Method and system for on-line detection of yeast adding uniformity

Abstract

The invention relates to a method and a system for detecting yeast adding uniformity on line. The method comprises the following steps: acquiring data; data processing: performing uniformity evaluation on the obtained data information related to the material; and a mixing scheme is optimized. During data processing, calculating a first uniformity evaluation index by using data which is extracted from the preprocessed image and is related to a first microscopic uniformity indicator; or when the false alarm rate evaluation index related to the first microcosmic uniformity indicator exceeds a preset threshold value, a second uniformity evaluation index is calculated by using data related to a second microcosmic uniformity indicator, so that uniformity evaluation is carried out based on the first uniformity evaluation index and / or the second uniformity evaluation index, the second microcosmic uniformity indicator is a texture feature extracted from an area delimited by taking the first microcosmic uniformity indicator as a center. The system comprises a mixing device, a monitoring module and a central control module, wherein the central control module can evaluate the uniformity of an image related to a material.

Description

Technical Field

[0001] The invention relates to the technical field of automated liquor brewing, and in particular to an online method and system for detecting koji addition uniformity. Background Art

[0002] In the production of liquor, Daqu plays a vital role. It not only contains a variety of microbial strains, but also contains brewing enzymes such as amylase, protease and esterase. These enzymes help to decompose starch, protein and fat in the raw materials, thereby promoting the formation of wine flavor. Adding qu is one of the key links in liquor brewing. It is required to evenly add qu powder after the mash is spread and cooled to ensure the fermentation quality. However, the traditional qu adding device has the problem of uneven qu adding and the inability to adaptively adjust the qu adding uniformity, which directly affects the fermentation effect of the mash and the quality of the final wine.

[0003] CN117142063A discloses a device for cooling and adding koji to fermented grains of white wine, comprising a conveyor, a feed hopper for adding koji to fermented grains on a conveyor belt, the front end of the feed hopper being the conveying entry direction of the fermented grains; an adding component, the adding component being arranged in the feed hopper, the adding component adding koji to the fermented grains on the conveyor belt at equal intervals; a measuring component, the measuring component being located at the front end of the feed hopper, the measuring component changing the amount of koji added by the adding component at the measuring point according to the thickness of the fermented grains at the measuring point, by arranging the adding component, when the fermented grains are transported by the conveyor belt, a transmission roller is controlled by a motor to quantitatively add koji to the fermented grains on the surface of the conveyor belt, so that the koji is evenly added inside the fermented grains, the thickness of the fermented grains on the surface of the conveyor belt is measured by the measuring component, and the adding component is controlled to change the amount of koji added, so that the addition of koji is more even.

[0004] CN221117383U discloses a material mixing device for brewing wine, comprising a frame, a mixing and cooling cylinder and a koji adding mechanism. The mixing and cooling cylinder comprises an outer cylinder and an inner cylinder. The outer cylinder is arranged on the frame and can rotate radially on the frame. The inner cylinder is arranged in the outer cylinder and can rotate axially in the outer cylinder. A feeding port is arranged at one end of the outer cylinder, and a feeding port is arranged at one end of the inner cylinder near the feeding port. The koji adding mechanism is rotatably arranged on the frame, and the koji adding mechanism can be connected with or separated from the feeding port by rotation. A main fan is arranged at the other end of the outer cylinder, and an air duct is formed between the outer cylinder and the inner cylinder. The air outlet of the main fan extends into the air duct. After the koji adding mechanism is connected with the feeding port, the gas transported by the main fan is discharged through the air duct and the koji adding mechanism in sequence.

[0005] The traditional process of adding koji often ignores the real-time monitoring of uniformity detection, resulting in uneven problems in the production process that cannot be discovered and solved in time, which greatly prolongs the mixing time and wastes a lot of resources. In addition, the existing image acquisition equipment is difficult to obtain high-quality images in complex industrial production environments, such as light changes, equipment vibrations, etc., which affects the accuracy of subsequent processing. Especially under the condition that the mash in the mixing tank has a high moisture content and a certain viscosity, it is a test for the image acquisition equipment to obtain high-quality images during the mixing process.

[0006] In addition, on the one hand, there are differences in understanding among those skilled in the art; on the other hand, the applicant studied a large number of documents and patents when making the present invention, but due to space limitations, not all details and contents are listed in detail. However, this does not mean that the present invention does not have the characteristics of these prior arts. On the contrary, the present invention already has all the characteristics of the prior art, and the applicant reserves the right to add relevant prior art to the background technology. Summary of the invention

[0007] In view of the deficiencies of the prior art, the present invention provides a method and system for online detection of bending uniformity to solve at least part of the above-mentioned technical problems.

[0008] The present invention discloses an online method for detecting the uniformity of added curves, which may include the following steps:

[0009] Data collection: Use the collection unit to obtain data information related to materials;

[0010] Data processing: Evaluate the uniformity of the acquired data information related to the material;

[0011] Mixing scheme optimization: Optimize the mixing scheme based on the uniformity evaluation results.

[0012] Preferably, when performing data processing, a first uniformity evaluation index is calculated using data related to a first microscopic uniformity indicator extracted from a preprocessed image, or a second uniformity evaluation index is calculated using data related to a second microscopic uniformity indicator when a false alarm rate evaluation index related to the first microscopic uniformity indicator exceeds a preset threshold, thereby performing uniformity evaluation based on the first uniformity evaluation index and / or the second uniformity evaluation index, wherein the second microscopic uniformity indicator is a texture feature extracted from an area defined with the first microscopic uniformity indicator as the center.

[0013] According to a preferred embodiment, in the grayscale image obtained after preprocessing of the image acquired by the acquisition unit, the area of ​​slight brightness change formed by the material structure and distribution characteristics based on the characteristics of the reflected light on the material surface in the mash accumulation area is extracted as the first microscopic uniformity indicator, wherein the data related to the first microscopic uniformity indicator includes: quantity, size, density, and distribution.

[0014] According to a preferred embodiment, a false alarm rate evaluation index calculated based on moisture content changes and / or composition differences is used to characterize the possibility of false alarm of a first uniformity evaluation index calculated using a first microscopic uniformity indicator, wherein when the false alarm rate evaluation index related to the first microscopic uniformity indicator does not reach a preset threshold, the first uniformity evaluation index can be directly used for uniformity evaluation.

[0015] According to a preferred embodiment, the preprocessed grayscale image is divided into multiple regions to extract the first microscopic uniformity indicator in each region, respectively, wherein when the false alarm rate evaluation index related to the first microscopic uniformity indicator exceeds a preset threshold, a random number of first microscopic uniformity indicators are identified and marked in each region, and corresponding windows are divided with these first microscopic uniformity indicators as the center, so as to extract the second microscopic uniformity indicator in the window.

