Milk beverage production monitoring method and system based on AI vision and medium

Abstract

The invention discloses a milk beverage production monitoring method and system based on AI vision and a medium, and relates to the technical field of AI vision, and the method comprises the steps: analyzing milk beverage component composition abnormal nodes, and generating an abnormal chain; production data are collected in the whole process, and instantaneous and time sequence image features are extracted; carrying out instantaneous and cumulative anomaly identification prediction on the dual-state features based on an anomaly chain, and outputting an instantaneous identification result and a cumulative prediction result; positioning and tracking the production process according to the instantaneous recognition result, and feeding back an instantaneous monitoring result; and blocking a tracking production flow according to the cumulative prediction result, and feeding back a cumulative compensation monitoring result. The technical problem that the product quality is affected due to the fact that the abnormities in the production process cannot be recognized and corrected in time due to the fact that traditional milk beverage production monitoring depends on artificial experience and discrete sensor data is solved, full-process real-time dynamic monitoring and accurate abnormity prediction are achieved through AI vision, the monitoring effect of the production process is improved, and the production efficiency is improved. And the product quality stability is ensured.

Description

Technical Field

[0001] This invention relates to the field of AI vision technology, and in particular to a method, system and medium for monitoring dairy beverage production based on AI vision. Background Technology

[0002] The production process of dairy beverages is characterized by multi-stage process coupling, multi-component physicochemical interactions, and susceptibility to environmental disturbances. Traditional quality monitoring mainly relies on manual sampling and offline testing, which presents significant technical bottlenecks: on the one hand, instantaneous anomalies such as sudden foreign object contamination or filling deviations on the production line are difficult to capture in real time, and manual inspections suffer from high lag and a high risk of missed detection; on the other hand, cumulative anomalies such as gradual deviations in raw material ratios and slow microbial growth are often only discovered during finished product sampling due to the lack of dynamic tracking capabilities throughout the entire process, leading to the scrapping of entire batches of products. With the development of AI vision technology, it has become possible to achieve full-process monitoring of production through image analysis, but existing technologies still have shortcomings. Relying solely on image features is insufficient to distinguish between occasional instantaneous anomalies and gradual cumulative anomalies, and it cannot dynamically correlate changes in component properties with visual appearances, resulting in low accuracy and insufficient predictability in early warning. Summary of the Invention

[0003] This invention provides a method, system, and medium for monitoring dairy beverage production based on AI vision. It addresses the technical problem that traditional dairy beverage production monitoring relies on human experience and discrete sensor data, which leads to the inability to identify and correct anomalies in the production process in a timely manner, thus affecting product quality. The invention achieves the technical effect of real-time dynamic monitoring and accurate anomaly prediction throughout the entire process through AI vision, thereby improving the monitoring effect of the production process and ensuring the stability of product quality.

[0004] In a first aspect, the present invention provides a method for monitoring dairy beverage production based on AI vision, wherein the method for monitoring dairy beverage production based on AI vision includes: The system analyzes the abnormal nodes in the composition of the target dairy beverage to obtain an abnormal feature change chain; it connects to an image acquisition module to collect data on the entire production process of the target dairy beverage, extracting instantaneous image features and time-series image features; based on the abnormal feature change chain, it identifies and predicts instantaneous and cumulative anomalies in the instantaneous and time-series image features, obtaining instantaneous anomaly identification results and cumulative prediction results; it performs production process location tracking based on the instantaneous anomaly identification results, generating instantaneous monitoring results for feedback; and it performs production process obstruction tracking based on the cumulative prediction results, generating cumulative compensation monitoring results for feedback.

[0005] Secondly, the present invention also provides an AI vision-based dairy beverage production monitoring system, wherein the AI ​​vision-based dairy beverage production monitoring system includes: Anomaly Node Analysis Unit: Analyzes anomaly nodes based on the composition of the target dairy beverage to obtain anomaly feature change chains; Full-Process Acquisition Unit: Connects to the image acquisition module to acquire the entire production process of the target dairy beverage, extracting instantaneous image features and time-series image features; Anomaly Identification and Prediction Unit: Based on the anomaly feature change chains, identifies and predicts instantaneous and cumulative anomalies in the instantaneous and time-series image features, obtaining instantaneous anomaly identification results and cumulative prediction results; Production Process Location Tracking Unit: Tracks the production process location based on the instantaneous anomaly identification results, generating instantaneous monitoring results for feedback; Production Process Obstruction Tracking Unit: Tracks production process obstructions based on the cumulative prediction results, generating cumulative compensation monitoring results for feedback.

[0006] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the AI ​​vision-based dairy beverage production monitoring method provided by the present invention.

