An artificial intelligence-based in-flight meal quality monitoring method

Abstract

The application provides an aviation meal quality monitoring method based on artificial intelligence, comprising: obtaining flight delay information, cabin configuration data and real-time production line task scheduling, combining a preset meal standard, evaluating the influence of the delay on the task scheduling and the meal holding time, generating a task reorganization requirement and a holding time extension prediction value; performing real-time image acquisition on the meal semi-finished product of the task reorganization requirement, performing edge detection and texture analysis, obtaining a raw material appearance quality index, and identifying a raw material attribute change trend based on the obtained raw material appearance quality index; performing cluster analysis on the raw material attribute change trend, identifying key feature parameters affecting meal quality, combining cabin meal standardization requirements and the holding time extension prediction value, determining the quality risk level of the current production batch, and realizing precise control of aviation meal quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to an aviation catering quality monitoring method based on artificial intelligence. BACKGROUND

[0002] As an important part of aviation service, aviation catering directly affects passenger experience and airline brand image, and the consistency of its production quality is crucial. With the rapid development of the aviation industry and the increasing demand for food quality from passengers, the traditional aviation catering quality control system is facing severe challenges. Currently, aviation catering production monitoring mainly relies on manual sampling and fixed-parameter automated equipment. This method is rigid and inefficient when dealing with flight dynamics. In particular, when flight delays occur, the existing system cannot flexibly adjust the production process and quality monitoring standards according to actual conditions, resulting in significant fluctuations in meal quality and failing to meet the standardized meal requirements for different cabin configurations. Changes in production plan caused by flight delays require real-time reorganization of kitchen production line task sequencing, which directly affects the holding time of semi-finished products, making them exhibit nonlinearly extended characteristics. Changes in holding time also cause dynamic changes in meal raw material properties, such as unpredictable fluctuations in key quality indicators such as taste, color, and nutritional content. This chain reaction makes it difficult for fixed-parameter quality monitoring systems to accurately assess the actual quality of meals and develop differentiated quality control strategies for different risk levels. Therefore, how to build an intelligent system that can perceive flight dynamics, adaptively adjust production monitoring parameters, and accurately monitor meal quality under different cabin configurations, and dynamically adjust detection parameters and threshold values according to quality risk levels in the event of flight delays and other abnormal situations, has become a key issue in ensuring the consistency of aviation catering production quality. SUMMARY

[0003] The present application provides an aviation catering quality monitoring method based on artificial intelligence, mainly including:

[0004] Obtain flight delay information, cabin configuration data, and real-time production line task sequencing, combine with preset meal standards, evaluate the impact of delays on task sequencing and meal holding time, generate task reorganization requirements and holding time extension prediction values;

[0005] Real-time image acquisition of meal semi-finished products for task reorganization requirements, edge detection and texture analysis, obtaining raw material appearance quality indicators, and identifying raw material property change trends based on the obtained raw material appearance quality indicators;

[0006] Cluster analysis of raw material property change trends, identification of key feature parameters affecting meal quality, combination of cabin meal standardization requirements and holding time extension prediction values, determination of the quality risk level of the current production batch;

[0007] For high-risk meals with a quality risk level greater than a preset level, image acquisition parameters are adjusted according to appearance deviation of the high-risk meals and change trend of the raw material attributes to obtain a new set of detection parameters; the preset level can be determined according to actual conditions;

[0008] The new set of detection parameters is optimized by using a machine learning algorithm to generate a dynamic threshold table adapted to current production conditions, real-time detection data is compared with the threshold table to obtain a meal quality consistency evaluation result, and the consistency evaluation result is used to evaluate whether the meal quality meets the aviation catering standard;

[0009] If the meal quality does not meet the aviation catering standard, the production line task sequence is adjusted according to the quality risk level, high-risk meals are preferentially processed, and the production process of the adjusted high-risk meals is monitored to extract standardized features of production images;

[0010] Deviation data is extracted according to the standardized features, the correlation between the deviation data and the holding time and the raw material attributes is analyzed, new production control instructions are generated and issued to the production line, the holding time and the raw material processing flow are adjusted, and meal quality consistency data meeting the aviation catering standard is obtained.

[0011] Further, the flight delay information, cabin configuration data and real-time production line task scheduling are acquired, the influence of the delay on the task scheduling and the meal holding time is evaluated in combination with the preset meal standard, the task reorganization requirement and the holding time extension prediction value are generated, including: according to the real-time flight delay time data and the cabin configuration data set, the flight corresponding meal preparation specification parameters are acquired from the preset meal specification library through data matching association rules, the delay time and the meal preparation specification parameters are associated and calculated by using a flight delay influence evaluation matrix, and delay task scheduling change data set is obtained. According to the delay task scheduling change data set and the real-time production capacity data of the production line, the airport meal loading time limit parameters are acquired from the preset task window database, and the candidate task sequence group meeting the loading time limit constraint is calculated by a time window matching algorithm. According to the candidate task sequence group and the meal preparation process data, the actual execution time of each process is calculated by using a process time prediction model, the corresponding holding time limit parameters are acquired from a process rule library, and process holding constraint data set is obtained. According to the process holding constraint data set, the process execution threshold is acquired from the production line capacity constraint database, the candidate task sequence group is screened by a multi-objective optimization algorithm, and a task sequence set meeting the holding constraint is obtained. According to the task sequence set meeting the holding constraint, the holding time extension value caused by the delay is calculated by using a task time compensation model, and the holding time extension value is filtered by a holding threshold judgment rule, and a task reorganization sequence and a corresponding holding time prediction value are obtained. According to the task reorganization sequence and the holding time prediction value, the task priority parameters are acquired from the preset task scheduling rule library, and the task execution time sequence table considering the holding extension is generated by a sequence compensation algorithm.

[0012] Further, the meal semi-finished product required for task reorganization is subjected to real-time image acquisition, edge detection and texture analysis to obtain raw material appearance quality indexes, and based on the obtained raw material appearance quality indexes, raw material attribute change trends are identified, including: obtaining a set of delayed flight numbers from a flight real-time state database, reading delay time data and cabin seat distribution data through a flight state monitoring interface, generating a delay time and seat distribution mapping table using a flight data mapper, and extracting task sequencing reference values from a preset task rule library. For the delay time and seat distribution mapping table, a delayed flight seat distribution matrix is generated by a cabin type distribution calculator, and a task sequencing evaluation function is used to calculate the impact of delays on the sequencing of each flight, resulting in a task impact score table. According to the task impact score table, a priority score is generated by a task priority calculation unit according to flight delay time, seat quantity, and cabin class, corresponding production specification requirements are read from a meal specification database, and a task priority order table is obtained. For the task priority order table, corresponding production line task parameters are extracted using a meal production rule matcher, a task timing arrangement table is generated by a production resource scheduler, and an initial task scheduling scheme is obtained. According to the initial task scheduling scheme, a delay time and delivery time correlation model is established using a support vector regression algorithm, and a set of time prediction parameters is obtained by training historical delivery data to correct the task scheduling scheme in time, and a meal delivery scheduling time table is obtained.

[0013] Further, the raw material attribute change trend is subjected to cluster analysis to identify key feature parameters that affect meal quality, and combined with cabin meal standardization requirements and extended prediction values of holding time, the quality risk level of the current production batch is determined, including: collecting meal semi-finished product images from the meal production line according to task reorganization requirements, collecting images in four directions, up, down, left and right, through high-definition cameras fixed around the production line, using an image preprocessing unit for Gaussian noise elimination and brightness equalization processing to obtain a set of preprocessed images. For the set of preprocessed images, a convolutional neural network is used to extract meal surface feature vectors, edge detection operations are performed by a Sobel operator, and texture feature parameters are calculated using a gray level co-occurrence matrix to obtain an image feature description matrix. According to the image feature description matrix, preset meal quality parameters are obtained from a quality standard database, and standard matching operations are performed on the feature description matrix by a feature similarity calculation unit to obtain semi-finished product quality indexes. For the semi-finished product quality indexes, a time window divider is used to construct a continuous monitoring data sequence, a standard reference interval is generated by a quality baseline calculation unit to obtain a semi-finished product quality baseline value. According to the semi-finished product quality baseline value and the continuous monitoring data sequence, a quality change prediction model is established using a random forest algorithm, and real-time data is updated through a sliding time window to obtain a semi-finished product attribute change trend graph.

