S2033: Based on edge intensity and edge direction, the edge points of the enhanced image are output through non-maximum suppression algorithm and hysteresis thresholding algorithm.

[0084] Non-maximum suppression (NVS) is a commonly used algorithm in edge detection. Its purpose is to retain the local maximum value of each pixel during edge detection and remove points that are not edges. In the Canny edge detector, NVS is used to refine edges, ensuring that each edge point retains only the strongest pixel value, thus making the edges more refined and clear. Hysteresis thresholding is a final step in edge detection used to determine edges. By setting two thresholds (a high threshold and a low threshold), pixels stronger than the high threshold are first marked as edge points. Then, pixels below the low threshold are tracked to determine if they are connected to the high threshold region. If connected, they are also marked as edge points. The hysteresis thresholding algorithm can remove falsely detected edge points, ensuring the accuracy of edge detection.

[0085] Specifically, by employing non-maximum suppression (NMS) and hysteresis thresholding algorithms, image edges can be effectively refined, and non-edge noise points can be removed. NMS ensures that only the strongest pixels are retained for each edge, thereby improving edge accuracy and continuity. Hysteresis thresholding further optimizes edge detection by connecting low-intensity edge points, reducing the possibility of false detections. The combination of these two methods significantly improves the accuracy and robustness of edge detection.

[0086] S2034: Merge edge points and output the edge result image.

[0087] S3: Extract features from the processed defect image to determine the defect image features.

[0088] It should be noted that by extracting features from the processed defect images, key defect features can be effectively identified from the images. This transforms the complex data of the images into simplified, analyzable numerical forms, improving the efficiency and accuracy of subsequent classification algorithms. By extracting useful features, detection accuracy can be significantly improved, and false positives and false negatives can be reduced.

[0089] The specific features of defective images include: contrast, homogeneity, maximum gray value, eccentricity, and average gray value.

[0090] S4: Calculate the fuzzy membership degree of the defect image features.

[0091] It should be noted that fuzzy membership is a concept in fuzzy set theory, used to measure the degree to which an element belongs to a certain set. In image processing and pattern recognition, fuzzy membership is used to represent the degree of fuzzy affiliation of a pixel or feature among different categories. Its value is between 0 and 1. The closer the value is to 1, the more likely the feature is to belong to a certain category (such as the defect category). The closer the value is to 0, the higher the probability that it belongs to other categories.

[0092] In one possible implementation, S4 specifically includes:

[0093] S401: Map the defect image features to a high-dimensional feature space.

[0094] In this context, a high-dimensional feature space refers to a space where data is mapped from its original low-dimensional space to a higher-dimensional space through feature mapping or transformation. In a high-dimensional space, data can be more easily separated and distinguished, especially when the data is not linearly separable.

[0095] Specifically, by mapping defect image features to a high-dimensional feature space, the relationships between image features are presented more clearly. This can effectively transform the original nonlinear data into linearly separable data in a high-dimensional space, making classification and detection easier.

[0096] S402: Determine the distance between defect image features and target class centers:

[0097]

[0098] Wherein, Φ( x r ) represents defect image features r Mapping representation in high-dimensional feature space This represents the representation of the target class center in the high-dimensional feature space, where || denotes the norm notation. Indicates the first m A set of defective image features for each cluster. Indicates the first m The total number of defect image features in each cluster, Φ( x k ) represents defect image features k Mapping representation in high-dimensional feature space K ( x r , x r ) represents defect image features r With defect image features r The kernel function values ​​between K ( x p ,x q ) represents defect image features p and defect image features q The kernel function values ​​between K ( x r , x k ) represents defect image features r and defect image features k The kernel function values ​​between.

[0099] S403: Calculate fuzzy membership degree based on distance:

[0100]

[0101] in, Representing defect image features r Fuzzy membership degree, τ This represents the lower bound of the membership degree, and max represents maximization.

[0102] In this invention, τ = ξ 0.3.

[0103] S5: Construct a hypersphere with the distribution center of the defect image features as the origin and the distribution range of the defect image features as the radius.

[0104] A hypersphere is a spherical region in a high-dimensional space with a uniform radius. Hyperspheres are used to describe the distribution of data in a high-dimensional feature space, with the goal of containing as much normal data as possible within the hypersphere.

[0105] It should be noted that by constructing a hypersphere with the distribution center of the defective image features as the origin and the distribution range as the radius, the distribution characteristics of the data in the feature space can be accurately described. Normal data can be effectively enclosed within the hypersphere, while abnormal or defective data can be excluded. The construction of the hypersphere simplifies the anomaly detection process and improves the accuracy of classification, especially when the feature distribution is relatively concentrated, it can more accurately distinguish between normal and defective samples.