[0007] This invention discloses an AI vision-based method, system, and medium for monitoring dairy beverage production. The method includes: analyzing the composition of the target dairy beverage to identify abnormal nodes and obtain an abnormal feature change chain; connecting an image acquisition module to collect data throughout the entire production process of the target dairy beverage, extracting instantaneous image features and time-series image features; identifying and predicting instantaneous and cumulative anomalies based on the abnormal feature change chain, obtaining instantaneous anomaly identification results and cumulative prediction results; locating and tracking the production process based on the instantaneous anomaly identification results, generating instantaneous monitoring results for feedback; and tracing production process obstructions based on the cumulative prediction results, generating cumulative compensation monitoring results for feedback. This invention solves the technical problem of traditional dairy beverage production monitoring relying on human experience and discrete sensor data, which leads to the inability to timely identify and correct anomalies during production, affecting product quality. It achieves real-time dynamic monitoring and accurate anomaly prediction throughout the entire process through AI vision, improving the monitoring effect of the production process and ensuring product quality stability. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the AI ​​vision-based dairy beverage production monitoring method of the present invention.

[0009] Figure 2 This is a schematic diagram of the AI ​​vision-based dairy beverage production monitoring system of the present invention.

[0010] Explanation of the attached diagram labels: 11. Abnormal node parsing unit, 12. Full-process acquisition unit, 13. Abnormal identification and prediction unit, 14. Production process positioning and tracking unit, 15. Production process obstruction tracking unit. Detailed Implementation

[0011] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0012] Example 1, as Figure 1 This is a flowchart illustrating the AI ​​vision-based dairy beverage production monitoring method of the present invention, wherein the AI ​​vision-based dairy beverage production monitoring method includes: The composition of the interactive target dairy beverage is analyzed for abnormal nodes to obtain the abnormal feature change chain.

[0013] Specifically, firstly, the composition data of the target dairy beverage is acquired, and historical information of each component is extracted, including its proportion, concentration, and trends. Then, through statistical analysis of this historical data, abnormal nodes are identified. For example, by setting a reasonable threshold, when the abnormal probability or deviation of a component deviates from the normal range, this abnormal change node is automatically marked. Next, based on these abnormal nodes, time-series information is connected to construct a complete chain of abnormal feature changes. This chain consists of a series of abnormal nodes occurring at different times connected according to temporal relationships. In this way, abnormal changes in various components during the dairy beverage production process can be tracked and predicted, providing a basis for subsequent anomaly identification and prediction.

[0014] In some embodiments, the composition of the interactive target dairy beverage is analyzed for abnormal nodes to obtain an abnormal feature change chain, including: Based on the composition of the target dairy beverage, historical data of each component is extracted to identify abnormal nodes, abnormal probabilities, and abnormal offsets; the abnormal probabilities and abnormal offsets are matched and filtered according to the target tolerance threshold to determine abnormal filtering nodes; according to the temporal relationship of the abnormal filtering nodes, the temporal nodes of the abnormal probabilities and abnormal offsets are mapped and connected to obtain the abnormal feature change chain.

[0015] Specifically, firstly, historical data on the composition of the target dairy beverage is extracted from the dairy beverage production database. This data includes the proportion, concentration, production batch information, and other key production parameters of each component (such as protein, fat, and sugar), covering detailed records for each production cycle. Subsequently, statistical analysis of the historical data is performed to identify nodes where a certain component exhibits abnormal fluctuations within a specific time period. The identification of abnormal nodes can be based on a set threshold or the statistical distribution of historical data (such as standard deviation or mean). If the value of a certain component exceeds a set range (e.g., the component concentration exceeds a preset range), then that time point is considered an abnormal node. After identifying outlier nodes, the standard deviation of each component is calculated based on historical data. Then, the mean of the corresponding component is subtracted from the component value of each outlier node, and the difference is divided by the corresponding standard deviation to obtain a standard score. This standard score is compared with a standard normal distribution table to determine the probability of the outlier node. For example, if the standard score is 2.5, the cumulative probability corresponding to this standard score is approximately 0.9938, and subtracting 0.9938 from 1 gives an outlier probability of 0.0062. Furthermore, the absolute difference between the actual value of the component at the outlier node and the median of its corresponding normal range is calculated to obtain the outlier offset. Subsequently, to eliminate anomalies caused by minor deviations that do not affect product quality from the abnormal nodes, a tolerance threshold is set based on the quality standards of the target dairy beverage. This threshold represents the maximum allowable range of anomalies, including an anomaly probability tolerance threshold and an anomaly offset tolerance threshold. By comparing the anomaly probability and anomaly offset of each abnormal node with their corresponding tolerance thresholds, if the anomaly probability or offset of a certain component exceeds the tolerance threshold, the node is identified as an anomaly screening node. This screening process ensures that only significant anomalies that have a quality impact are further analyzed and processed. After identifying multiple anomaly screening nodes, these nodes are mapped and connected in chronological order (i.e., time sequence relationship) to construct a complete anomaly characteristic change chain. This change chain reflects the time sequence of anomaly node occurrences, thus providing dynamic trend information on anomaly fluctuations. For example, if two anomaly nodes occur close together and in the same production stage, they are connected, indicating that these anomalies may be caused by the same reason. Through the above process, the abnormal changes of each component during the dairy beverage production process can be clearly displayed, forming a complete anomaly characteristic change chain, providing accurate technical basis for subsequent anomaly identification, prediction, and compensation control.