[0014] Further, the raw material attribute change trend is subjected to cluster analysis to identify key characteristic parameters affecting meal quality, and the current production batch quality risk level is determined in combination with the cabin meal standardization requirement and the extended holding time prediction value, including: the raw material color change curve, shape integrity data, and odor concentration value are acquired by the raw material attribute collector, the raw material surface feature data are acquired by the multi-spectral sensor, the k-means clustering algorithm is used to group the raw material feature data, and the raw material feature clustering table is obtained. According to the raw material feature clustering table, the cabin meal specification requirement is obtained from the meal standard database, the feature mapping unit is used to establish the mapping relationship between the raw material feature and the meal specification, the correlation degree calculator is used to generate the feature weight matrix, and the meal quality key feature table is obtained. According to the meal quality key feature table, the feature standard threshold is read from the quality evaluation rule library, the color, shape, and odor three indexes are scored by the feature quantization unit, and the raw material quality feature score matrix is obtained. According to the raw material quality feature score matrix and the extended holding time prediction data, the quality influence factor table is generated by the feature combination calculation unit, the factor weight distributor is used to determine the influence degree of each factor, and the quality risk assessment matrix is obtained. According to the quality risk assessment matrix, the risk level is divided by using the hierarchical clustering algorithm, the quality risk is graded by using the risk early warning rule library, and the production batch quality risk level judgment result is obtained.

[0015] Further, for high-risk meals with a quality risk level greater than a preset level, the image acquisition parameters are adjusted according to the appearance deviation of high-risk meals and the raw material attribute change trend to obtain a new detection parameter set, including: according to the quality risk level judgment result, if the quality risk level is higher than the preset early warning value of the preset level, the parameter optimization unit is used to calculate new detection reference data. According to the detection reference data, the standard sample and the measured image are compared by the appearance feature calculation unit, the support vector machine algorithm is used to establish a raw material attribute deviation model, and a feature deviation quantitative table is obtained. According to the feature deviation quantitative table, the image acquisition frequency sequence is generated by the time window calculation module, the feature threshold calibrator is used to set the new detection sensitivity parameter, and the optimized feature extraction rule set is obtained. According to the feature extraction rule set, the feature recognition accuracy is calculated by the detection parameter verification module, the deep convolution network is used to train and optimize the feature extraction parameter, and the calibrated parameter combination table is obtained. According to the calibrated parameter combination table, the new sampling instruction is issued to the image acquisition device by the task distribution module, the feature recognition module is used to execute the real-time detection task, and the new detection parameter set is obtained. According to the new detection parameter set, the detection effect score is calculated by the parameter evaluation unit, and the historical detection record is obtained from the quality monitoring database.

[0016] Further, the computer vision detection of meal appearance deviation data is obtained through the production line real-time monitoring system, the raw material attribute change data is extracted from the raw material quality monitoring system, the meal appearance deviation data and the raw material attribute change data are standardized, the exposure parameters of the image acquisition equipment are dynamically adjusted according to the standardized deviation data, including: obtaining the real-time image of meal appearance from the production line monitoring database, identifying the surface color, shape and texture abnormal area of the meal through the image feature extractor, using the multi-dimensional feature quantizer to perform area statistics and edge contour extraction on the abnormal area, and obtaining the appearance deviation data set. For the appearance deviation data set, read the real-time monitoring record from the raw material quality monitoring database, and through the attribute analysis unit, quantitatively calculate the meat fiber fracture degree, starch texture softening ratio, water loss value, oil oxidation content and spice residual intensity, and obtain the raw material quality change table. According to the raw material quality change table, the interval mapping conversion is performed on each index through the numerical normalization processor, the range standardization method is used to generate a normalized data sequence, and a standardized quality index set is obtained. For the standardized quality index set and the appearance deviation data set, a quality prediction model is established by using a support vector regression algorithm, a comprehensive deviation score is calculated by a deviation quantification unit, and a deviation level evaluation table is obtained. According to the deviation level evaluation table, the exposure intensity adjustment value, the contrast adjustment value and the color saturation adjustment value are generated by the image acquisition parameter calculator, the parameter mapping unit is used to convert the device control instruction, and an image acquisition parameter update scheme is obtained. For the image acquisition parameter update scheme, the historical adjustment record is obtained from the device parameter database, the effectiveness is evaluated by the parameter verification unit, and the optimized image acquisition parameter set is obtained.

[0017] Further, a machine learning algorithm is used to optimize the meal detection parameters based on a new detection parameter set, a dynamic threshold table that adapts to the current production conditions is generated, real-time detection data is compared with the threshold table to obtain a meal quality consistency evaluation result, and the consistency evaluation result is used to evaluate whether the meal quality meets the aviation catering standards, including: according to the new detection parameter set, shape, color, and texture feature data of the detection parameters are obtained through a parameter feature extractor, a random forest algorithm is used to calculate the feature weight distribution, temperature, humidity, and illumination intensity parameters are read from a production environment database to obtain a detection parameter optimization matrix. For the detection parameter optimization matrix, a detection threshold sequence is constructed through a threshold generation unit, a time window sliding method is used to dynamically update the threshold sequence, qualified sample parameters are obtained from a historical database to obtain a reference threshold table. According to the reference threshold table, the thresholds are temperature-corrected and humidity-calibrated through an environment compensator, a dynamic threshold interval is generated through a parameter self-adapting device to obtain a real-time detection threshold set. For the real-time detection threshold set, meal detection data is obtained through a data collector, a multi-dimensional comparator is used to calculate the matching degree of the detection value and the threshold to obtain a quality feature score table. According to the quality feature score table, a standardization score is generated through a score normalization processor, a quality classification model is established through a support vector machine algorithm to obtain a quality grade division result. For the quality grade division result, a judgment rule is obtained from a quality standard library, a compliance evaluation report is generated through a rule matching device to obtain a meal quality judgment result.

[0018] Further, if the meal quality does not meet the airline catering standards, the production line task order is adjusted according to the quality risk level, high-risk meals are preferentially processed, the production process of the adjusted high-risk meals is monitored, the standardized features of the production images are extracted, including: according to the quality consistency evaluation result, the quality risk value of each batch is calculated through a risk scoring unit, a priority calculator is used to generate a task priority sequence, a task list to be processed is extracted from a production task database, and a high-risk task priority table is obtained. According to the high-risk task priority table, a new production task sequence is generated through a task time sequence planner, a capacity allocator is used to allocate tasks according to the production line processing capacity, and an optimized task execution scheme is obtained. According to the optimized task execution scheme, an image acquisition device is used to obtain a production process image sequence, a feature extractor is used to obtain color distribution, shape contour, and texture structure data of the image, and an original feature data set is obtained. For the original feature data set, the feature data is mapped to a standard interval through a data normalization processor, key feature indicators are selected through a feature selector, and a standardized feature vector is obtained. According to the standardized feature vector, deep feature representation is extracted through a convolutional neural network, a feature comparator is used to calculate the matching degree with the quality standard template, and a feature matching score table is obtained. For the feature matching score table, a quality evaluator is used to generate a multi-dimensional quality score, a comprehensive judge is used to generate a monitoring report, and a production process quality feature set is obtained.

[0019] Further, according to the standardized feature extraction deviation data, the correlation between the deviation data and the holding time and the raw material attribute is analyzed, new production control instructions are generated and issued to the production line, the holding time and the raw material processing process are adjusted, and the meal quality consistency data meeting the aviation catering standard is obtained, including: according to the standardized feature data, the color uniformity, shape integrity and texture uniformity deviation values are extracted by the feature calculator, the corresponding relationship between the deviation values and the holding time and the raw material attribute change is constructed by the correlation analyzer, and the quality deviation mapping table is obtained. According to the quality deviation mapping table, the temperature influence model, the time influence model and the raw material influence model are established by the neural network algorithm, the temperature control value, the holding time value and the raw material processing value are generated by the parameter optimizer, and the production parameter adjustment scheme is obtained. According to the production parameter adjustment scheme, the equipment control instructions are generated by the parameter converter, the control instructions are verified by the instruction verifier, and the new production control instruction set is obtained. According to the standardized control instruction set, the equipment execution task sequence is generated by the task decomposer, the execution task is allocated by the task scheduler, and the production line execution scheme is obtained. According to the production line execution scheme, the temperature change curve, the holding time record and the raw material state data are obtained by the parameter collector, the threshold value is compared by the parameter verifier, and the process monitoring data set is obtained. According to the process monitoring data set, the meal color feature, shape feature and texture feature are extracted by the quality detector, the feature data is classified and evaluated by the random forest algorithm, and the quality consistency evaluation table is obtained.

