Crisp moon cake production safety quality monitoring method and system based on artificial intelligence

Abstract

The invention relates to the technical field of mooncake quality detection, and discloses a crisp mooncake production safety quality monitoring method and system based on artificial intelligence. The method comprises the following steps: carrying out adaptive enhancement and depth identification processing on multi-source sensing data to obtain multi-dimensional characterization feature data; collaborative quality detection processing is carried out on the multi-dimensional characterization feature data to obtain quality detection feature data; performing historical reference dynamic grading processing on an appearance quality index and an internal quality index in the quality detection characteristic data to obtain grade evaluation mapping data, performing real-time monitoring processing on key process parameter change and quality abnormity to obtain quality risk early warning data, and performing process correlation analysis and optimization processing on the quality risk early warning data and historical production data; and obtaining parameter regulation and control decision data. Intelligent monitoring and early warning of the whole mooncake production process are achieved, and the stability of product quality is improved through automatic optimization of technological parameters.

Description

Technical Field

[0001] The present application relates to the field of mooncake quality inspection, and in particular to an artificial intelligence-based crispy mooncake production safety and quality monitoring method and system. Background Art

[0002] In the mooncake production process, traditional quality monitoring methods mainly rely on manual visual inspection and simple machine vision inspection. Manual visual inspection has problems such as strong subjectivity, low efficiency, and poor consistency; while simple machine vision inspection is mainly based on basic algorithms such as edge detection and color analysis, and can only detect the basic appearance characteristics of the product, such as size and shape. At the same time, existing inspection equipment often separates appearance inspection from internal quality inspection, resulting in a lack of systematicity and relevance in the inspection results. Traditional production process parameter adjustments mainly rely on experience judgment, lacking data support and real-time feedback mechanisms.

[0003] The relevant technologies have the following deficiencies: Traditional detection methods cannot take into account both the appearance characteristics and internal quality of mooncakes at the same time, especially for products with complex hierarchical structures such as crispy mooncakes, whose subtle features such as wrinkle characteristics, hierarchical structure, and crack distribution are difficult to accurately identify. In addition, there is a lack of effective correlation analysis and feedback mechanism between the detection system and production process parameters, resulting in the inability to timely warn and deal with quality abnormalities, affecting the stability and consistency of product quality. Summary of the invention

[0004] The present application provides an artificial intelligence-based crispy mooncake production safety and quality monitoring method and system, which is used to establish an intelligent mooncake production safety and quality monitoring method that can deeply integrate multi-source perception data, quality inspection, and process parameter optimization, to achieve intelligent monitoring and early warning of the entire mooncake production process, and to improve the stability of product quality through automatic optimization of process parameters.

[0005] In the first aspect, the present application provides a mooncake production safety and quality monitoring method based on artificial intelligence, and the mooncake production safety and quality monitoring method based on artificial intelligence includes: performing time-series synchronous acquisition and ID association processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line to obtain multi-source perception data; performing adaptive enhancement and deep recognition processing on the multi-source perception data to obtain multi-dimensional characterization feature data; performing collaborative quality inspection processing on the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics in the multi-dimensional characterization feature data to obtain quality inspection feature data; performing historical reference dynamic grading processing on the appearance quality indicators and internal quality indicators in the quality inspection feature data to obtain grade evaluation mapping data; performing real-time monitoring processing on the changes in key process parameters and quality anomalies on the grade evaluation mapping data to obtain quality risk warning data; performing process correlation analysis and optimization processing on the quality risk warning data and historical production data to obtain parameter control decision data.

[0006] In a second aspect, the present application provides a mooncake production safety and quality detection system based on artificial intelligence, and the mooncake production safety and quality detection system based on artificial intelligence includes:

[0007] The acquisition module is used to perform time-series synchronous acquisition and ID association processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line to obtain multi-source perception data;

[0008] A recognition module, used to perform adaptive enhancement and deep recognition processing on the multi-source perception data to obtain multi-dimensional representation feature data;

[0009] A detection module, used to perform collaborative quality detection processing on the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics in the multi-dimensional characterization feature data to obtain quality detection feature data;

[0010] A grading module, used for performing historical reference dynamic grading processing on the appearance quality index and the internal quality index in the quality inspection feature data to obtain grade evaluation mapping data;

[0011] A monitoring module, used to perform real-time monitoring and processing of key process parameter changes and quality anomalies on the grade evaluation mapping data to obtain quality risk warning data;

[0012] The processing module is used to perform process correlation analysis and optimization processing on the quality risk warning data and historical production data to obtain parameter control decision data.

[0013] In the technical solution provided by the present application, the 360° multi-angle high-definition image group and internal structure infrared scanning data on the mooncake production line are collected in time series and ID-associated, so as to realize the all-round and blind-angle collection of the appearance and internal features of the mooncakes, and ensure the integrity and consistency of the data. By adaptively enhancing and deeply identifying multi-source perception data, complex feature information such as surface wrinkle features and hierarchical structure features are effectively extracted, overcoming the limitation of traditional detection methods relying on simple geometric features. When the multi-dimensional characterization feature data is processed for collaborative quality inspection, the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics are comprehensively analyzed to realize the integrated detection of internal and external quality. By dynamically grading the quality inspection feature data with historical reference, an adaptive evaluation system based on historical data is established, making the quality evaluation more objective and accurate. In the process of processing the grade evaluation mapping data, the changes in key process parameters and real-time monitoring of quality anomalies are realized to realize the timely discovery and early warning of quality anomalies in the production process. Finally, by performing process correlation analysis and optimization processing on the quality risk warning data and historical production data, a closed-loop quality control mechanism is formed, which effectively improves the stability of the production process and the consistency of product quality. It realizes intelligent management of the entire process from data collection, feature extraction, quality detection to early warning optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0015] Figure 1 This is a schematic diagram of an embodiment of a mooncake production safety and quality monitoring method based on artificial intelligence in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of an artificial intelligence-based mooncake production safety and quality inspection system in the embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application embodiment provides a method and system for monitoring the safety and quality of moon cake production based on artificial intelligence. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the mooncake production safety and quality monitoring method based on artificial intelligence includes:

[0019] Step S101, performing time-series synchronous acquisition and ID association processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line to obtain multi-source perception data;

[0020] Step S102, performing adaptive enhancement and deep recognition processing on the multi-source perception data to obtain multi-dimensional representation feature data;

[0021] Step S103, performing collaborative quality inspection processing on the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics in the multi-dimensional characterization feature data to obtain quality inspection feature data;

[0022] Step S104, performing historical reference dynamic grading processing on the appearance quality index and the internal quality index in the quality inspection feature data to obtain grade evaluation mapping data;

[0023] Step S105: Perform real-time monitoring of changes in key process parameters and quality anomalies on the grade evaluation mapping data to obtain quality risk warning data;

[0024] Step S106: Perform process correlation analysis and optimization processing on the quality risk warning data and historical production data to obtain parameter control decision data.