[0016] In some embodiments, obtaining the abnormal feature change chain includes: Based on the historical data and the temporal relationship of the anomaly filtering nodes, anomaly feature analysis is performed to obtain anomaly appearance features. The temporal changes of these anomaly appearance features are then extrapolated according to the temporal relationship to determine the anomaly feature type, including instantaneous anomaly features and cumulative anomaly features. Instantaneous anomaly features are those that reach a preset variable threshold within a single moment, while cumulative anomaly features are those that do not reach the preset variable threshold but exhibit cumulative changes over multiple time periods. The instantaneous and cumulative anomaly features are then mapped to the anomaly feature change chain according to the correspondence of the anomaly filtering nodes.

[0017] Specifically, firstly, historical data of anomaly screening nodes is extracted from historical data, and these nodes are arranged sequentially according to the timestamps recorded in the historical data. After obtaining the temporal relationship of the anomaly screening nodes, in-depth analysis of the historical data of these nodes is conducted to extract the anomaly-related features. These anomaly features are also sorted according to the temporal relationship of the nodes. Anomaly features typically include surface bubble features of dairy beverages, liquid layering features of dairy beverages, color features of dairy beverages, and surface texture features of dairy beverages. These features can be extracted using a pre-trained convolutional neural network (CNN) to describe the changes in the production process when anomalies occur at specific nodes. In addition, methods other than convolutional neural networks can be used to extract features. For example, the surface bubble features of milk beverages can be separated from other parts of the beverage using image segmentation techniques (such as thresholding and K-means clustering). Bubbles are usually bright areas in the image, which can be distinguished from the background through segmentation. Morphological operations (such as dilation and erosion) can then be used to further enhance the bubble edges and extract the bubble shape information to help detect the size and shape of the bubbles. By using edge detection algorithms (such as Canny edge detection) to capture the boundaries of the bubbles, the distribution, number, and size of the bubbles can be analyzed. In addition, Hough Transform can be used for circularity detection, which can accurately identify the number and size of the bubbles. The liquid layering features of milk beverages can be identified by converting the image to HSV (hue, saturation, brightness) or Lab color space to identify the color differences of the liquid. Layered regions usually show different color transitions, and layering can be analyzed based on color differences. The surface texture features of milk beverages can be obtained by extracting surface uniformity using texture analysis methods (such as gray-level co-occurrence matrix). Subsequently, the extracted abnormal features are sorted and organized according to the time series to form a feature dataset for each time point. Each data point represents an abnormal feature value at a timestamp, such as the number of bubbles on the surface of a milk beverage, color changes, or the extent of liquid stratification at a certain time. By obtaining the feature value at each time point, the feature value is compared with the corresponding preset variable threshold. If the value of a feature at a certain time point exceeds the threshold, it is considered a transient anomaly and marked as a transient anomaly feature, as an anomaly feature type.Next, a specific time window (e.g., every 30 seconds or every minute) is selected to divide the sorted results of abnormal appearance features, resulting in multiple abnormal appearance feature segments. For abnormal appearance feature segments where the anomaly at a single moment does not reach the threshold, the change in abnormal features of that segment is calculated, such as the number of bubbles, bubble size, color change, etc. By summing the changes in abnormal features of that segment, it is assessed whether it meets the standard for cumulative anomaly. If a feature does not exceed the corresponding preset variable threshold individually in multiple abnormal appearance feature segments, but the sum of its changes exceeds the set cumulative anomaly threshold, it is considered a cumulative anomaly and marked as a cumulative anomaly feature, as an anomaly feature type. Then, the instantaneous anomaly features and cumulative anomaly features are mapped to an empty queue or linked list according to the temporal relationship of the anomaly screening nodes, forming an anomaly feature change chain. This anomaly feature change chain shows the entire process from the initial anomaly node to the final anomaly result, helping to track the source and development trend of anomaly events and providing data support for subsequent production adjustments and quality control.

[0018] The image acquisition module is connected to collect data on the entire production process of the target dairy beverage, and extract instantaneous image features and time-series image features.

[0019] Specifically, before monitoring the production process of the target dairy beverage, an image acquisition module is connected and activated. Once activated, the module uses cameras or other imaging devices to collect image data at different stages of the dairy beverage production line. The acquisition process covers all production stages, from raw material mixing, filling, homogenization to final packaging. This module uses continuous image capture to ensure the acquisition of visual information at every moment of the production process, providing sufficient image data for subsequent analysis. Subsequently, instantaneous image information is obtained by extracting instantaneous frames from the acquired images. Then, instantaneous image features are obtained by performing image feature recognition on the instantaneous image information, accurately reflecting the production status of the dairy beverage at that moment. Simultaneously, time-series image information is obtained by extracting continuous frames from the acquired images. Then, time-series image features are obtained by performing image feature recognition on the time-series image information, helping to determine whether the production process is within the normal range. Both types of image features include visible features such as surface bubble characteristics, liquid layering characteristics, color characteristics, surface texture characteristics, and bottle markings, providing data support for subsequent anomaly detection, quality control, and production adjustments.

