Egg oiling quality detection method based on machine vision

Abstract

The invention belongs to the technical field of image analysis, and particularly relates to an egg oiling quality detection method based on machine vision, so as to solve the technical problems of low contrast, high light interference and single information dimension of the traditional machine vision in the prior art, and the method comprises the following steps: S1, obtaining a multispectral polarization image of an egg to be detected; s2, obtaining a contour mask and a highlight area mask of the egg; s3, obtaining an effective analysis area; s4, generating a linear polarization degree characteristic pattern representing the uniformity of the oil film; s5, generating a spectral angle characteristic pattern representing the oiling thickness; s6, constructing a region graph which takes each superpixel as a node and the adjacency relation as an edge; s7, fusing the initial feature vectors of the superpixel nodes; s8, identifying and outputting the super-pixel nodes which are judged as oiling defects; and S9, comprehensively evaluating the oiling quality grade of the egg. And automatic and high-precision comprehensive grading of egg oiling quality is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of image analysis, and in particular relates to a method for detecting egg oil coating quality based on machine vision. Background Art

[0002] Eggs, as an important agricultural product, have a shelf life significantly affected by their storage environment and their quality. To extend the shelf life of eggs and reduce moisture evaporation and microbial intrusion, a preservation technique commonly used in industrial production involves coating the surface of eggs with edible-grade mineral oil. This oiling treatment effectively seals pores on the eggshell surface, preserving the egg's internal quality. However, the effectiveness of the oiling process directly depends on the quality of the oil film, specifically its uniformity and thickness. Defects such as missing coating, excessive thickness, or uneven distribution (e.g., streaks or patches) will not achieve the desired preservation effect and may even affect the product's appearance. Currently, testing the quality of egg oiling primarily relies on manual visual inspection. This method is not only inefficient and labor-intensive, but also susceptible to subjective factors, lacking consistency and objectivity, and unable to meet the high-speed testing requirements of modern, large-scale production lines. Therefore, the development of an automated, highly accurate, and efficient oiling quality testing technology has become an urgent need within the industry.

[0003] To achieve automated inspection, several machine vision-based solutions have emerged. These solutions typically use conventional industrial cameras to capture egg images and then analyze oil film defects using image processing algorithms. However, these traditional methods face numerous challenges in practical application. Because the oil film itself is transparent and extremely thin, its contrast in ordinary visible light images is very low, resulting in faint defect features and difficulty distinguishing them from the eggshell background. Furthermore, the smooth and curved surface of an egg easily produces highlights (specular reflections) when illuminated by external light sources. These highlights can severely interfere with the effective extraction of information about the underlying oil film and may even be misidentified as defects, significantly reducing the accuracy and robustness of inspection. Furthermore, single imaging modalities (such as RGB imaging) struggle to simultaneously capture and distinguish information characterizing the physical uniformity of the oil film (such as coating morphology) and chemical information characterizing the amount of oil applied (such as coating thickness). This results in a single-dimensional inspection and an inability to provide a comprehensive and detailed assessment of oil coating quality. Therefore, effectively suppressing the interference from highlights and integrating multi-dimensional information to achieve simultaneous and accurate assessment of oil film uniformity and thickness remain key challenges in current egg coating quality inspection. Summary of the Invention

[0004] The present invention provides an egg oiling quality detection method based on machine vision to solve the technical problems faced by traditional machine vision in the prior art, such as low contrast, high light interference and single information dimension.

