Foaming state detection method and system based on image recognition

Abstract

The present invention relates to the field of image processing technology, and the present invention relates to a foaming state detection method and system based on image recognition, comprising: collecting a cross-sectional image of a polyurethane foam board and processing it to obtain edge pixel points; traversing the edge pixel points, obtaining its tangent and calculating the pixel aggregation degree; dividing the edge pixel points into first and second edge pixel points, and calculating the gradient mean and the distance mean respectively. The gradient mean and the distance mean are normalized, and the weighted sum with the pixel aggregation degree is rounded up to obtain the voting number threshold; the bubbles in the cross-sectional image are marked according to the voting number threshold, and a neural network model is trained based on the marked image to output the probability of the foaming state category. The bubbles in the image to be detected are marked using an edge detection algorithm and input into the neural network model, and the foaming state category with the highest probability is used as the detection result. The present invention solves the problem of accurately obtaining bubble information under different bubble distribution states.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more specifically, to a foaming state detection method and system based on image recognition. Background Art

[0002] Foaming state detection is mainly used to monitor and control the foam production process in real time to ensure product quality and consistency. The technology relies on high-resolution imaging systems and suitable lighting equipment to capture images of the foam formation process. First, the size, shape, distribution and density of the foam are analyzed through image preprocessing, feature extraction and image processing algorithms such as edge detection, segmentation algorithms and morphological processing. Then, machine learning and deep learning models such as support vector machines, random forests and deep convolutional neural network training models are used to identify and classify different types of foam defects, such as uneven bubble size, foam rupture and delamination. This technology is widely used in the food industry, chemical industry and building materials to detect and monitor the foaming state of products such as bread, cakes, foam plastics, polyurethane foam, concrete and insulation materials to improve the taste, structure and performance of the products. Although the technology faces challenges such as complex environmental conditions, real-time processing requirements and diverse defect types, by combining image recognition and machine learning methods, the foaming state detection system can achieve accurate monitoring and control of the production process, significantly improving product quality and production efficiency.

[0003] The patent document with the authorization number CN113436149B discloses a method and system for detecting the uniformity of the cable foam layer. The method obtains the internal imaging image of the cable through an X-ray flaw detection device; the internal imaging image is divided into multiple sub-domain images by the edge curve obtained by edge detection of the internal imaging image; the foaming layer area in the internal imaging image is identified according to the sub-domain image; the foam structure in the foaming layer area is detected and judged whether it is uniform; when the foam structure in the foaming layer area is uneven, the cable is judged to be a defective product, otherwise it is a qualified product, and the uniformity detection of cables of any length can be realized, thereby identifying the defective cables that cannot guarantee the uniform transmission impedance of the cable, and removing the limitations of the fixed length, large interference, and high distortion of the water capacitance measurement cable, and without damaging the cable, the capacitance stability problem of the non-destructive detection cable is solved, and the cable transmission bandwidth, transmission rate, and data transmission stability are guaranteed. However, although this method solves the detection of the foaming state to a certain extent, it still cannot accurately obtain the problem of bubble hole information. Summary of the invention

[0004] In order to solve the problem of inability to accurately obtain cell information raised in the above background technology, the present invention provides solutions in the following aspects.

[0005] In a first aspect, the present invention provides a foaming state detection method based on image recognition, comprising:

[0006] The cross-sectional image of the polyurethane foam board is collected and processed to obtain edge pixel points in the cross-sectional image; the edge pixel points are traversed to obtain the tangent passing through the edge pixel points, and the pixel concentration of the edge pixel points on the tangent is calculated; for a first edge pixel point that does not belong to an edge intersection and a second edge pixel point that belongs to an edge intersection, the gradient mean of the first edge pixel point and the distance mean of the second edge pixel point are obtained, the gradient mean and the distance mean are normalized to obtain a gradient weight and a distance weight, and the gradient weight and the distance weight are weightedly summed with the pixel concentration and then rounded up to obtain a voting number threshold; the bubbles in the cross-sectional image are marked according to the voting number threshold, and a neural network model is trained based on the cross-sectional image of the marked bubbles to obtain a trained neural network model; the bubbles in the cross-sectional image of the polyurethane foam board to be detected are marked using an edge detection algorithm and input into the trained neural network model, the probability of the foaming state category is output, and the foaming state category with the highest probability is used as the detection result of the foaming state.