[0025] It is understandable that the execution subject of the present application can be a mooncake production safety and quality inspection system based on artificial intelligence, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.

[0026] Specifically, when performing time-series synchronous acquisition and ID association processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line, a "3+2" layout scheme is adopted. Three high-definition cameras with an angle of 120° are arranged above the mooncake conveyor belt, and one camera with a 45° depression angle is arranged on both sides. Each camera uses a CMOS sensor with more than 10 million pixels. The camera takes 500 high-definition photos in 1 / 10 second, and the time sequence synchronization is performed through the industrial control network to ensure that the timestamps of all image data of each sample are aligned. The infrared sensor operates in the 800-2500nm band and uses diffuse reflection mode to collect the internal structure features of mooncakes. Each mooncake sample generates a unique ID code, and all sensor data are associated with the corresponding sample. When performing adaptive enhancement and depth recognition processing on multi-source perception data, the improved multi-scale Retinex algorithm is used to perform illumination normalization processing on the image. The infrared scanning data is denoised by wavelet transform, and the internal structure features are enhanced by adaptive threshold segmentation. Especially for the characteristics of crispy mooncakes, the focus is on extracting surface wrinkle characteristics, hierarchical structure characteristics, crack distribution characteristics and color uniformity characteristics.

[0027] In the collaborative quality inspection and processing link, the appearance specification parameters, surface texture integrity, internal structure uniformity and filling distribution characteristics are comprehensively analyzed. For example, for a batch of crispy mooncakes, the surface texture integrity is analyzed, the uniformity of the internal layers is evaluated, and the distribution status of the filling is detected by measuring its diameter, height and other specification parameters. When it is detected that the surface texture integrity of a mooncake is less than 90%, or the internal layer uniformity deviation exceeds 2mm, it is marked as a product that requires special attention. In the process of historical reference dynamic grading, quantitative evaluation indicators for appearance quality and internal quality are established. The high-quality product parameters in the historical production data are used as reference standards to dynamically adjust the grading threshold. For example, by analyzing the production data of the past month, the optimal range and acceptable range of surface wrinkle characteristics are determined, and the current product is dynamically graded.

[0028] During the real-time monitoring and processing stage of key process parameter changes and quality anomalies, key process parameters such as baking temperature and time are monitored. When the baking temperature fluctuation of a batch of mooncakes exceeds ±5°C, or the baking time deviation exceeds 30 seconds, quality risk warning information is generated in a timely manner. During process correlation analysis and optimization processing, the quality risk warning data is deeply correlated with historical production data for analysis. For example, when a batch of mooncakes had surface cracks, it was found through analysis that it was related to the too fast heating rate of the baking temperature, and based on this, process parameter optimization suggestions were generated, such as adjusting the heating rate from the original 15°C / minute to 12°C / minute.

[0029] In the embodiment of the present application, by performing time-series synchronous acquisition and ID association processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data on the mooncake production line, the full-range and blind-angle-free acquisition of the appearance and internal features of the mooncakes is achieved, ensuring the integrity and consistency of the data. By adaptively enhancing and deeply identifying multi-source perception data, complex feature information such as surface wrinkle features and hierarchical structure features is effectively extracted, overcoming the limitations of traditional detection methods that rely on simple geometric features. When performing collaborative quality inspection processing on multi-dimensional characterization feature data, the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics are comprehensively analyzed to achieve integrated detection of internal and external quality. By performing historical reference dynamic grading processing on quality inspection feature data, an adaptive evaluation system based on historical data is established, making quality evaluation more objective and accurate. In the process of processing the grade evaluation mapping data, the timely discovery and early warning of quality anomalies in the production process are achieved through real-time monitoring of changes in key process parameters and quality anomalies. Finally, by performing process association analysis and optimization processing on quality risk warning data and historical production data, a closed-loop quality control mechanism is formed, which effectively improves the stability of the production process and the consistency of product quality. It realizes intelligent management of the entire process from data collection, feature extraction, quality detection to early warning optimization.

[0030] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0031] (1) A high-definition camera array is configured above and around the mooncake production line to perform high-speed multi-angle image acquisition and processing to obtain all-round appearance image data;

[0032] (2) Performing illumination compensation and enhancement processing on the omnidirectional appearance image data to obtain a 360° multi-angle high-definition image group;

[0033] (3) Scanning and imaging the internal structure of the moon cake using an infrared sensor to obtain infrared scanning data of the internal structure;

[0034] (4) Perform industrial network timing synchronization processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data to obtain time-aligned data;

[0035] (5) Perform unique ID coding association processing on the time-aligned data to obtain sample identification data;

[0036] (6) The sample identification data and the time alignment data are packaged and processed to obtain multi-source perception data.

[0037] Specifically, in the image acquisition stage of the mooncake production line, a "3+2" HD camera array layout is adopted, that is, 3 HD cameras are evenly arranged at an angle of 120° above the conveyor belt for top-view shooting, and 1 camera with a 45° depression angle is arranged on each side for side-view shooting. Each HD camera uses a CMOS sensor with a resolution of 4096×3072 pixels and a sampling rate of 5000 frames per second. It completes the shooting of 500 photos within 1 / 10 second when the mooncake passes through the detection area, realizing all-round image acquisition without dead angles. Each camera is set with an aperture value of F4.0 and an exposure time of 1 / 10000 seconds to ensure image clarity under high-speed motion. The illumination compensation enhancement processing adopts a patented lighting system, which includes a main light source and an auxiliary light source. The main light source adopts an LED array with a luminous flux of 1000 lumens, a color temperature of 5500K, and a color rendering index of Ra≥95, which is synchronized with the camera exposure through pulse synchronization triggering. Auxiliary light sources are arranged around to eliminate shadows. Light compensation uses an adaptive algorithm to adjust the intensity and angle of the light source in real time according to the average brightness value and brightness variance of the image. For the collected images, histogram equalization is first performed, then Gaussian filtering is used to eliminate noise, and finally sharpening is performed to improve detail performance.

[0038] The internal structure scan uses a near-infrared sensor with an operating wavelength range of 800-2500nm and a diffuse reflection mode. During scanning, the sensor scans the mooncake layer by layer at a frequency of 300Hz, with a scanning resolution of 512×512 pixels and a scanning depth of up to 30mm, forming a 10-layer cross-sectional image. Each layer of scanning data includes reflection intensity values ​​and depth information, which are used to construct a three-dimensional model of the internal structure of the mooncake. During the scanning process, the sensor's transmitting power is 5mW, the receiving sensitivity is -60dBm, and the signal-to-noise ratio is greater than 40dB.