[0020] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0021] The application discloses an aviation catering quality monitoring method based on artificial intelligence, which obtains flight delay information, cabin configuration data and production line task sorting, evaluates the influence of delay on tasks and holding time, generates task reorganization requirements and holding time prediction values. Real-time collection of meal semi-finished product images is realized by using computer vision technology, image features are analyzed by combining a machine learning model, and raw material attribute change trends are identified. The quality risk level is determined by cluster analysis according to the attribute change trend, and the detection parameters are dynamically adjusted. The detection threshold value is optimized by using a machine learning algorithm, a dynamic threshold value table is generated, and meal quality consistency is evaluated. For meals that do not meet the standard, the production line task sorting is adjusted, secondary monitoring is performed, the correlation between deviation data and holding time and raw material attribute is analyzed, production parameters are optimized, holding time and raw material processing process are adjusted, and finally precise control and consistency guarantee of aviation meal quality are realized. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flowchart of an aviation catering quality monitoring method based on artificial intelligence.

[0023] Figure 2 A schematic diagram of an artificial intelligence-based aviation catering quality monitoring method of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be described in detail below in combination with the drawings and specific embodiments.

[0025] Aviation catering quality monitoring involves real-time supervision of meal production processes to ensure that meals meet preset quality standards. The catering production system includes a production line, monitoring equipment and a data processing module, wherein the production line is responsible for meal preparation, the monitoring equipment collects production data, and the data processing module analyzes data and generates control instructions. Meal data is divided into raw material data and production data, raw material data includes color, taste and other attributes, and production data includes holding time, process duration and other parameters. Flight delay information includes delay time, flight number, etc., and cabin configuration data includes seat distribution and meal grade. The monitoring method includes manual sampling and automated detection, and traditional automated detection relies on fixed parameters, which is difficult to adapt to dynamic changes.

[0026] As Figure 1 , the artificial intelligence-based aviation catering quality monitoring method of the present embodiment can specifically include:

[0027] S101 Obtain flight delay information, cabin configuration data and real-time production line task sequencing, combine preset meal standards, evaluate the impact of delays on task sequencing and meal holding time, generate task reorganization requirements and holding time extension prediction values.

[0028] In the present embodiment, the catering production system has a built-in monitoring function, which can be realized by system software or a special application program, and the specific form is determined according to the actual scene. The present embodiment does not make too many restrictions on the starting mode of the monitoring function, which can be manually operated by the user or triggered by sending instructions through external equipment. For example, when the monitoring function is integrated into the system software, the user can start the function through the production management interface, or quickly activate it through the shortcut icon. When the function is embedded in the application program, the developer can set the starting mode according to the needs, such as setting a monitoring button on the production scheduling interface, which the user can click to start.

[0029] S1011 Obtain delay information and cabin configuration data according to the real-time flight status interface, combine the preset meal standard library, and generate delay task sequencing change data set through correlation analysis.

[0030] In the embodiment of the present application, the delay time, flight number and cabin seat distribution are obtained through the flight status interface, and the corresponding meal preparation parameters such as holding time limit and process requirement are extracted from the preset meal standard library. The delay time and meal parameters are matched by using a data association algorithm to generate a task sequence change data set. For example, a certain flight is delayed for 90 minutes, and the business class meal holding time limit is 120 minutes. It is calculated that the task needs to be adjusted in advance to meet the loading time limit.

[0031] S1012 For the task sequence change data set, the loading time limit parameter is obtained from the preset task window database in combination with the real-time production capacity of the production line, and a candidate task sequence group is generated by a time matching algorithm.

[0032] In the embodiment of the present application, the airport meal loading time limit parameter is extracted from the task window database according to the task sequence change data set and the production line capacity data. The candidate task sequence group that meets the time limit constraint is calculated by a time matching algorithm. For example, the airport requires that the loading be completed 90 minutes before takeoff. The task sequence is adjusted according to the takeoff time after the delay to ensure that the production task meets the time limit requirement.

[0033] S1013 For the candidate task sequence group, the process execution threshold is obtained from the capacity constraint database by using a process prediction model to generate a process holding constraint data set.

[0034] In the embodiment of the present application, the candidate task sequence group is analyzed by a preset process prediction model, and the actual duration of each process is calculated in combination with the meal preparation process data. The process execution threshold is obtained from the capacity constraint database to generate a holding constraint data set. For example, the business class meal preparation takes 45 minutes, and the task completion time is determined according to the holding time limit requirement.

[0035] S1014 According to the holding constraint data set, the candidate task sequence group is screened by an optimization algorithm to generate a task sequence that meets the holding constraint and a holding time prediction value. In the embodiment of the present application, the process execution threshold is obtained from the production line capacity constraint database according to the process holding constraint data set, and the candidate task sequence group is screened by a multi-objective optimization algorithm to generate a task sequence that meets the holding constraint and a corresponding holding time prediction value. For example, the business class meal is arranged to start production at 9:50, and the holding time is extended to 135 minutes. The holding time extension value caused by the delay is calculated by a task time compensation model, and a task execution time sequence table is generated.

[0036] S1015 According to the task sequence and the holding time prediction value, the priority parameter is obtained from the task scheduling rule library, and the task execution time sequence table is generated by a sequence optimization algorithm.

[0037] In the embodiment of the present application, according to the task sequence meeting the heat preservation constraint, the priority parameter is extracted from the task scheduling rule library, and the sequence optimization algorithm is used to generate the task execution time sequence table. For example, the business class meal is produced in advance, and the production is arranged to be completed from 9:50 to 10:45, and the flexible time is reserved to deal with unexpected situations.

[0038] In the embodiment of the present application, the flight delay is closely related to the meal heat preservation time limit. Through real-time data analysis, the task sorting is dynamically adjusted to ensure the stability of the meal quality. For example, a flight is delayed for 90 minutes, and according to the heat preservation time limit of the business class and the economy class, the task sorting is optimized to complete the meal loading. Through intelligent task reorganization, the embodiment of the present application effectively reduces the influence of delay on meal quality, improves production efficiency, and ensures the consistency of aviation meal quality, providing passengers with high-quality meal experience.

[0039] S102 Real-time image acquisition is performed on the meal semi-finished product required for task reorganization, edge detection and texture analysis are performed, the appearance quality index of the raw material is obtained, and the attribute change trend of the raw material is identified based on the obtained appearance quality index of the raw material.

[0040] In the embodiment of the present application, for the task reorganization requirement, high-definition camera equipment deployed on the production line is used to collect meal semi-finished product images, image quality is optimized by combining preprocessing technology, key features are extracted by a deep learning model, product quality indexes are generated, and attribute change trends are predicted through time sequence analysis. This way can monitor the quality state of semi-finished products in real time, discover potential abnormalities in time, ensure that meal production meets the preset standards, and provide accurate protection for aviation meal quality.

[0041] As Figure 2 , S1021 collects meal semi-finished product images from multiple angles by high-definition camera devices fixed on the production line, and uses an image preprocessing module to perform noise elimination and brightness equalization to generate a standardized preprocessed image set.

[0042] In the embodiment of the present application, high-resolution camera devices are installed around the production line to collect meal semi-finished product images from four directions: up, down, left, and right. The camera devices use 12 million pixel sensors to ensure clear image details. In order to eliminate the influence of uneven environmental light or equipment vibration, random noise is removed by Gaussian filtering, and the image brightness is adjusted to a standard range, for example, a gray scale interval of 180 to 220, by using a brightness equalization algorithm, to generate a preprocessed image set suitable for subsequent analysis. The deployment of the camera devices is not limited in the embodiment of the present application, and can be adjusted according to the layout of the production line.

[0043] S1022 extracts a surface feature vector from the preprocessed image set using a convolutional neural network, and generates an image feature description matrix through edge detection and texture analysis algorithms.