[0016] According to a preferred embodiment, a second microscopic uniformity indicator is extracted using a local binary pattern, the obtained LBP code is used to generate a corresponding histogram, and then the normalized histogram is used as a feature vector calculation of the second microscopic uniformity indicator to obtain a second uniformity evaluation index for judging the consistency of the feature value, thereby achieving uniformity evaluation.

[0017] According to a preferred embodiment, when it is determined that there is at least one uneven area based on the uniformity evaluation result, the operating parameters of the mixing equipment are adjusted in real time until the mash and koji powder in the mixing tank are finally uniform, wherein the optimal mixing path is calculated using the least squares method so that the mixing equipment can be targetedly controlled in three dimensions, thereby obtaining the optimal mixing path.

[0018] According to a preferred embodiment, the acquisition unit is configured as an industrial camera to acquire original images related to materials in the mixing tank, wherein the materials refer to a mixture of mash and koji powder, and the mixing tank adopts a flipping device in cooperation with stirring teeth arranged in the tank to mix the mash and koji powder in the tank, wherein, under the action of the flipping device, the mixing tank rotates in a forward and reverse direction in sequence, so that the acquisition unit can acquire corresponding images each time the mixing tank flips back to the upright position.

[0019] The acquisition unit is configured as an industrial camera, which is specifically used to collect the original image of the mixture of mash and koji powder in the mixing tank. The high-definition and high-frame rate image data it provides can be used for subsequent image processing and uniformity evaluation. Since the camera can adapt to various ambient light conditions, good quality images can be obtained even in insufficient or excessive light conditions. The flipping and stirring mechanism of the mixing tank further ensures that the materials are fully mixed, and the industrial camera collects images every time the mixing tank flips back to the right position, providing an accurate time point for uniformity evaluation, thereby ensuring the accuracy and reliability of the evaluation results.

[0020] According to a preferred embodiment, the collection unit can be arranged in an isolation area on the inner side of the top cover of the mixing tank that is deviated from the center line of the mixing tank, so that when the mixing tank rotates, the collection unit located in the isolation area can collect the original image of the material in a way that the shooting area is not blocked by the material.

[0021] The acquisition unit is set in an isolated area on the inner side of the mixing tank cover that deviates from the center line. This innovative layout cleverly utilizes the structural characteristics of the mixing tank to create an area where materials will not enter, thereby ensuring that the shooting area of ​​the industrial camera will not be blocked by materials and that the industrial camera can always obtain clear and accurate original images. This clarity is crucial for subsequent uniformity assessment and can improve the accuracy of data processing. This design not only avoids interference from materials on image acquisition, but also protects the camera from contamination and damage, improving the stability of the system and the convenience of maintenance.

[0022] According to a preferred embodiment, the preprocessing operations include graying, filtering and denoising, and / or edge detection to improve image quality, wherein the graying process uses a weighted average method to comprehensively consider the contributions of different color channels; the filtering and denoising process uses a Gaussian filter to reduce noise in the image; and the edge detection process uses the Canny algorithm to identify the edges of objects in the image by finding areas in the image where the brightness changes dramatically.

[0023] Preprocessing operations include grayscale, filtering and denoising, and edge detection, which together improve the quality of the image and make subsequent feature extraction and uniformity assessment more accurate. Grayscale optimizes the brightness information of the image through weighted averaging, filtering and denoising reduces the noise of the image, and edge detection accurately identifies the outlines of objects in the image. The combination of these steps provides clear and accurate image data for uniformity assessment.

[0024] According to a preferred embodiment, threshold segmentation, region growing or clustering algorithm is used to segment the image, thereby dividing the material-related image into several meaningful regions so as to independently analyze the characteristics of each region, wherein the clustering algorithm includes a K-means clustering algorithm.

[0025] Use threshold segmentation, region growing or clustering algorithms to segment images to effectively divide the image into multiple representative regions, providing structured image data for subsequent feature extraction and uniformity assessment. In particular, the application of the K-means clustering algorithm divides the image into several clusters by calculating the similarity between pixels. The pixels in each cluster have a high degree of similarity, making the uniformity assessment more accurate.

[0026] According to a preferred embodiment, after the image is segmented into several regions, corresponding features are extracted for each region to evaluate the uniformity of the added curve, wherein the extracted features include one or more of the following features:

[0027] Color features: by calculating the color distribution of each area and expressing it in the form of a color histogram;

[0028] Texture features: extracted by using gray-level co-occurrence matrix or local binary pattern;

[0029] Shape features: obtained by calculating the bounding rectangle, circularity and other parameters of each area.

[0030] After the image is segmented into several regions, one or more of the features such as color, texture and shape are extracted for each region, providing multi-dimensional feature information for uniformity assessment. Color features help identify the distribution and proportion of different components, texture features describe the spatial arrangement and grayscale changes of the material, and shape features describe the appearance and geometry of the object. The comprehensive analysis of these features makes the assessment of the uniformity of the fermented grains and koji more comprehensive and in-depth.

[0031] According to a preferred embodiment, a statistical method or a machine learning model is used to evaluate the uniformity of features, wherein, when using a statistical method, a corresponding standard deviation and / or variance is set to judge the uniformity of the feature value distribution by judging the size of the standard deviation or variance of a certain feature; when using a machine learning model, a support vector machine or a random forest model is used to predict the features of a new image by learning the features of known uniform and non-uniform samples, thereby evaluating its uniformity.

[0032] Using statistical methods or machine learning models to evaluate the uniformity of features can accurately evaluate the uniformity of materials through quantitative analysis or pattern recognition. Statistical methods quantitatively describe the distribution of feature values ​​by calculating standard deviation and variance, while machine learning models predict the features of new images by learning the features of known samples to evaluate their uniformity. The application of this method not only improves the accuracy of the evaluation, but also enhances the level of intelligence.

[0033] According to a preferred embodiment, when it is determined that there is at least one uneven area based on the uniformity evaluation result, the operating parameters of the mixing equipment are adjusted in real time until the mash and koji powder in the mixing tank are finally uniform, wherein the optimal mixing path is calculated using the least squares method so that the mixing equipment can be targetedly controlled in three dimensions, thereby obtaining the optimal mixing path.

[0034] Based on the uniformity evaluation results, the operating parameters of the mixing equipment are adjusted in real time until uniformity is achieved, which realizes real-time optimization and control of the mixing process. By using the least squares method to calculate the optimal mixing path, targeted control is achieved in three dimensions to obtain the best mixing effect. This method not only improves production efficiency, but also ensures the consistency of product quality.

[0035] The present invention also discloses an online detection system for the uniformity of adding koji, which is characterized in that it includes: a mixing device for accommodating and mixing materials; a monitoring module for acquiring data information related to the adding koji process; and a central control module for processing the data information acquired by the monitoring module and generating a control signal for regulating the mixing device.