[0020] In some embodiments, the image acquisition module is connected to capture the entire production process of the target dairy beverage, extracting instantaneous image features and time-series image features, including: The entire process of production migration according to the composition is image-acquired to obtain target image information; instantaneous frames are extracted from the target image information to obtain instantaneous image information; continuous frames are extracted from the target image information to obtain temporal image information; image feature recognition is performed on the instantaneous image information and temporal image information respectively to obtain instantaneous image features and temporal image features.

[0021] Specifically, in the production process of dairy beverages, images are acquired throughout the entire process based on each stage of the production line and the composition characteristics of the dairy beverage. The image acquisition module follows each stage of dairy beverage production, capturing image data to ensure comprehensive monitoring of all key processes (such as raw material mixing, filling, homogenization, and packaging). This process obtains target image information covering all production stages, providing rich image data for subsequent analysis. Subsequently, instantaneous image frames are extracted from the acquired target image information. These instantaneous frames correspond to images at every moment in the production process, typically one frame of image data captured per second or less. Each instantaneous frame represents the specific state of the dairy beverage production process at that moment, such as the color of the dairy beverage and the distribution of surface bubbles. By summarizing these instantaneous frames, instantaneous image information is obtained. In addition to instantaneous frames, continuous frames are extracted from the target image information according to time windows (e.g., every 30 seconds or every minute) to obtain temporal image information. This temporal image information refers to the dynamic change information of the production process extracted from multiple consecutive image frames, which can be used to track the state changes of the dairy beverage in different production time periods. Subsequently, the instantaneous and time-series image information obtained is enhanced by a micro-change amplification algorithm. The enhanced instantaneous and time-series image information is then input into the corresponding convolutional neural networks for image feature recognition. This yields instantaneous and time-series image features, including visible features such as surface bubble features, liquid layering features, color features, surface texture features, and bottle markings. This helps the system determine whether any abnormalities occur during the production process and provides data support for subsequent quality control.

[0022] In some embodiments, obtaining the instantaneous image features includes: The instantaneous image information is enhanced using a micro-change amplification algorithm; the enhanced instantaneous image information is then input into a convolutional neural network model to extract the instantaneous image features.

[0023] Specifically, for the acquired instantaneous image information, the first step is to use a micro-change amplification algorithm to enhance the image information, so as to more clearly identify subtle changes in the image. Instantaneous image information enhances minute brightness fluctuations, edge perturbations, and curvature differences in the image, thereby highlighting subtle changes in the image. This algorithm typically uses processing formulas to amplify brightness differences, edge information, and curvature changes in the image, in order to more accurately capture detailed changes in the production process. For example, it may amplify the edge information of the image by performing spatial gradient processing or enhance the less changed parts of the image using time derivative calculations, making previously imperceptible features more obvious. The details of the processed image will be enhanced, providing clearer and more reliable image data for subsequent feature extraction and anomaly detection. Subsequently, the enhanced instantaneous image information is input into a pre-trained convolutional neural network model for feature extraction. Convolutional neural networks are deep learning models, particularly adept at processing image data. This convolutional neural network model uses historical images and labeled instantaneous image features, and iteratively trains through steps such as forward propagation, loss calculation, backpropagation, and parameter optimization, and then evaluates it using validation data until a preset accuracy is met. After receiving the enhanced instantaneous image information, the convolutional neural network (CNN) model uses convolutional kernels (filters) to locally scan the image, extracting low-level features (such as edges, corners, and textures). Pooling layers then reduce computational complexity while preserving key image features. Through multiple convolutional and pooling layers, the CNN model can extract instantaneous image features including color, shape, texture, bubble size, and morphology. This helps the system understand anomalies in the dairy beverage production process and is used for subsequent anomaly identification and quality control.

[0024] In some embodiments, the instantaneous image information is subjected to image enhancement processing using a micro-change amplification algorithm, including: By processing the formula: Image enhancement processing is performed to amplify subtle brightness fluctuations, edge perturbations, and curvature differences. For gain coefficient, For time differential weights, Spatial gradient weights, The spatial gradient term represents the gradient change of the image in space. The time derivative term represents the rate of change of the image between consecutive frames. The original image intensity represents the pixel value of the original image at spatial location x and time point t. This represents the pixel values ​​of the enlarged image, indicating the enhanced output image.

