A method, system, and electronic device for detecting foreign matter layering.

Abstract

This application discloses a method, system, and electronic device for foreign object layer detection, used to distinguish the category and layer of foreign objects based on feature engineering. The method for foreign object layer detection includes: acquiring a first original image, a second original image, a third original image, and a fourth original image of a display panel under test; performing homography transformation on the third original image and the fourth original image respectively to obtain a first transformed image and a second transformed image; extracting shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image respectively; performing standardization and normalization processing on the shape features, texture features, and grayscale features; evaluating and weighting the processed shape features, processed texture features, and processed grayscale features respectively, and fusing them to obtain a fusion result; inputting the fusion result into a pre-constructed classifier and outputting a classification result.

Description

Technical Field

[0001] This application relates to the field of image processing and detection technology, and in particular to a method, system, and electronic device for foreign object layer detection. Background Technology

[0002] The rapid development of display panel technology, especially the widespread application of liquid crystal displays (LCDs) and organic light-emitting diodes (OLEDs), has greatly enhanced the user's visual experience and interactive convenience. The manufacturing process of display panels involves numerous precise and complex steps, and the raw materials and production environment can introduce various foreign objects such as dust, dirt, and fibers. Taking OLED panels as an example, the key component, Cover Glass (CG), needs to be bonded to the pixel layer structure using optically clear adhesive (OCA). Before this bonding process, the CG itself is covered with a protective film (CG film), and the OCA layer is also usually covered with a transparent protective film called a backing film. Therefore, in actual production, each layer of the above structure may be contaminated with foreign object interference. Currently, the existing technologies for detecting foreign objects in display panels are mostly image acquisition schemes using a "single camera + single light source" or "fixed camera + light source combination," combined with simple grayscale contrast analysis and single shape feature extraction, and then foreign object identification is achieved through rule-based logical judgment.

[0003] However, only the presence of foreign matter in the OCA adhesive located between the heavy film and CG layer truly affects product quality. Foreign matter in other layers can be easily removed later without affecting product yield. However, the existing technology suffers from significant differences in the reflection and refraction characteristics of foreign matter at different layers. Images acquired from a single viewpoint or with a fixed light source cannot fully capture the appearance characteristics of foreign matter at different layers, making it difficult for the detection system to distinguish the actual location of the foreign matter. This easily leads to misjudging "non-influencing dirt on the heavy film surface" as "fatal foreign matter inside the OCA," or missing tiny foreign matter deep within the OCA, resulting in high false positive and false negative rates, failing to meet the requirements of high-precision quality control. Summary of the Invention

[0004] This application discloses a method, system, and electronic device for foreign object layer detection, which is used to distinguish the type and layer of foreign objects based on feature engineering.

[0005] The first aspect of this application discloses a method for detecting foreign matter layering, comprising:

[0006] A first original image of the display panel under test is acquired using an upper camera and an upper light source; a second original image of the display panel under test is acquired using an upper camera and a lower light source; a third original image of the display panel under test is acquired using a lower camera and an upper light source; and a fourth original image of the display panel under test is acquired using a lower camera and a lower light source.

[0007] The third original image and the fourth original image are respectively homography transformed to the viewpoint of the upper camera to obtain the first transformed image and the second transformed image;

[0008] Extract the shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image, respectively;

[0009] The shape features, texture features, and grayscale features are standardized and normalized.

[0010] The processed shape features, processed texture features, and processed grayscale features are evaluated and weighted based on a pre-defined evaluation model.

[0011] The processed shape features, processed texture features, and processed grayscale features are fused according to the assigned weights to obtain a fusion result;

[0012] The fusion result is input into a pre-built classifier, and the classification result is output.

[0013] Optionally, the standardization and normalization processing of the shape features, texture features, and grayscale features includes:

[0014] Obtain the standardization formula: z=(xi-μ) / σ, where z is the standard score, xi is the feature value corresponding to the shape feature, the texture feature and the grayscale feature respectively, μ is the mean of the feature value corresponding to the shape feature, the texture feature and the grayscale feature respectively, and σ is the standard deviation of the feature value corresponding to the shape feature, the texture feature and the grayscale feature respectively;

[0015] According to the standardization formula, calculate the standard score of the shape feature, the standard score of the texture feature, and the standard score of the grayscale feature respectively;

[0016] Obtain the normalization formula: x'=(z-zmin) / (zmax-zmin), where x' is the normalized eigenvalue, z is the standard score, zmin is the minimum of all standard scores, and zmax is the maximum of all standard scores;

[0017] According to the normalization formula, the normalized feature values ​​of the shape feature, the texture feature, and the grayscale feature are calculated respectively.

[0018] Optionally, the preset evaluation model is a pre-built XGBoost model, and the evaluation and weight assignment of the processed shape features, processed texture features, and processed grayscale features based on the preset evaluation model includes:

[0019] The saliency of the processed shape features, processed texture features, and processed grayscale features were evaluated based on the XGBoost model.

[0020] The weighting formulas are: t = im1 + im2 + im3, α = im1 / t, β = im2 / t, γ = im3 / t, where t is the overall saliency, im1 is the saliency of the processed shape feature, im2 is the saliency of the processed texture feature, im3 is the saliency of the processed grayscale feature, α is the weighting factor of the processed shape feature, β is the weighting factor of the processed texture feature, and γ is the weighting factor of the processed grayscale feature.

[0021] According to the weighting formula, the weighting factors of the processed shape features, the processed texture features, and the processed grayscale features are calculated respectively.

[0022] Optionally, the step of fusing the processed shape features, the processed texture features, and the processed grayscale features according to the assigned weights to obtain the fusion result includes:

[0023] Obtain the fusion formula: F=[α×F1,β×F2,γ×F3], where F is the fusion result, α is the weight factor of the processed shape feature, F1 is the processed shape feature, β is the weight factor of the processed texture feature, F2 is the processed texture feature, γ is the weight factor of the processed grayscale feature, and F3 is the processed grayscale feature.

