Food production control method, system and storage medium based on artificial intelligence

Abstract

This invention discloses an artificial intelligence-based food production control method, system, and storage medium, belonging to the field of factory control. The method includes: dividing the production process into production nodes and verification nodes; acquiring the operating status of food production equipment and scene images of its location; identifying the scene images to obtain the distance between production personnel and food production equipment; generating a visualization chart for each production node based on the distance and operating status; calculating the sub-efficiency value of each production node and the overall efficiency value of the production process based on the visualization chart; using a recognition model to identify food images, counting the number of times the recognition model identifies foreign objects, and calculating the risk value for the analysis period based on the number of times; generating a first suggestion and a second suggestion for optimizing the production process based on the overall efficiency value and the risk value. Managers can use this invention to understand the employee efficiency value and the risk value of foreign objects appearing in food in real time, thereby helping them to optimize the food production process.

Description

Technical Field

[0001] This invention belongs to the field of factory production control technology, specifically relating to a food production control method, system, and storage medium based on artificial intelligence. Background Technology

[0002] With increasing awareness of food safety and rising consumer demands for food quality, traditional food producers face challenges such as low production efficiency, difficulty in quality control, and high resource consumption. With the development of artificial intelligence, optimizing and controlling food production efficiency and safety through AI has become a future research direction in this field.

[0003] To integrate artificial intelligence (AI) technology with food production, existing technologies have proposed methods such as the following: For example, CN116681357A discloses an AI-based food quality analysis system. In this system, an improvement and optimization module receives collected food data, feature-combined food data, and analyzed food data. Based on these data, it performs improvements and optimizations to generate a more scientifically grounded food plan. Another example is CN117078105A, which discloses an AI-based production quality monitoring method and system. This method can analyze production process data, product quality data, and production environment data in real time, quickly identifying potential quality problems. By analyzing historical and real-time data, AI can predict potential quality problems and take preventative measures. Furthermore, it can automatically optimize production parameters to maximize production efficiency and product quality.

[0004] However, for the production of pasta and semi-cooked foods, the food formula and production environment data are relatively constant, and the production process relies on a large amount of manual operation. Therefore, the above-mentioned existing technologies cannot be applied to the optimization of this production process. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an artificial intelligence-based food production control method, system, and storage medium to help relevant personnel optimize the production process of semi-cooked foods.

[0006] To achieve the aforementioned objectives, this invention proposes an artificial intelligence-based food production control method, comprising:

[0007] Based on the food production process, a production flow is generated, which is divided into production nodes and verification nodes. The production nodes correspond to food production equipment, and the verification nodes correspond to X-ray inspection machines.

[0008] The system acquires the operating status of food production equipment and scene images of its location, with the scene images having a capture time, and marks the corresponding production node for each scene image;

[0009] The scene image captured within the analysis time period is defined as the target image. The target image is identified to obtain the distance between the production personnel and the food production equipment. A visualization chart of each production node is generated based on the distance and the operating status.

[0010] The sub-efficiency value of each production node is calculated based on the visualization chart, and the total efficiency value of the production process is calculated based on the efficiency values ​​of all production nodes.

[0011] The process involves acquiring food photos taken by the verification node during the analysis period, using a recognition model to identify the food images, counting the number of times the recognition model identifies foreign objects, and calculating the risk value for the analysis period based on the number of times the foreign objects are identified.

[0012] If the overall efficiency value is less than the first threshold, a first suggestion for optimizing the production process is generated; if the risk value is greater than the second threshold, a second suggestion for optimizing the production process is generated.

[0013] Furthermore, generating the visualization chart based on the distance and the operational data includes the following steps:

[0014] A third threshold is set. The operating state includes a start state and a stop state. When the food production equipment is in the start state, if the distance in the target image is greater than or equal to the third threshold, the target image is classified as a first state; otherwise, it is classified as a second state. When the food production equipment is in the stop state, if the distance is greater than or equal to the third threshold, the target image is classified as a third state; otherwise, it is classified as a fourth state.

[0015] The visualization chart includes a timeline, a first bar chart, and a second bar chart. The first bar chart and the second bar chart include the names of the production nodes. The first bar chart includes the duration of the start-up state and the stop state. The second bar chart includes the duration of the various states of the target image.