[0039] The timing synchronization of the industrial network adopts the IEEE 1588 Precision Time Protocol (PTP), and the network delay measurement accuracy is better than 100 nanoseconds. All acquisition devices are connected via industrial Ethernet with a transmission bandwidth of 1Gbps. Each data packet carries a 64-bit timestamp, and the clock synchronization accuracy is better than 1 microsecond. During the data alignment process, a unified time reference is first established, and then the data collected by different devices are time-matched according to the timestamp to ensure the timing consistency of the data. The unique ID code adopts a 32-bit structure: production date (8 bits) + production line number (4 bits) + batch number (8 bits) + serial number (12 bits). Coding example: 20240308-0001-000000001-000000000001 means the first product of the first batch of production line No. 1 on March 8, 2024. Each code is verified by CRC32 to ensure uniqueness and correctness.

[0040] The data packet encapsulation adopts a hierarchical structure, including: header information (128 bytes), including ID code, timestamp, checksum; image data area, storing 360° multi-angle images; scanning data area, storing infrared scanning data; tail checksum information (32 bytes). Data compression adopts lossless compression algorithm to ensure data integrity.

[0041] For example, a mooncake production line has a speed of 60 pieces / minute, and the above scheme is used for data acquisition and processing. Taking a mooncake with a diameter of 8cm and a height of 3cm as an example, 5 high-definition cameras collect a total of 2,500 original images (500 images / camera) in 1 / 10 second, each image is 12.6MB (4096×3072×3 bytes), and the total data volume of the original images is 31.5GB. After illumination compensation enhancement and screening, 500 valid images are retained, and the data volume is reduced to 6.3GB. Infrared scanning generates 10 layers of cross-sectional images, each layer is 262KB (512×512×1 byte), totaling 2.62MB. After time series synchronization processing, a multi-source perception data packet of about 6.31GB is generated for each mooncake. During the 8-hour production process, a total of 28,800 samples are processed. Through this scheme, the complete acquisition of the appearance characteristics and internal structure of the mooncakes is achieved, and the data acquisition accuracy reaches 99.9%.

[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0043] (1) Separating the security quality features of multi-source perception data to obtain a surface feature group and an internal feature group, and performing quality feature screening on the surface feature group and the internal feature group to obtain basic feature data;

[0044] (2) Performing stereoscopic reconstruction on the wrinkle texture in the basic feature data to obtain a wrinkle stereoscopic feature map, and performing deep layering on the wrinkle stereoscopic feature map to obtain multi-level structure data;

[0045] (3) performing fracture identification processing on the interlayer connection relationship of the multi-level structure data to obtain a crack feature distribution map, and performing feature parameter extraction processing on the crack feature distribution map to obtain crack morphological parameters;

[0046] (4) performing color brightness equalization processing on the crack morphology parameters to obtain color uniformity characteristics, and performing local contrast enhancement processing on the color uniformity characteristics to obtain enhanced color data;

[0047] (5) performing safety quality boundary extraction processing on the enhanced color data to obtain feature boundary data, and performing feature fusion processing on the feature boundary data to obtain candidate feature data;

[0048] (6) performing classification processing on the surface wrinkle degree, layer distribution state and crack distribution density in the candidate feature data to obtain quality feature grade data, and cross-mapping processing on the quality feature grade data and the internal color uniformity to obtain comprehensive feature mapping data;

[0049] (7) performing spatial correlation processing on the surface defect position and the internal structure change area in the comprehensive feature mapping data to obtain defect correlation data, and performing weight assignment processing on the defect correlation data to obtain feature weight data;

[0050] (8) sorting the quality influencing factors of the feature weight data to obtain an influencing factor sequence, and performing threshold determination on the influencing factor sequence to obtain associated determination data;

[0051] (9) Perform feature combination processing on the association judgment data to obtain multi-dimensional representation feature data.

[0052] Specifically, feature separation is performed to divide the multi-source perception data into a surface feature group and an internal feature group. The surface feature group contains the appearance information extracted from the high-definition image, and the internal feature group contains the internal structure information obtained by infrared scanning. The features are screened by a convolutional neural network, and the key feature points are extracted to form basic feature data. The feature screening adopts a double threshold method, setting the high threshold T1 = 0.85 and the low threshold T2 = 0.15, and retaining the feature points with confidence within this interval. The wrinkle texture information in the basic feature data is processed by a stereo reconstruction algorithm. First, the depth map is constructed using the parallax information of the multi-view image. The resolution of the depth map is consistent with the original image, and the depth accuracy reaches 0.1mm. Then, the three-dimensional structure of the wrinkle is reconstructed based on the depth map to form a three-dimensional feature map of the wrinkle. The reconstructed three-dimensional features are layered, and the spatial structure is evenly divided into 10 levels, with an interval of 0.5mm between each layer, to obtain multi-level structural data.

[0053] When identifying fractures in multi-level structural data, an edge detection algorithm is used to analyze the continuity between adjacent levels. When the inter-layer difference is detected to exceed the set threshold (standard deviation is greater than 0.3mm), it is marked as a potential fracture point. All fracture points are connected to form a crack feature distribution map, and the morphological parameters such as length, width, and direction of the crack are extracted. The crack length accuracy reaches 0.05mm, and the angle accuracy is 0.1 degrees. The color processing link first balances the color brightness of the crack morphological parameters. The adaptive histogram equalization algorithm is used to redistribute the image grayscale value to the [0,255] interval to obtain the color uniformity feature. The local contrast enhancement is performed on the balanced features, and the enhancement radius is set to 15 pixels and the gain coefficient is 1.5 to highlight the local details and form enhanced color data.

[0054] The safety quality boundary extraction adopts the regional growing algorithm, with the color mutation point as the seed point, and expands outward to form the feature boundary. The boundary extraction accuracy is controlled within 1 pixel, and the boundary smoothness coefficient is set to 0.8. The extracted boundary features are fused, and the weighted average method is used to merge the overlapping areas. The weight coefficient is dynamically assigned according to the feature significance to obtain the candidate feature data. In the process of quality feature grading, the surface wrinkle degree, layer distribution state and crack distribution density are divided into 5 levels. The wrinkle degree is divided according to the wrinkle depth (<0.5mm is mild, >2mm is severe); the layer distribution is evaluated by uniformity (standard deviation <0.1mm is excellent, >0.5mm is unqualified); the crack density is calculated by the number of cracks per unit area (<3 / cm 2 Normal: >10 / cm 2 The grading results are cross-mapped with the internal color uniformity to establish a comprehensive evaluation matrix.