[0024] The fusion result is calculated according to the fusion formula.

[0025] Optionally, the pre-built classifier is a plurality of support vector machine classifiers, each support vector machine classifier corresponding to a category. Before inputting the fusion result into the pre-built classifier and outputting the classification result, the method further includes:

[0026] Based on the PCA principal component analysis algorithm, the space of the fusion result is reduced to a preset dimension to obtain the dimensionality-reduced fusion result;

[0027] The step of inputting the fusion result into a pre-built classifier and outputting a classification result includes:

[0028] The dimensionality-reduced fusion results are input into the multiple support vector machine classifiers respectively, and the corresponding multiple decision values ​​are output;

[0029] The classification result is determined based on the multiple decision values, and the classification result includes the type of foreign object and the level at which the foreign object is located.

[0030] Optionally, before the third original image and the fourth original image are homogenized to the viewpoint of the upper camera, the method further includes:

[0031] Based on a pre-set calibration plate, the positions of the images captured by the upper camera and the images captured by the lower camera are calibrated to obtain calibration results;

[0032] Based on the calibration results, calculate the homography transformation matrix of the image captured by the lower camera and converted to the viewpoint of the upper camera;

[0033] The step of homography-transforming the third original image and the fourth original image to the viewpoint of the upper camera to obtain the first transformed image and the second transformed image includes:

[0034] Based on the homography transformation matrix, homography transformation is performed on the third original image and the fourth original image respectively to obtain the first transformed image and the second transformed image.

[0035] Optionally, before extracting the shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image, respectively, the method further includes:

[0036] Based on the point-line detection algorithm, candidate foreign object regions in the first original image, the second original image, the first transformed image, and the second transformed image are identified respectively.

[0037] Optionally, the step of extracting the shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image respectively includes:

[0038] The shape features corresponding to the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image are obtained respectively. The shape features include area, horizontal length, vertical length, aspect ratio, and roundness.

[0039] Based on the local binary pattern, gray-level co-occurrence matrix, or Gabor filter operator, extract the corresponding texture features in the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image, respectively.

[0040] The grayscale features corresponding to the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image are obtained respectively. The grayscale features include regional contrast, regional variance, average grayscale, maximum grayscale, minimum grayscale, and uniformity features.

[0041] The second aspect of this application discloses a system for detecting foreign object layering, comprising:

[0042] An image acquisition unit is used to acquire a first original image of the display panel under test through an upper camera and an upper light source, acquire a second original image of the display panel under test through an upper camera and a lower light source, acquire a third original image of the display panel under test through a lower camera and an upper light source, and acquire a fourth original image of the display panel under test through a lower camera and a lower light source.

[0043] The homography transformation unit is used to homography transform the third original image and the fourth original image to the viewpoint of the upper camera, respectively, to obtain a first transformed image and a second transformed image;

[0044] The extraction unit is used to extract the shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image, respectively.

[0045] The standardization and normalization processing unit is used to perform standardization and normalization processing on the shape features, texture features, and grayscale features;

[0046] The weight allocation unit is used to evaluate and assign weights to the processed shape features, processed texture features, and processed grayscale features based on a preset evaluation model.

[0047] The fusion unit is used to fuse the processed shape features, the processed texture features, and the processed grayscale features according to the assigned weights to obtain a fusion result;

[0048] A classification unit is used to input the fusion result into a pre-built classifier and output the classification result.

[0049] A third aspect of this application provides an electronic device, comprising:

[0050] Processor, memory, input / output units, and bus;

[0051] The processor is connected to memory, input / output units, and a bus;

[0052] The memory stores a program, which the processor calls to execute methods such as the first aspect and any optional foreign matter layering detection method of the first aspect.

[0053] The fourth aspect of this application provides a computer-readable storage medium on which a program is stored, which, when executed on a computer, performs the method of foreign matter layer detection as described in the first aspect and any optional method of the first aspect.

[0054] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0055] This application provides a method for foreign object layer detection, comprising: acquiring a first original image of a display panel under test using an upper camera and an upper light source; acquiring a second original image of the display panel under test using an upper camera and a lower light source; acquiring a third original image of the display panel under test using a lower camera and an upper light source; and acquiring a fourth original image of the display panel under test using a lower camera and a lower light source; homography transforming the third original image and the fourth original image to the viewpoint of the upper camera, respectively, to obtain a first transformed image and a second transformed image; extracting shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image, respectively; standardizing and normalizing the shape features, texture features, and grayscale features; evaluating and weighting the processed shape features, processed texture features, and processed grayscale features based on a preset evaluation model; fusing the processed shape features, processed texture features, and processed grayscale features according to the assigned weights to obtain a fusion result; and inputting the fusion result into a pre-constructed classifier and outputting a classification result. The foreign object layer detection method of this application first uses upper and lower cameras to capture four different combinations of images with upper and lower light sources, respectively. By using differentiated illumination angles, it fully captures the unique reflection, refraction, and scattering characteristics of light on foreign objects located at different layers. Second, it uses homography transformation to unify the lower camera's perspective with the upper camera's perspective, eliminating the feature misalignment problem caused by the difference in perspective between the upper and lower cameras, and ensuring that the four sets of images can be accurately compared based on a unified perspective. Next, it extracts shape features, texture features, and grayscale features, and eliminates the numerical scale interference of different features through standardization and normalization. Then, it combines the weight allocation of a preset evaluation model to achieve feature fusion. Finally, it outputs the classification result through a pre-built classifier, which can accurately identify the specific category and layer of the foreign object, effectively reducing the false detection rate and false negative rate, avoiding production waste caused by misjudging non-influencing foreign objects, and meeting the urgent need for high-precision quality control. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of an embodiment of the foreign matter layer detection method of this application;

