Intelligent baking control method for freshly-baked flavor nuts

Abstract

The invention discloses an intelligent baking control method for on-site baked flavor nuts, which comprises the following steps of: constructing space-time registration data by collecting nut surface reflectivity and two-dimensional temperature field data, extracting non-enzymatic browning reaction acceleration and a surface caramelization index from the space-time registration data, forming a comprehensive feature vector set in combination with the temperature field data, and carrying out surface caramelization on the non-enzymatic browning reaction acceleration and the surface caramelization index. Performing dimension reduction on the comprehensive feature vector set to obtain a key feature vector, constructing a multi-physics field coupling prediction model based on the key feature vector, inputting real-time space-time registration data into the model, obtaining a multi-dimensional data prediction value, and outputting the multi-dimensional data prediction value. And performing abnormality diagnosis according to the color difference value of the baked nut region detected in real time in combination with the predicted value of the multi-dimensional data of the nuts, and generating an adjustment control strategy based on an abnormality diagnosis result, thereby effectively improving the crisp taste and quality stability of the nuts.

Description

Technical Field

[0001] The present invention relates to the technical field of food processing, and in particular to an intelligent control method for baking freshly baked flavored nuts. Background Art

[0002] As demand for healthy foods grows, the processing and roasting of nuts, a nutritious food, are gaining increasing attention. Freshly roasted flavored nuts, with their unique aroma and taste, are highly sought after by consumers. However, quality control during the nut roasting process remains a technical challenge for the industry.

[0003] Currently, nut roasting processes rely primarily on empirical judgment or simple time and temperature control, lacking precise real-time monitoring and intelligent control. Traditional nut roasting methods typically rely on a single temperature parameter, making it difficult to effectively monitor surface color changes and internal component transformations, leading to unstable product quality.

[0004] During the nut roasting process, high-temperature roasting alone can cause the nut's skin to char while retaining moisture, affecting the nut's crisp texture. It also destroys volatile flavor compounds, including pyrazines and aldehydes, reducing the nut's aroma. Existing technologies lack multi-physics monitoring methods that combine hyperspectral imaging with infrared thermal imaging, making it impossible to accurately monitor color changes and temperature distribution during nut roasting. Furthermore, existing technologies lack predictive models based on multi-physics data, making it impossible to effectively predict and control key quality indicators such as the acceleration of non-enzymatic browning reactions and the surface caramelization index.

[0005] Furthermore, existing research on antioxidant treatments during nut roasting is insufficient, leading to the susceptibility of nuts to oxidation during roasting, which impacts the flavor and shelf life of the product. Furthermore, a lack of systematic evaluation methods and reprocessing technologies for roasted nuts hinders effective assurance of product consistency and high quality.

[0006] Therefore, there is an urgent need to develop a method that can monitor the changes in multiple physical fields during nut roasting in real time, accurately predict the roasting status and intelligently control it, so as to solve the problems of flavor loss and unstable quality in the existing nut roasting process. Summary of the Invention

[0007] In order to solve the above-mentioned technical problems, the present invention provides an intelligent control method for baking freshly baked flavored nuts.

[0008] The technical solution of the present invention is achieved as follows: an intelligent control method for baking freshly baked flavored nuts, comprising:

[0009] S1, collecting reflectivity distribution data and two-dimensional temperature field matrix data of the nut surface, unifying the reflectivity distribution data and the two-dimensional temperature field matrix data into the same coordinate system to obtain spatiotemporal registration data, and preprocessing the spatiotemporal registration data to obtain preprocessed spatiotemporal registration data;

[0010] S2, extracting non-enzymatic browning reaction acceleration and surface caramelization index based on preprocessed spatiotemporal registration data;

[0011] S3, combining the extracted non-enzymatic browning reaction acceleration and surface caramelization index with the temperature field data to form a comprehensive feature vector set, performing dimensionality reduction processing on the comprehensive feature vector set, and extracting key feature vectors;

[0012] S4. Constructing a multi-physics field coupling prediction model based on the extracted key feature vectors;

[0013] S5. Inputting the real-time spatiotemporal registration data into the multi-physics field coupling prediction model to obtain the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value, and surface caramelization index prediction value of the nuts;

[0014] S6. Perform abnormality diagnosis based on the real-time detected color difference value of the roasted nut area, combined with the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value, and surface caramelization index prediction value of the nuts, and generate an adjustment control strategy based on the abnormality diagnosis result.

[0015] Furthermore, in step S1, reflectivity distribution data and two-dimensional temperature field matrix data of the nut surface are collected, the reflectivity distribution data and the two-dimensional temperature field matrix data are unified into the same coordinate system to obtain spatiotemporal registration data, and the spatiotemporal registration data are preprocessed to obtain preprocessed spatiotemporal registration data:

[0016] The hyperspectral imaging system is used to collect the reflectance distribution data of the nut surface in the CIELAB color space in real time; the infrared thermal imager is used to collect the two-dimensional temperature field matrix data of the surface;

[0017] Using a spatiotemporal registration algorithm to coordinately align and time-synchronize the reflectivity distribution data and the temperature field matrix data to generate spatiotemporal registration data;

[0018] A bidirectional filtering algorithm is used to perform denoising on the spatiotemporal registration data to generate preprocessed spatiotemporal registration data.