[0055] During the spatial correlation analysis, the surface defect position is registered with the internal structural change area, and the registration accuracy is better than 0.2mm. According to the spatial distribution characteristics of the defects, the correlation coefficient is calculated to establish defect correlation data. Weights are assigned to the correlation data, and the weight values ​​are set according to the degree of influence of the defects on product quality (critical defects 1.0, minor defects 0.6, minor defects 0.3) to obtain feature weight data. The principal component analysis method is used to sort the quality influencing factors and extract the main influencing factors in the feature weight data. The influencing factors are sorted by weight, and the key threshold is set to 0.75 to screen out the main influencing factors and form an influencing factor sequence. The sequence is threshold judged, and the threshold is set to 1.5 times the weight mean. Factors exceeding the threshold are marked as key focus items to obtain correlation judgment data.

[0056] Finally, through the feature combination algorithm, the features in the associated judgment data are combined according to the preset rules to form the final multi-dimensional characterization feature data. For example, the puff pastry mooncakes on a production line have the following data after the above processing: the average surface wrinkle depth is 1.2mm, the standard deviation is 0.15mm; the layer uniformity deviation is 0.08mm; the crack density is 2.5 / cm 2 ; Color uniformity score is 92 points (out of 100 points). Through spatial correlation analysis, it is found that there is an 85% correlation between surface wrinkles and internal layer distribution, and the process parameters are optimized and adjusted accordingly. The acquisition and processing time of these data is controlled within 100ms, meeting the real-time monitoring requirements.

[0057] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0058] (1) performing dimensional feature extraction processing on the appearance specification parameters in the multi-dimensional characterization feature data to obtain specification feature data, and performing standard specification matching processing on the specification feature data to obtain specification compliance data;

[0059] (2) performing surface texture feature extraction processing on the specification conformity data to obtain a texture feature map, and performing integrity assessment processing on the texture feature map to obtain surface texture integrity data;

[0060] (3) performing internal structure feature extraction processing on the surface texture integrity data to obtain hierarchical uniformity data, and performing structural distribution evaluation processing on the hierarchical uniformity data to obtain internal structure uniformity data;

[0061] (4) performing sandwich area positioning processing on the internal structure uniformity data to obtain sandwich boundary data, and performing distribution feature extraction processing on the sandwich boundary data to obtain sandwich distribution feature data;

[0062] (5) The sandwich distribution characteristic data and the specification compliance data are collaboratively evaluated to obtain internal and external characteristic correlation data, and the internal and external characteristic correlation data are subjected to quality parameter extraction processing to obtain quality inspection characteristic data.

[0063] Specifically, the size features of the appearance specification parameters are extracted, focusing on the key size parameters of the mooncakes, such as diameter, height, and volume. The mooncake outline is extracted by edge detection algorithm, and the minimum circumscribed rectangle method is used to measure the outer dimensions, with a measurement accuracy of 0.1mm. The extracted size data is matched and compared with the standard specifications, and the deviation values ​​of various parameters are calculated. The specification compliance is calculated based on the deviation value, and the standard specification allows a deviation range of ±2%. Samples that exceed the range are marked as unqualified. The surface texture feature extraction of the specification compliance data uses the gray level co-occurrence matrix method to calculate the energy, entropy, contrast, correlation and other eigenvalues ​​of the texture. The texture feature extraction uses 16×16 pixels as the basic unit, with a step size of 8 pixels, and calculates the eigenvalues ​​of four directions (0°, 45°, 90°, 135°). After feature calculation, a texture feature map is formed, which contains information such as texture directionality, roughness, and regularity. The texture feature map is evaluated for integrity, mainly detecting abnormalities such as texture fracture, missing, and deformation. The evaluation results are expressed in 0-100 points, and 90 points or more are high-quality products.

[0064] The surface texture integrity data is further used to extract internal structural features, and the hierarchical structure of mooncakes is analyzed in combination with infrared scanning data. The image segmentation algorithm is used to identify the boundaries of different layers, and the interlayer distance and uniformity are calculated. The hierarchical uniformity data contains the mean and standard deviation of the interlayer distance, and the standard deviation is less than 0.2mm, which is considered to be structurally uniform. The hierarchical uniformity data is evaluated for structural distribution, the spatial distribution characteristics of the layers are analyzed, the continuity and integrity of each layer are evaluated, and the internal structural uniformity data is formed. In the process of sandwich area positioning, the regional growing algorithm is used to determine the boundary of the sandwich. The highlighted area in the infrared image is used as the seed point, and it is gradually expanded until the boundary threshold is reached. The sandwich boundary data records information such as the location, shape and area of ​​the boundary. The distribution feature of the sandwich boundary data is extracted, and the centering, shape regularity and filling uniformity of the sandwich are analyzed to obtain the sandwich distribution feature data. Samples with a sandwich centering deviation of more than 2mm or a filling unevenness of more than 10% need to be paid special attention.

[0065] Finally, a collaborative evaluation process is performed to correlate the sandwich distribution feature data with the specification compliance data. A feature correlation matrix is ​​constructed to analyze the correlation between the appearance specifications and the internal structure. Quality feature parameters are extracted from the correlation data, including appearance indicators (dimensional accuracy, surface integrity) and internal indicators (layer uniformity, sandwich distribution), to form the final quality inspection feature data.

[0066] For example, during the production of a batch of mooncakes with crispy skin, the standard product specifications are 80mm in diameter and 30mm in height. After dimensional feature extraction, the actual size of a sample was measured to be 79.8mm in diameter and 29.7mm in height, and the calculated specification compliance was 98%. Surface texture analysis showed that the texture integrity score was 92 points, and the texture direction consistency was good. Internal hierarchical analysis found that the sample had 12 clear crispy skin layers, with an average spacing of 2.1mm between layers, a standard deviation of 0.15mm, and good structural uniformity. Sandwich analysis showed that the sandwich area accounted for 23%, the centering deviation was 1.2mm, and the filling uniformity was 95%. The collaborative evaluation results showed that the appearance and internal features of the sample were coordinated by 96%, and the quality inspection feature parameters were all within the control range, and it was determined to be a high-quality product. The entire inspection process took 0.8 seconds, meeting the real-time inspection needs of the production line. Through this multi-level and multi-dimensional feature extraction and evaluation method, a comprehensive inspection of the quality of mooncakes was achieved to ensure that the product met the standard requirements in both appearance and internal quality. This method not only improves the accuracy and efficiency of detection, but also provides reliable data support for quality control of the production process.