[0025] Specifically, the micro-change amplification algorithm can enhance instantaneous image information through a processing formula. This formula is used to amplify subtle brightness fluctuations, edge perturbations, and curvature differences in the image. The specific processing formula is as follows: ,in, The gain coefficient is used to adjust the intensity of image enhancement. By adjusting the gain coefficient, the image enhancement effect can be amplified or reduced to ensure that small changes in the image are appropriately enhanced. The time differential weight is used to control the sensitivity of the image to changes between consecutive frames; Spatial gradient weights are used to control the sensitivity of the image to changes in spatial gradient. The spatial gradient term represents the gradient change of an image in space. It is commonly used to detect edge and texture features in an image. At the edges or textured parts of an image, the brightness changes more drastically. Using the spatial gradient term can highlight these changes and make the image edges more obvious. The time derivative term represents the rate of change of the image between consecutive frames. For rapid changes that occur during the production of dairy beverages, such as the formation of bubbles or changes in the color of the solution, the time derivative term can effectively amplify these rapid changes, making them easier for subsequent processing and analysis. It is the original image intensity, representing the original image pixel value at spatial location x and time point t. Each pixel value in the image corresponds to the light intensity or color value at that location. Through this pixel value, basic information such as the brightness and color of the image can be perceived. This represents the pixel values ​​of the enhanced output image after magnification. By processing the original image, minute brightness fluctuations and subtle edge changes are amplified to generate an enhanced image, making it easier to identify subtle anomalies or changes in the dairy beverage production process. This provides clearer and more accurate data support for subsequent feature extraction and anomaly detection.

[0026] Based on the abnormal feature change chain, instantaneous anomalies and cumulative anomalies are identified and predicted for the instantaneous image features and time-series image features, and instantaneous anomaly identification results and cumulative prediction results are obtained.

[0027] Specifically, after obtaining instantaneous image features and time-series image features, the instantaneous image features are compared with the instantaneous anomaly features of the corresponding nodes in the anomaly feature change chain to identify whether there are anomalies in the current instantaneous image features. If so, these anomalies are extracted to generate instantaneous anomaly identification results. After identifying an instantaneous anomaly, the time-series image features are also input into a Long Short-Term Memory (LSTM) network model for cumulative prediction of future time windows, obtaining cumulative prediction results. This LSTM network is constructed using historical time-series image features through the same training steps described above. Through these steps, the generated instantaneous anomaly identification results reflect the occurrence of immediate anomalies, while the cumulative prediction results show potential future anomalies. These results help the production line adjust operating parameters in a timely manner to ensure the quality and stability of the dairy beverage production process.

[0028] Based on the instantaneous anomaly identification results, the production process is located and tracked, and instantaneous monitoring results are generated and fed back.

[0029] Specifically, in the dairy beverage production process, after obtaining instantaneous anomaly identification results, these results are mapped to the workstation image acquisition number and bottle identification ID to generate instantaneous monitoring results. These results include detailed information such as the specific time, location, and type of the anomaly. For example, if excessive air bubbles appear in a dairy beverage bottle during filling at a certain time, the event will be recorded, and relevant monitoring results will be generated, indicating the bottle number, the workstation where the anomaly occurred, and the anomaly characteristics (such as the number of air bubbles). After generating the instantaneous monitoring results, these results are fed back to the on-site control system. A real-time alarm mechanism promptly alerts operators or the automated control system, prompting them to take necessary corrective measures to prevent the spread and escalation of quality problems, thereby improving the stability of the production process and product quality.

[0030] In some embodiments, production process location tracking is performed based on the instantaneous anomaly identification results, and instantaneous monitoring results are generated and fed back, including: Based on the workstation image acquisition number and the bottle identification ID in the instantaneous image features, the workstation and time location of the abnormal image frame are determined; based on the identified instantaneous anomaly identification result, it is mapped and combined with the workstation and time location to generate the instantaneous monitoring result; the instantaneous monitoring result is fed back to the field control system to trigger a location warning.

[0031] Specifically, the process begins by identifying the production stage where an anomaly occurred through the workstation image acquisition number. Each workstation and piece of equipment has a unique identifier, which pinpoints the specific production location where the anomaly occurred. Then, the bottle identification ID is obtained from the bottle markings in the instantaneous image features. This bottle identification ID serves as a unique identifier in dairy beverage production, determining which bottle the anomaly occurred on. Combined with timestamp information, the exact time and workstation of the anomaly event are accurately located, ensuring precise tracking of the anomaly information. After obtaining the workstation and time location information, this information is mapped and combined with the identified instantaneous anomaly results. That is, the identified anomaly features (such as the number of bubbles, layering, etc.) are combined with the location information and associated through time series, workstation number, and bottle identification ID to generate a complete instantaneous monitoring result. Finally, the generated instantaneous monitoring result is used to trigger an early warning mechanism, transmitting the anomaly data to operators or the automated control system. The on-site control system processes the received anomaly information in real time, including automatically adjusting production parameters, pausing the production line, and issuing alarms to prevent the spread of the anomaly and ensure the stability of dairy beverage production and product quality.

[0032] Based on the cumulative prediction results, production process disruption tracking is performed, and cumulative compensation monitoring results are generated and fed back.