[0016] Furthermore, calculating the overall efficiency value of the production process includes the following steps:

[0017] Locate the first region occupied by the start state in the first bar chart, define the region between the two first regions as the second region, count the first time length of the second region appearing in the first bar chart, locate the third region in the second bar chart, the third region is the region occupied by the third state, and the region before the third region is the region occupied by the second state, filter out the fourth region from the third region that has time overlap with the second region, and count the second time length of the fourth region appearing in the second bar chart;

[0018] Calculate the ratio of the first time length to the second time length for each production node, normalize the ratio for each production node to obtain the sub-efficiency value for each production node, calculate the average of all sub-efficiency values ​​for all production nodes, and use the average value as the total efficiency value of the production process.

[0019] Furthermore, generating the first suggestion and the second suggestion includes the following steps:

[0020] Calculate the first difference between the sub-efficiency value and the total efficiency value of each production node, define the production node whose first difference is less than the fifth threshold as the target node, generate the first suggestion for focusing on the work efficiency of the personnel at the target node, obtain the time point of foreign object appearance, determine the degree of dispersion of the appearance of the time point, if the degree of dispersion is large, generate the second suggestion for checking the raw materials, if the degree of dispersion is small, check each production node.

[0021] Furthermore, determining the degree of dispersion of the occurrence of the aforementioned time points includes the following steps:

[0022] Calculate the second difference between adjacent time points, arrange the second difference in ascending order into a numerical sequence, calculate the variance and median of the numerical sequence, if the variance is less than the sixth threshold and the median is greater than the seventh threshold, then the dispersion is judged as small; otherwise, it is judged as large.

[0023] Further, inspecting each of the production nodes includes the following steps:

[0024] A discrimination model is established, which includes a filtering model and a comparison model. The target image is input into the filtering model to obtain a filtered image. The target image and the filtered image are input into the comparison model. The comparison model calculates the similarity between the target image and the filtering model. If the similarity is less than a fourth threshold, the production node operation is determined to be non-standard.

[0025] Furthermore, establishing the training set for the filtering model includes the following steps:

[0026] A first image library and a second image library are established. The first image library includes a variety of standard wearable images, and the second image library includes a variety of occlusion patterns. The occlusion patterns in the second image library are used to cover the standard wearable images in the first image library to generate abnormal images. A training set is constructed based on all the standard wearable images and the abnormal images in the first image library.

[0027] Furthermore, constructing the filtering model includes the following steps:

[0028] A basic model is constructed based on a neural network, generating a first training set and a second training set. The first training set includes two identical standard wear images, and the second training set includes the anomalous image and the standard wear image used to generate the anomalous image. The basic model is trained using the first training set and the second training set, and the trained basic model is defined as the filtering model.

[0029] This invention also provides an artificial intelligence-based food production control system, which implements the aforementioned artificial intelligence-based food production control method. The system includes:

[0030] The acquisition module is used to acquire the operating status of food production equipment and scene images of the location. The scene images have the shooting time and the corresponding production node is marked for each scene image. The production process is generated based on the food production process and the production process is divided into the production nodes and verification nodes. The production nodes correspond to food production equipment and the verification nodes correspond to X-ray inspection machines.

[0031] The visualization module defines the scene image captured within the analysis time period as the target image, identifies the target image to obtain the distance between production personnel and food production equipment, and generates a visualization chart for each production node based on the distance and the operating status.

[0032] The calculation module calculates the sub-efficiency value of each production node based on the visualization chart, and calculates the total efficiency value of the production process based on the efficiency values ​​of all production nodes.

[0033] The identification module acquires food photos taken by the verification node during the analysis period, uses an identification model to identify the food images, counts the number of times the identification model identifies foreign objects, and calculates the risk value for the analysis period based on the number of times.

[0034] The suggestion module generates a first suggestion to optimize the production process if the total efficiency value is less than a first threshold, and generates a second suggestion to optimize the production process if the risk value is greater than a second threshold.

[0035] The present invention also discloses a computer storage medium storing program instructions, wherein the program instructions, when executed, control the device where the computer storage medium is located to perform the method described above.