[0036] Preferably, the mixing equipment includes a mixing tank, and the monitoring module includes a collection unit arranged on the top cover of the mixing tank to obtain images of the material in the mixing tank, so that the central control module can perform uniformity assessment on the images related to the material obtained by the collection unit.

[0037] The online detection system for adding koji uniformity provides a new way to comprehensively monitor and control the process of adding koji by integrating mixing equipment, monitoring module and central control module. The existence of mixing equipment provides a physical basis for the effective mixing of materials, and the monitoring module is set on the top cover of the mixing tank, which can obtain the image data of the materials in real time, ensuring the timeliness and accuracy of the monitoring data. This data acquisition method makes the monitoring process no longer dependent on manual intervention, reduces the possibility of human error, and provides a reliable basis for subsequent data analysis. The central control module processes and analyzes the collected image data, not only realizing the real-time evaluation of the uniformity of the materials, but also generating corresponding control signals to directly control the mixing equipment. This closed-loop control mechanism enables the system to take quick measures to adjust when it identifies unevenness, ensuring the stability and consistency of the production process, thereby effectively avoiding time delays and waste of resources caused by uneven mixing. This dynamic control capability is often difficult to achieve in traditional production systems, highlighting the advanced nature of the system. The integrated design of the system enhances the overall monitoring capability of the production process, allowing operators to control production dynamics in real time, respond quickly to emergencies, and improve production efficiency and safety. By combining material image processing with control signal generation, the system can also provide a rich basis for the accumulation and subsequent analysis of production data, making it possible to optimize and improve the production process. This data-driven decision-making mechanism not only improves the level of intelligent production, but also provides a new direction for future production management and technological innovation.

[0038] According to a preferred embodiment, the collection unit can be arranged in an isolation area on the inner side of the top cover of the mixing tank that is deviated from the center line of the mixing tank, wherein a transparent protective cover is arranged outside the collection unit for covering the isolation area, so as to block the collection unit from the space inside the mixing tank while ensuring light transmittance.

[0039] The acquisition unit is set in an isolated area deviating from the center line on the inner side of the mixing tank top cover and is equipped with a transparent protective cover. The design of the protective cover not only ensures the quality of image acquisition, but also protects the acquisition unit (such as industrial camera) from material contamination and damage to a large extent. Industrial cameras will inevitably be damaged by dust, liquid or other substances during normal mixing. Placing them in an isolated area can effectively extend the service life of the equipment and reduce the maintenance frequency and cost. This protection mechanism provides a guarantee for the long-term stable operation of the system and also makes the maintenance work of operators more convenient. It makes the acquisition unit not have to rely on complex protection devices or cleaning mechanisms, reducing the complexity of the system. Compared with traditional methods, this simplified design helps to reduce the overall cost of the system and improve reliability. Due to the enhanced stability of the acquisition unit, the system can provide continuous and real-time monitoring capabilities, which makes data collection more frequent and reliable during the addition process. The enhancement of real-time monitoring enables uneven problems in the production process to be identified and solved earlier, thereby improving overall production efficiency and product quality. This closed-loop feedback mechanism not only improves the intelligence level of the production process, but also promotes the optimization of production management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flowchart of the steps of a method for online detection of bending uniformity according to a preferred embodiment of the present invention;

[0041] Figure 2 It is a schematic diagram of the installation of a mixing tank according to a preferred embodiment of the present invention;

[0042] Figure 3 It is a schematic diagram of the installation of a collection unit according to a preferred embodiment of the present invention;

[0043] Figure 4 is an original image collected by a collection unit in a preferred embodiment of the present invention;

[0044] Figure 5 It is an image after data processing according to a preferred embodiment of the present invention;

[0045] Figure 6 is a processing diagram showing an uneven area according to a preferred embodiment of the present invention;

[0046] Figure 7 is another preferred embodiment provided by the present invention, showing a processing diagram with an uneven area;

[0047] Figure 8 It is a hardware connection diagram of an online detection and curvature uniformity system according to a preferred embodiment of the present invention.

[0048] Reference numerals list

[0049] 100: mixing equipment; 110: mixing tank; 111: top cover; 112: isolation area; 113: protective cover; 120: turning device; 200: monitoring module; 210: collection unit; 300: central control module. DETAILED DESCRIPTION

[0050] The following is a detailed description with reference to the accompanying drawings.

[0051] Example 1

[0052] like Figure 1 As shown, the present invention discloses an online method for detecting the uniformity of added curves, which includes one or more of the following steps:

[0053] S1. Data collection: using the collection unit 210 to obtain data information related to the material;

[0054] S2. Data processing: uniformity evaluation of acquired data information related to materials;

[0055] S3. Mixing scheme optimization: Optimize the mixing scheme based on the uniformity evaluation results.

[0056] Preferably, in step S1, the acquisition unit 210 may be configured as an industrial camera to collect original images related to the material in the mixing tank 110, wherein the industrial camera generally has high resolution, high frame rate and good anti-interference ability. Preferably, in the present invention, the material may refer to a mixture of fermented grains and koji powder. Preferably, when configuring the industrial camera, selection may be made according to the production environment and requirements. For example, a camera with more than 2 million pixels may be selected to ensure image clarity; if it is necessary to collect images during the material mixing process, a suitable shutter speed may be selected according to the material mixing speed to avoid motion blur; LED lights may be configured for lighting to ensure that suitable image quality can be obtained under various ambient light conditions.

[0057] Preferably, if Figure 2 As shown, the fermented grains and the koji powder can be stirred and mixed in the mixing tank 110, wherein the mixing tank 110 can use the turning device 120 to cooperate with the stirring teeth (not shown in the figure) provided in the tank to fully mix the fermented grains and the koji powder in the tank. Further, the turning device 120 can be provided with a turning motor to provide corresponding turning power according to the matching gear, and then transmit the power to the turning mechanism through the reducer, thereby driving the mixing tank 110 to rotate. Preferably, as the mixing tank 110 rotates, the fermented grains and the koji powder in the tank can be further mixed under the breaking action of the stirring teeth.