[0005] To solve the above problems, the present invention provides a method for detecting the quality of egg oil coating based on machine vision, which adopts the following technical solution: The method for detecting the quality of egg oil coating based on machine vision comprises the following steps: S1, acquiring a multispectral polarization image of the egg to be tested by imaging at multiple spectral bands and at least three different polarization angles; S2 uses a multi-task convolutional neural network to simultaneously segment the multispectral polarization image to obtain the egg contour mask and highlight area mask; S3, removing pixels corresponding to the highlight area mask within the area defined by the contour mask to obtain a valid analysis area; S4, calculating the linear polarization degree of each pixel point in the effective analysis area based on the image data at different polarization angles, and generating a linear polarization degree characteristic map representing the uniformity of the oil film; S5, calculating the spectral angle based on the multispectral data of each pixel point in the effective analysis area and a pre-built standard oil spectral library, and generating a spectral angle characteristic map representing the oil thickness; S6, performing superpixel segmentation on the effective analysis area and constructing a regional graph with each superpixel as a node and the adjacent relationship as an edge; S7: For each superpixel node, extract the statistical moments of all pixels in the linear polarization feature map and the mean and variance of all pixels in the spectral angle feature map, and fuse them into the initial feature vector of the superpixel node; S8, inputting the region map with the initial feature vector into the pre-trained graph neural network model, identifying and outputting superpixel nodes determined to be oil defects; S9, comprehensively evaluate the oiling quality grade of eggs based on the location, number and total area of ​​superpixel nodes with oiling defects.

[0006] Furthermore, when obtaining a multispectral polarization image, the following steps are included: A multispectral imaging system is used to perform imaging using multiple bands covering the visible to near-infrared spectrum. In each wavelength band, at least three images with different polarization angles were collected by rotating the linear polarizer in front of the camera.

[0007] Furthermore, when collecting images at three different polarization angles, the three polarization angles are 0°, 45°, and 90°, respectively.

[0008] Furthermore, when calculating the linear polarization degree of each pixel point in the effective analysis area based on the image data at different polarization angles, the following steps are included: using the image intensity collected at three polarization angles of 0°, 45°, and 90° 、 、 , calculate the Stokes component for each pixel according to the following formula 、 、 Then, the linear polarization degree of each pixel is calculated: ; ; ; .

[0009] Furthermore, in S5, when preparing the standard sample, standard oil films of different known thicknesses are coated on the eggshell, and their reflectivity in different bands is measured to form a standard oil spectral library, wherein the standard oil spectral library includes standard spectral vectors corresponding to different thicknesses.

[0010] Furthermore, for each effective pixel in the effective analysis area, its grayscale value in different bands is extracted to form the spectral vector of the pixel; The spectral angle of the pixel is obtained by calculating the inverse cosine function value of the angle between the spectral vector of the pixel and the standard spectral vector.

[0011] Furthermore, superpixel segmentation is performed on the effective analysis area, and the construction of the area map includes the following steps: A simple linear iterative clustering algorithm or its variants is used to divide the effective analysis area into multiple super-pixel regions of uniform size; By traversing all superpixels, if two superpixels have a shared boundary, an edge is added between the nodes representing them to build a region graph.

[0012] Furthermore, the shared boundary includes a contact or overlapped portion of boundaries of two superpixels.

[0013] Furthermore, when extracting the initial feature vector for each superpixel node, the following steps are included: Calculate the mean, variance, skewness, and kurtosis of all pixels in the superpixel node on the linear polarization feature map as the linear polarization statistical moment feature; Calculate the mean and variance of all pixels in the superpixel node on the spectral angle feature map as the spectral angle statistical features; The linear polarization degree statistical moment feature and the spectral angle statistical feature are concatenated to form the initial feature vector of the superpixel node.

[0014] Furthermore, based on the location, number, and total area of ​​superpixel nodes with oil coating defects, a comprehensive assessment of the oil coating quality of the egg includes the following steps: Preset defect quantity threshold and defect area percentage threshold; The oiling quality is rated as excellent, qualified, or unqualified based on whether the total number of superpixel nodes with oiling defects exceeds the defect number threshold and whether the percentage of the total defect area to the total egg area exceeds the defect area percentage threshold.