[0007] The invention not only calculates the gradient mean and distance mean of edge pixels, but also normalizes these features. The normalized gradient mean and distance mean are weighted summed with the pixel concentration to obtain the voting threshold. By comprehensively utilizing multiple features, the features of edge pixels can be more comprehensively reflected, thereby improving the reliability and accuracy of foaming state detection.

[0008] Furthermore, the collecting and processing of the cross-sectional image of the polyurethane foam board includes collecting the image using a complementary metal oxide semiconductor camera and performing grayscale processing.

[0009] By using complementary metal oxide semiconductor cameras for image acquisition and grayscale processing of images, high-quality image data can be obtained under different lighting conditions. This feature enables the solution to adapt to complex production environments and ensure the stability and consistency of detection results.

[0010] Furthermore, the tangent line is obtained by using a Sobel operator algorithm.

[0011] By using advanced image processing techniques such as edge detection and Sobel operator, the edge pixels and their tangents in the cross-sectional image of the polyurethane foam board can be accurately obtained. In particular, traversing the edge pixels and calculating the pixel concentration on their tangents makes the capture of edge features more detailed. This method effectively improves the accuracy of the detection results and avoids the errors caused by inaccurate edge detection in traditional methods.

[0012] Furthermore, the pixel concentration of the edge pixels on the tangent line is calculated, and the formula is as follows:

[0013]

[0014] In the formula, J(W i ) represents the pixel concentration of edge pixels on the tangent line of edge pixel i, Δd i,p Represents the first-order difference value of the distance between the pth edge pixel and the adjacent edge pixel on the tangent line of edge pixel i, m i Represents the number of edge pixels on the tangent line of edge pixel i.

[0015] Furthermore, the gradient mean of the first edge pixel is obtained, and the formula is as follows:

[0016]

[0017] Where W j 1 ′ represents the gradient mean of the jth first edge pixel, Indicates the edge B where the jth first edge pixel is located j The mean gradient of the sth edge pixel on , Indicates the edge B where the jth first edge pixel is located j The number of edge pixels on the image.

[0018] Furthermore, the formula for calculating the mean distance of the second edge pixel point is as follows:

[0019]

[0020] In the formula, represents the mean distance of the kth second edge pixel, d k,z represents the distance between the kth second edge pixel point and its left adjacent edge pixel point z in the vertical direction of the gradient, d k,y Represents the distance between the kth second edge pixel point and its adjacent edge pixel point y on the right in the vertical direction of the gradient.

[0021] Furthermore, the voting threshold is calculated as follows:

[0022]

[0023] In the formula, YZ represents the voting threshold, W j 1 represents the jth first edge pixel, represents the kth second edge pixel, J(W j 1 ) represents the pixel concentration of edge pixels on the tangent line of the jth first edge pixel point, represents the gradient weight of the jth first edge pixel after the gradient mean normalization process, m 1 Indicates the number of first edge pixels, Indicates the pixel concentration of edge pixels on the tangent line of the kth second edge pixel, m 2 Indicates the number of second edge pixels, represents the distance weight after the distance mean normalization of the k-th second edge pixel point, Represents the ceiling function.

[0024] The method for calculating the voting threshold cleverly combines the weighted sum of the gradient mean, distance mean and pixel concentration, and rounds it up. This process ensures the rigor of cell marking, effectively reduces the possibility of mislabeling, and further improves detection accuracy.

[0025] Furthermore, the neural network model is a convolutional neural network model.