[0044] In the embodiment of the present application, the pre-processed image set is subjected to feature extraction by using a convolutional neural network model to generate a feature vector containing the shape, color and arrangement regularity of the food surface. To further analyze the image details, edge detection is performed using a Sobel operator to calculate an edge integrity score, for example, the grain edge score of the rice semi-finished product needs to reach 0.85 or above. At the same time, texture features are calculated through a gray level co-occurrence matrix, including contrast, correlation and entropy value, to form an 8x8 feature description matrix. The contrast reflects the difference in brightness between pixels, the correlation represents the arrangement regularity, and the entropy value represents the texture complexity. These parameters together constitute the feature description matrix, providing a data basis for quality evaluation.

[0045] S1023 According to the feature description matrix, the food standard parameters are obtained from the pre-set quality standard database, and the semi-finished product quality index is calculated by a feature matching algorithm.

[0046] In the embodiment of the present application, the pre-set food quality parameters are extracted from the quality standard database, for example, the rice semi-finished product needs to meet the requirements that the shape integrity rate is not less than 95% and the color uniformity is not less than 0.9. The cosine similarity algorithm is used to match the feature description matrix with the standard parameters to generate the semi-finished product quality index. If the similarity score is less than 0.8, it indicates that the semi-finished product has quality deviation, and the deviation degree of the quantitative index is recorded. The embodiment of the present application does not make too many limitations on the specific implementation of the matching algorithm, which can be optimized by the technician according to the actual scene.

[0047] S1024 Combined with the semi-finished product quality index, a time series analysis technique is used to construct a continuous monitoring data sequence to generate a quality benchmark interval.

[0048] In the embodiment of the present application, the time window segmentation module is used to construct a continuous monitoring data sequence of the semi-finished product quality index at intervals of 5 minutes. By calculating the mean and standard deviation of the index within 1 hour, a quality benchmark interval is generated, for example, the standard interval of rice color uniformity is 0.85 to 0.95. When the index of the continuous three time windows is lower than the lower limit of the interval, it is determined that there is a quality fluctuation risk. The embodiment of the present application ensures the real-time quality monitoring through time series analysis.

[0049] S1025 According to the continuous monitoring data sequence, a quality change prediction model is constructed using a random forest algorithm, and the data is updated through a sliding time window to generate a semi-finished product attribute change trend chart.

[0050] In the embodiment of the present application, based on the random forest algorithm, a quality change prediction model is trained in combination with historical data, and the input variables include raw material batch, processing time and environmental temperature, etc. The model updates data in real time through a 60-minute sliding time window, and predicts the change trend of the quality index in the next 30 minutes. For example, if the uniformity of rice color decreases by more than 15%, it is determined to be a quality abnormality. The generated change trend graph takes time as the horizontal axis and the standardized quality score as the vertical axis, intuitively displays the dynamic changes of attributes such as color and shape, and facilitates production management personnel to monitor the semi-finished product state in real time. The embodiment of the present application effectively warns the quality risk through the prediction model, and improves the stability of meal production.

[0051] In the embodiment of the present application, the quality monitoring of meal semi-finished products involves multi-dimensional feature analysis. Taking rice semi-finished products as an example, images are collected from different angles by four high-definition cameras to ensure that the complete shape and color distribution of rice grains are captured. The preprocessed images are extracted by a convolutional neural network, combined with edge detection and texture analysis, and a high-precision feature description matrix is generated. The quality standard database provides clear quantitative indicators, and the deviation degree is calculated through feature matching to generate a monitoring data sequence in real time. The random forest model predicts the quality trend based on time series data, and timely discovers abnormalities, such as color changes caused by prolonged holding time. This multi-level analysis method can accurately identify the causes of quality fluctuations, ensure that meals meet the aviation catering standards, and improve production efficiency and passenger satisfaction.

[0052] S103 Cluster analysis is performed on the change trend of the properties of raw materials to identify key feature parameters that affect meal quality, and the quality risk level of the current production batch is determined in combination with the cabin meal standardization requirements and the predicted value of the holding time extension.

[0053] In the embodiment of the present application, for the dynamic changes of raw material properties, multi-spectral sensors are used to collect feature data, and clustering algorithms are used to analyze key indicators such as color, shape and odor to generate a quality matrix, and in combination with cabin meal standards and holding time prediction values, the quality risk level of the production batch is evaluated. This method can accurately identify the causes of quality fluctuations, dynamically divide the risk level, provide a basis for production optimization, and ensure the stability and consistency of aviation catering quality.

[0054] S1031 The surface feature data of the raw materials are collected by multi-spectral sensors, and the k-means clustering algorithm is used to group the feature data to generate a raw material feature classification table.

[0055] In the embodiment of the present application, the multispectral sensor is used to collect the surface data of the raw materials, covering the wavelength range of 400 to 1000 nanometers, and obtaining the characteristics such as color and texture. For example, the color data of beef meals is analyzed by the main wavelength, and the main wavelength of fresh beef is about 650 nanometers, and the deteriorated beef is shifted to below 580 nanometers. The k-means clustering algorithm is used to divide the characteristic data into three categories: A class represents high quality, B class represents general, and C class represents unqualified, and a characteristic classification table is generated. The embodiment of the present application realizes the preliminary screening of the quality of raw materials through clustering analysis, and provides data support for subsequent evaluation.

[0056] S1032 According to the characteristic classification table, the cabin meal specification requirements are extracted from the meal standard database, the corresponding relationship between the raw material characteristics and the meal demand is established through the characteristic mapping technology, and the key characteristic weight table is generated.

[0057] In the embodiment of the present application, the cabin meal specification is obtained from the meal standard database, for example, the first class requires A class raw materials, the business class accepts A class or B class, and the economy class uses B class. Through the characteristic mapping algorithm, the color, shape and smell are associated with the meal specification, and the weight of each characteristic is assigned by using the correlation degree analysis, for example, the color weight is 0.4, the shape is 0.3, and the smell is 0.3, and the key characteristic weight table is generated. The embodiment of the present application does not make too many limitations on the specific implementation of the mapping algorithm, which can be adjusted by the technician according to the actual demand.

[0058] S1033 According to the key characteristic weight table, the standard threshold is obtained from the quality evaluation rule library, the color, shape and smell indexes are scored through quantitative analysis, and the raw material characteristic score matrix is generated.

[0059] In the embodiment of the present application, the standard threshold is extracted from the quality evaluation rule library, for example, the uniformity of beef color needs to reach 0.9 or more, the shape integrity needs to reach 0.92 or more, and the volatile odor concentration needs to be lower than the preset value. Through the quantitative analysis module, the color, shape and smell are scored in percentage, and an 8x8 characteristic score matrix is generated. The color score includes freshness, uniformity and glossiness sub-items, the shape score includes integrity, regularity and thickness uniformity, and the smell score includes odor detection value and volatile concentration. Through multi-dimensional scoring, the comprehensiveness of the evaluation is ensured.

[0060] S1034 Combined with the characteristic score matrix and the holding time prediction value, the quality influence factor table is generated through risk factor analysis, the weight of each factor is determined, and the quality risk evaluation matrix is generated.

[0061] In the embodiment of the present application, according to the feature score matrix and the holding time prediction value, the quality influencing factor table is generated by the factor analysis module. If the holding time is extended beyond 90 minutes, the odor factor weight is increased to 0.4 and the shape weight is reduced to 0.2. Based on historical data, the influence degree of each factor under different holding time is calculated to generate a quality risk assessment matrix. The embodiment of the present application adjusts the dynamic weight to adapt to the real-time changes of the production conditions and improves the accuracy of risk assessment.

[0062] S1035 According to the quality risk assessment matrix, the hierarchical clustering algorithm is used to divide the risk level, and the production batch quality risk level judgment result is generated by combining the risk early warning rule library.

[0063] In the embodiment of the present application, the hierarchical clustering algorithm is used to divide the quality risk into four levels of green, yellow, orange and red. Green represents safe quality, yellow represents slight risk, orange represents medium risk, and red represents major risk. The judgment criteria are extracted from the risk early warning rule library, for example, when the color score is less than 75 points and the odor score is less than 80 points, it is judged as orange risk. The generated risk level result directly guides the subsequent quality control measures, for example, the orange risk batch needs to adjust the production parameters in priority. The embodiment of the present application ensures that the high-risk batches are processed in time through fine risk division.