[0058] Preferably, the acquisition unit 210 configured in the industrial camera can be installed in the mixing tank 110 at a key position where the shooting area can fully cover the material, so as to capture the state of each area thereof. However, unlike the conventional installation position, the present invention also makes a special design for the industrial camera, wherein the industrial camera of the present invention can be set at a position close to the center line of the mixing tank 110 but slightly offset on the inner side of the top cover 111 of the mixing tank 110. Further, the direction of the offset is related to the structure of the mixing tank 110 and the stirring teeth arranged therein, wherein, based on the difference in the structure of the mixing tank 110 and the stirring teeth arranged therein, there is an isolation area 112 on the top cover 111 of the mixing tank 110 that the fermented grains and the koji powder will not enter when mixing in the tank, so that the industrial camera is set in the isolation area 112 to avoid the fermented grains and the koji powder from blocking the shooting area of ​​the industrial camera as much as possible, and due to the top-view angle of the industrial camera, it can fully cover the materials in the tank, so as to more accurately perform data analysis and processing. Furthermore, the mixing tank 110 may rotate in a forward and reverse direction in sequence. During the rotation of the mixing tank 110, since the stirring teeth arranged therein have a certain rotation direction, the isolation area 112 may appear at a position close to the center line of the mixing tank 110 but slightly to the left or right. Figure 3 The schematic diagram shows that an industrial camera is arranged in an isolated area 112 on the left side of the top cover 111 of the mixing tank 110. Preferably, the industrial camera can collect corresponding images each time the mixing tank 110 is flipped back to the right position, wherein the mixing tank 110 flipping back to the right position means that the mixing tank 110 is flipped to a position where its top cover 111 is located on the top of the tank body.

[0059] Preferably, the isolation area 112 can be determined based on field experiments and / or model simulations. Preferably, a protective cover 113 made of a transparent crystal material can be added to the industrial camera disposed in the isolation area 112. The refractive index characteristics of the protective cover 113 can affect the propagation of light, which can help improve the quality and clarity of the image captured by the industrial camera. The transparent protective cover 113 not only provides physical protection to prevent damage and contamination, but also optimizes optical performance to ensure more accurate image capture.

[0060] Preferably, if Figure 1 As shown, in step S2, one or more of the following sub-steps may be included:

[0061] S2.1, Preprocessing: The acquired images are preprocessed to improve image quality and highlight specific features;

[0062] S2.2, Region segmentation: The image is divided into multiple regions, each region represents a part of the material, and any white point in a segmented region is taken as the center to perform real-time online detection operations;

[0063] S2.3, Feature extraction: Extract features from each region;

[0064] S2.4. Uniformity assessment: Evaluate the uniformity of the mash and koji by comparing the characteristics of different areas.

[0065] Preferably, in step S2.1, preprocessing the acquired image can improve the image quality for subsequent analysis, wherein the preprocessing operation may include processing means such as grayscale, filtering and denoising, and / or edge detection.

[0066] Preferably, grayscale processing refers to converting a color image into a grayscale image, so that most image processing algorithms can have better results when processing single-channel images. Such an operation can reduce computational complexity while retaining important information in the image. Preferably, when processing images related to materials, grayscale helps to highlight brightness changes and facilitate subsequent edge detection and region segmentation. For example, functions in image processing libraries such as OpenCV can be used to achieve grayscale. Preferably, grayscale processing can use the following function:

[0067] I gray (x,y)=0.299·I R (x,y)+0.587·I G (x,y)+0.114·I B (x,y)

[0068] Among them, I gary (x, y) is the pixel value of the grayscale image at position (x, y). The grayscale image is a single-channel image. The value of each pixel represents its brightness, usually ranging from 0 (black) to 255 (white); R (x,y),I G (x,y),I B (x, y) represents the pixel values ​​of the red, green and blue channels at the position (x, y), respectively. Preferably, in the above formula, 0.299, 0.587 and 0.114 are the weights of the red channel, the green channel and the blue channel, respectively, representing the human eye's perception of red, green and blue. These coefficients are used to adjust the contribution of each color channel to the final grayscale value. Since the human eye has different sensitivities to different colors, these coefficients are selected to better reflect the visual perception of the human eye, among which the human eye has the strongest perception of green and relatively weak perception of blue. By weighted averaging, the contributions of different color channels are comprehensively considered, so that the generated grayscale image can better reflect the actual perception of the human eye. For example, due to the strong perception of green light, the weight of green is set to the highest.

[0069] Preferably, the filtering and denoising process can use Gaussian filtering to remove noise in the image, wherein Gaussian filtering is a linear smoothing filter that reduces the noise in the image by calculating the weighted average of the pixel neighborhood. The Gaussian function has a bell-shaped curve shape in the spatial domain, which can effectively remove high-frequency noise while maintaining the edge features of the image. Preferably, when processing images related to materials, the noise may come from environmental factors (such as light changes) or limitations of the image sensor, and the filtering and denoising process can improve the accuracy of subsequent analysis, wherein the grayscale image is filtered to generate a smooth image for use in subsequent steps. Preferably, the present invention can use a two-dimensional Gaussian filter for smoothing and denoising, wherein the formula used is as follows:

[0070]

[0071] Among them, I filtered (x, y) is the pixel value of the output image at position (x, y), the result after Gaussian filtering; I(x+m, y+n) is the pixel value of the input image at position (x+m, y+n), and the variables m and n represent the offset of the filter in the image; σ 2 is the variance of the Gaussian function, which controls the width of the Gaussian distribution.

[0072] Preferably, the edge detection process can use the Canny algorithm to identify the edge of the object in the image by finding the area with drastic brightness change in the image. Preferably, the Sobel operator or other gradient operators (such as Prewitt or Laplacian) can be used to calculate the gradient amplitude G and direction θ of the image, where the calculation formula is as follows:

[0073]

[0074] Among them, G x and G y are the gradients of the image in the horizontal and vertical directions, respectively.

[0075] Preferably, the specific configuration of the Sobel operator can select a 3x3 convolution kernel. Preferably, by setting two thresholds (a high threshold and a low threshold), the pixels are divided into three categories: strong edges with gradient amplitudes greater than the high threshold; weak edges with gradient amplitudes between the low threshold and the high threshold; non-edges with gradient amplitudes less than the low threshold. Furthermore, if a pixel is marked as a strong edge, it is retained unconditionally; if it is a weak edge, it is retained only if there is a strong edge in its adjacent pixels. In the edge connection stage, strong edge pixels are used to connect weak edges to form a complete edge. A connection algorithm (such as an eight-connected domain or a four-connected domain) is used to detect weak edges and share the adjacency relationship of strong edges.

[0076] Preferably, in step S2.2, threshold segmentation, region growing or clustering algorithm (such as K-means) may be used to segment the image, thereby dividing the image into several meaningful regions so as to independently analyze the characteristics of each region.

[0077] Preferably, the threshold segmentation method divides the image into different regions by setting one or more thresholds, wherein the part with pixel values ​​higher than the threshold is divided into one region, and the part with pixel values ​​lower than the threshold is divided into another region, thereby quickly identifying the region of interest. For each pixel I (x, y), if its gray value is greater than a certain set threshold T, the pixel is marked as the foreground (or target region), otherwise it is marked as the background, thereby obtaining the following formula:

[0078]

[0079] Among them, R(x,y) is the segmented image, 1 represents the foreground, and 0 represents the background. Furthermore, an adaptive threshold or a global threshold can be used to select an appropriate threshold.