[0058] Figure 2 This is a schematic diagram of another embodiment of the foreign matter layer detection method of this application;

[0059] Figure 3 This is a schematic diagram of another embodiment of the foreign matter layer detection method of this application;

[0060] Figure 4 This is a schematic diagram of another embodiment of the foreign matter layer detection method of this application;

[0061] Figure 5 This is a schematic diagram of another embodiment of the foreign matter layer detection method of this application;

[0062] Figure 6 This is a schematic diagram of another embodiment of the foreign matter layer detection method of this application;

[0063] Figure 7 This is a schematic diagram of an embodiment of the foreign object layer detection system of this application;

[0064] Figure 8 This is a schematic diagram of one embodiment of the electronic device of this application. Detailed Implementation

[0065] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0066] It should be understood that, when used in this application specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0067] It should also be understood that the term “and / or” as used in this application specification means any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0068] As used in this application specification, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."

[0069] Furthermore, in the description of this application, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0070] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0071] Based on the above, this application discloses a method, system, and electronic device for detecting foreign objects in layers, which is used to distinguish the type and layer of foreign objects based on feature engineering.

[0072] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0073] The method described in this application can be applied to servers, devices, terminals, or other devices with logical processing capabilities; therefore, this application does not limit its application. For ease of description, the following description uses a terminal as the executing entity.

[0074] Please see Figure 1 This application provides an embodiment of a method for detecting foreign object layering, comprising:

[0075] 101. Acquire a first original image of the display panel under test using an upper camera and an upper light source; acquire a second original image of the display panel under test using an upper camera and a lower light source; acquire a third original image of the display panel under test using a lower camera and an upper light source; and acquire a fourth original image of the display panel under test using a lower camera and a lower light source.

[0076] In this embodiment, the upper and lower cameras can be industrial cameras with a resolution of no less than 5 megapixels and a frame rate of no less than 30fps. The lens focal length is set to 8-25mm depending on the installation distance between the platform and the camera to ensure complete coverage of the field of view of the display panel under test. The upper and lower light sources can be LED surface light sources with adjustable light intensity, with an adjustment range of 500-5000 lux, to adapt to OLED or LCD display panels of different thicknesses. The specific shooting process can be as follows: First, the upper camera is triggered and the upper light source is lit simultaneously to capture the first original image, which can highlight the surface reflection characteristics. Then, while keeping the upper camera on, the lower light source is switched on to capture the second original image, which can highlight the internal obstruction using transmitted light. Next, the lower camera is switched on and the upper light source is lit to capture the third original image, which is an observation of the transmission effect of the upper light source from a bottom perspective. Finally, while keeping the lower camera on and the lower light source is triggered, the fourth original image is captured, which is an observation of the mixed reflection and transmission characteristics of the lower light source from a bottom perspective.

[0077] 102. Homography transform the third original image and the fourth original image to the viewpoint of the upper camera, respectively, to obtain the first transformed image and the second transformed image;

[0078] In this embodiment, a homography transformation matrix pre-calculated using a calibration board is used to perform a geometric transformation on the third and fourth original images captured by the lower camera. Specifically, for each pixel in the third and fourth original images, its new coordinates under the target viewpoint, i.e., the new coordinates under the viewpoint of the upper camera, are calculated according to the homography transformation matrix. Since the calculated new coordinates are usually floating-point numbers, a bilinear interpolation algorithm should also be used to calculate the pixel value at that position, thereby generating two distortion-free first and second transformed images that are perfectly aligned with the viewpoint of the upper camera, thus maintaining the spatial consistency of the four images.

[0079] 103. Extract the shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image, respectively;

[0080] In this embodiment, feature mining is performed on the four images after the viewpoint is unified: the first original image, the second original image, the first transformed image, and the second transformed image. This process extracts shape features, texture features, and grayscale features. Shape features effectively distinguish the macroscopic morphological differences between the circular shape of bubbles and the elongated shape of fibers generated by foreign objects within different layers. Texture features capture the textural details of the foreign object's surface or background, demonstrating strong characterization of the microstructure of the foreign object's surface and insensitivity to changes in illumination. Grayscale features reflect the differences in light reflection or refraction at different layers of foreign objects. The combination of these three types of features comprehensively covers the appearance characteristics of the foreign object.

[0081] 104. Standardize and normalize the shape features, texture features, and grayscale features;

[0082] In this embodiment, to eliminate the numerical scale differences between different types of features and thus avoid these differences interfering with subsequent feature evaluation, fusion, and classification processes, the different types of features should first undergo standardization and normalization. The shape features, texture features, and grayscale features in step 103 all exhibit significant scale differences. If the original features are used directly, features with large numerical scales will have their importance overemphasized in subsequent weight evaluations, causing the model to mistakenly equate "numerical size" with "feature importance," ignoring some crucial small-scale features. Specifically, standardization first converts the features into a distribution with a mean of 0 and a standard deviation of 1, thereby eliminating the magnitude differences in features. Then, normalization maps the features to the [0,1] interval, further avoiding interference from extreme values ​​and ensuring that all features are placed on a fair and comparable numerical platform, thus solving the problem of low classification accuracy caused by chaotic feature scales in existing technologies.

[0083] 105. Based on the preset evaluation model, evaluate and assign weights to the processed shape features, processed texture features, and processed grayscale features respectively;

[0084] In this embodiment, the importance of features from different categories in the final classification task of distinguishing the type and level of foreign objects varies. Therefore, a data-driven approach must be used to adaptively quantify their significance, thereby highlighting key features and suppressing secondary or interfering features during fusion to maximize the discriminative power of the classification model. If only simple feature concatenation is performed, all features will be treated equally, and features with strong discriminative power but smaller numerical ranges compared to other feature dimensions are easily overlooked, while some weakly correlated features may introduce noise. Specifically, a preset evaluation model is used to assess the importance of standardized shape features, texture features, and grayscale features, and weights are then assigned according to importance to dynamically construct an optimal feature combination, ensuring that the final fused feature set input to the classifier contains the richest and most effective discriminative information.

