Method for automatically identifying and judging oyster powder coating effect

Abstract

The invention provides a method for automatically identifying and judging an oyster powder coating effect, and relates to the technical field of aquatic product processing, and the method comprises the following steps: freezing, shaping and screening oyster meat before coating processing; the method comprises the following steps: scanning the upper surface layer of oyster meat through laser to obtain point cloud data of the upper surface of unprocessed oysters, and generating an ideal powder coating layer data model; performing laser scanning on the surface of the powder-coated oyster again by adopting the same equipment parameters and scanning mode as the original oyster scanning, and obtaining three-dimensional point cloud data of an actual powder-coated layer to obtain an actual powder-coated layer data model; and performing integrated calculation and comparative analysis on the actual powder coating layer and the ideal powder coating layer to judge whether the quality of the powder-coated product is qualified or not.

Description

Technical Field

[0001] The present invention relates to the technical field of aquatic product processing, in particular to a method for automatically identifying and judging the effect of oyster powder coating. Background Art

[0002] Breaded oysters, also known as breaded oysters, are a semi-finished food product made from oysters. A uniform layer of breadcrumbs coats the outer surface of the oyster meat, trapping its juices within the coating and retaining them during frying, ensuring the oysters' freshness and tenderness. Breaded oysters are prepared by first freezing the shelled oysters individually, thawing them to a semi-thawed state with a surface temperature no higher than 10°C, then coating each oyster with an even layer of batter. The oysters are then immediately coated with a thick layer of breadcrumbs, packaged, and frozen again for storage. In actual production, the coating process of breaded oysters is completed by mechanized equipment. The oyster meat first passes through a prepared batter pool, is immersed in the batter, and is directly sent to a conveyor belt covered with breadcrumbs. During the conveying process, the feeding mechanism above simultaneously feeds a certain amount of breadcrumbs to completely cover the oyster meat. After being sent out with the conveyor belt, the oyster meat enters the subsequent mesh chain conveyor belt to remove the excess breadcrumbs that are not fixed on the surface, completing the mechanized coating process.

[0003] Because the surface of oyster meat is not completely flat, but rather exhibits a unique, irregular, and undulating surface for each individual oyster, mechanized battering and breading processes can quantitatively apply batter and breading materials by controlling process parameters. However, this cannot guarantee the same total amount of batter remaining on the oyster surface and the amount of breading subsequently applied. This results in inconsistent product quality and uncertainty regarding the breading effect of the finished product, prone to quality issues such as incomplete coating, uneven coating thickness, and surface unevenness. Furthermore, since this product requires high-temperature frying before consumption, the quality of the breading directly impacts the product's taste. If the breading is incomplete, loose, or too thin, the oyster meat will release significant amounts of juice as the temperature rises during frying. This excess water, when in contact with the hot oil, can cause splashing and injury, posing a safety hazard. Furthermore, the loss of oyster flavor along with the water also significantly reduces the freshness of the breaded oyster. If the coating is too thick, the juice will not leak out, but the oyster meat inside will not be cooked thoroughly. Prolonging the frying time will cause the coating to burn.

[0004] At present, the domestic standards for breaded oyster processing do not define a specific value for the coating thickness. In actual production, quality inspectors use visual inspection to inspect the finished products after coating with the naked eye. This method can only identify obvious uncovered areas, and cannot quantify parameters such as completeness and thickness. In addition, the oyster meat is irregular and has a concave and convex surface with visual blind spots, making it impossible to effectively identify the thickness. Summary of the Invention

[0005] In order to solve the defects of the prior art, the purpose of the present invention is to provide a method for automatically identifying and judging the coating effect of oysters based on point cloud data.

[0006] In order to solve the above problems, the technical solution of the present invention is:

[0007] A method for automatically identifying and judging the effect of oyster powder coating comprises the following steps:

[0008] Freeze-shape and sort the oyster meat before coating;

[0009] Laser scanning of the upper surface of oyster meat is used to obtain point cloud data of the unprocessed upper surface of the oyster and generate an ideal breading layer data model.

[0010] Using the same equipment parameters and scanning methods as those used to scan the original oysters, the surface of the powdered oysters was laser scanned again to obtain the 3D point cloud data of the actual powder coating layer and obtain the actual powder coating layer data model;

[0011] By integrating the calculation and comparative analysis of the actual powder coating layer and the ideal powder coating layer, the quality of the powder coated product can be judged.