[0015] The beneficial effect is that the present invention obtains multispectral polarization images of eggs and uses a multi-task convolutional neural network to simultaneously segment highlight areas, thereby overcoming the problem of difficulty in extracting defect features caused by highlight interference and oil film transparency in traditional visual inspection, and improving the robustness of detection. At the same time, the present invention combines two physical mechanisms, namely, the linear polarization degree calculated by polarization information is used to characterize the physical uniformity of the oil film, and the spectral angle calculated by multispectral information is used to characterize the thickness of the oil coating, thereby achieving a simultaneous quantitative evaluation of the two key dimensions of the oil coating quality, making the detection results more comprehensive and accurate. By introducing superpixel segmentation and graph neural network technology, it is possible to fuse local features and spatial context information between regions for comprehensive judgment, thereby improving the recognition accuracy of complex morphological defects such as stripes and plaques, and achieving automated, high-precision comprehensive grading of egg oil coating quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The flowchart of the egg oiling quality inspection method based on machine vision. DETAILED DESCRIPTION

[0017] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Those skilled in the art should know that the embodiments described below are part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0018] Embodiments of the egg oiling quality detection method based on machine vision provided by the present invention: like Figure 1 As shown, the egg oiling quality detection method based on machine vision includes the following steps: S1, obtaining a multispectral polarization image of the egg to be tested by imaging in multiple spectral bands and at least three different polarization angles.

[0019] In an optional embodiment, obtaining a multispectral polarization image includes the following steps: A multispectral imaging system is used to perform imaging using multiple bands covering the visible to near-infrared spectrum. In each wavelength band, at least three images with different polarization angles were collected by rotating the linear polarizer in front of the camera.

[0020] Specifically, the multispectral imaging system can capture image information of eggs at multiple preset narrowband wavelengths. For example, three filters with central wavelengths of 530nm, 660nm, and 850nm can be selected. These wavelengths correspond to the optical response characteristics of different substances on the egg surface and inside, respectively, and can capture subtle differences before and after oiling. The visible light band is helpful for observing surface color and texture, while the near-infrared band is more sensitive to the absorption and scattering properties of the oil film, thereby enhancing detection contrast.

[0021] In each selected wavelength band, such as the 660nm band, the linear polarizer is controlled to rotate to three or more preset angles in sequence for photography. Typical angle settings are 0°, 45°, and 90°. The first image is captured when the linear polarizer is at 0°, the second image is captured when it is rotated to 45°, and the third image is captured when it is rotated to 90°. In this way, not only is the intensity information of the egg obtained under different spectra, but also the polarization state information after the light wave interacts with the egg surface, providing the necessary data foundation for the subsequent accurate calculation of the polarization degree.

[0022] S2 uses a multi-task convolutional neural network to simultaneously segment the multispectral polarization image to obtain the egg contour mask and highlight area mask.

[0023] In an optional embodiment, the nine acquired images are stacked into a nine-channel input tensor. A multi-task learning model based on the U-Net architecture is employed. The multi-task learning model shares a single encoder backbone network but has two independent decoder branches. The first decoder branch outputs a binary outline mask of the egg, and the second decoder branch outputs a binary mask of the highlight region. By performing end-to-end training on a dataset with manually labeled outlines and highlight regions, the multi-task learning model is able to learn the mapping relationship from multi-dimensional input to the two segmentation tasks.

[0024] S3, removing pixels corresponding to the highlight area mask within the area defined by the contour mask to obtain a valid analysis area.

[0025] S4, calculating the linear polarization degree of each pixel point in the effective analysis area based on the image data at different polarization angles, and generating a linear polarization degree characteristic map representing the uniformity of the oil film.

[0026] In an optional embodiment, when three different polarization angles are selected as 0°, 45° and 90°, the degree of linear polarization (DoLP) of each pixel point in the effective analysis area is calculated based on the image data at different polarization angles, including the following steps: using the image intensity collected at the three polarization angles of 0°, 45° and 90° 、 、 , calculate the Stokes component for each pixel according to the following formula 、 、 Then, the linear polarization degree of each pixel is calculated: ; ; ; .

[0027] Specifically, the light intensity value measured at each pixel at three polarization angles is converted into a physical quantity that describes its polarization state. represents the total light intensity, reflects the difference between the horizontal and vertical polarization components, and It reflects the difference of ±45° diagonal polarization components. For example, for a certain pixel point on the surface of an egg, the image intensity values ​​collected in a specific band are =180, =150, =100. Then according to the formula, is 280, is 80, .