[0085] 106. Based on the assigned weights, fuse the processed shape features, processed texture features, and processed grayscale features to obtain the fusion result;

[0086] In this embodiment, a weighted, asymmetric fusion strategy is used to complementarily integrate the advantages of three heterogeneous features: shape, texture, and grayscale. This generates a robust comprehensive feature representation with discriminative power far exceeding any single feature or simply concatenated feature, thus addressing complex and ever-changing foreign object detection scenarios. While the three types of features extracted in step 103 each have value, using any single feature alone cannot fully cover the requirements for foreign object differentiation. Furthermore, without weighted fusion, simply concatenating the three types of features and inputting them directly into the classifier would dilute the effectiveness of high-value features due to low-value features. Through the weights assigned in step 105, the fusion process allows each type of feature to contribute its actual value. Specifically, it retains the ability of shape and texture features to identify foreign object types while emphasizing the role of grayscale features in determining the category and level of the foreign object, making the fusion result both comprehensive and targeted. In addition, the fused single feature vector can also adapt to the input logic of the classifier, avoiding computational redundancy and judgment confusion caused by independent input of multiple features.

[0087] 107. Input the fusion result into the pre-built classifier and output the classification result.

[0088] In this embodiment, the high-dimensional features fused in step 106 are input into a pre-trained multi-classifier for decision-making. This classifier typically employs a support vector machine model based on a one-to-many strategy and uses a Gaussian radial basis function as the kernel function to handle nonlinear problems. Specifically, the classifier calculates the decision values ​​of the sample to multiple classification hyperplanes, and finally, by comparing these decision values, selects the category with the highest confidence as the output result. This simultaneously provides the type of the foreign object and its hierarchical level, achieving automated and accurate classification. This overcomes the shortcomings of traditional rule-based logical judgment methods, such as poor generalization ability and difficulty in handling complex features. Ultimately, it systematically solves the problems of high false positive and false negative rates in the background technology, outputting stable and reliable quality control conclusions.

[0089] In this embodiment, firstly, four combinations of shooting are performed using upper and lower cameras with upper and lower light sources, respectively. By using differentiated lighting angles, the unique reflection, refraction, and scattering characteristics of foreign objects located at different levels are fully captured. Secondly, homography transformation is used to unify the lower camera's perspective to the upper camera's perspective, eliminating the feature misalignment problem caused by the difference in perspective between the upper and lower cameras, and ensuring that the four sets of images can be accurately compared based on a unified perspective. Next, shape features, texture features, and grayscale features are extracted, and the numerical scale interference of different features is eliminated through standardization and normalization. Then, feature fusion is achieved by combining the weight allocation of a preset evaluation model. Finally, the classification result is output through a pre-built classifier, which can accurately identify the specific category and level of the foreign object, effectively reducing the false detection rate and false negative rate, avoiding production waste caused by misjudging non-influencing foreign objects, and meeting the urgent need for high-precision quality control.

[0090] Please see Figure 2 The standardization and normalization of shape features, texture features, and grayscale features in step 104 may include, but are not limited to, the following:

[0091] 201. Obtain the standardization formula, and calculate the standard scores of shape features, texture features, and grayscale features according to the standardization formula;

[0092] 202. Obtain the normalization formula, and calculate the normalized feature values ​​of shape features, texture features, and grayscale features according to the normalization formula.

[0093] In this embodiment, the standardization formula is: z = (xi - μ) / σ, where z is the standard score, xi is the feature value corresponding to each of the shape feature, texture feature, and grayscale feature, μ is the mean of the feature value corresponding to each of the shape feature, texture feature, and grayscale feature, and σ is the standard deviation of the feature value corresponding to each of the shape feature, texture feature, and grayscale feature. Specifically, for shape feature, texture feature, and grayscale feature, the mean μ and standard deviation σ of all feature values ​​in each category are calculated respectively. μ and σ are calculated based on all samples of the corresponding feature itself (e.g., the mean of shape feature is obtained only from the shape feature values ​​of all samples), ensuring that the standardization result can eliminate the magnitude difference within the same category of features; then, for each original feature value xi in each category of features, the standardization formula is applied to calculate so that the transformed standard score presents a distribution with a mean of 0 and a standard deviation of 1.

[0094] After standardization, normalization is required to further unify the data scale to a fixed interval. The normalization formula is: x'=(z-zmin) / (zmax-zmin), where x' is the normalized feature value, z is the standard score, zmin is the minimum of all standard scores, and zmax is the maximum of all standard scores. Specifically, based on all standard scores obtained from standardization, that is, all standard scores covering shape features, texture features, and grayscale features, the overall minimum value zmin and maximum value zmax are determined. Then, using the formula x'=(z-zmin) / (zmax-zmin), the standard score z of each feature class is mapped to the interval [0,1] to obtain the normalized feature value x'. After these two steps, all features are transformed to the same numerical range, providing a stable and fair data foundation for subsequent feature weighted fusion and classifier decision-making.

[0095] It should be noted that, Figures 1 to 8 In any of the embodiments shown, the shape features, texture features, and grayscale features before and after standardization and normalization, as well as the fusion features represented by the fusion results before and after dimensionality reduction, can each be converted into shape feature vectors, texture feature vectors, and grayscale feature vectors before and after standardization and normalization, and the fusion feature vector represented by the fusion results before and after dimensionality reduction. That is, when performing specific formula and algorithm calculations, the transformed shape feature vectors, texture feature vectors, and grayscale feature vectors before and after standardization and normalization, and the fusion feature vector represented by the fusion results before and after dimensionality reduction, are used in the calculations.