[0026] Based on the labeled cell images, the neural network model is trained, and the powerful feature extraction capability of the convolutional neural network model is fully utilized to achieve the probabilistic output of the foaming state category. This not only improves the intelligence and automation level of foaming state detection, but also makes the detection results more reliable and stable. The use of the neural network model for detection can achieve rapid annotation of the cross-sectional image of the polyurethane foam board to be detected and real-time detection of the foaming state. This feature greatly improves the detection efficiency, making this method suitable for real-time monitoring and quality control in production environments.

[0027] In a second aspect, the present invention provides a foaming state detection system based on image recognition, comprising a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, any of the above-mentioned foaming state detection methods based on image recognition is implemented.

[0028] The present invention not only proposes a detection method, but also designs a corresponding detection system, including a memory and a processor. The memory stores computer program instructions, and when these instructions are executed by the processor, a foaming state detection method based on image recognition can be implemented. This system design takes into account the combination of software and hardware, has good integration and practicality, and is easy to promote in practical applications.

[0029] The beneficial effects of the present invention are:

[0030] This invention not only can detect the foaming state of polyurethane foam board efficiently and accurately by introducing precise image processing technology, comprehensive utilization of multiple features, optimized threshold calculation method and neural network model, but also provides a highly intelligent and automated detection system. Its advantages in real-time detection, system integration, adaptability to complex environments and future scalability provide strong technical support for production quality control and have broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0032] Figure 1 is a flow chart schematically showing a foaming state detection method based on image recognition according to an embodiment of the present invention;

[0033] Figure 2 is a grayscale diagram schematically showing a cross-section of a polyurethane foam board according to an embodiment of the present invention;

[0034] Figure 3 is an exemplary diagram schematically showing the relationship between the cell distribution and its tangent according to an embodiment of the present invention;

[0035] Figure 4 is a diagram schematically showing a comparison between a first edge pixel point and a second edge pixel point according to an embodiment of the present invention;

[0036] Figure 5 FIG. 4 is a flowchart schematically showing a foaming state detection system based on image recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0038] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0039] Embodiment of the foaming state detection method based on image recognition.

[0040] like Figure 1 As shown, the flow chart of the foaming state detection method based on image recognition of an embodiment of the present invention includes the following steps:

[0041] S101: collecting a cross-sectional image of a polyurethane foam board and processing it to obtain edge pixel points in the cross-sectional image, traversing the edge pixel points and obtaining a tangent line passing through the edge pixel points, and calculating a pixel concentration degree of the edge pixel points on the tangent line.

[0042] like Figure 2 Shown is a grayscale image of a cross section of a polyurethane foam board according to an embodiment of the present invention.

[0043] Among them, the polyurethane foam board is a common thermal insulation and structural material, and its cross-sectional image is an image obtained by cutting the foam board and photographing or scanning its cross section. In this embodiment, the cross-sectional image is obtained by photographing with a complementary metal oxide semiconductor camera, and this image shows the pore structure and distribution inside the foam board. Edge pixels are points in the image where the grayscale value changes significantly. They can be extracted by edge detection algorithms and are used to describe the boundaries of objects. Pixel concentration is a concept in image processing that measures the density of pixels in a certain area of ​​the image. Specifically, it describes the concentration of pixels in a certain area or on a certain edge in the image. And the tangent of the edge pixel point is obtained using the Sobel operator.

[0044] Specifically, the cross-sectional image is processed, including graying, filtering, edge detection and other operations. Through the above steps, the acquisition and processing of the cross-sectional image of the polyurethane foam board are specifically implemented, and the edge pixels in the image are successfully extracted. These processing steps lay the foundation for the subsequent processing of edge pixels.

[0045] like Figure 3 As shown, an exemplary diagram of the relationship between the cell distribution and its tangent line of an embodiment of the present invention.

[0046] Figure 3 The straight line shown in the figure is the tangent of cell No. 1, which passes through cells No. 6, 4, 2, 7, and 9. The change in the distance between two adjacent edge pixels on the tangent of cell No. 1 can reflect the distribution of the cells. The smaller the change in the distance between two adjacent edge pixels on the tangent of cell No. 1, the more concentrated the cell distribution is.