[0080] Preferably, the region growing method is a segmentation method based on pixel similarity, which mainly selects seed points and gradually expands the region according to the similarity. Starting from one or more seed points, check whether its neighboring pixels meet specific similarity conditions (such as color, grayscale, etc.), and if so, include them in the current region.

[0081] Preferably, the K-means clustering algorithm is an unsupervised learning algorithm that groups data points into K clusters with high similarity within the clusters. The algorithm iteratively calculates the distance between each point and the cluster center and updates the cluster center until convergence. The K-means clustering algorithm regards image pixels as points in a high-dimensional space and iteratively optimizes the cluster center to reduce the distance from each point to the center of its cluster. In material image segmentation, the K-means clustering algorithm can divide the image into multiple regions to ensure that each region represents a part of the material. Preferably, the objective function of the K-means clustering algorithm is:

[0082]

[0083] Among them, S i is the i-th cluster, μ i is the center of the cluster, ∥x-μ i ∥ is the distance from pixel x to cluster center μ iEuclidean distance. Preferably, when using the K-means clustering algorithm for regional segmentation, K initial cluster centers can be randomly selected first. These centers can be randomly selected pixel values ​​in the image, representing different color or brightness features; then each pixel is assigned to the cluster center closest to it to form a preliminary regional division; after completing the pixel assignment, the algorithm will recalculate the center of each cluster by calculating the average value of all pixels in each cluster, and using this new average value as the new cluster center; finally, repeat the above pixel assignment and cluster center update steps until the cluster center no longer changes significantly, or reaches a preset number of iterations, to ensure the stability of each cluster, thereby forming several meaningful clusters that can represent different areas of the material.

[0084] Preferably, in step S2.3, features extracted from each region are used for the evaluation of the uniformity of the curve, for example, color features: by calculating the color distribution of each region and representing it in the form of a color histogram; texture features: extracted using a gray level co-occurrence matrix (GLCM) or a local binary pattern (LBP); shape features: calculating the bounding rectangle, circularity, etc. of each region.

[0085] Preferably, the color histogram can convert the color information in the image into digital features by calculating the frequency of each color appearing in the image. In the analysis of fermented mash and koji, the color histogram can help identify the distribution and proportion of different components. Further, the calculation of color distribution may include: for each area, traversing each pixel therein to obtain its color value; mapping the color value to the corresponding bin (interval) of the histogram. For example, in the RGB color space, the value of each color channel (red, green, blue) can be divided into several intervals, and then the number of pixels in each interval is calculated.

[0086] Preferably, texture features are used to describe the spatial arrangement and grayscale changes of pixels in an image, and are usually used to identify materials, patterns or structures in an image. In the analysis of mash and koji, texture features can identify the uniformity and separation of materials and help determine the mixing effect.

[0087] Preferably, shape features are used to describe the appearance and geometry of an object, which can help identify the shape features of a material and determine its uniformity and distribution.

[0088] Preferably, in step 2.4, a statistical method or a machine learning model may be used to evaluate the uniformity of the features.

[0089] Preferably, the following standard deviation may be set in the statistical method, which is a statistic used to describe the degree to which the data deviates from the mean and reflects the degree of dispersion of the data:

[0090]

[0091] Where fi is the i-th eigenvalue, μ is the mean of the eigenvalues, and N is the number of eigenvalues.

[0092] Preferably, a variance Var(f) may also be provided in the statistical method, which is the square of the standard deviation and is used to measure the degree of dispersion of data.

[0093] Preferably, if the standard deviation or variance of a feature is small, it indicates that the feature value distribution in the region is relatively uniform; conversely, a larger standard deviation or variance indicates that there is a large change in the feature value distribution, indicating that the region is uneven.

[0094] Preferably, a machine learning model can be used to evaluate uniformity in a more complex way, by learning the features of known uniform and non-uniform samples for prediction, wherein the following models can be used: Support Vector Machine (SVM): suitable for classification of high-dimensional spatial data; Random Forest: classification by building multiple decision trees, suitable for processing complex features. Preferably, the model can be trained using labeled training data (known uniform and non-uniform samples) so that the uniformity of a new image can be evaluated after the features of the new image are input into the trained model. Preferably, a machine learning framework such as Scikit-learn can be used to build and train the model, and parameters can be adjusted according to the characteristics of the data set and the requirements of the model, such as: the type of kernel function of SVM, the number of trees of random forest, etc.

[0095] In one embodiment, the original image collected by the industrial camera is as follows: Figure 4 As shown, the method of step S2 is performed on the original image acquired by the industrial camera to evaluate its uniformity. Figure 5 As shown in the figure, the original image can first undergo preprocessing operations such as grayscale, filtering and denoising, and edge detection. The grayscale processing can convert the color image into a grayscale image to reduce the computational complexity while retaining important structural information. The filtering and denoising processing can remove the fog and random noise points in the original image, thereby improving the overall quality of the image. The edge detection processing can effectively identify important features and structures in the image, such as the outline, boundary and corner points of the object; it is then divided into 64 matrices, and the size of each matrix is ​​reasonably divided according to the overall size of the image to facilitate the analysis of local uniformity, so that each area can be evaluated independently; then in each matrix, random white points are identified and marked as the center of uniformity calculation, and uniformity calculation is performed on the areas centered on these white points to evaluate the uniformity of the area. The calculation results include uniformity values, which will reflect the consistency of the area in terms of texture, brightness, color distribution, etc.; finally, the calculated uniformity values ​​will be returned to the corresponding processed image for visual comparison.

[0096] Preferably, the white dots used as markers for judging uniformity in the grayscale image can be defined as "first microscopic uniformity indicators" in the present invention, wherein the first microscopic uniformity indicator refers to a small brightness change area formed by the material structure and distribution characteristics based on the characteristics of the reflected light on the surface of the material in the mash accumulation area. Due to the different refractive indices and scattering characteristics of different components in the material, the formation of this area shows the change of local optical properties, reflects the uniformity of material distribution at the microscopic level, and reveals the interaction force between particles and material migration characteristics. Further, in addition to the visual physical uniformity, the first microscopic uniformity indicator also represents the distribution state of different components in the material, especially different components in the mash (such as sugar, yeast, acidic substances, etc.), to characterize the local chemical reaction active area, thereby reflecting the chemical uniformity of the material in the microstructure, wherein the presence or absence of the first microscopic uniformity indicator and its distribution can affect the reaction rate in the fermentation process and the taste level of the final product. Exemplarily, Table 1 shows experimental data related to the first microscopic uniformity indicator measured in a preferred embodiment.