[0064] In the embodiment of the present application, the multi-dimensional analysis of raw material attributes is the core of the quality control of aviation catering. Taking beef meals as an example, the color change is captured by a multi-spectral sensor, and the odor volatile is detected by an electronic nose sensor to ensure the comprehensiveness of the feature data. The combination of k-means clustering and hierarchical clustering algorithm realizes progressive analysis from feature grouping to risk division. The introduction of cabin meal standards ensures that the evaluation results match the actual needs. The dynamic weight adjustment mechanism can optimize the factor influence according to the change of holding time to generate an accurate risk assessment matrix. This method effectively identifies the quality abnormality reasons, such as odor changes caused by prolonged holding, and timely triggers production adjustment to ensure that the meals meet the aviation service standards and improve the passenger experience.

[0065] S104 For high-risk meals with a quality risk level greater than a preset level, the image acquisition parameters are adjusted according to the appearance deviation of the high-risk meals and the change trend of the raw material attributes to obtain a new set of detection parameters; the preset level can be determined according to actual conditions.

[0066] In the embodiment of the present application, for high-risk batches, the detection parameters are dynamically optimized by multi-dimensional analysis and machine learning algorithm combined with appearance deviation and raw material attribute change to improve the sensitivity and accuracy of abnormal detection. This method can adapt to changes in the production environment in real time, discover quality problems in time and guide process adjustment, thereby ensuring the stability of aviation catering and passenger satisfaction.

[0067] S1041 If the quality risk level exceeds the preset level, a new detection reference data is calculated using a preset parameter optimization method.

[0068] In the embodiment of the present application, the quality risk level of the real-time image is obtained, and if the quality risk level exceeds the preset level, the appearance integrity, color uniformity and shape regularity are analyzed by the risk scoring module. For example, in the case of salad making, the standard lettuce leaf integrity needs to reach 0.85 or more, the color uniformity needs to reach 0.8, and the shape regularity needs to reach 0.75. If the detection value is lower than the standard, the risk score exceeds the warning threshold of 0.8, triggering the parameter optimization process. The parameter optimization method is based on the current detection data and historical standard samples to calculate new detection reference data. The preset level can be determined according to the actual situation.

[0069] S1042 For the detection reference data, the support vector machine algorithm is used to compare the standard sample and the measured image to generate a feature deviation quantitative table, and the image acquisition frequency and detection sensitivity are optimized through time series analysis.

[0070] In the embodiment of the present application, the support vector machine algorithm is used to extract the edge contour and color distribution features, compare the measured image with the standard sample, and generate a feature deviation quantitative table. For example, the edge defect rate of lettuce leaf is 15%, the browning area ratio is 12%, and the cutting size error is 8mm. Through the time window calculation module, an image acquisition frequency sequence is generated at intervals of 60 seconds, and the frequency is increased from 180 seconds / time to 60 seconds / time. At the same time, the feature threshold calibration module adjusts the edge detection threshold to 0.8, the color segmentation threshold to 0.85, and the shape matching threshold to 0.9, to ensure that the detection sensitivity adapts to the high-risk scenario.

[0071] S1043 According to the feature deviation quantitative table, the parameter features are extracted, a calibrated parameter combination table is generated, and is issued to the image acquisition device for real-time detection.

[0072] In the embodiment of the present application, the deep convolutional neural network is used to train the feature extraction model based on 5000 historical samples, including normal, slight defect and severe defect samples. After training, the model recognition accuracy is 97%, the missed detection rate is reduced to 2%, and the false detection rate is reduced to 1%. The generated parameter combination table includes exposure parameters, focusing parameters, etc., such as adjusting the exposure intensity to 200-220, and the sharpness is not less than 0.95. The task distribution module issues the parameters to 4 high-definition cameras, and the feature recognition module detects the changes of lettuce leaves in real time. The embodiment of the present application optimizes the parameters through deep learning to improve the detection accuracy.

[0073] S1044 For the changes in raw material properties, multi-dimensional feature quantization and support vector regression algorithm are used to generate a standardized quality index set and a deviation level evaluation table to guide further adjustment of image acquisition parameters.

[0074] In the embodiment of the application, the fiber breakage rate, starch texture softening ratio, water loss value, oil oxidation content and spice residual intensity of beef meals are analyzed by the multi-dimensional feature quantization module. For example, the fiber breakage rate is 25%, the water loss is 8%, and the oil oxidation is 0.35. The numerical normalization module maps the indicators to the 0-1 interval, and generates a set of standardized quality indicators using the range standardization method. The support vector regression algorithm trains a quality prediction model based on historical data, and the bias quantization module calculates a comprehensive bias score. If the score is 0.82, it is determined to be high risk, and the exposure intensity is adjusted to 220, the contrast to 1.5, and the color saturation to 1.3.

[0075] In the embodiment of the application, the aviation catering quality monitoring needs to cope with complex production scenarios. Taking beef meals as an example, surface features are captured by high-resolution cameras to detect color, shape, and texture abnormalities. When the area of abnormal regions exceeds 10% or the edge loss exceeds 15%, it is recorded as an appearance deviation. The raw material attribute analysis shows that the starch texture softening ratio rises to 0.45 and the spice residual intensity drops to 65%. The support vector regression model predicts the quality trend, and the detection parameters are optimized by combining the deep convolutional network, which significantly improves the abnormal detection rate. After parameter adjustment, the detection record of the past 7 days shows that the abnormal detection rate has increased by 25% and the accuracy has increased by 8%. The detection results are fed back to the pretreatment link to guide the adjustment of the sharpness of the cutting tool and the pre-cooling temperature, ensuring the stability of meal quality and meeting the standardization needs of different passenger cabins.

[0076] S105 uses a machine learning algorithm to optimize meal detection parameters for a new set of detection parameters, generates a dynamic threshold table that adapts to the current production conditions, compares real-time detection data with the threshold table, and obtains meal quality consistency evaluation results. The consistency evaluation results are used to evaluate whether the meal quality meets the aviation catering standards.

[0077] In the embodiment of the application, for high-dynamic production scenarios, multi-dimensional feature extraction and environmental calibration technology are used in combination with machine learning algorithms to optimize detection thresholds, ensuring the accuracy and adaptability of detection results. This method can respond to environmental changes and raw material differences in real time, dynamically adjust detection standards, effectively identify quality abnormalities, and improve the efficiency and reliability of aviation catering quality control, providing passengers with meals that meet their expectations.

[0078] S1051 obtains shape, color and texture feature data of the detection parameters through the feature extraction module, combines production environment parameters, generates a feature weight distribution using a random forest algorithm, and constructs a detection parameter optimization matrix.

[0079] In the embodiment of the present application, high-resolution image processing technology is used to extract shape, color and texture features from the food sample by the feature extraction module. Taking a western-style steak as an example, the shape features include edge integrity, area regularity and thickness uniformity; the color features cover color uniformity, browning degree and charring area ratio; the texture features involve texture directionality, surface roughness and crack density. The temperature, humidity and light intensity data are obtained from the production environment database, such as temperature 25 degrees Celsius, humidity 75%, and light intensity 750 lux. The random forest algorithm analyzes the feature importance and generates a weight distribution, such as color uniformity weight 0.35, edge integrity 0.25, and texture directionality 0.15. The weight distribution is combined with the environmental parameters to generate a multi-dimensional detection parameter optimization matrix, providing a data basis for threshold adjustment.

[0080] S1052 According to the detection parameter optimization matrix, a detection threshold sequence is constructed by using a threshold generation algorithm, and a time window sliding method is used for dynamic updating to generate a reference threshold table.

[0081] In the embodiment of the present application, a threshold sequence containing multiple detection parameters is constructed based on the optimization matrix by the threshold generation algorithm, such as 36 feature thresholds. The time window sliding method updates the threshold data every 5 minutes with a 30-minute cycle, combined with the historical qualified sample parameters, to ensure that the threshold sequence reflects the latest production state. For example, the standard steak color uniformity threshold is 0.85 to 0.95, and the sequence range is dynamically adjusted according to recent detection data. The generated reference threshold table provides a reference for environmental calibration.

[0082] S1053 For the reference threshold table, temperature and humidity calibration is performed by the environmental compensation algorithm, and a dynamic threshold interval is generated combined with an adaptive adjustment mechanism to construct a real-time detection threshold set.