[0036] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0037] This invention divides the production process into production nodes and verification nodes. It calculates sub-efficiency values ​​and the overall efficiency value of the entire production process by analyzing the operating status of equipment at each production node and the distance between employees and the food production equipment. It calculates the risk value of foreign objects in the food by analyzing the number of times foreign objects are detected at the verification nodes. By displaying these values ​​to management personnel, they can understand which production node has a lower efficiency value and which periods have a higher frequency of foreign object occurrences. After understanding these values, management personnel can more effectively optimize the food production process. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the steps of the artificial intelligence-based food production control method of the present invention.

[0039] Figure 2 This is a schematic diagram illustrating the principle of the visualization chart of this invention;

[0040] Figure 3 This is a schematic diagram of the structure of the artificial intelligence-based food production control system of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] It is understood that the terms "first," "second," etc., used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are used only to distinguish one element from another. For example, without departing from the scope of this application, a first script may be referred to as a second script, and similarly, a second script may be referred to as a first script.

[0043] like Figure 1 As shown, an artificial intelligence-based food production control method includes:

[0044] S1: Generate a production process based on the food production process, and divide the production process into production nodes and verification nodes. Production nodes correspond to food production equipment, and verification nodes correspond to X-ray inspection machines.

[0045] The food production process in this embodiment includes mincing meat, mixing auxiliary materials, marinating, filling, heat processing, and X-ray inspection. Except for X-ray inspection, all other nodes are production nodes, X-ray inspection is a verification node, the food production equipment used for mixing is a mixer, and the X-ray inspection uses an X-ray inspection machine. Other production nodes will not be listed one by one.

[0046] S2: Obtain the operating status of the food production equipment and the scene images of its location. The scene images have the shooting time, and the corresponding production node is marked for each scene image.

[0047] The operating status includes start-up and stop status. The scene image is an overall image of the workshop captured by the camera, which includes food production equipment and related operators. Each scene image also has the attribute of the shooting time and the attribute of which production node the scene image corresponds to. By assigning attributes to the scene images, it is easier to capture and analyze data later.

[0048] S3: Define scene images captured within the analysis time period as target images, identify target images to obtain the distance between production personnel and food production equipment, and generate a visualization chart for each production node based on the distance and operating status.

[0049] S4: Calculate the sub-efficiency value of each production node based on the visualization chart, and calculate the total efficiency value of the production process based on the efficiency values ​​of all production nodes.

[0050] Specifically, the analysis period is set from 8:00 AM to 6:00 PM. Scene images acquired between 8:00 AM and 6:00 PM are used as the target images for analysis. In this embodiment, a CNN deep learning network is used to identify the target images, thereby obtaining the first contour representing the production personnel and the second contour representing the food production equipment. The centroids of the first and second contours are then located, and the distance between the two centroids is taken as the distance between the production personnel and the food production equipment. Based on the distance and the operating status of the food production equipment, a visualization chart is generated for that equipment, i.e., the production node. The generation method and specific form of the visualization chart will be introduced later. The lower the sub-efficiency value of a production node, the higher its working efficiency. After calculating the sub-efficiency value of each production node, the overall efficiency value of the entire production process is also calculated. By displaying the overall efficiency value, production personnel can easily understand the working efficiency of the entire production process.

[0051] S5: Obtain food photos taken by the verification node during the analysis period, use the recognition model to identify the food images, count the number of times the recognition model identifies foreign objects, and calculate the risk value for the analysis period based on the number of times.

[0052] S6: If the overall efficiency value is less than the first threshold, generate the first suggestion for optimizing the production process; if the risk value is greater than the second threshold, generate the second suggestion for optimizing the production process.

[0053] The recognition model is also a CNN deep learning network model. After the food is prepared, an X-ray inspection machine takes pictures of the food to obtain X-ray images. Then, the recognition model is used to identify the X-ray images to determine whether there are foreign objects, such as hair, paper scraps, and various particles. If a foreign object is found, it is counted in a counter. Then, the number of times a foreign object is detected during the analysis period is counted, for example, 25, and the total number of food photos identified is counted, for example, 500. In this embodiment, when calculating the risk value, the ratio of the number of times a foreign object appears to the total number of food photos is first calculated, and then multiplied by the corresponding weight to obtain the risk value. Here, the weight is set to 10. For example, the risk value in this case is 25 / 500*10 = 0.5.