[0047] In one embodiment, the pixel concentration of edge pixels on the tangent line is calculated using the following formula:

[0048]

[0049] In the formula, J(W i ) represents the pixel concentration of edge pixels on the tangent line of edge pixel i, Δd i,p Represents the first-order difference value of the distance between the pth edge pixel and the adjacent edge pixel on the tangent line of edge pixel i, mi Represents the number of edge pixels on the tangent line of edge pixel i.

[0050] In another embodiment, the pixel concentration of edge pixels on the tangent line is calculated using the following formula:

[0051]

[0052] In the formula, J(W i ) represents the pixel concentration of edge pixels on the tangent line of edge pixel i, Δd i,p Represents the first-order difference value of the distance between the pth edge pixel and the adjacent edge pixel on the tangent line of edge pixel i, represents the variance of the first-order difference value, m i Represents the number of edge pixels on the tangent line of edge pixel i.

[0053] S102: For a first edge pixel point that does not belong to an edge intersection and a second edge pixel point that belongs to an edge intersection, obtain a gradient mean and a distance mean, and perform normalization processing to obtain a gradient weight and a distance weight.

[0054] Among them, the edge intersection refers to the location where two or more edges meet at the same point or area. The edge intersection usually has complex features because multiple edges are intertwined there to form a more complex local structure. The first edge pixel points refer to those pixels that are not located at the edge intersection. In image processing, these pixels usually represent relatively simple or single edge parts without the interweaving of other edges. The second edge pixel points refer to those pixels located at the edge intersection. In image processing, these pixels usually involve the intersection of multiple edges and have more complex features. The gradient mean represents the average value of the gradient mean calculated on the edge pixels. The gradient mean is an important indicator in image processing, which indicates the degree of pixel grayscale change and is usually used to describe the strength of the edge. The distance mean refers to the average value of the distance mean calculated between the pixels at the edge intersection. These distance means can reflect the geometric characteristics of the edge intersection. Normalization to obtain gradient weights and distance weights is the process of converting gradient data and distance data into a standard range to ensure that the final sum of weights is 1.

[0055] like Figure 4 As shown, a comparison diagram of the first edge pixel point and the second edge pixel point of an embodiment of the present invention.

[0056] Specifically, in a cross-sectional image of a polyurethane foam board, the pixels on the straight portion of a single edge are first edge pixels, which are located in the non-intersection area of ​​the edge. The pixels at the intersection or corner are second edge pixels, which are located in the edge intersection area. In the edge detection process, the gradient mean of each edge pixel is calculated, and then these gradient means are averaged to obtain the gradient mean. When detecting the intersection of the cell edges, the distance from each pixel at the intersection to its adjacent edge pixel is calculated, and these distances are averaged to obtain the distance mean.

[0057] In one embodiment, the gradient mean of the first edge pixel is obtained using the following formula:

[0058]

[0059] Where W j 1 ′ represents the gradient mean of the jth first edge pixel, Indicates the edge B where the jth first edge pixel is located j The mean gradient of the sth edge pixel on , Indicates the edge B where the jth first edge pixel is located j The number of edge pixels on the image.

[0060] The formula for obtaining the mean distance of the second edge pixel is as follows:

[0061]

[0062] In the formula, represents the mean distance of the kth second edge pixel, d k,z represents the distance between the kth second edge pixel point and its left adjacent edge pixel point z in the vertical direction of the gradient, d k,y Represents the distance between the kth second edge pixel point and its adjacent edge pixel point y on the right in the vertical direction of the gradient.

[0063] S103: performing weighted summation of the gradient weight, the distance weight and the pixel concentration, and rounding up to obtain a voting number threshold.

[0064] Among them, rounding up is rounding the value up to the nearest integer. That is, for any non-integer value, rounding up returns the smallest integer greater than or equal to the value. The vote threshold is used to determine whether a result meets a certain condition or standard. In classification or detection tasks, it helps determine whether to accept a specific decision or result. This threshold is usually used to determine the validity of the result by comparing it with the calculated result to determine whether to accept a decision. Optimizing performance: Reasonable setting of the threshold can help the system perform better under different conditions. For example, in the task of detecting bubbles, correctly setting the threshold can improve the accuracy of detection.