[0083] In the embodiment of the present application, the threshold is adjusted according to the change of production environment by using the environmental compensation algorithm. For example, for every 1 degree Celsius increase in temperature, the texture feature threshold is lowered by 4% to adapt to the softening of the meat; for every 1% increase in humidity, the color threshold is increased by 2% to offset the influence of surface moisture. If the light intensity is less than 800 lux, increase the image contrast threshold by 3%. The adaptive adjustment mechanism generates a dynamic threshold interval based on the environmental trend, such as the color uniformity interval adjustment to 0.82 to 0.92. The real-time detection threshold set ensures that the detection parameters match the production conditions. The embodiment of the present application improves the robustness of detection through environmental calibration.

[0084] S1054 According to the real-time detection threshold set, the food detection data is obtained by the data acquisition module, the matching degree between the detection value and the threshold is calculated by using the multi-dimensional matching algorithm, the quality feature score table is generated, and the quality grade is divided by the classification model.

[0085] In the embodiment of the application, real-time meal detection data is obtained by the data acquisition module, covering shape, color and texture characteristics. The cosine similarity algorithm is used to calculate the matching degree of the detection value and the threshold set, and a quality characteristic score table is generated. For example, the uniformity score of the color of the steak sample is 0.87, the edge integrity is 0.90, and the texture directionality is 0.84. The score is mapped to the 0-1 interval by the normalization processing, and input into the support vector machine classification model. The model is trained based on historical data, and the quality is divided into excellent, qualified and unqualified three grades. First-class steak requires all characteristic scores to be higher than 0.9 for excellent and lower than 0.8 for unqualified. The embodiment of the application realizes accurate quality evaluation through multi-dimensional matching and classification.

[0086] In the embodiment of the application, the dynamic threshold optimization of aviation catering is the key to ensure the consistency of quality. Taking steak as an example, multi-dimensional features are extracted through high-resolution images, and the feature weights are determined by combining the random forest algorithm to adapt to the changes in the production environment. The environment compensation mechanism dynamically adjusts the threshold according to the temperature and humidity, for example, the color threshold is adjusted upward when the humidity increases, to ensure the detection sensitivity. The support vector machine model divides the quality level based on the standardized score, and when the color uniformity decreases to 0.75, it is determined as unqualified and the reason is recorded. The detection results are fed back to the production link to guide the adjustment of processing temperature or tool parameters, and the process flow is optimized. This dynamic and intelligent detection scheme effectively responds to quality fluctuations, ensures that meals meet the strict standards of different passenger cabins, and improves the overall quality of aviation services.

[0087] S106 If the meal quality does not meet the aviation catering standards, adjust the production line task sequence according to the quality risk level, prioritize high-risk meals, monitor the production process of the adjusted high-risk meals, and extract standardized features of the production images.

[0088] In the embodiment of the application, for batches of meals that do not meet the quality standards, the production plan is dynamically adjusted through risk assessment and task optimization algorithms, and the production process is monitored in real time by combining high-precision image analysis technology to generate a multi-dimensional quality feature set. This method can quickly respond to quality abnormalities, optimize resource allocation, and accurately identify production deviations, thereby ensuring the quality consistency of aviation catering and improving production efficiency and passenger satisfaction.

[0089] S1061 Based on the quality consistency evaluation results of the meals, calculate the quality risk value of each batch through the risk analysis module, generate a task priority sequence using the priority allocation algorithm, and construct a high-risk task priority table.

[0090] In the embodiment of the present application, key indicators of each batch are extracted from the food quality consistency evaluation data, and the quality risk value is calculated through the risk analysis module. Taking fruit platter as an example, if the freshness is less than 0.85, the cutting regularity is less than 0.8, or the color uniformity is less than 0.75, the risk value exceeds the warning threshold 0.8. The priority allocation algorithm sorts according to the risk value, extracts the to-be-processed task from the production task database, and generates a high-risk meal task priority list. For example, 300 portions of first-class cabin, 500 portions of business class, and 1200 portions of economy class platters are marked as high risk and need to be processed in priority. The embodiment of the present application ensures that the high-risk batch is processed in time through risk quantification.

[0091] S1062 For the high-risk meal task priority list, a production task time sequence is generated using a task scheduling algorithm, and the task is allocated to the production line capacity to generate an optimized task execution plan.

[0092] In the embodiment of the present application, a new production task time sequence is generated by the task scheduling algorithm according to the production line processing capacity and the task urgency. For example, the first-class cabin platter task is scheduled to start at 9:00, the business class at 10:30, and the economy class at 13:00. The capacity allocation module allocates the high-precision processing first-class cabin task to the No. 1 production line, and the remaining tasks to the No. 2 and No. 3 production lines, to maximize resource utilization. The optimized task execution plan can effectively shorten the processing cycle of the high-risk batch. The embodiment of the present application improves production flexibility through task optimization.

[0093] S1063 According to the task execution plan, a high-frequency image acquisition device is used to obtain a sequence of meal production process images, generate a standardized feature vector, and use a deep learning model for feature analysis.

[0094] In the embodiment of the present application, a high-definition image acquisition device deployed on the production line takes a production process image every 30 seconds to generate a continuous image sequence. The feature extraction module extracts color distribution, shape contour, and texture structure data from the image, such as color saturation, edge browning degree, cutting size consistency, and flesh texture clarity of fruit platter. Data normalization processing maps the feature values to the 0-1 interval, and the feature selection algorithm selects 6 key indicators, such as the color saturation of apple slices reaching 0.85 and the cutting consistency reaching 0.9. The standardized feature vector is input into a convolutional neural network to extract a 32-dimensional deep feature representation. The deep learning model analyzes the features to improve the robustness of feature extraction. The embodiment of the present application ensures the traceability of the production process through high-frequency monitoring.

[0095] S1064 For the deep feature representation, a feature matching algorithm is used to calculate the similarity with the quality standard template to generate a multi-dimensional quality score and a production process quality feature set.

[0096] In the embodiment of the application, by using a feature matching algorithm, a feature matching score table is generated by comparing deep feature representations with quality standard templates through a cosine similarity method. If the matching score of the fruit platter batch is lower than 0.8, the deviation reason is analyzed from three aspects of color, shape and freshness, for example, the apple slice edge browning exceeds the standard or the slice size is uneven. The quality evaluation module generates a multi-dimensional score, and the comprehensive judgment module outputs a detailed monitoring report containing the index score and deviation description. The production process quality feature set records complete feature data, supporting subsequent process optimization and quality traceability. The embodiment of the application improves the accuracy of quality monitoring through multi-dimensional analysis.

[0097] In the embodiment of the application, efficient task adjustment and real-time monitoring are needed for high-risk batch management of in-flight catering. Taking fruit platter as an example, quality abnormal batches are quickly identified through risk analysis, and production timing is arranged in priority to ensure that first-class cabin tasks are completed in priority. Production line allocation considers capacity differences, and fine task allocation is made to high-precision equipment to improve processing quality. High-frequency image acquisition combined with convolutional neural networks extracts key features in real time, for example, detects that the browning area of a batch of apple slices exceeds the standard, records the deviation and triggers process adjustment suggestions, such as optimizing tool cleaning frequency or adjusting pre-cooling time. The monitoring report provides detailed data support for production managers to guide process improvement, ensuring that meals meet the strict standards of different passenger cabins and enhancing the quality assurance capability of aviation services.

[0098] S107 According to the standardized feature extraction deviation data, the correlation between the deviation data and the holding time, the raw material attribute is analyzed, the new production control instruction is generated and sent to the production line, the holding time and the raw material processing process are adjusted, and the meal quality consistency data meeting the in-flight catering standards is obtained.

[0099] In the embodiment of the application, for meal quality deviation, multi-dimensional feature analysis and machine learning algorithm are used to accurately identify the deviation reason and optimize the production parameters, and through real-time monitoring and instruction adjustment, the meal quality meets the in-flight catering standards. This method can effectively deal with complex dynamic changes in the production process, improve the accuracy of quality control and production efficiency, and provide high-quality meal experience for air passengers.

[0100] S1071 The color uniformity, shape integrity and texture uniformity deviation values of the meal sample are extracted through the feature analysis module, the correlation analysis algorithm is used to construct the mapping relationship between the deviation and the holding time and the raw material attribute, and the quality deviation mapping table is generated.