[0054] The first threshold is set to 0.8, and the second threshold is set to 0.4. When the calculated efficiency value is less than 0.8, the system generates a first suggestion to pay attention to the employee's production efficiency. If the risk value is greater than 0.4, a second suggestion is generated to check the production raw materials or production process.

[0055] This invention divides the production process into production nodes and verification nodes. It calculates sub-efficiency values ​​and the overall efficiency value of the entire production process by analyzing the operating status of equipment at each production node and the distance between employees and the food production equipment. It calculates the risk value of foreign objects in the food by analyzing the number of times foreign objects are detected at the verification nodes. By displaying these values ​​to management personnel, they can understand which production node has a lower efficiency value and which periods have a higher frequency of foreign object occurrences. After understanding these values, management personnel can more effectively optimize the food production process.

[0056] Of particular note is that managers can use this invention to understand employee efficiency values ​​and the risk of foreign objects appearing in food in real time, which can help them optimize the food production process.

[0057] This embodiment generates a visualization chart based on distance and operational data, including the following steps:

[0058] A third threshold is set. The operating states include start-up and stop states. When the food production equipment is in the start-up state, if the distance in the target image is greater than or equal to the third threshold, the target image is classified as the first state; otherwise, it is classified as the second state. When the food production equipment is in the stop state, if the distance is greater than or equal to the third threshold, the target image is classified as the third state; otherwise, it is classified as the fourth state.

[0059] Before generating the visualization, a third threshold is first set, for example, 2 meters. If the distance between the operator and the food production equipment in the target image exceeds 2 meters and the equipment is running, the target image is marked as the first state, indicating that the operator is not near the equipment while the machine is running. If the distance is less than 2 meters and the equipment is running, the target image is marked as the second state, indicating that the operator is monitoring the equipment while it is running. If the distance between the operator and the food production equipment in the target image exceeds 2 meters and the equipment is stopped, the target image is marked as the third state, indicating that the operator is not near the equipment when the machine is stopped. If the distance is less than 2 meters and the equipment is stopped, the target image is marked as the fourth state.

[0060] The visualization charts include a timeline, a first bar chart, and a second bar chart. The first and second bar charts include the names of the production nodes. The first bar chart includes the duration of the start and stop states, and the second bar chart includes the duration of various states of the target image.

[0061] Since each target image has a capture time, the duration of each state can be generated. For example, target image 1 is the first state, and the capture time is 8:00. Subsequent images up to target image 20 are all in the first state, and the capture time of target image 20 is 8:20. The duration of the first state can be calculated to be 20 minutes. The calculation method for other states is the same, and no further examples will be given here.

[0062] like Figure 2 The diagram shows a visualization, including a first bar chart (C1) and a second bar chart (C2). The first bar chart (C1) includes the durations of the start-up state (A1, A3, and A5) and the durations of the stop state (A2 and A4). The second bar chart (C2) includes the durations of the first state (B1), the second state (B2), the third state (B3), and the fourth state (B4). The remaining portions are unlabeled. This visualization allows for a more detailed display of the duration of each state within the analysis period. It provides insight into the specific working status of the production nodes and enables the calculation of their efficiency values.

[0063] This embodiment calculates the overall efficiency value of the production process using the following steps:

[0064] Locate the first region occupied by the start state in the first bar chart, define the region between the two first regions as the second region, count the first time length of the second region appearing in the first bar chart, locate the third region in the second bar chart, the third region is the region occupied by the third state, and the region before the third region is the region occupied by the second state, filter out the fourth region from the third region that has time overlap with the second region, and count the second time length of the fourth region appearing in the second bar chart.

[0065] Calculate the ratio of the first time length to the second time length for each production node. Normalize the ratio of the production nodes to obtain the sub-efficiency value of each production node. Calculate the average of the sub-efficiency values ​​of all production nodes and use the average value as the total efficiency value of the production process.

[0066] Continue to refer to Figure 2 First, locate the first regions A1, A3, and A5 occupied by the start state in the first bar chart C1. Then, locate the second regions A2 and A4 between them. Then, calculate the first time length occupied by the second regions A2 and A4. Here, only the second region located between the two first regions is located. The reason is that if the first bar chart shows a stop state at the beginning or end, it may be due to the employee's personal reasons for not arriving on time or leaving early. By removing such regions, the time of such regions will no longer be calculated in the future, thus improving the humanization level of system management.