[0106] S6: Based on fuzzy membership, construct the objective function and constraints for the hypersphere.

[0107] It should be noted that by constructing the objective function and constraints of the hypersphere based on fuzzy membership, the uncertainty and fuzziness in the data can be handled more accurately. This allows for dynamic adjustment of the boundary between normal and abnormal data, providing a certain tolerance for samples in marginal regions, thereby reducing false positives and false negatives. Furthermore, the introduction of fuzzy membership makes the algorithm more robust, better adaptable to complex or incomplete sample data, and improves the accuracy and stability of defect detection.

[0108] The objective function is as follows:

[0109]

[0110] in, f Describe the objective function. R This represents the radius of the hypersphere. C 1 and C 2 represents the penalty factor. Representing defect image features r Fuzzy membership degree, iter r Let 'min' represent the slack variable, and 'min' represent minimization. r =1,2,…, N , N This represents the total number of defective image features.

[0111] The specific constraints are as follows:

[0112]

[0113] Wherein, Φ( x r ) represents defect image features r Mapping representation in high-dimensional feature space a It represents the center of the hypersphere.

[0114] S7: Under the constraints, with the objective function as the goal, the optimal hypersphere parameters are determined by using an optimization algorithm.

[0115] It should be noted that by minimizing the objective function under constraints, the model can be ensured to be both accurate and efficient when processing complex data. The optimization algorithm can automatically adjust the radius and center position of the hypersphere so that the hypersphere can best adapt to the distribution of data, while maximizing the encirclement of normal data and excluding defective data, thereby improving the automation of the model and ensuring the accuracy and robustness of the detection process.

[0116] In one possible implementation, S7 specifically includes:

[0117] S701: Introducing Lagrange multipliers, we obtain the dual objective function and dual constraints.

[0118] Lagrange multipliers are mathematical tools used to solve constrained optimization problems. By introducing multipliers, the constraints are combined with the objective function, transforming the problem into an unconstrained optimization problem, thus facilitating its solution. The dual objective function is obtained from the primal optimization problem using the Lagrange multiplier method. The dual constraints are the representation of the constraints of the primal problem within the Lagrange multiplier method.

[0119] Specifically, by introducing Lagrange multipliers, the constraints are transformed into a dual problem, which effectively simplifies the optimization process. The dual problem avoids explicit computation of high-dimensional eigenmaps and reduces computational complexity by utilizing kernel function techniques. Furthermore, by maximizing the dual objective function, the optimal hypersphere parameters can be solved more stably and efficiently.

[0120] The dual objective function and dual constraints are as follows:

[0121]

[0122] in, f d Describe the dual objective function. α r Representing defect image features r The corresponding Lagrange multipliers, α s Representing defect image features s The corresponding Lagrange multipliers, K ( x r , x s ) represents defect image features r With defect image features s The kernel function values ​​between r = s =1,2,…, N , N This represents the total number of defective image features.

[0123]

[0124]

[0125] in, C 1 and C 2 represents the penalty factor. Representing defect image features r Fuzzy membership degree.

[0126] S702: Under the constraints of duality, the optimal hypersphere parameters are determined by minimizing the dual objective function using the crow optimization algorithm.

[0127] In one possible implementation, S702 specifically includes:

[0128] S7021: Set the initial parameters of the crow search algorithm, including: population size, maximum number of iterations, and search range.

[0129] S7022: Initialize the position and velocity of all crows, where the crow's position represents the hypersphere parameter.

[0130] S7023: Using the dual objective function as the fitness function, calculate the fitness value of each crow.

[0131] Update the positions of each crow based on their fitness values:

[0132]

[0133]

[0134]

[0135] in, h p,iter+1 Indicates the first p Only crows in the first iter Position in one iteration h p,iter Indicates the first p Only crows in the first ω Position in the next iteration iter Indicates inertia weight, c This represents a random number in the interval [0,1]. fl p,iter Indicates the first p Only crows in the first iter Flight distance in the next iteration m q,iter Indicates the first iter During the nth iteration q The best spot for a crow ω max Indicates the maximum number of iterations. AP max This represents the maximum inertia weight value. t Indicates the current iteration number. AP Indicates the probability of consciousness. AP 1 and λ 2 represents the upper and lower bounds of the probability of consciousness, respectively. gbest Represents a random variable. Figure 2 Let Γ represent the global optimal position, and let Γ represent the gamma function. C (0,1) represents a vector of random numbers between 0 and 1. C (0,1) is a perturbation factor, often used to enhance jumpiness and diversity.

[0136] S7024: Calculate the fitness value of each crow after the update.