[0019] Furthermore, in step S2, the non-enzymatic browning reaction acceleration and surface caramelization index are extracted based on the pre-processed spatiotemporal registration data:

[0020] Based on the preprocessed spatiotemporal registration data obtained in step S1, the time series corresponding to the preprocessed spatiotemporal registration data is decomposed into high-frequency detail features and low-frequency trend features using the db4 wavelet basis function, wherein the high-frequency detail features correspond to noise and instantaneous fluctuations in the data, and the low-frequency trend features reflect the browning reaction trend, and the high-frequency noise is filtered out by threshold processing;

[0021] Aiming at the decomposed low-frequency trend features, the adaptive moving window algorithm is used to calculate the local derivative features;

[0022] Based on the local derivative features and in combination with the historical data distribution, the probability distribution of the current local derivative features is compared with the feature distribution of the historical qualified data through the KL divergence analysis method. If the distribution difference exceeds a preset threshold, it is determined to be an abnormal data point;

[0023] A bidirectional Kalman filter algorithm is used to perform time series smoothing correction on the abnormal points to generate a continuous and stable correction feature sequence;

[0024] For the corrected characteristic sequence, the ratio of the reflectance spectrum in a specific band to the reference band is extracted, a reflectance ratio model is constructed, and the surface caramelization index is calculated.

[0025] Furthermore, in step S3, the extracted non-enzymatic browning reaction acceleration and surface caramelization index are combined with the temperature field data to form a comprehensive feature vector set, and the comprehensive feature vector set is subjected to dimensionality reduction processing to extract the key feature vector:

[0026] The baking area is divided into several grids, and the average temperature, temperature gradient, non-enzymatic browning reaction acceleration and caramelization index in each grid are calculated to form the original feature vector;

[0027] Calculating regional global statistical features of the original feature vector, wherein the regional global statistical features include the mean value, standard deviation, maximum value, minimum value, and median of the entire region;

[0028] Extracting time series features from the original feature vector, wherein the time series features include the rate of change, fluctuation amplitude, and trend coefficient of each parameter over a period of time;

[0029] Calculating spatial distribution characteristics for the original eigenvectors, the spatial distribution characteristics including spatial autocorrelation coefficients, entropy values, and inhomogeneity indexes of various parameters;

[0030] Forming a comprehensive feature vector set based on the global statistical features, the temporal features, and the spatial distribution features;

[0031] Performing zero-mean normalization on the comprehensive feature vector set, and calculating the covariance matrix to obtain the feature variance contribution rate after dimensionality reduction;

[0032] According to the feature variance contribution rate, the principal component whose cumulative contribution reaches a preset threshold is screened out as the key feature vector.

[0033] Furthermore, in step S4, based on the extracted key feature vectors, a multi-physics field coupling prediction model is constructed as follows:

[0034] Based on the extracted key feature vectors, the historical roasting data was cleaned, outliers were removed, missing values ​​were filled, and multi-objective labels such as chromaticity, browning acceleration, and caramelization index were annotated to construct a training dataset containing multi-physics field information.

[0035] Based on the temporal and spatial feature distribution of the training dataset, a convolutional neural network architecture that integrates residual connections and attention mechanisms is designed, and the network hierarchy and parameter configuration are clarified;

[0036] Based on the designed convolutional neural network architecture, the model is trained using an adaptive moment estimation optimizer. The training set, validation set, and test set are divided. Batch training and iterative optimization are implemented to adjust network parameters.

[0037] For the trained model, the temperature-related characteristic components and the corresponding color prediction values ​​are extracted from the model output. Multiple sampling points are selected within the temperature variation range, and the temperature-color nonlinear relationship is fitted using the cubic spline interpolation algorithm to generate a response surface model for multi-physics field coupling.

[0038] Furthermore, the spatiotemporal registration data in step S5 is input into the multi-physics field coupling prediction model to obtain the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value and surface caramelization index prediction value of the nuts:

[0039] Perform feature standardization and missing value interpolation on the spatiotemporal registration data collected in real time to obtain real-time preprocessed spatiotemporal registration data;

[0040] Based on the feature distribution of the real-time preprocessed spatiotemporal registration data, a convolutional neural network architecture including residual connections and an attention mechanism is called to perform prediction to obtain a chromaticity prediction value for a period of time in the future;

[0041] According to the prediction results of the convolutional neural network, the chromaticity value predicted by the neural network is compared with the temperature-color mapping relationship in the response surface model, and the prediction results of the non-enzymatic browning reaction acceleration and the caramelization index are adjusted;

[0042] For the prediction results adjusted by the response surface model, multi-dimensional prediction data including chromaticity, non-enzymatic browning reaction acceleration and surface caramelization index are generated.