[0067] In the second bar chart C2, locate the third region B3, which represents the state where the equipment is stopped and personnel are not nearby. Prior to the third region is region B2, which represents the state where the equipment is running and personnel are not nearby. The third region actually appears when personnel leave the equipment during operation and do not return when the equipment stops. Finally, calculate the ratio of the first time length to the second time length. The second time length will always be less than or equal to the first time length, and the larger the second time length is used as the denominator, the smaller the calculated ratio. Finally, normalize the ratios of all production nodes and use the normalized values ​​as the sub-efficiency values ​​for each production node. After obtaining the sub-efficiency values, calculate the average of all sub-efficiency values ​​and use the average as the total efficiency value.

[0068] This embodiment generates the first and second recommendations through the following steps:

[0069] Calculate the first difference between the sub-efficiency value of each production node and the total efficiency value. Define the production node whose first difference is less than the fifth threshold as the target node. Generate the first suggestion for focusing on the work efficiency of the personnel at the target node. Obtain the time point when the foreign object appears and judge the degree of dispersion of the time point. If the dispersion is large, generate the second suggestion for checking the raw materials. If the dispersion is small, check each production node.

[0070] Specifically, firstly, a first difference is calculated between the sub-efficiency value of each production node and the overall efficiency value of the production process. A negative first difference indicates that the sub-efficiency value of the production node is lower than the overall average. Furthermore, if the negative first difference is too large, it indicates low efficiency for that production node. This first suggestion serves to remind relevant personnel to pay attention to the work efficiency of employees at this node. Before generating the second suggestion, the discrete values ​​of the times when foreign objects appear are calculated. If the discrete values ​​are too large, it indicates that the times when foreign objects appear are scattered, which may be due to employee negligence. Therefore, a second suggestion is generated to monitor the employee's production process. If the discrete values ​​are too small, it indicates that the times when foreign objects appear are concentrated, which may be due to poor raw material handling. Therefore, a second suggestion is generated to monitor the raw materials.

[0071] This embodiment determines the degree of dispersion of time points by the following steps:

[0072] Calculate the second difference between adjacent time points, arrange the second difference in ascending order into a numerical sequence, calculate the variance and median of the numerical sequence. If the variance is less than the sixth threshold and the median is greater than the seventh threshold, the dispersion is judged as small; otherwise, it is judged as large.

[0073] If the variance is small, it indicates that the time intervals between time points are relatively even. If the median is large, it indicates that the time points are relatively evenly distributed throughout the analysis period. In this case, the dispersion of the time points is judged as small. If this condition is not met, the dispersion of the time points is judged as large.

[0074] This embodiment checks each production node, including the following steps:

[0075] A discrimination model is established, which includes a filtering model and a comparison model. The target image is input into the filtering model to obtain a filtered image. The target image and the filtered image are input into the comparison model. The comparison model calculates the similarity between the target image and the filtering model. If the similarity is less than the fourth threshold, the production node operation is judged to be non-standard.

[0076] This embodiment uses an automated method to detect employee production processes. Specifically, two models are set up. After the target image is input into the filtering model, if the personnel in the target image are not wearing protective clothing properly, such as not wearing hats and gloves, the filtering model will not show the part they are wearing, thus obtaining a filtered image. Then, the recognition model compares the filtered image with the target image before it was input into the filtering model and calculates the similarity between the two. If the similarity is less than the fourth threshold, for example, 90%, it is considered that the clothing is not properly worn.

[0077] Traditional identification methods involve building a database of non-standard clothing images, training a discrimination model using these images, and then identifying the target image, only recognizing the non-standard clothing postures trained on. This invention, however, first filters out the standard clothing portion during identification, as images of standard clothing are relatively fixed, and then compares it with the images before filtering. This significantly improves the model's applicability and accuracy. Notably, both the filtering and comparison models can be constructed using CNN deep learning algorithms.

[0078] The steps for establishing the training set for the filtering model in this embodiment are as follows:

[0079] A first image library and a second image library are established. The first image library includes a variety of standard wearable images, and the second image library includes a variety of occlusion patterns. The occlusion patterns in the second image library are used to cover the standard wearable images in the first image library to generate abnormal images. A training set is constructed based on all the standard wearable images and abnormal images in the first image library.