[0028] After obtaining the Stokes component, the degree of linear polarization (DoLP) is further calculated. The degree of linear polarization is a normalized value between 0 and 1, which quantifies the proportion of linear polarization components in the light wave. Continuing with the previous example, the degree of linear polarization of the pixel is A high DoLP value in an area indicates that the light reflected from the surface in that area has strong polarization characteristics, which is usually related to uniform oil film coverage; conversely, a low DoLP value means that the surface is rough or has no oil film coverage, thus achieving quantitative distinction between oil-defective areas.

[0029] S5, calculating the spectral angle based on the multispectral data of each pixel point in the effective analysis area and a pre-built standard oil spectral library, and generating a spectral angle characteristic diagram representing the oil thickness.

[0030] Specifically, a standard sample was prepared by coating eggshells with standard oil films of varying known thicknesses. The reflectance at 750nm, 850nm, and 940nm was measured to construct a standard oil spectral library (containing standard spectral vectors corresponding to different thicknesses). For each valid pixel in the effective analysis area, the grayscale values ​​at the three bands were extracted to form the pixel's spectral vector. The pixel's spectral angle was determined by calculating the inverse cosine function of the angle between the pixel's spectral vector and the standard spectral vector. This calculation was performed for all pixels within the effective analysis area, generating a single-channel spectral angle map (SAM). The pixel values ​​in the map reflect the difference between the oil thickness at that point and the standard thickness. Smaller pixel values ​​indicate a closer match to the standard thickness at that point; larger pixel values ​​indicate a more significant difference.

[0031] S6, perform superpixel segmentation on the effective analysis area and construct a regional graph with each superpixel as a node and the adjacent relationship as an edge.

[0032] In an optional embodiment, performing superpixel segmentation on the effective analysis area and constructing the area map includes the following steps: The simple linear iterative clustering (SLIC) algorithm or its variants are used to divide the effective analysis area into multiple super-pixel regions of uniform size; By traversing all superpixels, if two superpixels have a shared boundary, an edge is added between the nodes representing them to build a region graph.

[0033] Specifically, this method performs superpixel segmentation only on the valid analysis area to avoid background interference and improve computational efficiency. For an effective analysis area with approximately 200,000 pixels, for example, the SLIC algorithm can target 500 superpixels. The SLIC algorithm iteratively clusters the 200,000 pixels into 500 compact regions of roughly 400 pixels each, each with irregular shapes, taking into account both color similarity and spatial distance between pixels. These regions are known as superpixels.

[0034] Superpixels can fit the local texture and color boundaries of the egg surface well, serving as the basic unit for subsequent analysis.

[0035] After segmentation, the 500 superpixels are abstracted into 500 nodes in a region graph. Each pair of superpixels, such as superpixel 123 and superpixel 124, is examined. If any portion of their boundaries touch or overlap, a connecting edge is established between the nodes representing them. By traversing all superpixel pairs, an undirected graph is ultimately formed. This region graph not only contains the internal information of each superpixel but also encodes the adjacency relationships between them, laying the foundation for contextual analysis using advanced models such as graph neural networks.

[0036] S7, for each superpixel node, extract the statistical moments of all its internal pixels on the linear polarization feature map and the mean and variance on the spectral angle feature map, and fuse them into the initial feature vector of the superpixel node.

[0037] In an optional embodiment, when extracting the initial feature vector for each superpixel node, the following steps are included: Calculate the mean, variance, skewness, and kurtosis of all pixels in the superpixel node on the DoLP feature map as the DoLP statistical moment features; Calculate the mean and variance of all pixels in the superpixel node on the SAM feature map as SAM statistical features; The DoLP statistical moment features are concatenated with the SAM statistical features to form the initial feature vector of the superpixel node.

[0038] Specifically, this step assigns quantitative descriptive information to each superpixel node in the region map. For a superpixel node representing a uniformly oiled area, the DoLP values ​​of its internal pixels may be highly concentrated around 0.8, so the calculated DoLP mean will be close to 0.8, and the variance will be very small, such as 0.01. The two higher-order statistics, skewness and kurtosis, can describe the shape of the data distribution in more detail and reflect the smoothness of the oil film. At the same time, the SAM mean of the superpixel node may be 0.1 and the variance is 0.005, indicating that its spectral characteristics are very close to those of the standard oil film and the internal material is uniform.