[0033] Specifically, after obtaining the cumulative prediction results, the system analyzes them based on the aforementioned cumulative anomaly thresholds to determine if any risks may arise during production. For example, if the cumulative prediction results show that the color change of the milk beverage exceeds the corresponding cumulative anomaly threshold, or the number of bubbles exceeds the corresponding cumulative anomaly threshold, the production batch is considered to have a potential quality risk and requires process isolation. In this case, an isolation command is generated, automatically suspending the batch of milk beverage from entering downstream processes (such as ingredient preparation and filling) to prevent the problematic product from being further transmitted downstream and to prevent contamination or quality issues from spreading to mass production. After the isolation tracking mechanism is executed, cumulative compensation monitoring results are generated. These results include the bottle identification ID and compensation measures (such as suspending production or initiating a manual sampling and re-inspection process). The cumulative compensation monitoring results are fed back to the on-site control system and operators in real time to ensure that problems in the production process can be adjusted and compensated in a timely manner. The system also simultaneously marks abnormal situations for subsequent tracking and data recording, ensuring that the abnormalities do not recur, thereby effectively reducing production losses and ensuring the quality stability of the milk beverage.

[0034] In some embodiments, it also includes: A 3D line laser scanner is used to acquire point cloud images of the bottom of the storage tank, and the boundary and height data of the deposition area are extracted. Based on the height data at different time points, the deposition growth rate is calculated by point cloud difference to construct a deposition thickness change curve. When the growth rate of the deposition thickness change curve reaches a preset threshold, an equipment warning message is output.

[0035] Specifically, in the dairy beverage production process, a 3D line laser scanner is used to perform high-precision scanning of the bottom of the storage tank, acquiring point cloud image data of the tank bottom. The point cloud image reflects the geometry of the tank bottom, especially the shape and structure of the sedimentation area. By performing cluster analysis on the spatial distribution of the point cloud data, the sedimentation areas at the bottom of the tank are identified. This cluster analysis can be achieved through region growing algorithms or density-based spatial clustering methods (such as DBSCAN), grouping all densely distributed points into a group to determine the boundaries of the sedimentation. The sedimentation area is usually represented by an irregular boundary on the bottom surface, and the outline of the sedimentation can be defined through these boundary points. In addition, the thickness of the sediment at each location is calculated using the z-coordinate of each point in the point cloud data, forming the height data of the sedimentation area. Subsequently, at different time points, the sedimentation height data for each time point is extracted from the height data of the sedimentation area. By performing point cloud difference calculation on the height data of multiple time points, that is, by calculating the height difference between adjacent times at the same location, and then dividing the height difference by the time difference, the deposition growth rate is obtained, which is used to identify the thickness change of the sediment within a specific time period. Next, a deposition thickness variation curve is constructed by using time as the horizontal axis and deposition growth rate as the vertical axis. This curve illustrates the accumulation trend of sediments and helps predict whether sediments will exceed a predetermined threshold. Then, the constructed deposition thickness variation curve is monitored and analyzed in real time. Specifically, the deposition growth rate is calculated by dividing the difference in deposition growth rate between adjacent time periods by the time difference. The calculated deposition growth rate is then evaluated. If the deposition growth rate is greater than or equal to a preset threshold, it indicates rapid sediment growth, which may affect the normal operation of the storage tank and could even lead to blockages in the filling section, bacterial growth, or other equipment malfunctions. In this case, an automatic early warning mechanism is triggered, outputting equipment warning information, including shutdown suggestions or cleaning instructions. An alarm is also sent to operators or the automated control system to alert them to the risk of blockages in the filling section or bacterial growth, and appropriate measures are taken to ensure smooth production and minimize downtime and equipment damage risks.

[0036] In summary, the AI ​​vision-based dairy beverage production monitoring method provided by this invention has the following technical effects: The system analyzes the composition of the target dairy beverage to identify abnormal nodes and obtain abnormal feature change chains. It then connects to an image acquisition module to capture the entire production process of the target dairy beverage, extracting instantaneous and temporal image features. Based on the abnormal feature change chains, it identifies and predicts instantaneous and cumulative anomalies in the instantaneous and temporal image features, obtaining instantaneous anomaly identification results and cumulative prediction results. Based on the instantaneous anomaly identification results, it tracks and locates the production process, generating instantaneous monitoring results for feedback. Based on the cumulative prediction results, it tracks and tracks obstacles in the production process, generating cumulative compensation monitoring results for feedback. This achieves the technical effect of using AI vision to realize real-time dynamic monitoring and accurate anomaly prediction throughout the entire process, improving the monitoring effect of the production process and ensuring product quality stability.

[0037] Example 2, as Figure 2 This is a schematic diagram of the AI ​​vision-based dairy beverage production monitoring system of the present invention. For example, Figure 1 The flowchart of the AI ​​vision-based dairy beverage production monitoring method of this invention can be seen as follows: Figure 2 The structure shown is implemented.