[0080] The occlusion patterns include various patterns or solid color frames, or various common occlusion objects. Then, an occlusion pattern is randomly selected and placed on the standard wearable image to generate an abnormal image. The occlusion position can be a hand, head, or other position. Then, the standard wearable image and the abnormal image are combined into a training set.

[0081] This embodiment constructs the filtering model through the following steps:

[0082] A basic model is built based on a neural network, generating a first training set and a second training set. The first training set includes two identical standard wear images, and the second training set includes an abnormal image and a standard wear image used to generate the abnormal image. The basic model is trained using the first training set and the second training set, and the trained basic model is defined as a filtering model.

[0083] First, the model is trained using the same standard wear image, so that when the base model receives a standard wear image without occlusion, it outputs the image directly without processing it. Then, the model is trained using anomalous images and standard wear images that generate anomalous images, so that when the base model receives a standard wear image with occlusion, it filters out the occlusion and outputs only the standard wear portion.

[0084] In this embodiment, the number of the first training set and the second training set are the same. This ensures that the base model receives balanced training.

[0085] like Figure 3 As shown, the present invention also provides an artificial intelligence-based food production control system, which is used to implement the above-mentioned artificial intelligence-based food production control method. The system includes:

[0086] The acquisition module is used to acquire the operating status of food production equipment and scene images of the location. The scene images have the shooting time and the corresponding production node is marked for each scene image. The production process is generated based on the food production process and the production process is divided into production nodes and verification nodes. The production nodes correspond to food production equipment and the verification nodes correspond to X-ray inspection machines.

[0087] The visualization module defines scene images captured within the analysis period as target images, identifies target images to obtain the distance between production personnel and food production equipment, and generates visualization charts for each production node based on the distance and operating status.

[0088] The calculation module calculates the sub-efficiency value of each production node based on a visual chart, and calculates the total efficiency value of the production process based on the efficiency values ​​of all production nodes.

[0089] The identification module acquires food photos taken by the verification node during the analysis period, uses the identification model to identify the food images, counts the number of times the identification model identifies foreign objects, and calculates the risk value for the analysis period based on the number of times.

[0090] The suggestion module generates a first suggestion to optimize the production process if the overall efficiency value is less than the first threshold, and a second suggestion to optimize the production process if the risk value is greater than the second threshold.

[0091] The present invention also discloses a computer storage medium storing program instructions, wherein the program instructions, when executed, control the device where the computer storage medium is located to perform the method described above.

[0092] It should be understood that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A food production control method based on artificial intelligence, characterized in that, include: Based on the food production process, a production flow is generated, which is divided into production nodes and verification nodes. The production nodes correspond to food production equipment, and the verification nodes correspond to X-ray inspection machines. The system acquires the operating status of food production equipment and scene images of its location, with the scene images having a capture time, and marks the corresponding production node for each scene image; The scene image captured within the analysis time period is defined as the target image. The target image is identified to obtain the distance between the production personnel and the food production equipment. A visualization chart of each production node is generated based on the distance and the operating status. The sub-efficiency value of each production node is calculated based on the visualization chart, and the total efficiency value of the production process is calculated based on the efficiency values ​​of all production nodes. The process involves acquiring food photos taken by the verification node during the analysis period, using a recognition model to identify the food images, counting the number of times the recognition model identifies foreign objects, and calculating the risk value for the analysis period based on the number of times the foreign objects are identified. If the overall efficiency value is less than the first threshold, a first suggestion for optimizing the production process is generated; if the risk value is greater than the second threshold, a second suggestion for optimizing the production process is generated.

2. The method according to claim 1, characterized in that, Generating the visualization chart based on the distance and the operating status includes the following steps: A third threshold is set. The operating state includes a start state and a stop state. When the food production equipment is in the start state, if the distance in the target image is greater than or equal to the third threshold, the target image is classified as a first state; otherwise, it is classified as a second state. When the food production equipment is in the stop state, if the distance is greater than or equal to the third threshold, the target image is classified as a third state; otherwise, it is classified as a fourth state. The visualization chart includes a timeline, a first bar chart, and a second bar chart. The first bar chart and the second bar chart include the names of the production nodes. The first bar chart includes the duration of the start-up state and the stop state. The second bar chart includes the duration of various states of the target image.