[0039] Connecting the statistics together constitutes the initial features of the superpixel node. Taking the above data as an example, the initial feature vector of the superpixel node can be expressed as a six-dimensional vector, with specific values ​​of (0.8, 0.01, corresponding skewness and kurtosis values, 0.1, 0.005). In contrast, a superpixel containing a missed area may have its DoLP mean reduced to 0.3, its variance increased to 0.2, and its SAM mean increased accordingly. Its initial feature vector will be significantly different from the aforementioned uniform area node. By generating a unique initial feature vector for each superpixel node, the local physical characteristics of the image are converted into digital signals that can be processed by the machine learning model.

[0040] S8, inputs the region map with the initial feature vector into the pre-trained graph neural network model, identifies and outputs the superpixel nodes determined to be oil defects.

[0041] In an optional embodiment, a graph convolutional network (GCN) model, pre-trained on a large dataset of defect-labeled egg images, is used. The previously constructed region graph, with each superpixel node bearing a six-dimensional initial feature vector, is used as input to the GCN model. Through multiple layers of graph convolution within the model, the features of each superpixel node aggregate the features of its neighboring nodes, thereby learning the spatial contextual patterns of defect regions, such as streaks or patches. The final layer of the model outputs a binary classification result for each node, determining whether the superpixel node belongs to an oil-defect region.

[0042] S9, comprehensively evaluate the oiling quality grade of eggs based on the location, number and total area of ​​superpixel nodes with oiling defects.

[0043] In an optional embodiment, comprehensively evaluating the egg oil coating quality grade based on the location, number, and total area of ​​superpixel nodes with oil coating defects includes the following steps: Preset defect quantity threshold and defect area percentage threshold; The oiling quality is rated as excellent, qualified, or unqualified based on whether the total number of superpixel nodes with oiling defects exceeds the defect number threshold and whether the percentage of the total defect area to the total egg area exceeds the defect area percentage threshold.

[0044] Specifically, the assessment criteria achieve automated and standardized quality grading by setting clear digital thresholds. For example, two sets of thresholds can be pre-defined: the first set is the premium threshold, which means the number of defects must not exceed 3 and the total area of ​​the defects must not exceed 1% of the total area of ​​the egg; the second set is the acceptable threshold, which means the number of defects must not exceed 10 and the total area of ​​the defects must not exceed 5% of the total area of ​​the egg. The thresholds can be flexibly adjusted to meet production process requirements.

[0045] After inspection is complete, the total number of superpixel nodes identified as oil defects and their total pixel area are counted. For example, let's assume an egg has a total area of ​​250,000 pixels and two defective superpixel nodes are detected, totaling 2,000 pixels. At this point, the number of defects (2) is less than the high-quality threshold of 3, and the defect area percentage is 0.8%, which is also less than the high-quality threshold of 1%. Therefore, the egg is rated as high-quality.

[0046] If another egg is detected with 8 defective superpixel nodes with a total area of ​​10,000 pixels, and the number of defects and area percentage do not exceed the bottom line threshold of qualified products, but exceed the high-quality product threshold, the egg will be rated as qualified.

[0047] If any indicator of the number or area of ​​defects exceeds the bottom line threshold of qualified products, for example, the total area of ​​defects reaches 15,000 pixels, which is 6% of the total area, the eggs will be directly rated as unqualified.

[0048] In addition, in the description of this specification, “a plurality of” means at least two, for example, two, three or more, etc., unless otherwise clearly and specifically defined.