[0096] Please see Figure 3 In step 105, the processed shape features, processed texture features, and processed grayscale features are evaluated and weighted based on a preset evaluation model. In step 106, the processed shape features, processed texture features, and processed grayscale features are fused according to the assigned weights to obtain a fusion result. Specifically, this may include, but is not limited to, the following:

[0097] 301. Evaluate the saliency of the processed shape features, processed texture features, and processed grayscale features based on the XGBoost model;

[0098] 302. Obtain the weight formula, and calculate the weight factors of the processed shape features, the processed texture features, and the processed grayscale features according to the weight formula.

[0099] 303. Obtain the fusion formula and calculate the fusion result based on the fusion formula.

[0100] In this embodiment, a pre-trained XGBoost model is first used to adaptively determine the relative importance of different feature types in the classification task in a data-driven manner. Specifically, the shape features, texture features, and grayscale features, after standardization and normalization, are input into the XGBoost model. The XGBoost model analyzes the information gain or coverage brought by each feature when constructing the decision tree and quantifies the significance of each feature. The significance is in the form of a score, with im1 being the significance of the processed shape feature, im2 being the significance of the processed texture feature, and im3 being the significance of the processed grayscale feature.

[0101] Subsequently, weighting factors are calculated according to the weighting formula: t = im1 + im2 + im3, α = im1 / t, β = im2 / t, γ = im3 / t, where t is the overall saliency, im1 is the saliency of the processed shape feature, im2 is the saliency of the processed texture feature, im3 is the saliency of the processed grayscale feature, α is the weighting factor of the processed shape feature, β is the weighting factor of the processed texture feature, and γ is the weighting factor of the processed grayscale feature. Specifically, the three saliency scores are first added together to obtain the overall saliency t. Then, the saliency of each feature is divided by the sum t to obtain the weighting factors α, β, and γ, which represent the importance of shape, texture, and grayscale features, respectively. This ensures that the sum of the weights of the three types of features is 1, thus achieving weight allocation according to importance.

[0102] After obtaining the weighting factors, a feature weighted concatenation method is used for fusion. The fusion formula is: F=[α×F1,β×F2,γ×F3], where F is the fusion result, α is the weighting factor of the processed shape feature, F1 is the processed shape feature, β is the weighting factor of the processed texture feature, F2 is the processed texture feature, γ is the weighting factor of the processed grayscale feature, and F3 is the processed grayscale feature. Specifically, according to the fusion formula, the processed shape feature F1, texture feature F2, and grayscale feature F3 are multiplied by their corresponding weighting factors and then combined to obtain the fusion result F. This fusion strategy is not a simple concatenation, but rather uses weighting factors to non-linearly scale various features, thereby highlighting the influence of high-salience features in the final feature representation and suppressing the interference of low-salience features, generating an optimized feature set with stronger discriminative power and more suitable for subsequent classifier decisions.

[0103] Please see Figure 4 In this embodiment, the pre-built classifier consists of multiple support vector machine (SVM) classifiers, each corresponding to one category. To improve the accuracy and efficiency of the classification process, the dimensionality of the fusion result should be reduced. Specific steps may include, but are not limited to, the following:

[0104] 401. Based on the PCA principal component analysis algorithm, the space of the fusion result is reduced to a preset dimension to obtain the dimensionality-reduced fusion result;

[0105] 402. Input the dimensionality-reduced fusion results into multiple support vector machine classifiers and output the corresponding decision values.

[0106] 403. Determine the classification result based on multiple decision values. The classification result includes the type of foreign object and the level at which the foreign object is located.

[0107] In this embodiment, firstly, the dimensionality reduction of the fusion result is performed based on the PCA principal component analysis algorithm: firstly, the various features in the fusion result are decentered, that is, each feature is subtracted from the average of its own feature values ​​to obtain the decentered result; then, the covariance matrix of the fused features is calculated: CM=XX T / n, where CM is the covariance matrix, and X is the decentralized result (an n-dimensional square matrix). T Let be the transpose of X, and n be the dimension of X. Then, use Singular Value Decomposition (SVD) to obtain the corresponding eigenvalues ​​and eigenvectors. Next, sort the eigenvalues ​​in descending order, select the k largest eigenvalues ​​(the value of k depends on the actual work requirements), and then use the corresponding k eigenvectors as column vectors to form an eigenvector matrix P. Finally, map the fusion result to the eigenvector matrix P. This achieves the mapping of high-dimensional fusion results to a preset dimension, reducing data redundancy, avoiding classifier overfitting, and retaining the core information for distinguishing different objects, thus obtaining a dimensionality-reduced fusion result.

[0108] The dimensionality-reduced feature vectors are fed in parallel into multiple support vector machine classifiers constructed using a one-to-many strategy. Each classifier is responsible for distinguishing a specific category, such as "OCA layer fibers" or "CG membrane surface contamination." Each classifier outputs a decision value, which reflects the confidence that the input sample belongs to its corresponding category. Finally, the system compares the decision values ​​output by all classifiers and selects the category with the highest confidence as the final classification result, thereby simultaneously determining the specific type of foreign object and its physical level.