[0012] Preferably, the step of freezing, shaping and screening the oyster meat before coating processing specifically includes: placing individual oyster meats one by one into a shaping mold and then freezing them, the shaping mold adopts a waist-round shape that imitates the appearance of oyster meat, with a horizontal bottom surface and four walls perpendicular to the bottom surface, and the soft oyster meat is placed in to fully fill the interior of the mold and freeze it for shaping; after freezing and demolding, the bottom surface and the four sides of the oyster meat are flat, and only the upper surface is an irregular curved surface.

[0013] Preferably, the step of obtaining point cloud data of the unprocessed upper surface of the oyster meat by laser scanning and generating an ideal coating layer data model specifically includes: obtaining point cloud data of the unprocessed upper surface of the oyster, fixing the unprocessed oyster and scanning it with a laser scanner, if the scanned point cloud has many holes, performing multiple scans, and aligning the multiple point clouds obtained by using an improved ICP algorithm to obtain the original unprocessed oyster surface point cloud layer P0.

[0014] Preferably, the improved ICP algorithm specifically includes:

[0015] Initial alignment: Calculate the center of gravity of the source point cloud using the formula: Calculate the center of gravity of the target point cloud using the formula: Move the source point cloud to coincide with the center of gravity of the target point cloud;

[0016] Calculate the eigenvalues ​​of each point in the source point cloud, including the normal vector, curvature, surface change rate, and fast point feature histogram, as the input of the neural network, and the x, y, and z coordinates of the matching point corresponding to the target point cloud as the output to build the neural network;

[0017] The feature values ​​extracted from the newly scanned source point cloud are input into the trained neural network model to obtain the coordinate values ​​of the target point cloud. The rotation matrix R and translation vector t are calculated based on the corresponding point pairs. Let and For the corresponding point set, by calculating the covariance matrix in

[0018] Perform singular value decomposition on C and get C=U∑V T , rotation matrix R = VU T , the translation vector

[0019] According to the calculated R and t, the point cloud is transformed.

[0020] Preferably, the step of laser scanning the surface of the powdered oyster again using the same equipment parameters and scanning method as those used to scan the original oysters, obtaining three-dimensional point cloud data of the actual powder coating layer, and obtaining a data model of the actual powder coating layer specifically includes: after completing the coating process, placing the powdered oyster again on the scanning workbench, using the same equipment parameters and scanning method as those used to scan the original oysters, laser scanning the surface of the powdered oyster again, obtaining three-dimensional point cloud data of the actual powder coating layer, and recording it as the P1 surface.

[0021] Preferably, the step of evaluating whether the quality of the powder coated product is qualified by integrating, calculating and comparing the actual powder coating layer with the ideal powder coating layer specifically includes:

[0022] Generate the oyster meat fitting surface I0 based on P0, move I0 vertically upward to obtain the ideal upper surface I1 of the breading, and move I0 vertically downward to obtain the ideal lower surface I2 of the breading;

[0023] Traverse all points on the P1 surface. If any point is below the ideal upper surface I1 of the powder coating, it means that the powder coating thickness of the oyster is not enough to ensure a complete shell, and the oyster is judged as unqualified.

[0024] Traverse all points in the P1 surface and record the number of points above the ideal powder coating lower surface I2, and define:

[0025]

[0026] Compare a and b, where b is the judgment value generated by the system based on the outer surface of the oyster. If a>b, the oyster has too much thick coating, which will make the inner layer of the oyster difficult to cook during frying, and is therefore judged as unqualified. If a≤b, the oyster is judged to have an appropriate coating thickness, and is therefore qualified.

[0027] The value of b is related to the mean square error error when P0 generates the oyster meat fitting surface I0. The x-axis is the error value and the y-axis is b. M in this method is the over-thickness coefficient. The default value of M is 0.5. Before production, the M threshold can also be set according to the process requirements for the thickness of the powder coating or the qualified rate judgment index.