[0097] Table 1 Experimental data

[0098]

[0099] Preferably, the appearance of the first microscopic uniformity indicator indicates that the material is unevenly distributed or has some characteristics different from the surrounding conditions, which reflects the uniformity of the material mixing in the image, wherein the data related to the first microscopic uniformity indicator can be used to implement the evaluation of the uniformity of the material mixing. Preferably, the data related to the first microscopic uniformity indicator may include: quantity, size (i.e., area or diameter), density, and distribution, wherein the more the number of first microscopic uniformity indicators, the more uneven the material distribution; the larger the size of the first microscopic uniformity indicator means the aggregation of the material, and the smaller the size means that the material is in a relatively uniform dispersion state. If the size difference is large, it may also indicate that the material distribution is more uneven; the more the number of first microscopic uniformity indicators in a unit area and the higher the density, the more uneven the material distribution in the local area. Preferably, the first uniformity evaluation index U based on the first microscopic uniformity indicator is 1 It can be calculated using the following formula:

[0100]

[0101] Where N w is the number of the first microscopic uniformity indicator, that is, the total number of white spots detected, N t is the total number of pixels, S avg is the average size (area) of the first microscopic uniformity indicator, S maxis the size of the largest first microscopic uniformity indicator (can be set according to the image characteristics), D w is the density of the first microscopic uniformity indicator, C d is the distribution coefficient of the first microscopic uniformity indicator, w 1 、w 2 、w 3 、w 4 is the weight coefficient of each item, reflecting the importance of each parameter in the final evaluation.

[0102] Preferably, the first uniformity evaluation index U 1 In the calculation formula, the first term is the ratio of the number of first microscopic uniformity indicators to the total number of pixels, which indicates the overall level of unevenness in the image, wherein the higher the ratio, the more uneven the mixing; the second term is the ratio of the average size to the maximum size of the first microscopic uniformity indicator, which reflects the effect of the size of the first microscopic uniformity indicator on the uniformity, wherein the closer the value is to 1, the smaller the white spots are and the more uniform the mixing is; the third term is the density of the first microscopic uniformity indicator, which is the concentration of white spots in the local area, and the higher the density, the greater the unevenness; the fourth term is the white spot distribution coefficient, and the closer it is to 1, the higher the uniformity.

[0103] Further preferably, the distribution coefficient of the first microscopic uniformity indicator can be calculated using the following formula:

[0104]

[0105] Where D is the inhomogeneity measure of the distribution of the first microscopic uniformity indicator, such as the standard deviation, which reflects the distribution difference of the first microscopic uniformity indicator in different regions. max is the maximum possible value of the uneven distribution.

[0106] Preferably, if the first uniformity evaluation index U calculated by the first microscopic uniformity indicator is 1 To evaluate the uniformity of material mixing, there is a possibility of false alarm in some cases, which may include: (1) Moisture content change: The moisture content of wet mash may cause the gray value to change, and some areas show brighter white spots due to uneven moisture distribution, which is misjudged as unevenness; (2) Component difference: Different components in mash (such as fiber, starch, protein) have different light reflection characteristics, which may produce obvious brightness differences in the image. Therefore, the present invention can characterize the possibility of false alarm by the false alarm rate evaluation index B, wherein the false alarm rate evaluation index B can be calculated using the following formula:

[0107]

[0108] Where M max is the maximum value of moisture content, Mmin is the minimum value of moisture content, M avg is the average moisture content, C d is the component difference coefficient, which calculates the difference between the light reflectance coefficients of different components (such as fiber, starch, protein), C max is the maximum possible value of the composition difference, which can be obtained based on experimental data or literature values, w 5 、w 6 is the weight coefficient.

[0109] Preferably, when the calculated false alarm rate evaluation index B does not reach the preset threshold, the first uniformity evaluation index U calculated based on the first microscopic uniformity indicator can be used. 1 Rapidly evaluate the uniformity of material mixing to save computing power and improve evaluation efficiency; when the calculated false alarm rate evaluation index B exceeds the preset threshold, it can be further evaluated based on other microscopic uniformity indicators to ensure the accuracy of the evaluation results. Preferably, when the computing power is sufficient or the accuracy of the evaluation results is required to be high, a variety of microscopic uniformity indicators can also be used for comprehensive evaluation. When a variety of microscopic uniformity indicators are used for comprehensive evaluation, the importance of the evaluation indicators calculated by each microscopic uniformity indicator can be flexibly adjusted according to the relationship between the false alarm factors of different microscopic uniformity indicators and the existing conditions.

[0110] Preferably, when the calculated false alarm rate evaluation index B exceeds the preset threshold, the first microscopic uniformity indicator can be used to further obtain the second microscopic uniformity indicator, wherein the second microscopic uniformity indicator is a texture feature extracted from an area demarcated with the first microscopic uniformity indicator as the center. The mash contains a variety of components (such as residual starch, protein, cellulose, etc.), which will show different texture features in the image. Among them, the heterogeneity of the mash makes uniformity analysis by texture more effective. The size, shape and distribution of particles in the mash will directly affect its texture features. The presence of particles will cause different local structures to appear in the image, reflecting the uniformity or inhomogeneity of the substance. The moisture content of the mash will also affect its surface texture. Uneven moisture distribution may cause different brightness and texture features in the image. The reason why texture features are used as the second microscopic uniformity indicator is that texture features can effectively reflect the local structure and distribution of different components in the mash, so that the uniformity of the components can be more clearly identified. In addition, texture features can extract a lot of information, such as contrast, directionality, roughness, etc., which can comprehensively evaluate the uniformity of the mash, while other features cannot provide such rich information, especially compared with color features (converted into brightness information in grayscale images) and shape features (which have high requirements on the shape of particles), texture features are more robust to noise and small local changes, and can provide a more stable uniformity assessment.

[0111] Preferably, the area demarcated with the first microscopic uniformity indicator as the center can be set according to specific needs, and can be set as a fixed-size window. For example, a 3×3, 5×5 or 7×7 square window can be selected, and the boundary of the set window can be expanded around the first microscopic uniformity indicator to ensure that all calculations are within the window. Preferably, in the above-set window, the local binary pattern (LBP) can be used to extract the second microscopic uniformity indicator. The reason is that: LBP focuses more on capturing the local structure and pattern in the image, can effectively reflect the local characteristics of the texture, and performs well in processing areas with large brightness changes; LBP has good robustness to illumination changes, and its calculation is based on relative changes (i.e., comparison of adjacent pixels), which can better capture the local characteristics of the object without being affected by illumination; LBP calculation is relatively simple and fast, suitable for real-time processing and large-scale image analysis, and the feature vector it generates is relatively small, easy to store and process. Therefore, when processing images of materials such as fermented grains, especially when the texture of the fermented grains in the image shows strong local structural changes (such as granularity, block structure, etc.), LBP will be a more appropriate choice because it can effectively capture these local features. Exemplarily, for a 3×3 neighborhood window, the grayscale values ​​of the center pixel of the window and the surrounding 8 pixels are compared. If the grayscale value of the surrounding pixels is greater than the center pixel, the corresponding position is marked as 1, otherwise it is marked as 0, and then the obtained binary number is converted into a decimal number to form an LBP code, and then a corresponding histogram is generated for the formed LBP code to represent the frequency of different LBP values.