[0109] It's worth noting that Support Vector Machine (SVM) classifiers, compared to traditional logical reasoning, can handle data with higher feature dimensions. Furthermore, compared to deep learning network models, they require fewer samples, rely less on GPUs, and are easier to deploy in more real-world scenarios. The goal of an SVM classifier is to find an optimal hyperplane that maximizes the margin between classes. The hyperplane can be represented as: f(x) = w T x+b, where w represents the weight vector and b is the bias term; in addition, a Gaussian kernel function should be selected to suit this scenario, which is: K(xi,xj)=e -γ||xi-xj||²γ is the parameter of the kernel function, used to control the smoothness of the function, and ||xi-xj||² represents the square of the Euclidean distance between vectors xi and xj. The pre-construction steps of the Support Vector Machine (SVM) classifier are as follows: First, collect all n category images, denoted as C1~Cn, such as CG interior, under the membrane, above and below the CG membrane, foreign objects in the OCA interlayer, bubbles, fibers, dirt, scratches, debris, etc.; then, for each category k, train an SVM to distinguish the current category from other categories, ultimately obtaining n SVM classifiers; subsequently, during classification validation, the current data feature vector is used as the input value for each classifier model to obtain the corresponding hyperplane, and then the distance from the current data to the hyperplane is used as the decision value, finally obtaining the calculated decision values ​​f1(x), f2(x),...,f k (x); Finally, the current sample data is sorted according to the decision value to determine which category it belongs to.

[0110] Please see Figure 5 Before performing homography transformation on an image, the basis for the homography transformation should be obtained. Specific steps may include, but are not limited to, the following:

[0111] 501. Based on the pre-set calibration board, perform position calibration on the images captured by the upper camera and the images captured by the lower camera to obtain the calibration results;

[0112] 502. Based on the calibration results, calculate the homography transformation matrix of the image captured by the lower camera and converted to the viewpoint of the upper camera;

[0113] 503. Based on the homography transformation matrix, homography transformation is performed on the third original image and the fourth original image respectively to obtain the first transformed image and the second transformed image.

[0114] In this embodiment, to eliminate the difference in perspective between the upper and lower cameras and provide a unified benchmark for subsequent feature extraction, the perspective of the lower camera image should be converted to that of the upper camera. Specifically, a pre-set calibration board is placed on a platform coplanar with the display panel under test. The upper and lower cameras are controlled to capture images of the calibration board, ensuring that the calibration board completely covers the fields of view of both cameras. By extracting the coordinates of the calibration points on the calibration board, the positional association between corresponding points in the upper and lower camera images is established, ultimately obtaining a calibration result characterizing the positional relationship between the two cameras. Then, based on this calibration result, the homography transformation matrix is ​​further calculated, i.e., obtained by solving the spatial mapping equation of the calibration point coordinate pairs in the calibration result. This matrix describes the mapping relationship between pixels in the lower camera image and pixels in the upper camera image. Finally, the homography transformation matrix is ​​applied to the third and fourth original images captured by the lower camera, respectively, to map and transform the coordinates of each pixel in the image. Pixel values ​​are then filled in by interpolation, ultimately obtaining a first transformed image and a second transformed image that are completely consistent with the perspective of the upper camera, laying a unified perspective foundation for subsequent comparison and extraction of foreign object features among multiple images.

[0115] Please see Figure 6 Before performing shape, texture, and grayscale feature extraction, the area containing the foreign object can be narrowed down to allow for more accurate extraction of the three types of features. Specific steps may include, but are not limited to, the following:

[0116] 601. Based on the point and line detection algorithm, identify the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image, respectively.

[0117] 602. Obtain the shape features corresponding to the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image, respectively. The shape features include area, horizontal length, vertical length, aspect ratio, and roundness.

[0118] 603. Based on the local binary pattern, gray-level co-occurrence matrix, or Gabor filter operator, extract the corresponding texture features from the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image, respectively.

[0119] 604. Obtain the grayscale features corresponding to the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image, respectively. The grayscale features include regional contrast, regional variance, average grayscale, maximum grayscale, minimum grayscale, and uniformity features.

[0120] In this embodiment, a point-line detection algorithm, typically a combination of blob analysis and edge detection, is first used to perform a full-domain scan on four viewpoint-aligned images. By calculating gradient magnitude, intensity threshold, and connected component analysis, all potential candidate foreign object regions are identified and located, thereby effectively separating the target from the background and providing accurate candidate foreign object regions for subsequent feature extraction.

[0121] Based on the candidate foreign object regions extracted in the previous step, the corresponding feature information is calculated. Geometric attributes are calculated through morphological contour analysis to obtain shape features such as the area represented by the total number of connected pixels, the length and width (horizontal and vertical length) of the minimum bounding rectangle, the ratio of the two (aspect ratio), and roundness similar to a circle. Texture features can be extracted using LBP, GLCM, or Gabor filter operators. The texture features selected here are not sensitive to illumination and have good translation and rotation invariance. At the same time, the pixel intensity distribution within the region is statistically analyzed, and its grayscale contrast with the surrounding background, the variance within the region, the arithmetic mean grayscale, the maximum / minimum grayscale values, and the uniformity features reflecting the distribution uniformity are calculated. This constructs a multi-dimensional feature set that comprehensively represents the morphology, structure, and optical properties of the foreign object, providing a differentiated information basis for subsequent fusion and classification.

[0122] The above embodiments illustrate the method for foreign object layer detection provided in this application. The system, electronic device, and storage medium for foreign object layer detection provided in this application are described below:

[0123] Please see Figure 7 This application provides an embodiment of a system for detecting foreign object layering, comprising:

[0124] Image acquisition unit 701 is used to acquire a first original image of the display panel under test through an upper camera and an upper light source, acquire a second original image of the display panel under test through an upper camera and a lower light source, acquire a third original image of the display panel under test through a lower camera and an upper light source, and acquire a fourth original image of the display panel under test through a lower camera and a lower light source.

[0125] The homography transformation unit 702 is used to homography transform the third original image and the fourth original image to the viewpoint of the upper camera, respectively, to obtain the first transformed image and the second transformed image;

[0126] The extraction unit 703 is used to extract the shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image, respectively.

[0127] The standardization and normalization processing unit 704 is used to perform standardization and normalization processing on shape features, texture features and grayscale features;

[0128] The weight allocation unit 705 is used to evaluate and assign weights to the processed shape features, processed texture features, and processed grayscale features based on a preset evaluation model.