3. The method according to claim 2, characterized in that, Calculating the overall efficiency of a production process involves the following steps: Locate the first region occupied by the start state in the first bar chart, define the region between the two first regions as the second region, count the first time length of the second region appearing in the first bar chart, locate the third region in the second bar chart, the third region is the region occupied by the third state, and the region before the third region is the region occupied by the second state, filter out the fourth region from the third region that has time overlap with the second region, and count the second time length of the fourth region appearing in the second bar chart; Calculate the ratio of the first time length to the second time length for each production node, normalize the ratio for each production node to obtain the sub-efficiency value for each production node, calculate the average of all sub-efficiency values ​​for all production nodes, and use the average value as the total efficiency value of the production process.

4. The method according to claim 3, characterized in that, Generating the first and second recommendations includes the following steps: Calculate the first difference between the sub-efficiency value and the total efficiency value of each production node, define the production node whose first difference is less than a fifth threshold as a target node, generate a first suggestion for focusing on the work efficiency of the personnel at the target node, obtain the time point of foreign object appearance, determine the degree of dispersion of the appearance of the time point, if the degree of dispersion is large, generate a second suggestion for checking raw materials, if the degree of dispersion is small, check each production node.

5. The method according to claim 4, characterized in that, Determining the degree of dispersion of the time points includes the following steps: Calculate the second difference between adjacent time points, arrange the second difference in ascending order into a numerical sequence, calculate the variance and median of the numerical sequence, if the variance is less than the sixth threshold and the median is greater than the seventh threshold, then the dispersion is judged as small; otherwise, it is judged as large.

6. The method according to claim 4, characterized in that, Inspecting each of the production nodes includes the following steps: A discrimination model is established, which includes a filtering model and a comparison model. The target image is input into the filtering model to obtain a filtered image. The target image and the filtered image are input into the comparison model. The comparison model calculates the similarity between the target image and the filtering model. If the similarity is less than a fourth threshold, the production node operation is determined to be non-standard.

7. The method according to claim 6, characterized in that, Building the training set for the filtering model includes the following steps: A first image library and a second image library are established. The first image library includes a variety of standard wearable images, and the second image library includes a variety of occlusion patterns. The occlusion patterns in the second image library are used to cover the standard wearable images in the first image library to generate abnormal images. A training set is constructed based on all the standard wearable images and the abnormal images in the first image library.

8. The method according to claim 7, characterized in that, Constructing the filtering model includes the following steps: A basic model is constructed based on a neural network, generating a first training combination and a second training combination. The first training combination includes two identical standard wearable images, and the second training combination includes the abnormal image and the standard wearable image used to generate the abnormal image. The basic model is trained using the first training combination and the second training combination, and the trained basic model is defined as the filtering model.

9. An artificial intelligence-based food production control system, used to implement the method as described in any one of claims 1-8, characterized in that, include: The acquisition module is used to acquire the operating status of food production equipment and scene images of the location. The scene images have the shooting time and the corresponding production node is marked for each scene image. The production process is generated based on the food production process and the production process is divided into the production nodes and verification nodes. The production nodes correspond to food production equipment and the verification nodes correspond to X-ray inspection machines. The visualization module defines the scene image captured within the analysis time period as the target image, identifies the target image to obtain the distance between production personnel and food production equipment, and generates a visualization chart for each production node based on the distance and the operating status. The calculation module calculates the sub-efficiency value of each production node based on the visualization chart, and calculates the total efficiency value of the production process based on the efficiency values ​​of all production nodes. The identification module acquires food photos taken by the verification node during the analysis period, uses an identification model to identify the food images, counts the number of times the identification model identifies foreign objects, and calculates the risk value for the analysis period based on the number of times. The suggestion module generates a first suggestion to optimize the production process if the total efficiency value is less than a first threshold, and generates a second suggestion to optimize the production process if the risk value is greater than a second threshold.

10. A computer storage medium, characterized in that, The computer storage medium stores program instructions, wherein when the program instructions are executed, they control the device where the computer storage medium is located to perform the method described in any one of claims 1-8.

Images