[0112] Preferably, the normalized histogram is regarded as a feature vector of the second microscopic uniformity indicator, and the feature vector has the same size as the histogram, wherein the feature vector can be expressed as:

[0113] F=[f 1 ,f 2 ,…,f n ]

[0114] In the formula, f i It is a second microscopic uniformity indicator for each extraction.

[0115] Furthermore, the second uniformity evaluation index U for judging the consistency of the characteristic value can be calculated using the characteristic vector of the second microscopic uniformity indicator. 2 , and then realize the uniformity evaluation, where the second uniformity evaluation index U 2 The calculation formula is as follows:

[0116]

[0117] Wherein, μ is the average value of the characteristic value. Preferably, the second uniformity evaluation index U 2 The closer the value is to 1, the higher the uniformity.

[0118] Preferably, the calculated second uniformity evaluation index U 2 The uniformity values ​​are mapped to the image for visual comparison, wherein a heat map or other visualization techniques may be used to more intuitively observe the uniformity differences in different areas.

[0119] For example, Figure 6 As shown, for areas with lower uniformity of features (for example, the brightness of the middle area of ​​the material is lower than that of the evenly mixed material), they can be identified as non-uniform areas and displayed on the processing diagram, so that users can intuitively check whether the uniformity is qualified, and the non-uniform results will be recorded to facilitate subsequent optimization of the mixing scheme. Preferably, in the present invention, the industrial camera can capture images of two consecutive adjacent moments when capturing the original image, so as to correct the evaluation results, such as Figure 6 and Figure 7 Two processing diagrams are shown.

[0120] Preferably, the uneven area can be determined based on the uniformity evaluation result obtained in step S2, wherein step S3 can be initiated when there is at least one uneven area to optimize the mixing scheme according to the distribution of the uneven area until there is no uneven area.

[0121] Preferably, during the mixing process, the uniformity evaluation results can be monitored in real time, and the mixing scheme can be adjusted in real time according to the evaluation value, such as adjusting the mixing time, speed or other parameters, to ensure the uniformity of the final result. In step S3, the main goal of the mixing scheme optimization is to eliminate the non-uniformity in the material, and by adjusting the operating parameters of the mixing equipment 100, the mash and koji powder in the mixing tank 110 are finally uniform, wherein the optimization goals may include: improving the overall uniformity of the material: ensuring that all areas meet the set uniformity standards; reducing the mixing time: optimizing the mixing process to improve production efficiency while ensuring uniformity; reducing energy consumption: reducing the energy consumption of equipment operation through intelligent control to achieve economic benefits. Preferably, when adjusting the operating parameters of the mixing equipment 100, the underlying control logic may include three dimensions, namely, lateral control, longitudinal control and vertical control, which are independent of each other and closely integrated to ensure the comprehensiveness and effectiveness of the mixing. Preferably, in the optimization scheme, the least square method can be used to calculate the optimal mixing path, wherein the optimal operation path is found by minimizing the objective function (i.e., the sum of squares of the error between the actual mixing effect and the expected effect). Furthermore, by constructing a mathematical model that represents the relationship between various parameters and uniformity in the mixing process, an error function is defined to calculate the error between the current mixing scheme and the target uniformity, and then the path of the mixing device 100 is adjusted through iterative calculation to minimize the error function, thereby obtaining the optimal mixing path.

[0122] Mixing scheme optimization can quickly respond to and handle material inhomogeneities through real-time monitoring and feedback, while combining big data analysis and machine learning to provide intelligent support for decision-making. This process significantly improves the uniformity of the final product, ensures production quality, and effectively reduces production costs, unnecessary repeated mixing and energy consumption. In addition, the method of the present invention exhibits good stability in a complex production environment, reduces human intervention, and realizes automated production. At the same time, as the production environment and raw material characteristics change, the system can quickly adapt and adjust the mixing scheme to maintain high production efficiency, thereby providing companies with a more efficient and energy-saving solution.

[0123] Preferably, by establishing a data feedback mechanism, the result of each mixing (qualified or unqualified) can be used as training data and continuously input into the algorithm, so that after each operation, the model is updated according to the current mixing result to form a closed-loop learning. Preferably, according to historical data and the current mixing situation, the K value and the selection of cluster centers are dynamically adjusted to adapt to the characteristics of different batches of mash, so as to optimize the clustering process through practical feedback, thereby improving the overall segmentation accuracy. With the continuous acquisition of data, historical successful cases can be used to extract empirical formulas, which can be used to predict future mixing effects, reduce trial and error costs, and improve efficiency. Combined with machine learning, an intelligent system can be designed to provide suggestions for optimizing mixing solutions based on real-time data and historical experience, thereby further improving the uniformity and quality of mixing.

[0124] Example 2

[0125] This embodiment is a further improvement on Embodiment 1, and the repeated contents will not be repeated here.

[0126] like Figure 8 As shown, the present invention discloses an online detection system for the uniformity of adding koji, which can execute the online detection method for the uniformity of adding koji as described in Example 1. Preferably, the system of the present invention may include: a mixing device 100, which is used to accommodate and mix materials; a monitoring module 200, which is used to obtain data information related to the process of adding koji; and a central control module 300, which is used to process the data information obtained by the monitoring module 200 and generate a control signal for regulating the mixing device 100.

[0127] Preferably, if Figure 2 As shown, the mixing device 100 may include a mixing tank 110, wherein the mixing tank 110 may use a turning device 120 to cooperate with stirring teeth (not shown in the figure) provided in the tank to fully mix the fermented grains and koji powder in the tank. Further, the turning device 120 may be provided with a turning motor to provide corresponding turning power according to the matching gear, and then transmit the power to the turning mechanism through a reducer, thereby driving the mixing tank 110 to rotate. Preferably, as the mixing tank 110 rotates, the fermented grains and koji powder in the tank can be further mixed under the breaking action of the stirring teeth.

[0128] Preferably, if Figure 3As shown, the monitoring module 200 may include a collection unit 210 for acquiring images of materials, wherein the collection unit 210 may be arranged in an isolation area 112 on the top cover 111 of the mixing tank 110. Preferably, a protective cover 113 made of a transparent crystal material may be added to the industrial camera arranged in the isolation area 112. The refractive index characteristics of the protective cover 113 can affect the propagation of light, which can help improve the quality and clarity of the image taken by the industrial camera. The transparent protective cover 113 not only provides physical protection to prevent damage and contamination, but also optimizes the optical performance to ensure more accurate image capture. Preferably, the protective cover 113 of the present invention can be made of a material with high light transmittance to minimize the absorption and scattering of light, thereby maintaining the clarity of the image. Further preferably, the protective cover 113 can also be made of a material with a moderate refractive index to reduce the refraction distortion of light, wherein the moderate refractive index can be, for example, between 1.4 and 1.6.