[0129] The fusion unit 706 is used to fuse the processed shape features, processed texture features and processed grayscale features according to the assigned weights to obtain the fusion result;

[0130] Classification unit 707 is used to input the fusion result into a pre-built classifier and output the classification result.

[0131] Optionally, the normalization and standardization processing unit 704 is specifically used for:

[0132] Obtain the standardization formula: z=(xi-μ) / σ, where z is the standard score, xi is the feature value corresponding to each of the shape feature, texture feature and grayscale feature, μ is the mean of the feature value corresponding to each of the shape feature, texture feature and grayscale feature, and σ is the standard deviation of the feature value corresponding to each of the shape feature, texture feature and grayscale feature.

[0133] Based on the standardization formula, calculate the standard scores for shape features, texture features, and grayscale features respectively;

[0134] Obtain the normalization formula: x'=(z-zmin) / (zmax-zmin), where x' is the normalization eigenvalue, z is the standard score, zmin is the minimum of all standard scores, and zmax is the maximum of all standard scores;

[0135] Based on the normalization formula, the normalized feature values ​​of shape features, texture features, and grayscale features are calculated respectively.

[0136] Optionally, the weight allocation unit 705 is specifically used for:

[0137] The saliency of the processed shape features, processed texture features, and processed grayscale features were evaluated based on the XGBoost model.

[0138] The weighting formulas are: t = im1 + im2 + im3, α = im1 / t, β = im2 / t, γ = im3 / t, where t is the overall saliency, im1 is the saliency of the processed shape feature, im2 is the saliency of the processed texture feature, im3 is the saliency of the processed grayscale feature, α is the weighting factor of the processed shape feature, β is the weighting factor of the processed texture feature, and γ is the weighting factor of the processed grayscale feature.

[0139] Based on the weighting formula, the weighting factors of the processed shape features, the processed texture features, and the processed grayscale features are calculated respectively.

[0140] Optionally, the fusion unit 706 is specifically used for:

[0141] Obtain the fusion formula: F=[α×F1,β×F2,γ×F3], where F is the fusion result, α is the weight factor of the processed shape feature, F1 is the processed shape feature, β is the weight factor of the processed texture feature, F2 is the processed texture feature, γ is the weight factor of the processed grayscale feature, and F3 is the processed grayscale feature.

[0142] The fusion result is calculated based on the fusion formula.

[0143] Optionally, the pre-built classifier consists of multiple support vector machine classifiers, each corresponding to one class. Prior to classification unit 707, the following is also included:

[0144] Dimensionality reduction unit 708 is used to reduce the space of the fusion result to a preset dimension based on the PCA principal component analysis algorithm, so as to obtain the dimensionality-reduced fusion result;

[0145] Classification unit 707 is specifically used for:

[0146] The fusion results after dimensionality reduction are input into multiple support vector machine classifiers, and the corresponding decision values ​​are output.

[0147] The classification result is determined based on multiple decision values. The classification result includes the type of foreign object and the level at which the foreign object is located.

[0148] Optionally, prior to the homography transformation unit 702, the following may also be included:

[0149] The calibration unit 709 is used to perform position calibration on the images captured by the upper camera and the images captured by the lower camera according to the pre-set calibration plate, and obtain the calibration results;

[0150] The calculation matrix unit 710 is used to calculate the homography transformation matrix of the image captured by the lower camera to the viewpoint of the upper camera based on the calibration results.

[0151] The homography transformation unit 702 is specifically used for:

[0152] Based on the homography transformation matrix, homography transformation is performed on the third original image and the fourth original image respectively to obtain the first transformed image and the second transformed image.

[0153] Optionally, prior to extraction unit 703, the following may also be included:

[0154] The recognition unit 711 is used to identify candidate foreign object regions in the first original image, the second original image, the first transformed image, and the second transformed image, respectively, according to the point and line detection algorithm.

[0155] Optionally, the extraction unit 703 is specifically used for:

[0156] The shape features corresponding to the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image are obtained respectively. The shape features include area, horizontal length, vertical length, aspect ratio, and roundness.

[0157] Based on the local binary pattern, gray-level co-occurrence matrix, or Gabor filter operator, extract the corresponding texture features from the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image, respectively.

[0158] The grayscale features corresponding to the candidate foreign object regions of the first original image, the second original image, the first transformed image, and the second transformed image are obtained respectively. The grayscale features include regional contrast, regional variance, average grayscale, maximum grayscale, minimum grayscale, and uniformity features.

[0159] Please see Figure 8 This application provides an electronic device, including:

[0160] The processor 801, memory 802, input / output unit 803, and bus 804.

[0161] The processor 801 is connected to the memory 802, the input / output unit 803, and the bus 804.

[0162] The memory 802 stores a program, and the processor 801 calls the program to execute it, such as... Figures 1 to 6 The method for detecting foreign matter layering as shown in any embodiment.

[0163] This application provides a computer-readable storage medium on which a program is stored, and when the program is executed on a computer, it performs the following... Figures 1 to 6 The method for detecting foreign matter layering as shown in any embodiment.

[0164] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0165] In the several embodiments provided in this application, it should be understood that the disclosed systems, 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 system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0166] 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.

[0167] Furthermore, the functional units in the various embodiments of this application 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0168] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part 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 application. 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.