[0067] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0068] (1) performing standard parameter extraction processing on the appearance quality index in the quality inspection feature data to obtain appearance quality benchmark data, and performing historical data comparison processing on the appearance quality benchmark data to obtain appearance reference data;

[0069] (2) extracting characteristic parameters of internal quality indicators in the quality inspection characteristic data to obtain internal quality benchmark data, and performing historical data comparison processing on the internal quality benchmark data to obtain internal reference data;

[0070] (3) performing dynamic threshold division processing on the appearance reference data and the internal reference data to obtain reference grade interval data, and performing grading standard matching processing on the reference grade interval data to obtain standard grade data;

[0071] (4) Performing quality grade mapping processing on the standard grading data to obtain grade corresponding data, and performing historical trend analysis processing on the grade corresponding data to obtain trend characteristic data;

[0072] (5) Dynamically classify the trend feature data to obtain classified reference data, and perform quality grade evaluation on the classified reference data to obtain grade evaluation mapping data.

[0073] Specifically, standard parameters are extracted for the appearance quality indicators, including the diameter, height, surface texture integrity, color uniformity and other key parameters of the mooncakes. The appearance quality indicators use a multi-dimensional scoring method to standardize each parameter to a range of 0-100 points. The scores for diameter and height are calculated based on the deviation between the actual measured value and the standard value. The deviation is within ±1% and the score is 95-100 points, ±1%-2% and the score is 85-94 points, and ±2%-3% and the score is 75-84 points. The surface texture integrity is obtained by image analysis to evaluate the regularity and uniformity of the folds. The integrity is above 98% and the score is 95-100 points, and 95%-98% and the score is 85-94 points. Color uniformity is calculated based on color variance. The variance is less than 5% and the score is 95-100 points, and 5%-10% and the score is 85-94 points. When comparing the appearance quality benchmark data with historical data, the sliding window method is used, and the window size is set to the first 30 batches of sample data. Calculate the mean and standard deviation of each indicator of the current sample and historical data to form the appearance reference data. The same method is used to process internal quality indicators, including layer uniformity, filling distribution, filling density and other parameters. In the internal quality scoring standard, layer uniformity is evaluated based on the standard deviation of the inter-layer distance. If the standard deviation is less than 0.2mm, the score is 95-100 points, and 0.2-0.4mm is 85-94 points; the filling distribution is based on the center offset. If the offset is less than 1mm, the score is 95-100 points, and 1-2mm is 85-94 points; the filling density is obtained through infrared scanning data analysis. If the density is greater than 95%, the score is 95-100 points, and 90%-95% is 85-94 points.

[0074] The appearance reference data and internal reference data are divided into four levels: superior, first-class, qualified, and unqualified. The threshold is determined by comprehensively considering the distribution of historical data and the current production status, and the rationality of the grading standard is maintained through adaptive adjustment. The grade interval after division is matched with the standard grading requirements to ensure that the grading results meet the requirements of industry standards. The standard grading data contains the scores of each sample on various indicators and the final grade judgment. The quality grade mapping adopts the weighted scoring method, and the weight ratio of appearance quality and internal quality is 4:6. Sub-weights are set according to the importance of different quality indicators, such as 40% for layer uniformity, 35% for sandwich distribution, and 25% for filling density in internal quality. The comprehensive score is obtained through weight calculation, and the corresponding grade data is generated. The historical trend analysis of the corresponding grade data is carried out, and the exponential smoothing method is used to predict the quality change trend, and the change law of each indicator is observed to obtain the trend characteristic data.

[0075] In the process of dynamic grading based on trend feature data, a time series weight factor is introduced, and the weight of recent data is larger and decays over time. The grading reference data generated in this way can better reflect the current production status. Finally, the quality grade is evaluated, and the static indicator score and dynamic change trend are comprehensively considered to form the final grade evaluation mapping data.

[0076] Take the actual application of a crispy mooncake production line as an example: during the 8-hour continuous production process, a total of 1,800 samples were tested. The standard specifications are diameter 80mm±1mm, height 30mm±0.5mm, surface wrinkles 12-15, and layers 10-12. The specific testing process of a sample is as follows: First, the appearance parameters are measured, and the diameter is 80.3mm (deviation 0.37%, score 98 points), height 29.8mm (deviation 0.67%, score 96 points), surface texture integrity 97% (score 90 points), color uniformity variance 4.2% (score 96 points). Internal testing shows that the standard deviation of layer uniformity is 0.18mm (score 96 points), the sandwich offset is 0.8mm (score 97 points), and the filling density is 96.5% (score 97 points). After weighted calculation, the comprehensive score is 96.5 points. Compared with the data of the first 30 batches, all indicators of this sample are within the normal fluctuation range (within ±2 standard deviations). Through dynamic threshold classification, it is judged as a superior product, and the quality trend remains stable. The entire evaluation process takes 0.5 seconds, meeting the real-time monitoring needs of the production line.

[0077] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0078] (1) Process parameter extraction is performed on the grade evaluation mapping data to obtain key parameter data, and change trend analysis is performed on the key parameter data to obtain parameter change trend data;

[0079] (2) performing abnormal fluctuation detection processing on the baking temperature and time parameters in the parameter change trend data to obtain process fluctuation data, and performing quality impact analysis processing on the process fluctuation data to obtain quality fluctuation data;

[0080] (3) Calculating the deviation between the quality fluctuation data and the quality benchmark standard to obtain quality deviation data, and performing abnormality determination processing on the quality deviation data to obtain abnormal characteristic data;

[0081] (4) Performing quality risk assessment on the abnormal feature data to obtain risk level data, and performing time series correlation analysis on the risk level data to obtain risk correlation data;

[0082] (5) Perform early warning threshold determination on risk-related data to obtain early warning indicator data, and perform real-time monitoring on the early warning indicator data to obtain quality risk early warning data.

[0083] Specifically, key process parameters are extracted, with a focus on parameters such as baking temperature, baking time, molding pressure, and proofing time. Time series data with a unit of 10 minutes is established for each parameter, and the parameter value and its change are recorded. The sampling interval of the baking temperature is 5 seconds, the time parameter is accurate to the second level, and the sampling frequency of the pressure parameter is 10Hz. The extracted parameter data is statistically analyzed, and statistical quantities such as the mean, standard deviation, and coefficient of variation are calculated to evaluate the stability of the parameters. The sliding time window method is used for trend analysis, and the window width is set to 30 minutes, sliding every 5 minutes. For baking temperature changes, the standard temperature range is set to 180±5℃, and the temperature rise rate, holding time, and decline rate are recorded. Time parameter monitoring includes the duration of each process and its fluctuations. For example, the standard baking time is 25 minutes, and the allowable fluctuation range is ±2 minutes. By analyzing the change curve of the parameter over time, the mutation point and abnormal fluctuation are identified.