[0038] Based on the same concept as the AI ​​vision-based dairy beverage production monitoring method in the embodiments described above, the present invention also provides an AI vision-based dairy beverage production monitoring system comprising: Anomaly Node Analysis Unit 11: Analyzes anomaly nodes based on the composition of the target dairy beverage to obtain anomaly feature change chains; Full-Process Acquisition Unit 12: Connects to the image acquisition module to collect the entire production process of the target dairy beverage, extracting instantaneous image features and time-series image features; Anomaly Recognition and Prediction Unit 13: Based on the anomaly feature change chains, identifies and predicts instantaneous and cumulative anomalies in the instantaneous and time-series image features, obtaining instantaneous anomaly recognition results and cumulative prediction results; Production Process Positioning and Tracking Unit 14: Tracks the production process based on the instantaneous anomaly recognition results, generating instantaneous monitoring results for feedback; Production Process Obstruction Tracking Unit 15: Tracks production process obstructions based on the cumulative prediction results, generating cumulative compensation monitoring results for feedback.

[0039] In some embodiments, the abnormal node parsing unit 11 includes: Based on the composition of the target dairy beverage, historical data of each component is extracted to identify abnormal nodes, abnormal probabilities, and abnormal offsets; the abnormal probabilities and abnormal offsets are matched and filtered according to the target tolerance threshold to determine abnormal filtering nodes; according to the temporal relationship of the abnormal filtering nodes, the temporal nodes of the abnormal probabilities and abnormal offsets are mapped and connected to obtain the abnormal feature change chain.

[0040] In some embodiments, the abnormal node parsing unit 11 includes: Based on the historical data and the temporal relationship of the anomaly filtering nodes, anomaly feature analysis is performed to obtain anomaly appearance features. The temporal changes of these anomaly appearance features are then extrapolated according to the temporal relationship to determine the anomaly feature type, including instantaneous anomaly features and cumulative anomaly features. Instantaneous anomaly features are those that reach a preset variable threshold within a single moment, while cumulative anomaly features are those that do not reach the preset variable threshold but exhibit cumulative changes over multiple time periods. The instantaneous and cumulative anomaly features are then mapped to the anomaly feature change chain according to the correspondence of the anomaly filtering nodes.

[0041] In some embodiments, the full-process acquisition unit 12 further includes: The entire process of production migration according to the composition is image-acquired to obtain target image information; instantaneous frames are extracted from the target image information to obtain instantaneous image information; continuous frames are extracted from the target image information to obtain temporal image information; image feature recognition is performed on the instantaneous image information and temporal image information respectively to obtain instantaneous image features and temporal image features.

[0042] In some embodiments, the full-process acquisition unit 12 includes: The instantaneous image information is enhanced using a micro-change amplification algorithm; the enhanced instantaneous image information is then input into a convolutional neural network model to extract the instantaneous image features.

[0043] In some embodiments, the full-process acquisition unit 12 includes: By processing the formula: Image enhancement processing is performed to amplify subtle brightness fluctuations, edge perturbations, and curvature differences. For gain coefficient, For time differential weights, Spatial gradient weights, The spatial gradient term represents the gradient change of the image in space. The time derivative term represents the rate of change of the image between consecutive frames. The original image intensity represents the pixel value of the original image at spatial location x and time point t. This represents the pixel values ​​of the enlarged image, indicating the enhanced output image.

[0044] In some embodiments, the production process location tracking unit 14 includes: Based on the workstation image acquisition number and the bottle identification ID in the instantaneous image features, the workstation and time location of the abnormal image frame are determined; based on the identified instantaneous anomaly identification result, it is mapped and combined with the workstation and time location to generate the instantaneous monitoring result; the instantaneous monitoring result is fed back to the field control system to trigger a location warning.

[0045] In some embodiments, the production process disruption tracking unit 15 includes: A 3D line laser scanner is used to acquire point cloud images of the bottom of the storage tank, and the boundary and height data of the deposition area are extracted. Based on the height data at different time points, the deposition growth rate is calculated by point cloud difference to construct a deposition thickness change curve. When the growth rate of the deposition thickness change curve reaches a preset threshold, an equipment warning message is output.

[0046] In embodiment three, the present invention also provides a computer-readable storage medium that can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the AI ​​vision-based dairy beverage production monitoring method in the embodiments of the present invention, thereby realizing the above-mentioned AI vision-based dairy beverage production monitoring method.

[0047] It should be understood that the embodiments disclosed in this invention and the above description enable those skilled in the art to implement this invention. However, this invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this invention.

Claims

1. A method for monitoring dairy beverage production based on AI vision, characterized in that, include: The composition of the interactive target dairy beverage is analyzed to identify abnormal nodes and obtain abnormal feature change chains. The image acquisition module is connected to collect data on the entire production process of the target dairy beverage, and extract instantaneous image features and time-series image features. Based on the abnormal feature change chain, instantaneous anomalies and cumulative anomalies are identified and predicted for the instantaneous image features and time-series image features to obtain instantaneous anomaly identification results and cumulative prediction results; Based on the instantaneous anomaly identification results, the production process is located and tracked, and instantaneous monitoring results are generated and fed back. Based on the cumulative prediction results, production process disruption tracking is performed, and cumulative compensation monitoring results are generated and fed back.