[0129] Preferably, the monitoring module 200 may also include other types of sensors for acquiring physicochemical parameters of the material, such as a temperature sensor, a humidity sensor, and the like.

[0130] Preferably, the central control module 300 may perform uniformity assessment on the material-related images acquired by the acquisition unit 210 , wherein the central control module 300 may execute the uniformity assessment steps as described in Example 1.

[0131] Preferably, the central control module 300 can determine the uneven area based on the obtained uniformity evaluation results. When there is at least one uneven area, the central control module 300 can adjust the mixing scheme of the mixing device in real time according to the evaluation value, such as adjusting the mixing time, speed or other parameters to ensure the uniformity of the final result.

[0132] It should be noted that the above-mentioned specific embodiments are exemplary, and those skilled in the art can come up with various solutions inspired by the disclosure of the present invention, and these solutions also belong to the disclosure scope of the present invention and fall within the protection scope of the present invention. Those skilled in the art should understand that the present invention specification and its drawings are illustrative and do not constitute a limitation of the claims. The scope of protection of the present invention is defined by the claims and their equivalents. The present invention specification contains multiple inventive concepts, such as "preferably" or "according to a preferred embodiment", which means that the corresponding paragraph discloses an independent concept, and the applicant reserves the right to file a divisional application based on each inventive concept. Throughout the text, the features guided by "preferably" are only an optional method and should not be understood as a must-have setting. Therefore, the applicant reserves the right to abandon or delete the relevant preferred features at any time.

Claims

1. A method for online detection of bending uniformity, characterized in that: It includes: Data collection: using a collection unit (210) to obtain data information related to the material; Data processing: Evaluate the uniformity of the acquired data information related to the material; Mixing scheme optimization: Optimize the mixing scheme based on the uniformity evaluation results. Wherein, when performing data processing, the first uniformity evaluation index is calculated using the data related to the first microscopic uniformity indicator extracted from the preprocessed image, or the second uniformity evaluation index is calculated using the data related to the second microscopic uniformity indicator when the false alarm rate evaluation index related to the first microscopic uniformity indicator exceeds a preset threshold, thereby performing uniformity evaluation based on the first uniformity evaluation index and / or the second uniformity evaluation index. The second microscopic uniformity indicator is a texture feature extracted from a region defined with the first microscopic uniformity indicator as the center.

2. The method according to claim 1, characterized in that: In the grayscale image obtained after preprocessing of the image acquired by the acquisition unit (210), a small brightness change area formed by the material structure and distribution characteristics based on the characteristics of the reflected light on the material surface in the mash accumulation area is extracted as a first microscopic uniformity indicator, wherein the data related to the first microscopic uniformity indicator includes: quantity, size, density, and distribution.

3. The method according to claim 1 or 2, characterized in that: The false alarm rate evaluation index calculated based on the moisture content change and / or composition difference is used to characterize the possibility of false alarm of the first uniformity evaluation index calculated using the first microscopic uniformity indicator, wherein when the false alarm rate evaluation index related to the first microscopic uniformity indicator does not reach a preset threshold, the first uniformity evaluation index can be directly used for uniformity evaluation.

4. The method according to any one of claims 1 to 3, characterized in that: The preprocessed grayscale image is divided into multiple regions to extract the first microscopic uniformity indicator in each region, wherein when the false alarm rate evaluation index related to the first microscopic uniformity indicator exceeds a preset threshold, a number of random first microscopic uniformity indicators are identified and marked in each region, and corresponding windows are divided with these first microscopic uniformity indicators as the center, so as to extract the second microscopic uniformity indicator in the window.

5. The method according to any one of claims 1 to 4, characterized in that: The second microscopic uniformity indicator is extracted using the local binary pattern, the obtained LBP code is used to generate a corresponding histogram, and then the normalized histogram is used as the eigenvector calculation of the second microscopic uniformity indicator to obtain a second uniformity evaluation index for judging the consistency of the eigenvalues, thereby achieving uniformity evaluation.

6. The method according to any one of claims 1 to 5, characterized in that: When it is determined based on the uniformity evaluation result that there is at least one uneven area, the operating parameters of the mixing device (100) are adjusted in real time until the mash and koji powder in the mixing tank (110) are finally uniform, wherein the optimal mixing path is calculated using the least squares method so that the mixing device (100) can be controlled in a targeted manner in three dimensions, thereby obtaining the optimal mixing path.

7. The method according to any one of claims 1 to 6, characterized in that: The acquisition unit (210) is used to acquire original images related to materials in the mixing tank (110), wherein the materials refer to a mixture of fermented grains and koji powder. The mixing tank (110) uses a flipping device (120) in cooperation with stirring teeth provided in the tank to mix the fermented grains and koji powder in the tank, wherein, under the action of the flipping device (120), the mixing tank (110) rotates in a forward and reverse direction in sequence, so that the acquisition unit (210) can acquire corresponding images each time the mixing tank (110) flips back to a normal position.

8. The method according to any one of claims 1 to 7, characterized in that: The acquisition unit (210) can be arranged in an isolation area (112) on the inner side of a top cover (111) of the mixing tank (110) and deviated from the center line of the mixing tank (110), so that when the mixing tank (110) is rotating, the acquisition unit (210) located in the isolation area (112) can acquire an original image of the material in a manner in which a shooting area is not blocked by the material.

9. An online detection system for adding bending uniformity, characterized in that: It includes: A mixing device (100) for containing and mixing materials; A monitoring module (200) is used to obtain data information related to the process of adding music; The central control module (300) is used to process the data information acquired by the monitoring module (200) and generate a control signal for regulating the mixing device (100). The mixing device (100) comprises a mixing tank (110), and the monitoring module (200) comprises a collection unit (210) arranged on a top cover (111) of the mixing tank (110) to obtain an image of the material in the mixing tank (110), so that the central control module (300) can perform uniformity evaluation on the image related to the material obtained by the collection unit (210).

10. The system according to claim 9, characterized in that The collection unit (210) can be arranged inside an isolation area (112) on the inner side of a top cover (111) of the mixing tank (110) and deviated from the center line of the mixing tank (110), wherein a transparent protective cover (113) is arranged outside the collection unit (210) for covering the isolation area (112), so as to isolate the collection unit (210) from the inner space of the mixing tank (110) while ensuring light transmittance.

Images