[0137] S7025: If the fitness value of the updated crow is greater than or equal to the fitness value of the current crow, update the current crow's position. If the fitness value of the updated crow is less than the fitness value of the current crow, keep the current crow's position unchanged.

[0138] S7026: Repeat steps S7022 to S7025 until the maximum number of iterations is reached, and determine the optimal hypersphere parameters.

[0139] It should be noted that by minimizing the dual objective function under dual constraints and using the crow optimization algorithm, the optimal solution for the hypersphere parameters can be effectively searched, avoiding getting trapped in local optima and providing good convergence. It can quickly determine the optimal hypersphere parameters under multiple constraints, thereby improving the accuracy and stability of defect detection.

[0140] S8: Construct a decision function based on the optimal hypersphere parameters.

[0141] The decision function is a function used in classification algorithms to determine the category of a sample. It maps the features of a sample to a numerical value and determines which category the sample belongs to based on that numerical value.

[0142] It should be noted that by constructing a decision function based on the optimal hypersphere parameters, the samples to be detected can be accurately classified, ensuring that defective samples are accurately contained within the hypersphere, while normal samples are correctly identified as anomalies. This can significantly improve the accuracy and stability of defect detection and reduce false positives and false negatives.

[0143] The decision function is as follows:

[0144]

[0145] in, D 2 ( z ) represents the decision function, Φ( z ) represents image features z Mapping representation in high-dimensional feature space K ( z , z ) represents image features z and image features z The kernel function values ​​between K ( x r , z ) represents defect image features x r and image features z The kernel function values ​​between T This indicates the transpose operation.

[0146] S9: Acquire image features of the gallium oxide transistor to be detected.

[0147] S10: Based on the image features, a decision function is used to perform defect detection on the gallium oxide transistor to be tested, and the defect detection result is obtained.

[0148] It should be noted that by using decision functions to detect defects in image features, it is possible to accurately determine whether gallium oxide transistors have defects, ensuring the stability and reliability of the detection results and improving the efficiency and accuracy of automated detection systems.

[0149] In one possible implementation, S10 specifically includes:

[0150] Determine whether the decision function is valid. If it is, determine that the gallium oxide transistor under test is a defective gallium oxide transistor; otherwise, determine that the gallium oxide transistor under test is a normal gallium oxide transistor.

[0151] In this invention, defect images of gallium oxide transistors are first acquired using a scanning electron microscope. Image processing techniques such as diffusion processing, image enhancement, and edge detection are then used to preprocess the images, highlighting defect features. Next, through feature extraction and fuzzy membership calculation, image features are mapped to a high-dimensional feature space, and a hypersphere model is constructed. An optimization algorithm determines the optimal parameters of the hypersphere by minimizing the objective function, thereby constructing a decision function. Finally, the transistors under inspection are classified based on the image features and the decision function to determine whether defects exist. This invention offers advantages such as high precision, high robustness, and automation, effectively improving defect detection efficiency and reducing false positives and false negatives.

[0152] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0153] In this embodiment of the invention, by automatically acquiring defect images of gallium oxide transistors and performing image processing, detailed information about the transistor surface can be quickly obtained, eliminating the time and effort consumed by manual inspection and avoiding interference from human factors. Through feature extraction, detailed features can be extracted, greatly improving the detection accuracy. Based on the fuzzy membership degree calculation of image features, it can handle fuzzy and uncertain situations. The construction of the hypersphere utilizes the distribution center and distribution range of image features, thereby accurately constructing the boundary of the defect category in the high-dimensional feature space, effectively distinguishing normal and defective transistors, and improving the accuracy of classification. By establishing objective functions and constraints, combined with fuzzy membership degrees, the parameters of the hypersphere are optimized, making the construction of the hypersphere not only more accurate, but also more robust in noisy or uncertain environments, reducing the occurrence of missed detections and false detections. The decision function constructed based on the optimal hypersphere parameters can accurately determine whether there are defects in the transistor according to the features of the image to be detected, avoiding the problems of false and missed detections, and accurately classifying defective transistors and normal transistors, improving the accuracy and efficiency of detection.

[0154] Reference manual attached ​ The diagram shows a schematic of the structure of a gallium oxide transistor defect detection system based on image processing provided in an embodiment of the present invention.