[0084] Statistical process control methods are used to detect abnormal fluctuations in baking temperature and time parameters. The 3σ principle is used for temperature control. When the temperature deviation exceeds 3 times the standard deviation, it is judged to be abnormal. The fluctuation control of time parameters is more stringent, and the 2σ standard is used. Abnormal fluctuation detection also takes into account the rate of change of parameters. An alarm is triggered when the temperature change rate exceeds 2℃ / minute or the accumulated time deviation exceeds 5 minutes. The process fluctuation data records information such as the time when the abnormality occurs, the duration, and the deviation amplitude. In the quality impact analysis stage, a correlation model between process parameters and quality indicators is established. Temperature fluctuations have a significant impact on the color and crispness of the product, and time deviations mainly affect the degree of ripening of the product. The process fluctuation data is compared with the quality data collected in real time to quantify the impact of process fluctuations on quality. The quality fluctuation data includes the change amount of each quality indicator and its correlation coefficient with the process parameters.

[0085] During the deviation calculation process, the quality fluctuation data is compared with the preset quality benchmark standards. The benchmark standards include: product surface goldenness 90±5%, crispness score ≥85 points, and moisture content 18±2%. The absolute deviation and relative deviation between the actual value and the standard value are calculated to generate quality deviation data. The deviation data is judged as abnormal, and a multi-indicator comprehensive evaluation method is used to set different levels of warning thresholds.

[0086] The quality risk assessment adopts the hierarchical analysis method to assign weights to each risk indicator according to its importance. The weight of temperature anomaly is 0.4, the weight of time anomaly is 0.3, and the weight of other parameter anomalies is 0.3. Products are divided into four levels according to the degree of risk: normal, slight risk, medium risk and severe risk. The risk level data records the risk score and level determination results of each sample. The time series correlation analysis studies the propagation law of risk events, and uses a minimum time window of 5 minutes to analyze the changing trend of risk levels. The periodic characteristics of risks are discovered through autocorrelation analysis, and the transmission relationship between different risk factors is discovered through cross-correlation analysis. Risk association data describes the distribution characteristics of risks in time and space.

[0087] The early warning threshold is determined by the dynamic threshold method, and the early warning ladder is set based on the distribution characteristics of historical data. The threshold for minor risks is set at the mean ± 2σ, the threshold for medium risks is set at the mean ± 3σ, and the threshold for severe risks is set at the mean ± 4σ. The early warning indicator data includes the triggering conditions and current status of each level of early warning. During real-time monitoring, data collection and analysis are performed every 10 seconds, and an early warning signal is generated immediately once the early warning threshold is exceeded.

[0088] For example, during the 8-hour continuous production process of a mooncake production line, the temperature and time parameters of the baking process fluctuated abnormally. The temperature record showed that the temperature suddenly rose in the third hour, from 180°C to 188°C, and the heating rate reached 4°C / minute, exceeding the standard limit of 2°C / minute. At the same time, the baking time of this batch of products was shortened to 22 minutes, 25 minutes lower than the standard time. Through quality impact analysis, it was found that the abnormal temperature caused the surface color of the product to deepen, and the golden degree reached 98%, exceeding the standard range; the shortened time caused the crispness of the product to decrease to 82 points, lower than the required 85 points. After risk assessment, this batch of products was judged to be of medium risk and triggered an orange warning. The warning information includes abnormal parameters, deviation values, impact range and processing suggestions. The production line then adjusted the process parameters to bring the product quality back to the control range. The entire monitoring process is fully automated, and it only takes 30 seconds from abnormal detection to warning generation, effectively preventing the production of batch defective products.

[0089] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0090] (1) Separate the quality risk warning data by process characteristics to obtain a key process parameter sequence, and perform time series matching between the key process parameter sequence and historical production data to obtain process parameter mapping data;

[0091] (2) performing parameter correlation analysis on the baking temperature, baking time, and molding pressure in the process parameter mapping data to obtain process parameter correlation data, and performing quality influence weight calculation on the process parameter correlation data to obtain parameter influence weight data;

[0092] (3) Compare and analyze the parameter influence weight data with the historical good product production parameters to obtain parameter deviation data, and calculate the compensation factor of the parameter deviation data to obtain process compensation data;

[0093] (4) performing parameter boundary constraint processing on the process compensation data to obtain safety parameter range data, and performing control sequence generation processing on the safety parameter range data to obtain a process control sequence;

[0094] (5) The process control sequence is processed to match the production working conditions to obtain working condition adaptation data, and the working condition adaptation data is evaluated and processed to obtain parameter control decision data.

[0095] Specifically, the key process parameters are extracted through process feature separation. The key process parameters include baking temperature (range 175-185℃), baking time (standard 25 minutes), molding pressure (standard 1.5MPa), etc. The parameters are sampled in 5-minute time units to establish parameter time series. The parameter acquisition frequency is set as: temperature once every 10 seconds, time accurate to seconds, and pressure once per second. These parameter sequences are time-series matched with historical production data, and the matching window is set to the first 30 batches of data, with a sliding step of 1 batch. Through timestamp alignment, the time sequence consistency of the data is ensured to form process parameter mapping data. The correlation analysis of process parameter mapping data adopts multivariate statistical methods. First, the Pearson correlation coefficient matrix between parameters is calculated to analyze the correlation between baking temperature, time, and pressure. A correlation coefficient greater than 0.8 indicates a strong correlation, 0.5-0.8 indicates a moderate correlation, and less than 0.5 indicates a weak correlation. For example, the correlation coefficient between baking temperature and baking time is 0.75, indicating that there is a significant medium-strength correlation between the two. The generated process parameter association data records the degree of mutual influence between the parameters.

[0096] The weight calculation of quality influence is based on the analytic hierarchy process, and a judgment matrix is ​​constructed based on expert experience and historical data. For crispy mooncakes, the weight of the baking temperature on crispness is 0.5, the weight on color is 0.3, and the weight on taste is 0.2. The weight of the baking time on the degree of maturity is 0.6, and the weight on the moisture content is 0.4. The weight of the molding pressure on the hierarchy is 0.7, and the weight on the appearance is 0.3. The parameter influence weight data is obtained through weight calculation. During the comparative analysis, the parameter influence weight data is compared with the historical good product data. The good product standards include: appearance goldenness 90±5%, hierarchy clarity ≥95%, and crispness score ≥90 points. Calculate the deviation value between the current parameter and the good product parameter to obtain the parameter deviation data. The compensation factor is calculated based on the deviation data, and the compensation factor calculation takes into account the sensitivity and adjustment difficulty of the parameter. The temperature compensation factor is 0.8 (considering thermal inertia), the time compensation factor is 1.0 (direct compensation), and the pressure compensation factor is 0.9 (considering material properties).