[0028]

[0029] Compared to existing technologies, this method uses machine vision technology to freeze-shape and screen the oysters before coating, then performs laser scanning to obtain point cloud data of the sample's upper surface contour, generating a reference layer that meets process requirements. After coating, the oyster's upper surface is laser scanned again to obtain point cloud data of the surface contour, creating a model of the actual coating layer. By integrating, calculating, and comparing the actual coating layer with the reference layer, quality indicators such as coating integrity and appropriate thickness are determined, ultimately assessing the quality of the coated product. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0031] Figure 1 This is a flowchart of the method for automatically identifying and judging the effect of oyster flour coating according to the present invention;

[0032] Figure 2 is a neural network model diagram;

[0033] Figure 3 Schematic diagram of the value of b. DETAILED DESCRIPTION

[0034] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0035] Specifically, the present invention provides a method for automatically identifying and judging the effect of oyster powder coating, such as Figure 1 As shown, the method includes the following steps:

[0036] S1: Freeze-forming and screening of oyster meat before coating;

[0037] Specifically, individual oysters are placed individually into a mold and then frozen. The mold is designed to mimic the shape of an oyster, with a horizontal bottom and vertical walls. The soft-bodied oysters are placed until the mold is fully filled and frozen to set their shape. After being removed from the mold, the bottom and sides of the oysters are flat, with only the top surface showing irregular curves. Partially thawed oysters are screened to remove any visible defects or distortions caused by thermal expansion and contraction.

[0038] S2: Laser scanning the upper surface of the oyster meat is used to obtain point cloud data of the unprocessed oyster upper surface and generate an ideal breading layer data model;

[0039] Specifically, the point cloud data of the upper surface of the unprocessed oyster is obtained. The unprocessed oyster is fixed and scanned using a line laser scanner. If the scanned point cloud has many holes, multiple scans are performed. The multiple point clouds obtained are aligned using the improved ICP algorithm to obtain the original unprocessed oyster surface point cloud layer, which is defined as P0.

[0040] The oyster meat fitting surface I0 is generated based on P0. The oyster meat fitting surface is obtained by fitting the original oyster layer. It describes the point cloud layer of the oyster surface after being coated with powder under ideal conditions. The specific fitting method is as follows:

[0041] Set the quadratic polynomial z = a0x 2 +a1y 2 +a2xy+a3x+a4y+a5, by minimizing the sum of squares of errors from the point P0 to the surface, the 6 coefficients of the polynomial are determined to obtain I0.

[0042] Shifting I0 vertically by μ1 from the highest point of P0 yields surface I1. Shifting I0 vertically by μ2 from the lowest point of P0 yields surface I2. With μ1 as the minimum constant thickness, layer I1 represents the lowest coating thickness. If the coating thickness is below I1, the shell will crack during frying. With μ2 as the maximum constant thickness, layer I2 represents the highest coating thickness. If the coating thickness is above I2, the shell will be too thick, potentially undercooking the oysters. Therefore, in theory, a coating thickness between surfaces I1 and I2 meets process requirements.

[0043] v1 and μ2 are process coefficients set manually based on the thickness of the breading. Based on process requirements, when using breadcrumbs for coating, the default values ​​for μ1 are 3mm and μ2 are 6mm; when using breadcrumbs for coating, the default values ​​for μ1 are 5mm and μ2 are 8mm. Before production, the μ1 and μ2 thresholds can be customized based on actual coating thickness requirements or pass rate criteria.

[0044] The improved ICP algorithm specifically includes the following:

[0045] 1. Initial alignment: Calculate the center of gravity of the source point cloud using the formula: Calculate the center of gravity of the target point cloud using the formula: Move the source point cloud to coincide with the center of gravity of the target point cloud.

[0046] 2. Calculate the eigenvalues ​​of each point in the source point cloud, including the normal vector, curvature, surface change rate, and FPFH (Fast Point Feature Histogram), which are used as the input of the neural network. The x, y, and z coordinates of the matching points corresponding to the target point cloud are output. The neural network is constructed. The specific structure is as follows: Figure 2 As shown:

[0047] Among them, x1, x2, x3 and x4 in the figure are the four eigenvalues ​​of the input respectively. is the hidden neuron of the first layer, is the hidden neuron of the second layer, y1, y2, y3 are the outputs, which are the x, y, z coordinate values ​​of the target point cloud respectively.