[0101] In the embodiment of the application, the feature analysis module is used to extract key quality indicators from the standardized feature data. Taking western-style beefsteak as an example, the color uniformity deviation shows that the browning area accounts for about 15%, the shape integrity deviation shows that the edge loss rate is about 12%, and the texture uniformity deviation shows that the hardness fluctuation exceeds the standard deviation by 0.3. The correlation analysis algorithm quantifies the relationship between the deviation and the holding time and the raw material properties through regression analysis, for example, the browning area increases by about 5% for every 30-minute extension of the holding time, and the raw material fiber looseness increases by about 8%. The generated quality deviation mapping table clearly shows the correlation between each deviation indicator and the production condition, providing data basis for parameter optimization. The embodiment of the application ensures accurate identification of the cause of the deviation through correlation analysis.

[0102] S1072 For the quality deviation mapping table, a neural network algorithm is used to construct a temperature, time and raw material influence model, a parameter optimization algorithm is used to generate new production control parameters, and a production adjustment scheme is developed.

[0103] In the embodiment of the application, a three-dimensional influence model is constructed based on historical data and the mapping table through a deep neural network algorithm: the temperature influence model shows that the browning speed increases by about 20% for every 5-degree Celsius increase in holding temperature; the time influence model shows that holding time exceeding 120 minutes will cause meat fiber denaturation; and the raw material influence model indicates that insufficient marinating time increases water loss by about 10%. The parameter optimization algorithm uses the gradient descent method to iteratively calculate the optimal parameter combination to generate new control parameters, for example, reducing the holding temperature to 65 degrees Celsius, controlling the holding time to 100 minutes, and extending the marinating time to 4 hours. The production adjustment scheme specifies the target values of each parameter, providing guidance for equipment control. The embodiment of the application improves the scientificity of parameter optimization through multi-model analysis.

[0104] S1073 According to the production adjustment scheme, the instruction generation module converts the optimized parameters into equipment control instructions, and a verification mechanism is used to ensure the compliance of the instructions to generate a standardized control instruction set.

[0105] In the embodiment of the application, the instruction generation module converts the optimized parameters into specific equipment instructions, for example, adjusting the heating power of the temperature control equipment to 65%, the ventilation frequency of the holding box to 6 times per hour, and the raw material pretreatment time to 240 minutes. The verification mechanism compares the instructions with the equipment operating range to ensure that the temperature control accuracy is within ±1 degree Celsius and the time error is less than 5 minutes. If the instructions exceed the range, they are automatically corrected to the compliant value. The generated standardized control instruction set ensures the executability and safety of the instructions. The embodiment of the application reduces the execution risk through strict verification.

[0106] S1074 For the standardized control instruction set, the task allocation module generates a device execution task sequence, and real-time monitoring technology is used to obtain production process data to generate a process monitoring data set.

[0107] In the embodiment of the application, the task allocation module decomposes the instructions into tasks such as temperature adjustment, timing control, and raw material processing. The task scheduling algorithm allocates tasks in the order of the process to ensure that the raw material processing is completed before the heat preservation program is started. The parameter acquisition module records temperature changes, heat preservation time, and raw material state data every 5 minutes to generate a temperature curve and a state log. For example, the temperature curve shows that the heat preservation process is stable at 65 degrees Celsius, and the raw material state sensor records that the fiber looseness has decreased to 5%. The parameter verification module compares the data with the preset threshold to generate a process monitoring data set to support quality evaluation. The embodiment of the application ensures the controllability of the production process through real-time monitoring.

[0108] S1075 According to the process monitoring data set, the quality analysis module extracts meal characteristics, uses a random forest algorithm for classification evaluation, and generates a meal quality consistency evaluation table that meets the aviation catering standards.

[0109] In the embodiment of the application, the quality analysis module extracts color, shape, and texture characteristics from the monitoring data, such as a decrease in the browning area of a steak to 8%, an increase in edge integrity to 0.95, and a decrease in hardness standard deviation to 0.15. The random forest algorithm is based on 100 decision trees to classify feature data into three levels: excellent, qualified, and unqualified. The algorithm confirms that color uniformity has the greatest impact on quality through feature importance analysis, with a weight of about 0.4. The evaluation results show that the qualified rate of the optimized batch has increased from 85% to 95%, and the excellent product rate has increased from 40% to 60%. The meal quality consistency evaluation table that meets the aviation catering standards records the scores of each indicator and the classification results to provide a reference for process improvement. The embodiment of the application realizes fine management of quality through classification evaluation.

[0110] In the embodiment of the application, the quality optimization of aviation catering relies on a closed-loop mechanism of deviation analysis and parameter adjustment. Taking a steak as an example, the browning problem caused by excessive heat preservation time is identified through correlation analysis, and the neural network model further quantifies the influence of temperature and time to guide parameter optimization. The optimized instructions are strictly checked and allocated to tasks to ensure efficient execution of the production line. Real-time monitoring data shows that temperature control is accurate, raw material state is improved, and quality characteristics are significantly improved. The evaluation table provides detailed data for production managers, such as recording a batch that is upgraded to an excellent level due to reduced browning, to guide the adjustment of marinating process or heat preservation equipment settings. This intelligent and data-driven optimization scheme effectively reduces quality fluctuations, ensures that meals meet the high standards of different cabins, and improves the overall competitiveness of aviation services.

[0111] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to only the specific embodiments. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to provide the best illustration of the application and its practical application to those skilled in the art and to enable those skilled in the art to best utilize the application. The application is therefore limited only by the claims and their full scope and equivalents.

Claims

1. An artificial intelligence-based in-flight meal quality monitoring method, characterized by, The method comprises: Obtaining flight delay information, cabin configuration data and real-time production line task scheduling, combining preset meal standards, evaluating the influence of delay on task scheduling and meal holding time, generating task reorganization requirements and holding time extension prediction values; Real-time image acquisition is performed on meal semi-finished products of task reorganization requirements, edge detection and texture analysis are performed, material appearance quality indexes are obtained, and material attribute change trend is identified based on the obtained material appearance quality indexes; Cluster analysis is performed on the material attribute change trend, key feature parameters affecting meal quality are identified, and quality risk level of the current production batch is determined in combination with cabin meal standardization requirements and holding time extension prediction values; For high-risk meals with a quality risk level greater than a preset level, the image acquisition parameters are adjusted according to the appearance deviation of the high-risk meals and the material attribute change trend to obtain a new set of detection parameters; The new set of detection parameters is optimized by using a machine learning algorithm to generate a dynamic threshold table suitable for the current production conditions, and the real-time detection data is compared with the threshold table to obtain a meal quality consistency evaluation result, which is used to evaluate whether the meal quality meets the aviation catering standards; If the meal quality does not meet the aviation catering standards, the production line task scheduling is adjusted according to the quality risk level, the high-risk meals are preferentially processed, the production process of the adjusted high-risk meals is monitored, and standardized features of production images are extracted; Deviation data is extracted according to the standardized features, the correlation between the deviation data and the holding time and the material attribute is analyzed, new production control instructions are generated and sent to the production line, the holding time and the material processing process are adjusted, and meal quality consistency data meeting the aviation catering standards are obtained.

2. The method of claim 1, wherein the method is based on artificial intelligence. The method comprises: According to the flight delay time data and the cabin configuration data set, the flight corresponding meal making specification parameters are obtained from the preset meal specification library through data matching association rules, and the delay task scheduling change data set is obtained; For the delay task scheduling change data set, the airport catering loading time limit parameters are obtained from the preset task window database through the time window matching algorithm, and the candidate task sequence group meeting the loading time limit constraint is obtained; For the candidate task sequence group, the process execution threshold is obtained from the production line capacity constraint database, and the process holding constraint data set is obtained; According to the process holding constraint data set, the candidate task sequence group is screened to obtain the task reorganization requirements and the holding time extension prediction values.