[0097] Parameter boundary constraint processing determines the safe operating range of each parameter. The safe range of temperature is 180±5℃, the safe range of time is 25±2 minutes, and the safe range of pressure is 1.5±0.2MPa. A control sequence is generated within these ranges, with the control step size being: temperature 0.5℃, time 30 seconds, and pressure 0.1MPa. The process control sequence includes the direction, amplitude, and order of parameter adjustment. When matching production conditions, the actual status of the current production line is considered, including factors such as equipment performance, raw material characteristics, and environmental conditions. The operating condition adaptation data records the optimal values ​​and adjustment schemes of various parameters under different conditions. The control scheme evaluation adopts a multi-objective optimization method, while considering the balance between product quality improvement and production efficiency.

[0098] For example, a mooncake production line experienced quality fluctuations during the production process. Through parameter analysis, it was found that the baking temperature was 183°C (3°C higher), the time was 23 minutes (2 minutes shorter), and the molding pressure was 1.3MPa (0.2MPa lower). Correlation analysis showed that high temperature shortened the time (correlation coefficient 0.75), and low pressure affected the hierarchy (impact weight 0.7). By comparing the parameters of historical good products, the temperature compensation value was calculated to be -2.4°C (considering the compensation factor of 0.8), the time compensation value was +2 minutes, and the pressure compensation value was +0.18MPa (considering the compensation factor of 0.9). According to the safety parameter range, a control sequence was generated: first reduce the temperature to 180.6°C, extend the baking time to 25 minutes, and finally adjust the pressure to 1.48MPa. After implementing the control plan, the product qualification rate increased from the original 92% to 98%, especially the hierarchy and crispness indicators were significantly improved. The entire parameter optimization process took 5 minutes, achieving rapid response and precise control.

[0099] The above describes the mooncake production safety and quality monitoring method based on artificial intelligence in the embodiment of the present application. The following describes the mooncake production safety and quality detection system based on artificial intelligence in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the mooncake production safety and quality detection system based on artificial intelligence includes:

[0100] The acquisition module is used to perform time-series synchronous acquisition and ID association processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line to obtain multi-source perception data;

[0101] A recognition module, used to perform adaptive enhancement and deep recognition processing on the multi-source perception data to obtain multi-dimensional representation feature data;

[0102] A detection module, used to perform collaborative quality detection processing on the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics in the multi-dimensional characterization feature data to obtain quality detection feature data;

[0103] A grading module, used for performing historical reference dynamic grading processing on the appearance quality index and the internal quality index in the quality inspection feature data to obtain grade evaluation mapping data;

[0104] A monitoring module, used to perform real-time monitoring and processing of key process parameter changes and quality anomalies on the grade evaluation mapping data to obtain quality risk warning data;

[0105] The processing module is used to perform process correlation analysis and optimization processing on the quality risk warning data and historical production data to obtain parameter control decision data.

[0106] Through the collaborative cooperation of the above components, the 360° multi-angle high-definition image group and internal structure infrared scanning data on the mooncake production line are collected in time series and ID-associated, which realizes the all-round and blind-angle collection of the appearance and internal features of the mooncakes, ensuring the integrity and consistency of the data. By adaptively enhancing and deeply identifying multi-source perception data, complex feature information such as surface wrinkle features and hierarchical structure features are effectively extracted, overcoming the limitations of traditional detection methods that rely on simple geometric features. When the multi-dimensional characterization feature data is processed for collaborative quality inspection, the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics are comprehensively analyzed to achieve integrated detection of internal and external quality. By dynamically grading the quality inspection feature data with historical reference, an adaptive evaluation system based on historical data is established, making the quality evaluation more objective and accurate. In the process of processing the grade evaluation mapping data, the changes in key process parameters and quality anomalies are monitored in real time, which realizes the timely discovery and early warning of quality anomalies in the production process. Finally, by performing process correlation analysis and optimization processing on the quality risk warning data and historical production data, a closed-loop quality control mechanism is formed, which effectively improves the stability of the production process and the consistency of product quality. It realizes intelligent management of the entire process from data collection, feature extraction, quality detection to early warning optimization.

[0107] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A mooncake production safety and quality monitoring method based on artificial intelligence, characterized in that: The mooncake production safety and quality monitoring method based on artificial intelligence includes: The 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line are collected in time series and ID-associated to obtain multi-source perception data; Performing adaptive enhancement and deep recognition processing on the multi-source perception data to obtain multi-dimensional representation feature data; Performing collaborative quality inspection on the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics in the multi-dimensional characterization feature data to obtain quality inspection feature data; Performing historical reference dynamic grading processing on the appearance quality index and the internal quality index in the quality inspection feature data to obtain grade evaluation mapping data; Perform real-time monitoring and processing of key process parameter changes and quality anomalies on the grade evaluation mapping data to obtain quality risk warning data; The quality risk warning data and historical production data are subjected to process correlation analysis and optimization processing to obtain parameter control decision data.

2. The method for monitoring mooncake production safety and quality based on artificial intelligence according to claim 1, characterized in that: The 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line are collected in time sequence synchronously and ID associated to obtain multi-source perception data, including: A high-definition camera array is configured above and around the mooncake production line to perform high-speed multi-angle image acquisition and processing to obtain all-round appearance image data; Performing illumination compensation and enhancement processing on the omnidirectional appearance image data to obtain a 360° multi-angle high-definition image group; The internal structure of the moon cake is scanned and imaged by an infrared sensor to obtain infrared scanning data of the internal structure; Performing industrial network timing synchronization processing on the 360° multi-angle high-definition image group and the internal structure infrared scanning data to obtain time-aligned data; Performing unique ID coding association processing on the time-aligned data to obtain sample identification data; The sample identification data and the time alignment data are encapsulated in data packets to obtain the multi-source perception data.