[0048] Select the Tahn activation function, the specific formula is:

[0049]

[0050] Where x is the input value and e is a natural constant. sinh(x) and cosh(x) are the hyperbolic sine and cosine functions, respectively, and their formulas are:

[0051]

[0052] Select the Adam optimizer, which takes into account the first-order moment estimate mean and second-order moment estimate variance of the gradient when updating the model parameters. The parameter update steps are as follows:

[0053] Let g t is the gradient of step t, θ t is the model parameter of the t-th step, α is the learning rate, β1 and β2 are exponential decay rates, set to β1 = 0.9, β2 = 0.999 respectively, ∈ is a constant, set to 10 -8 .

[0054] Calculate the first moment of the gradient to estimate momentum: m t =β1m t-1 +(1-β1)g t

[0055] Compute the variance of the second moment estimate of the gradient:

[0056] Apply bias correction to the first- and second-order moment estimates:

[0057]

[0058] Update the model:

[0059] The mean square error (MSE) is selected as the loss function. Among them, y i is the true value of the i-th sample, is the predicted value of the i-th sample.

[0060] The training process of the model: collect the source point cloud and the target point cloud, first use the point cloud registration software to manually match the source point cloud and the target point cloud one by one, then extract the feature values ​​from the source point cloud as the input value, and the coordinate value of the target point cloud as the output value to train the neural network.

[0061] 3. Input the feature values ​​extracted from the newly scanned source point cloud into the trained neural network model to obtain the coordinate values ​​of the target point cloud. Calculate the rotation matrix R and translation vector t based on the corresponding point pairs. Let and For the corresponding point set, by calculating the covariance matrix in

[0062] 4. Perform singular value decomposition on C and obtain C = U∑V T , rotation matrix R = VU T , the translation vector

[0063] 5. Transform the point cloud based on the calculated R and t.

[0064] S3: Using the same equipment parameters and scanning method as for scanning the original oysters, the surface of the powdered oysters is laser scanned again to obtain the 3D point cloud data of the actual powder coating layer and obtain the actual powder coating layer data model;

[0065] Specifically, after the coating process is completed, the powdered oysters are placed on the scanning workbench again, and the surface of the powdered oysters is laser scanned again using the same equipment parameters and scanning method as those used to scan the original oysters to obtain the three-dimensional point cloud data of the actual powder coating layer, which is recorded as the P1 surface.

[0066] S4: By integrating the actual powder coating layer with the ideal powder coating layer and conducting comparative analysis, the quality of the powder coated product is judged to be qualified.

[0067] Specifically, by integrating the actual powder coating layer with the ideal powder coating layer and conducting comparative analysis, we can judge whether the powder coating layer is complete, whether the thickness is reasonable, and other quality indicators, and evaluate whether the product quality after powder coating is qualified.

[0068] Generate the oyster meat fitting surface I0 based on P0, move I0 upward in the vertical direction to obtain the ideal breading upper surface I1, and move I0 downward in the vertical direction to obtain the ideal breading lower surface I2.

[0069] Traversing all the points in the P1 surface, if there is a point below the ideal upper surface of the powder coating I1, it means that the thickness of the powder coating of the oyster is not enough to ensure a complete powder shell, and it is therefore judged to be unqualified.

[0070] Traverse all points in the P1 surface and record the number of points above the ideal powder coating lower surface I2, and define:

[0071]

[0072] Compare a and b. If a>b, where b is the judgment value generated by the system based on the outer surface of the oyster, then the oyster has too many areas with too thick batter coating, which will make the inner layer of oyster meat difficult to cook during frying, so it is judged to be unqualified; if a≤b, then the thickness of the oyster batter coating is judged to be appropriate, so it is judged to be qualified.

[0073] The method of selecting the value of b is related to the mean square error error when P0 generates the oyster meat fitting surface I0. The generation method is as follows: Figure 3 As shown in the figure, the x-axis is the error value and the y-axis is b. M in this method is the over-thickness coefficient, and its default value is 0.5. Before production, the M threshold can also be set based on the process requirements for the coating thickness or the pass rate judgment index.

[0074]

[0075] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. A method for automatically identifying and judging the coating effect of oysters, characterized in that: The method comprises the following steps: Freeze-shape and sort the oyster meat before coating; Laser scanning of the upper surface of oyster meat is used to obtain point cloud data of the unprocessed upper surface of the oyster and generate an ideal breading layer data model. Using the same equipment parameters and scanning methods as those used to scan the original oysters, the surface of the powdered oysters was laser scanned again to obtain the 3D point cloud data of the actual powder coating layer and obtain the actual powder coating layer data model; By integrating the calculation and comparative analysis of the actual powder coating layer and the ideal powder coating layer, the quality of the powder coated product can be judged.