[0065] In one embodiment, the vote count threshold is calculated using the following formula:

[0066]

[0067] In the formula, YZ represents the voting threshold, W j 1 represents the jth first edge pixel, represents the kth second edge pixel, J(W j 1 ) represents the pixel concentration of edge pixels on the tangent line of the jth first edge pixel point, represents the gradient weight of the jth first edge pixel after the gradient mean normalization process, m 1 Indicates the number of first edge pixels, Indicates the pixel concentration of edge pixels on the tangent line of the kth second edge pixel, m 2 Indicates the number of second edge pixels, represents the distance weight after the distance mean normalization of the k-th second edge pixel point, Represents the ceiling function.

[0068] In another embodiment, the vote count threshold is calculated using the following formula:

[0069]

[0070] In the formula, YZ represents the voting threshold, W j 1 represents the jth first edge pixel, represents the kth second edge pixel, Represents the minimum value of the mean gradient of the first edge pixel, represents the maximum value of the mean gradient of the first edge pixel, J(W j 1 ) represents the pixel concentration of edge pixels on the tangent line of the jth first edge pixel point, represents the gradient weight of the jth first edge pixel after the gradient mean normalization process, m 1 Indicates the number of first edge pixels, Indicates the pixel concentration of edge pixels on the tangent line of the kth second edge pixel, m 2 Indicates the number of second edge pixels, represents the distance weight after the distance mean normalization of the k-th second edge pixel point, Represents the ceiling function Indicates the minimum value of the mean distance of the second edge pixel point, Indicates the maximum value of the mean distance of the second edge pixels.

[0071] S104: marking the cells in the cross-sectional image according to the voting number threshold, training a neural network model based on the cross-sectional image of the marked cells, and outputting the probability of the foaming state category.

[0072] Among them, marking the bubbles refers to identifying and marking the location of the bubbles in the image. This is a key step and is usually used for subsequent analysis or training models. The marking process includes determining the boundaries or locations of the bubbles and marking them as specific categories or areas. Training a neural network model is to input labeled data, such as labeled bubble images, and adjust the parameters of the model so that the model can learn the features in the data, thereby improving its classification or prediction capabilities for similar data. The probability of the foaming state category refers to the probability value output by the neural network model, which indicates that the bubbles in the image belong to different foaming state categories. This is one or more probability values ​​that indicate the model's confidence in each category.

[0073] Specifically, the neural network model is trained using labeled images as training sets. The model adjusts weights through the back-propagation algorithm to minimize the error between the predicted results and the actual labels. Commonly used neural network types include convolutional neural networks, etc. The trained neural network model classifies the input image and outputs the probability of each possible foaming state category. For example, the model may output "Bubble state A: 70%" and "Bubble state B: 30%", indicating that the bubbles in the image are more likely to belong to state A.

[0074] S105: using an edge detection algorithm to mark the cells in the cross-sectional image of the polyurethane foam board to be detected and inputting the mark into the neural network model, and taking the foaming state category with the highest probability as the foaming state detection result.

[0075] Among them, the edge detection algorithm is a technology in image processing that is used to identify areas in an image where the grayscale changes significantly, which usually represent the boundaries of objects. Commonly used edge detection algorithms include the Sobel operator and the Laplace operator. The bubble state category with the highest probability refers to the category with the largest predicted probability among all possible categories output by the neural network model. The model calculates a probability value for each category and selects the category with the highest probability value as the final result.

[0076] Specifically, the category with the highest probability is selected as the final result to ensure that the classification result that the model is most confident in is selected, thereby improving the accuracy of classification and solving the problem of accurately obtaining pore information under different pore distribution states.

[0077] Embodiment of the foaming state detection system based on image recognition:

[0078] like Figure 5 As shown, a flow chart of a foaming state detection system based on image recognition according to an embodiment of the present invention includes a processor and a memory.