Claims

1. A method for detecting egg oil quality based on machine vision, characterized in that: The following steps are involved: S1, acquiring a multispectral polarization image of the egg to be tested by imaging at multiple spectral bands and at least three different polarization angles; S2 uses a multi-task convolutional neural network to simultaneously segment the multispectral polarization image to obtain the egg contour mask and highlight area mask; S3, removing pixels corresponding to the highlight area mask within the area defined by the contour mask to obtain a valid analysis area; S4, calculating the linear polarization degree of each pixel point in the effective analysis area based on the image data at different polarization angles, and generating a linear polarization degree characteristic map representing the uniformity of the oil film; S5, calculating the spectral angle based on the multispectral data of each pixel point in the effective analysis area and a pre-built standard oil spectral library, and generating a spectral angle characteristic map representing the oil thickness; S6, performing superpixel segmentation on the effective analysis area and constructing a regional graph with each superpixel as a node and the adjacent relationship as an edge; S7: For each superpixel node, extract the statistical moments of all pixels in the linear polarization feature map and the mean and variance of all pixels in the spectral angle feature map, and fuse them into the initial feature vector of the superpixel node; S8, inputting the region map with the initial feature vector into the pre-trained graph neural network model, identifying and outputting superpixel nodes determined to be oil defects; S9, comprehensively evaluate the oiling quality grade of eggs based on the location, number and total area of ​​superpixel nodes with oiling defects.

2. The egg oiling quality detection method based on machine vision according to claim 1, characterized in that: Acquiring a multispectral polarization image involves the following steps: A multispectral imaging system is used to perform imaging using multiple bands covering the visible to near-infrared spectrum. In each wavelength band, at least three images with different polarization angles were collected by rotating the linear polarizer in front of the camera.

3. The egg oiling quality detection method based on machine vision according to claim 2, characterized in that: When images with three different polarization angles are collected, the three polarization angles are 0°, 45°, and 90°, respectively.

4. The method for detecting egg oil coating quality based on machine vision according to claim 3, wherein: The calculation of the linear polarization degree of each pixel in the effective analysis area based on the image data at different polarization angles includes the following steps: using the image intensity collected at three polarization angles of 0°, 45°, and 90° 、 、 , calculate the Stokes component for each pixel according to the following formula 、 、 Then, the linear polarization degree of each pixel is calculated: ； ； ； 。 5. The method for detecting egg oil quality based on machine vision according to claim 1, wherein: In S5, when preparing the standard sample, standard oil films of different known thicknesses are coated on the eggshell, and their reflectivity in different bands is measured to form a standard oil spectral library, wherein the standard oil spectral library includes standard spectral vectors corresponding to different thicknesses.

6. The method for detecting egg oil quality based on machine vision according to claim 5, characterized in that: For each valid pixel in the effective analysis area, its grayscale value in different bands is extracted to form the pixel's spectral vector; The spectral angle of the pixel is obtained by calculating the inverse cosine function value of the angle between the spectral vector of the pixel and the standard spectral vector.

7. The method for detecting egg oil quality based on machine vision according to claim 1, wherein: Performing superpixel segmentation on the effective analysis area and constructing the area map includes the following steps: A simple linear iterative clustering algorithm or its variants is used to divide the effective analysis area into multiple super-pixel regions of uniform size; By traversing all superpixels, if two superpixels have a shared boundary, an edge is added between the nodes representing them to build a region graph.

8. The method for detecting egg oil coating quality based on machine vision according to claim 7, wherein: The shared boundary includes a contact or overlapped portion of the boundaries of two superpixels.

9. The method for detecting egg oil quality based on machine vision according to claim 1, wherein: When extracting the initial feature vector for each superpixel node, the following steps are included: Calculate the mean, variance, skewness, and kurtosis of all pixels in the superpixel node on the linear polarization feature map as the linear polarization statistical moment feature; Calculate the mean and variance of all pixels in the superpixel node on the spectral angle feature map as the spectral angle statistical features; The linear polarization degree statistical moment feature and the spectral angle statistical feature are concatenated to form the initial feature vector of the superpixel node.

10. The method for detecting egg oil coating quality based on machine vision according to any one of claims 1 to 9, characterized in that: Based on the location, number, and total area of ​​superpixel nodes with oiling defects, the comprehensive assessment of the oiling quality of eggs includes the following steps: Preset defect quantity threshold and defect area percentage threshold; The oiling quality is rated as excellent, qualified, or unqualified based on whether the total number of superpixel nodes with oiling defects exceeds the defect number threshold and whether the percentage of the total defect area to the total egg area exceeds the defect area percentage threshold.

Images