2. The method for automatically identifying and judging the effect of oyster powder coating according to claim 1, characterized in that: The steps of freezing, shaping, and screening the oyster meat before coating specifically include: placing individual oyster meats one by one into a shaping mold and then freezing them, wherein the shaping mold is a waist-round shape that mimics the shape of the oyster meat, with a horizontal bottom and four walls perpendicular to the bottom. After the soft oyster meat is placed in, the mold is fully filled and frozen to shape; after being demolded from the frozen state, the bottom surface and the surrounding sides of the oyster meat are flat, and only the upper surface is an irregular curved surface.

3. The method for automatically identifying and judging the effect of oyster powder coating according to claim 1, characterized in that: The step of obtaining point cloud data of the unprocessed oyster upper surface by laser scanning the upper surface of the oyster meat and generating an ideal coating layer data model specifically includes: obtaining point cloud data of the unprocessed oyster upper surface, fixing the unprocessed oyster and scanning it with a laser scanner; if the scanned point cloud has many holes, performing multiple scans; and registering the multiple point clouds using an improved ICP algorithm to obtain the original unprocessed oyster surface point cloud layer P0.

4. The method for automatically identifying and judging the effect of oyster powder coating according to claim 3, characterized in that: The improved ICP algorithm specifically includes: Initial alignment: Calculate the center of gravity of the source point cloud using the formula: Calculate the center of gravity of the target point cloud using the formula: Move the source point cloud to coincide with the center of gravity of the target point cloud; Calculate the eigenvalues ​​of each point in the source point cloud, including the normal vector, curvature, surface change rate, and fast point feature histogram, as the input of the neural network, and the x, y, and z coordinates of the matching point corresponding to the target point cloud as the output to build the neural network; The feature values ​​extracted from the newly scanned source point cloud are input into the trained neural network model to obtain the coordinate values ​​of the target point cloud. The rotation matrix R and translation vector t are calculated based on the corresponding point pairs. Let and For the corresponding point set, by calculating the covariance matrix in Perform singular value decomposition on C and get C=U∑V T , rotation matrix R = VU T , the translation vector According to the calculated R and t, the point cloud is transformed.

5. The method for automatically identifying and judging the effect of oyster powder coating according to claim 1, characterized in that: The steps of laser scanning the surface of the powdered oyster again using the same equipment parameters and scanning method as those used for scanning the original oyster, obtaining three-dimensional point cloud data of the actual powder coating layer, and obtaining a data model of the actual powder coating layer specifically include: after completing the coating process, placing the powdered oyster again on the scanning workbench, using the same equipment parameters and scanning method as those used for scanning the original oyster, laser scanning the surface of the powdered oyster again, obtaining three-dimensional point cloud data of the actual powder coating layer, and recording it as the P1 surface.

6. The method for automatically identifying and judging the effect of oyster powder coating according to claim 5, characterized in that: The step of evaluating whether the quality of the powder coated product is qualified by integrating, calculating, and comparing the actual powder coating layer with the ideal powder coating layer specifically includes: Generate the oyster meat fitting surface I0 based on P0, move I0 vertically upward to obtain the ideal upper surface I1 of the breading, and move I0 vertically downward to obtain the ideal lower surface I2 of the breading; Traverse all points on the P1 surface. If any point is below the ideal upper surface I1 of the powder coating, it means that the powder coating thickness of the oyster is not enough to ensure a complete shell, and the oyster is judged as unqualified. Traverse all points in the P1 surface and record the number of points above the ideal powder coating lower surface I2, and define: Compare a and b, where b is the judgment value generated by the system based on the outer surface of the oyster. If a>b, the oyster has too much thick coating, which will make the inner layer of the oyster difficult to cook during frying, and is therefore judged as unqualified. If a≤b, the oyster is judged to have an appropriate coating thickness, and is therefore qualified. The value of b is related to the mean square error error when P0 generates the oyster meat fitting surface I0. The x-axis is the error value and the y-axis is b. M in this method is the over-thickness coefficient. The default value of M is 0.

5. Before production, the M threshold can also be set according to the process requirements for the thickness of the powder coating or the qualified rate judgment index.

Images