3. The method for monitoring mooncake production safety and quality based on artificial intelligence according to claim 1 is characterized in that: The adaptive enhancement and deep recognition processing of the multi-source perception data to obtain multi-dimensional representation feature data includes: Performing security quality feature separation processing on the multi-source perception data to obtain a surface feature group and an internal feature group, and performing quality feature screening processing on the surface feature group and the internal feature group to obtain basic feature data; Performing stereoscopic reconstruction processing on the wrinkle texture in the basic feature data to obtain a wrinkle stereoscopic feature map, and performing depth layering processing on the wrinkle stereoscopic feature map to obtain multi-level structure data; Performing fracture identification processing on the interlayer connection relationship of the multi-level structure data to obtain a crack feature distribution map, and performing feature parameter extraction processing on the crack feature distribution map to obtain crack morphological parameters; Performing color brightness equalization processing on the crack morphology parameters to obtain color uniformity characteristics, and performing local contrast enhancement processing on the color uniformity characteristics to obtain enhanced color data; Performing safety quality boundary extraction processing on the enhanced color data to obtain feature boundary data, and performing feature fusion processing on the feature boundary data to obtain candidate feature data; Performing grade processing on the surface wrinkle degree, layer distribution state and crack distribution density in the candidate feature data to obtain quality feature grade data, and cross-mapping processing on the quality feature grade data and internal color uniformity to obtain comprehensive feature mapping data; Performing spatial correlation processing on the surface defect position and the internal structure change area in the comprehensive feature mapping data to obtain defect correlation data, and performing weight assignment processing on the defect correlation data to obtain feature weight data; Performing a quality influencing factor sorting process on the feature weight data to obtain an influencing factor sequence, and performing a threshold determination process on the influencing factor sequence to obtain associated determination data; The association determination data is subjected to feature combination processing to obtain the multi-dimensional representation feature data.

4. The method for monitoring mooncake production safety and quality based on artificial intelligence according to claim 1 is characterized in that: The collaborative quality inspection process of the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics in the multi-dimensional characterization feature data to obtain quality inspection feature data includes: Performing a dimensional feature extraction process on the appearance specification parameters in the multi-dimensional characterization feature data to obtain specification feature data, and performing a standard specification matching process on the specification feature data to obtain specification conformity data; Performing surface texture feature extraction processing on the specification compliance data to obtain a texture feature map, and performing integrity assessment processing on the texture feature map to obtain surface texture integrity data; Performing internal structure feature extraction processing on the surface texture integrity data to obtain hierarchical uniformity data, and performing structural distribution evaluation processing on the hierarchical uniformity data to obtain internal structure uniformity data; Performing sandwich area positioning processing on the internal structure uniformity data to obtain sandwich boundary data, and performing distribution feature extraction processing on the sandwich boundary data to obtain sandwich distribution feature data; The sandwich distribution characteristic data and the specification compliance data are collaboratively evaluated to obtain internal and external characteristic correlation data, and the internal and external characteristic correlation data are subjected to quality parameter extraction processing to obtain the quality detection characteristic data.

5. The method for monitoring mooncake production safety and quality based on artificial intelligence according to claim 1 is characterized in that: The process of performing historical reference dynamic grading processing on the appearance quality index and the internal quality index in the quality inspection feature data to obtain the grade evaluation mapping data includes: Performing standard parameter extraction processing on the appearance quality index in the quality inspection feature data to obtain appearance quality benchmark data, and performing historical data comparison processing on the appearance quality benchmark data to obtain appearance reference data; Performing feature parameter extraction processing on the internal quality indicators in the quality detection feature data to obtain internal quality benchmark data, and performing historical data comparison processing on the internal quality benchmark data to obtain internal reference data; Performing dynamic threshold division processing on the appearance reference data and the internal reference data to obtain reference grade interval data, and performing grading standard matching processing on the reference grade interval data to obtain standard grading data; Performing quality grade mapping processing on the standard grading data to obtain grade corresponding data, and performing historical trend analysis processing on the grade corresponding data to obtain trend characteristic data; The trend characteristic data is dynamically graded to obtain graded reference data, and the graded reference data is subjected to quality grade evaluation to obtain the grade evaluation mapping data.

6. The method for monitoring mooncake production safety and quality based on artificial intelligence according to claim 1 is characterized in that: The step of performing real-time monitoring and processing on the changes in key process parameters and quality anomalies on the grade evaluation mapping data to obtain quality risk warning data includes: Performing process parameter extraction processing on the grade evaluation mapping data to obtain key parameter data, and performing change trend analysis processing on the key parameter data to obtain parameter change trend data; Performing abnormal fluctuation detection processing on the baking temperature and time parameters in the parameter change trend data to obtain process fluctuation data, and performing quality impact analysis processing on the process fluctuation data to obtain quality fluctuation data; Performing deviation calculation processing on the quality fluctuation data and the quality reference standard to obtain quality deviation data, and performing abnormality determination processing on the quality deviation data to obtain abnormal characteristic data; Performing quality risk assessment processing on the abnormal feature data to obtain risk level data, and performing time series correlation analysis processing on the risk level data to obtain risk association data; The risk-related data is subjected to early warning threshold determination processing to obtain early warning indicator data, and the early warning indicator data is subjected to real-time monitoring processing to obtain the quality risk early warning data.

7. The method for monitoring mooncake production safety and quality based on artificial intelligence according to claim 1 is characterized in that: The process correlation analysis and optimization processing of the quality risk warning data and historical production data to obtain parameter control decision data includes: Performing process feature separation processing on the quality risk warning data to obtain a key process parameter sequence, and performing time series matching processing on the key process parameter sequence and the historical production data to obtain process parameter mapping data; Performing parameter correlation analysis on the baking temperature, baking time, and molding pressure in the process parameter mapping data to obtain process parameter correlation data, and performing quality influence weight calculation on the process parameter correlation data to obtain parameter influence weight data; Comparative analysis is performed on the parameter influence weight data and historical good product production parameters to obtain parameter deviation data, and compensation factor calculation is performed on the parameter deviation data to obtain process compensation data; Performing parameter boundary constraint processing on the process compensation data to obtain safety parameter range data, and performing control sequence generation processing on the safety parameter range data to obtain a process control sequence; The process control sequence is processed for production condition matching to obtain condition adaptation data, and the condition adaptation data is processed for control scheme evaluation to obtain the parameter control decision data.

8. An artificial intelligence-based mooncake production safety and quality detection system, used to implement the artificial intelligence-based mooncake production safety and quality monitoring method as described in any one of claims 1 to 7, characterized in that: The mooncake production safety and quality detection system based on artificial intelligence includes: The acquisition module is used to perform time-series synchronous acquisition and ID association processing on the 360° multi-angle high-definition image group and internal structure infrared scanning data collected on the mooncake production line to obtain multi-source perception data; A recognition module, used to perform adaptive enhancement and deep recognition processing on the multi-source perception data to obtain multi-dimensional representation feature data; A detection module, used to perform collaborative quality detection processing on the appearance specification parameters, surface texture integrity, internal structure uniformity and sandwich distribution characteristics in the multi-dimensional characterization feature data to obtain quality detection feature data; A grading module, used for performing historical reference dynamic grading processing on the appearance quality index and the internal quality index in the quality inspection feature data to obtain grade evaluation mapping data; A monitoring module is used to perform real-time monitoring and processing of key process parameter changes and quality anomalies on the grade evaluation mapping data to obtain quality risk warning data; The processing module is used to perform process correlation analysis and optimization processing on the quality risk warning data and historical production data to obtain parameter control decision data.

Images