Claims

1. A method of foreign object layer separation detection, characterized by, The method comprises the following steps: obtaining a first original image of a display panel to be tested by an upper camera and an upper light source, obtaining a second original image of the display panel to be tested by the upper camera and a lower light source, obtaining a third original image of the display panel to be tested by a lower camera and the upper light source, and obtaining a fourth original image of the display panel to be tested by the lower camera and the lower light source; positionally calibrating the images captured by the upper camera and the images captured by the lower camera according to a pre-set calibration plate to obtain a calibration result; calculating a homography transformation matrix for converting the images captured by the lower camera to the visual angle of the upper camera according to the calibration result; performing homography transformation on the third original image and the fourth original image respectively according to the homography transformation matrix to obtain a first transformed image and a second transformed image; identifying candidate foreign matter regions of the first original image, the second original image, the first transformed image and the second transformed image respectively according to a point-line detection algorithm; obtaining shape features corresponding to the candidate foreign matter regions of the first original image, the second original image, the first transformed image and the second transformed image respectively, wherein the shape features include area, horizontal length, vertical length, horizontal-vertical ratio and circularity; extracting texture features corresponding to the candidate foreign matter regions of the first original image, the second original image, the first transformed image and the second transformed image respectively according to a local binary pattern, a gray level co-occurrence matrix or a Gabor filter operator; obtaining gray features corresponding to the candidate foreign matter regions of the first original image, the second original image, the first transformed image and the second transformed image respectively, wherein the gray features include region contrast, region variance, average gray, maximum gray, minimum gray and uniformity feature; performing standardization processing and normalization processing on the shape features, the texture features and the gray features; performing evaluation and weight distribution on the processed shape features, the processed texture features and the processed gray features respectively based on a pre-set evaluation model; performing fusion on the processed shape features, the processed texture features and the processed gray features according to the distributed weights to obtain a fusion result; inputting the fusion result into a pre-constructed classifier and outputting a classification result.

2. The method of claim 1, wherein, The standardization processing and normalization processing on the shape features, the texture features and the gray features comprise: obtaining a standardization formula: z=(xi-μ) / σ, wherein z is a standard score, xi is a feature value corresponding to the shape features, the texture features and the gray features respectively, μ is a mean value of the feature values corresponding to the shape features, the texture features and the gray features respectively, and σ is a standard deviation of the feature values corresponding to the shape features, the texture features and the gray features respectively. According to the standardization formula, the standard score of the shape feature, the standard score of the texture feature and the standard score of the gray feature are calculated respectively; A normalization formula is obtained: x'=(z-zmin) / (zmax-zmin), wherein the x' is a normalized feature value, the z is a standard score, the zmin is the minimum value of all standard scores, and the zmax is the maximum value of all standard scores; According to the normalization formula, the normalized feature value of the shape feature, the normalized feature value of the texture feature and the normalized feature value of the gray feature are calculated respectively.

3. The method of claim 1, wherein, The preset evaluation model is an XGBoost model constructed in advance, and the evaluation and weight distribution of the processed shape feature, the processed texture feature and the processed gray feature based on the preset evaluation model include: The significance of the processed shape feature, the processed texture feature and the processed gray feature is evaluated based on the XGBoost model respectively; A weight formula is obtained: t=im1+im2+im3, α=im1 / t, β=im2 / t, γ=im3 / t, wherein t is the overall significance, im1 is the significance of the processed shape feature, im2 is the significance of the processed texture feature, im3 is the significance of the processed gray feature, α is the weight factor of the processed shape feature, β is the weight factor of the processed texture feature, and γ is the weight factor of the processed gray feature; According to the weight formula, the weight factor of the processed shape feature, the weight factor of the processed texture feature and the weight factor of the processed gray feature are calculated respectively.

4. The method of claim 3, wherein, The fusion of the processed shape feature, the processed texture feature and the processed gray feature according to the assigned weight to obtain a fusion result includes: A fusion formula is obtained: F=[α×F1, β×F2, γ×F3], wherein F is the fusion result, α is the weight factor of the processed shape feature, F1 is the processed shape feature, β is the weight factor of the processed texture feature, F2 is the processed texture feature, γ is the weight factor of the processed gray feature, and F3 is the processed gray feature; The fusion result is calculated according to the fusion formula.

5. The method according to any one of claims 1-4, characterized in that, The pre-constructed classifier is a plurality of support vector machine classifiers, each support vector machine classifier corresponding to a category, and before the fusion result is input into the pre-constructed classifier and a classification result is output, it further includes: According to the PCA principal component analysis algorithm, the space of the fusion result is reduced to a preset dimension to obtain a reduced dimension fusion result; The input of the fusion result into the pre-constructed classifier and the output of the classification result include: The reduced dimension fusion result is input into the plurality of support vector machine classifiers respectively, and a plurality of decision values corresponding thereto are output; The classification result is determined according to the plurality of decision values, and the classification result includes a foreign object type and a foreign object layer.

6. A system for foreign object layer detection, characterized in that The method for performing the foreign matter layering detection as claimed in any one of claims 1-5, the system comprising: An image acquisition unit configured to acquire a first original image of a display panel to be tested by an upper camera and an upper light source, a second original image of the display panel to be tested by the upper camera and a lower light source, a third original image of the display panel to be tested by a lower camera and the upper light source, and a fourth original image of the display panel to be tested by the lower camera and the lower light source; A homographic transformation unit configured to homographically transform the third original image and the fourth original image to a perspective of the upper camera to obtain a first transformed image and a second transformed image, respectively; An extraction unit configured to extract shape features, texture features, and grayscale features corresponding to the first original image, the second original image, the first transformed image, and the second transformed image, respectively; A standardization and normalization processing unit configured to perform standardization processing and normalization processing on the shape features, the texture features, and the grayscale features; A weight distribution unit configured to evaluate and distribute weights to the processed shape features, the processed texture features, and the processed grayscale features based on a preset evaluation model, respectively; A fusion unit configured to fuse the processed shape features, the processed texture features, and the processed grayscale features according to the distributed weights to obtain a fusion result; A classification unit configured to input the fusion result into a pre-constructed classifier and output a classification result.

7. An electronic device, comprising: comprising: a processor, a memory, an input / output unit, and a bus; the processor is connected with the memory, the input / output unit, and the bus; the memory stores a program, and the processor invokes the program to execute the method for foreign matter layering detection as claimed in any one of claims 1-5.

Images