[0079] The present invention also provides a foaming state detection system based on image recognition. The system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the foaming state detection method based on image recognition according to the present invention is implemented.

[0080] The foaming state detection system based on image recognition also includes other components well known to those skilled in the art, such as a communication interface, and the configuration and functions of the components are known in the art, so they will not be described in detail here.

[0081] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that may be used to store the required information and may be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0082] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0083] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A foaming state detection method based on image recognition, characterized in that: include: Collecting a cross-sectional image of the polyurethane foam board and processing it to obtain edge pixel points in the cross-sectional image; Traversing the edge pixel points, obtaining a tangent line passing through the edge pixel points, and calculating the pixel concentration of the edge pixel points on the tangent line; For the first edge pixel point that does not belong to the edge intersection and the second edge pixel point that belongs to the edge intersection, the gradient mean of the first edge pixel point and the distance mean of the second edge pixel point are obtained. The gradient mean of the first edge pixel point is expressed as: In the formula, represents the gradient mean of the jth first edge pixel, Indicates the edge B where the jth first edge pixel is located j The mean gradient of the sth edge pixel on , Indicates the edge B where the jth first edge pixel is located j The number of edge pixels on ; The mean distance of the second edge pixel points is expressed as: In the formula, represents the mean distance of the kth second edge pixel, d k,z represents the distance between the kth second edge pixel point and its left adjacent edge pixel point z in the vertical direction of the gradient, d k,y represents the distance between the kth second edge pixel point and its adjacent edge pixel point y on the right in the vertical direction of the gradient; The gradient mean and the distance mean are normalized to obtain a gradient weight and a distance weight, and the gradient weight, the distance weight, the pixel concentration of the edge pixel points on the tangent line of the first edge pixel point, and the pixel concentration of the edge pixel points on the tangent line of the second edge pixel point are weighted and summed up to obtain a voting number threshold after rounding up; the cells in the cross-sectional image are marked according to the voting number threshold, and a neural network model is trained based on the cross-sectional image of the marked cells to obtain a trained neural network model; The foam cells in the cross-sectional image of the polyurethane foam board to be detected are annotated using an edge detection algorithm and input into the trained neural network model, the probability of the foaming state category is output, and the foaming state category with the highest probability is taken as the detection result of the foaming state.

2. The foaming state detection method based on image recognition according to claim 1, characterized in that: The cross-sectional image of the polyurethane foam board is collected and processed, Including, using complementary metal oxide semiconductor camera for acquisition and grayscale processing.

3. The foaming state detection method based on image recognition according to claim 1, characterized in that: The tangent line is obtained by using the Sobel operator algorithm.

4. The foaming state detection method based on image recognition according to claim 1, characterized in that: Calculate the pixel concentration of edge pixels on the tangent line. The formula is as follows: In the formula, J(W i ) represents the pixel concentration of edge pixels on the tangent line of edge pixel i, Δd i,p Represents the first-order difference value of the distance between the pth edge pixel and the adjacent edge pixel on the tangent line of edge pixel i, m i Represents the number of edge pixels on the tangent line of edge pixel i.

5. The foaming state detection method based on image recognition according to claim 1, characterized in that: Calculate the voting threshold using the following formula: In the formula, YZ represents the voting threshold, represents the jth first edge pixel, represents the kth second edge pixel, Indicates the pixel concentration of edge pixels on the tangent line of the jth first edge pixel, represents the gradient weight of the jth first edge pixel after the gradient mean normalization processing, m1 represents the number of first edge pixels, represents the pixel concentration of edge pixels on the tangent line of the kth second edge pixel, m2 represents the number of second edge pixels, represents the distance weight after the distance mean normalization of the k-th second edge pixel point, Represents the ceiling function.

6. The foaming state detection method based on image recognition according to claim 1, characterized in that: The neural network model is a convolutional neural network model.

7. The foaming state detection system based on image recognition is characterized in that: The invention comprises a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the foaming state detection method based on image recognition according to any one of claims 1 to 6 is implemented.

Images