3. The method of claim 1, wherein, The method further comprises: obtaining flight delay data and cabin configuration information, calculating the influence coefficient of delay on production line task scheduling, if the task scheduling influence coefficient is greater than the target value, adjusting the task priority according to the production line task scheduling and the preset meal standard, and predicting the meal distribution scheduling time according to the adjusted task priority by using a linear regression algorithm. 4.The method of claim 1, wherein, The meal semi-finished product requiring task reorganization is subjected to real-time image acquisition, edge detection and texture analysis to obtain raw material appearance quality indexes, and based on the obtained raw material appearance quality indexes, raw material attribute change trend is identified, including: Images in four directions of up, down, left and right of the meal semi-finished product are acquired, Gaussian noise elimination and brightness equalization processing are performed to obtain a pre-processing image group; According to the pre-processing image group, meal surface feature vectors are extracted, and a feature description matrix is calculated through Sobel operator and gray level co-occurrence matrix; For the feature description matrix, preset meal quality parameters are obtained from a quality standard database, and standard matching operation is performed through a feature similarity calculation unit to obtain semi-finished product quality index; According to the semi-finished product quality index and the continuous monitoring data sequence constructed by the time window divider, a quality change prediction model is established by using a random forest algorithm, real-time data is updated through a sliding time window, and a semi-finished product attribute change trend is obtained. 5.The method of claim 1, wherein, The raw material attribute change trend is subjected to cluster analysis to identify key feature parameters affecting meal quality, combined with cabin meal standardization requirements and extended prediction value of holding time, the quality risk level of the current production batch is determined, including: Raw material surface feature data is acquired, the raw material surface feature data is grouped to obtain a raw material feature clustering table; For the raw material feature clustering table, cabin meal specification requirements are obtained from a meal standard database, and a raw material feature and meal specification mapping relationship is established through a feature mapping unit to obtain a meal quality key feature table; According to the meal quality key feature table, feature standard thresholds are read from a quality evaluation rule library, and scores of three indexes of color, shape and odor are calculated through a feature quantization unit to obtain a raw material quality feature score matrix; For the raw material quality feature score matrix, a hierarchical clustering algorithm is used for risk level division, preset risk level thresholds are read from a risk early warning rule library, and the quality risk level of the current production batch is obtained.

6. The method of claim 1, wherein the method is based on artificial intelligence. For high-risk meals with a quality risk level greater than a preset level, image acquisition parameters are adjusted according to the appearance deviation of high-risk meals and the raw material attribute change trend to obtain a new detection parameter set, including: Detection data of appearance integrity, color uniformity and shape regularity are obtained through a risk scoring unit, if the risk score of the detection data is higher than a preset early warning value, new detection reference data is calculated; According to the detection reference data, a support vector machine algorithm is used to compare standard samples and measured images to obtain a feature deviation quantitative table; The feature deviation quantitative table is processed by a time window calculation module to generate an image acquisition frequency sequence, a detection sensitivity parameter is set, and a feature extraction rule set is obtained; For the feature extraction rule set, a real-time detection task is performed by a feature recognition module to obtain a new detection parameter set.

7. The method of claim 1, wherein, The method further comprises: obtaining computer vision detection meal appearance deviation data through a production line real-time monitoring system, extracting raw material attribute change data from a raw material quality monitoring system, standardizing the meal appearance deviation data and the raw material attribute change data, and dynamically adjusting exposure parameters of an image acquisition device according to the standardized deviation data. 8.The method of claim 1, wherein the method further comprises: The method further comprises: obtaining shape feature data, color feature data and texture feature data of the detection parameters, calculating the weight distribution of each feature data by using a random forest algorithm to obtain a detection parameter optimization matrix; constructing a detection threshold sequence through a threshold generation unit for the detection parameter optimization matrix, updating the detection threshold sequence by using a time window sliding method to obtain a baseline threshold table; performing temperature correction and humidity calibration on the threshold value through an environment compensator according to the baseline threshold table, generating a dynamic threshold interval by using a parameter self-adaptor to obtain a real-time detection threshold set; and obtaining meal quality consistency evaluation results by using a data collector to obtain meal detection data for the real-time detection threshold set, and calculating the matching degree between the detection data and the real-time detection threshold set by using a multi-dimensional comparator. If the meal quality does not meet the aviation catering standard, the production line task scheduling is adjusted according to the quality risk level, high-risk meals are preferentially processed, the production process of the adjusted high-risk meals is monitored, and standardized features of production images are extracted, including: calculating the quality risk value of each batch by using a risk scoring unit according to the quality consistency evaluation data, generating a task priority sequence for the quality risk value by using a priority calculator to obtain a high-risk task priority table; generating a production task sequence by using a task time sequence planner for the high-risk task priority table, and distributing the production task sequence according to the production line processing capacity by using a production capacity allocator to obtain a task execution scheme; obtaining a production process image sequence by using an image acquisition device according to the task execution scheme, obtaining color distribution, shape contour and texture structure data from the image sequence by using a feature extractor to obtain a standardized feature vector; and extracting deep feature representations for the standardized feature vector by using a convolutional neural network to obtain standardized features of the production images. The method further comprises: obtaining computer vision detection meal appearance deviation data through a production line real-time monitoring system, extracting raw material attribute change data from a raw material quality monitoring system, standardizing the meal appearance deviation data and the raw material attribute change data, and dynamically adjusting exposure parameters of an image acquisition device according to the standardized deviation data. The method further comprises: obtaining shape feature data, color feature data and texture feature data of the detection parameters, calculating the weight distribution of each feature data by using a random forest algorithm to obtain a detection parameter optimization matrix; constructing a detection threshold sequence through a threshold generation unit for the detection parameter optimization matrix, updating the detection threshold sequence by using a time window sliding method to obtain a baseline threshold table; performing temperature correction and humidity calibration on the threshold value through an environment compensator according to the baseline threshold table, generating a dynamic threshold interval by using a parameter self-adaptor to obtain a real-time detection threshold set; and obtaining meal quality consistency evaluation results by using a data collector to obtain meal detection data for the real-time detection threshold set, and calculating the matching degree between the detection data and the real-time detection threshold set by using a multi-dimensional comparator. The method further comprises: obtaining computer vision detection meal appearance deviation data through a production line real-time monitoring system, extracting raw material attribute change data from a raw material quality monitoring system, standardizing the meal appearance deviation data and the raw material attribute change data, and dynamically adjusting exposure parameters of an image acquisition device according to the standardized deviation data. 9.The method of claim 1, wherein, The method further comprises: obtaining shape feature data, color feature data and texture feature data of the detection parameters, calculating the weight distribution of each feature data by using a random forest algorithm to obtain a detection parameter optimization matrix; constructing a detection threshold sequence through a threshold generation unit for the detection parameter optimization matrix, updating the detection threshold sequence by using a time window sliding method to obtain a baseline threshold table; performing temperature correction and humidity calibration on the threshold value through an environment compensator according to the baseline threshold table, generating a dynamic threshold interval by using a parameter self-adaptor to obtain a real-time detection threshold set; and obtaining meal quality consistency evaluation results by using a data collector to obtain meal detection data for the real-time detection threshold set, and calculating the matching degree between the detection data and the real-time detection threshold set by using a multi-dimensional comparator. The method further comprises: obtaining computer vision detection meal appearance deviation data through a production line real-time monitoring system, extracting raw material attribute change data from a raw material quality monitoring system, standardizing the meal appearance deviation data and the raw material attribute change data, and dynamically adjusting exposure parameters of an image acquisition device according to the standardized deviation data. The method further comprises: obtaining shape feature data, color feature data and texture feature data of the detection parameters, calculating the weight distribution of each feature data by using a random forest algorithm to obtain a detection parameter optimization matrix; constructing a detection threshold sequence through a threshold generation unit for the detection parameter optimization matrix, updating the detection threshold sequence by using a time window sliding method to obtain a baseline threshold table; performing temperature correction and humidity calibration on the threshold value through an environment compensator according to the baseline threshold table, generating a dynamic threshold interval by using a parameter self-adaptor to obtain a real-time detection threshold set; and obtaining meal quality consistency evaluation results by using a data collector to obtain meal detection data for the real-time detection threshold set, and calculating the matching degree between the detection data and the real-time detection threshold set by using a multi-dimensional comparator. ​ ​ 10.The method of claim 1, wherein, ​ ​ For the quality deviation mapping table, a temperature influence model and a time influence model are established through a neural network algorithm, and a parameter optimizer is used to generate temperature control values and holding time values according to the influence models; According to the temperature control values and holding time values, a device control instruction is generated, and an instruction verifier is used to verify the compliance of the device control instruction to generate a new production control instruction; For the production control instruction, a temperature change curve and a holding time record are obtained through a parameter collector, and a random forest algorithm is used to classify and evaluate the temperature change curve and the holding time record to obtain final food quality consistency data.

Images