2. The AI ​​vision-based method for monitoring dairy beverage production according to claim 1, characterized in that, The composition of the interactive target dairy beverage is analyzed for abnormal nodes to obtain abnormal feature change chains, including: Based on the composition of the target milk beverage, historical data of each component is extracted to identify abnormal nodes, abnormal probabilities, and abnormal offsets. Based on the target tolerance threshold, anomaly probability and anomaly offset are used for anomaly matching and filtering to determine anomaly filtering nodes; Based on the temporal relationship of the anomaly screening nodes, the anomaly probability and anomaly offset temporal nodes are mapped and connected to obtain the anomaly feature change chain.

3. The method for monitoring dairy beverage production based on AI vision according to claim 2, characterized in that, Obtaining the abnormal feature change chain includes: Based on the historical data and the temporal relationship of the anomaly filtering nodes, anomaly feature analysis is performed to obtain anomaly appearance features. The abnormal appearance features are analyzed according to the temporal relationship to determine the abnormal feature type, including instantaneous abnormal features and cumulative abnormal features. Instantaneous abnormal features are features that reach a preset variable threshold within a single moment, while cumulative abnormal features are features that do not reach the preset variable threshold within multiple time periods but have cumulative changes. The instantaneous abnormal features and cumulative abnormal features are mapped to the abnormal feature change chain according to the correspondence of the abnormal filtering nodes.

4. The AI ​​vision-based method for monitoring dairy beverage production according to claim 2, characterized in that, The image acquisition module is connected to capture the entire production process of the target dairy beverage, extracting instantaneous image features and time-series image features, including: The entire process of production migration based on component composition is image-acquired to obtain target image information; Instantaneous frame extraction is performed on the target image information to obtain instantaneous image information; The target image information is subjected to continuous frame extraction to obtain temporal image information; Image feature recognition is performed on the instantaneous image information and the time-series image information respectively to obtain the instantaneous image features and the time-series image features.

5. The AI ​​vision-based method for monitoring dairy beverage production according to claim 4, characterized in that, Obtaining the instantaneous image features includes: The instantaneous image information is enhanced using a micro-change amplification algorithm; The enhanced instantaneous image information is input into a convolutional neural network model to extract the instantaneous image features.

6. The AI ​​vision-based method for monitoring dairy beverage production according to claim 5, characterized in that, The instantaneous image information is subjected to image enhancement processing using a micro-change amplification algorithm, including: By processing the formula: Image enhancement processing is performed to amplify subtle brightness fluctuations, edge perturbations, and curvature differences. For gain coefficient, For time differential weights, Spatial gradient weights, The spatial gradient term represents the gradient change of the image in space. The time derivative term represents the rate of change of the image between consecutive frames. The original image intensity represents the pixel value of the original image at spatial location x and time point t. This represents the pixel values ​​of the enlarged image, indicating the enhanced output image.

7. The method for monitoring dairy beverage production based on AI vision according to claim 6, characterized in that, Based on the instantaneous anomaly identification results, the production process is located and tracked, and instantaneous monitoring results are generated and fed back, including: Based on the workstation image acquisition number and the bottle identification ID in the instantaneous image features, the workstation and time location of the image frame where the anomaly occurred are determined; Based on the identified instantaneous anomaly, the results are mapped and combined with the workstation and time location to generate the instantaneous monitoring results; The instantaneous monitoring results are fed back to the field control system to trigger a positioning warning.

8. The method for monitoring dairy beverage production based on AI vision according to claim 1, characterized in that, Also includes: A 3D line laser scanner was used to acquire point cloud images of the bottom of the storage tank, and the boundary and height data of the sedimentation area were extracted. Based on the height data at different time points, the deposition growth rate is calculated by point cloud difference, and a deposition thickness variation curve is constructed. When the rate of increase of the deposition thickness variation curve reaches a preset threshold, the device outputs a warning message.

9. A dairy beverage production monitoring system based on AI vision, characterized in that, The method for monitoring dairy beverage production based on AI vision as described in any one of claims 1-8 includes: Anomaly Node Analysis Unit: The composition of the interactive target milk beverage is analyzed for anomaly nodes to obtain the chain of changes in anomaly features; Full-process acquisition unit: Connects to the image acquisition module to acquire the entire production process of the target dairy beverage, and extracts instantaneous image features and time-series image features; Anomaly identification and prediction unit: Based on the anomaly feature change chain, it performs instantaneous anomaly and cumulative anomaly identification and prediction on the instantaneous image features and time-series image features to obtain instantaneous anomaly identification results and cumulative prediction results; Production process positioning and tracking unit: performs production process positioning and tracking based on the instantaneous anomaly identification results, and generates instantaneous monitoring results for feedback; Production process obstruction tracking unit: Based on the cumulative prediction results, it performs production process obstruction tracking and generates cumulative compensation monitoring results for feedback.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the AI ​​vision-based dairy beverage production monitoring method as described in any one of claims 1 to 8.

Images