[0155] This invention provides a gallium oxide transistor defect detection system 20 based on image processing, comprising: a processor 201 and a memory 202;

[0156] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described gallium oxide transistor defect detection method based on image processing and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0157] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0158] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0159] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0160] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0161] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0162] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0163] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0164] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0165] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0166] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0167] This invention provides a readable storage medium comprising: storing a program or instructions on the readable storage medium, wherein when the program or instructions are executed by a processor, the program or instructions implement the steps of the above-described gallium oxide transistor defect detection method based on image processing, and can achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0168] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. A method for detecting defects in gallium oxide transistors based on image processing, characterized by, The method comprises the following steps: S1: acquiring a defect image of a gallium oxide transistor; S2: sequentially performing diffusion processing, image enhancement processing and edge detection processing on the defect image; S3: performing feature extraction on the defect image after image processing to determine defect image features; S4: calculating fuzzy membership degrees of the defect image features; S5: constructing a hypersphere with the distribution center of the defect image features as the origin and the distribution range of the defect image features as the radius; S6: constructing a target function and constraint conditions related to the hypersphere based on the fuzzy membership degrees; S7: under the constraint of the constraint conditions, adopting an optimization algorithm to determine optimal hypersphere parameters by minimizing the target function; S8: constructing a decision function according to the optimal hypersphere parameters; S9: acquiring image features of a gallium oxide transistor to be detected; S10: performing defect detection on the gallium oxide transistor to be detected through the decision function according to the defect image features to obtain a defect detection result.

2. The image processing based gallium oxide transistor defect detection method of claim 1, wherein, The S2 specifically comprises: S201: constructing a PM model and performing diffusion processing on the defect image through the PM model to obtain a diffusion image; S202: performing image enhancement processing on the diffusion image through Fourier transform and a high-pass filter to determine an enhanced image; S203: performing edge detection processing on the enhanced image through a Canny edge detector to determine an edge result image.

3. The image processing based gallium oxide transistor defect detection method of claim 2, wherein, The S201 specifically comprises: S2011: initializing the defect image to obtain an initial image; S2012: calculating gradient values of the initial image in multiple directions; S2013: calculating local gray variance of the initial image; S2014: performing normalization processing on the local gray variance; S2015: determining diffusion coefficients according to the normalized local gray variance and the gradient values in the directions; S2016: constructing a PM model according to the diffusion coefficients and updating the initial image through the PM model; S2017: repeating steps S2012 to S2016 until a maximum iteration number is reached, and outputting the diffusion image.

4. The image processing based gallium oxide transistor defect detection method of claim 3, wherein, After the S2016, the method further comprises: controlling the diffusion intensity of the updated initial image through a flux function.

5. The image processing based gallium oxide transistor defect detection method of claim 2, wherein, The S202 specifically comprises: S2021: converting the diffusion image from a spatial domain to a frequency domain through the Fourier transform to obtain a frequency domain image; S2022: performing image enhancement processing on the frequency domain image through the high-pass filter to determine the enhanced image, and a calculation formula of the high-pass filter is specifically as follows: wherein B u v represents the value of the high-pass filter at the point u v in the frequency domain, D 0 represents the cut-off frequency, D u v represents the distance of the point u v in the frequency domain to the center of the Fourier spectrum, n represents the order of the filter.​​​​​​ 6. The image processing based gallium oxide transistor defect detection method of claim 2, wherein, The S203 specifically comprises: S2031: determining the gradient of the enhanced image through a differential operator; S2032: calculating the edge intensity and edge direction of the enhanced image according to the gradient; S2033: outputting edge points of the enhanced image through a non-maximum suppression algorithm and a hysteresis threshold processing algorithm according to the edge intensity and the edge direction; S2034: fusing the edge points to output the edge result image.

7. The image processing based gallium oxide transistor defect detection method of claim 1, wherein, The S4 specifically comprises: S401: map the defect image feature into a high-dimensional feature space; S402: determine a distance between the defect image feature and a target class center in the high-dimensional feature space; S403: calculate the fuzzy membership degree according to the distance.

8. The image processing based gallium oxide transistor defect detection method of claim 1, wherein, The S7 specifically comprises: S701: introduce a Lagrange multiplier to obtain a dual objective function and a dual constraint condition; S702: under the constraint of the dual constraint condition, determine the optimal hypersphere parameter by a crow optimization algorithm, with the objective of minimizing the dual objective function.

9. The image processing based gallium oxide transistor defect detection method of claim 1, wherein, The S10 specifically comprises: judging whether the decision function is established; if yes, determining that the to-be-detected gallium oxide transistor is a defective gallium oxide transistor; otherwise, determining that the to-be-detected gallium oxide transistor is a normal gallium oxide transistor.

10. An image processing based gallium oxide transistor defect detection system, comprising: Comprise: A processor and a memory; The memory stores a program or instructions executable on the processor, and the program or instructions, when executed by the processor, implement the steps of the gallium oxide transistor defect detection method based on image processing according to any one of claims 1 to 9.

Images