[0043] Furthermore, in step S6, an abnormality diagnosis is performed based on the real-time detected color difference value of the roasted nut area, combined with the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value, and surface caramelization index prediction value of the nuts, and an adjustment control strategy is generated based on the abnormality diagnosis result:

[0044] Based on the real-time detected color difference value of the roasted nuts region, when the real-time detected regional color difference value exceeds a preset threshold, the current chromaticity prediction value, the non-enzymatic browning reaction acceleration prediction value, and the surface caramelization index prediction value are retrieved;

[0045] Comparing and analyzing the chromaticity prediction value, the non-enzymatic browning reaction acceleration prediction value, and the surface caramelization index prediction value at the current moment with the preset standard values ​​of the corresponding indicators, determining the direction and magnitude of the deviation of the prediction value, and locating the specific cause of the deviation in combination with the change trends of the chromaticity prediction value, the non-enzymatic browning reaction acceleration prediction value, and the surface caramelization index prediction value at the current moment, to generate an abnormality diagnosis result;

[0046] If the real-time regional color difference value does not exceed the preset threshold, the current baking state is determined to be within the normal range, and subsequent data will continue to be monitored. At the same time, the current non-enzymatic browning reaction acceleration and surface caramelization index will be recorded and included in the preset historical database;

[0047] Analyze the causes and impact of the deviations in the abnormal diagnosis results, and select the corresponding adaptive control strategies for the non-enzymatic browning reaction acceleration and surface caramelization index from the pre-set rules and strategy library;

[0048] The regulation control strategy is converted into executable instructions for the device, and the device parameter settings are adjusted. The real-time data and predicted value changes after adjustment are continuously monitored, and the control strategy is optimized based on actual feedback.

[0049] Furthermore, the method further includes reducing the degree of oxidation during the baking process of the freshly baked flavored nuts through an antioxidant process, wherein the antioxidant process is as follows:

[0050] Before baking the raw nuts, soaking the raw nuts in a glucose oxidase solution, wherein the enzyme activity of the glucose oxidase solution is 50 U / g-100 U / g, and the soaking time is 30 minutes;

[0051] During the baking process, nitrogen is used as the protective gas in the gradient baking process, and a natural antioxidant is sprayed, wherein the natural antioxidant contains 0.1%-0.3% of rosemary extract and 0.05% of vitamin E.

[0052] Furthermore, the method further includes performing a quality inspection on the baked batch of freshly baked flavored nut samples after completing the baking process of steps S1 to S6, specifically:

[0053] Sampling the freshly roasted flavored nuts to obtain nut samples;

[0054] Determining the flavor index of the nut sample; the flavor index includes crispness, acetaldehyde content and sensory score; the sensory score is obtained by a review team evaluating the appearance, smell, taste and aftertaste of the nut sample;

[0055] If the flavor index is qualified, the batch of freshly roasted flavored nuts will be packaged; if the flavor index is unqualified, the batch of freshly roasted flavored nuts will be re-roasted or subjected to negative pressure formaldehyde removal operation;

[0056] Furthermore, the re-baking or negative pressure formaldehyde removal operation is:

[0057] If the crispness is within 600gf-800gf, the batch of freshly roasted flavored nuts is re-roasted by regularly raising the temperature;

[0058] If the acetaldehyde content is within 0.5 mg / kg-1.2 mg / kg, the aldehyde substances in the batch of freshly roasted flavored nuts are removed by vacuum negative pressure adsorption.

[0059] Beneficial effects:

[0060] The present invention collects surface reflectance and two-dimensional temperature field data of nuts, obtains spatiotemporal registration data after spatiotemporal registration and preprocessing, extracts non-enzymatic browning reaction acceleration and surface caramelization index from the spatiotemporal registration data, combines them with temperature field data to form a comprehensive feature vector set, performs dimensionality reduction on the comprehensive feature vector set to obtain key feature vectors, constructs a multi-physics field coupling prediction model based on the key feature vectors, inputs real-time spatiotemporal registration data into the model to obtain multi-dimensional data prediction values, and performs abnormality diagnosis based on the real-time detected color difference values ​​of the roasted nut regions combined with the multi-dimensional data prediction values ​​of the nuts. Based on the abnormality diagnosis results, a regulation and control strategy is generated. In addition, the present invention reduces oxidation through an antioxidant process, tests roasted nuts, and performs corresponding operations on unqualified nut samples. Compared with the existing technology, the present invention uses a hyperspectral imaging system and an infrared thermal imager to monitor the nut roasting process in real time. Combined with the multi-physics field coupling prediction model, it can accurately predict the roasting state and output the optimal regulation and control strategy. At the same time, the antioxidant process is used to reduce the degree of oxidation during the nut production process, thereby preserving the volatile flavor substances in the nuts and effectively improving the crisp taste and quality stability of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a structural block diagram of an intelligent control method for freshly baked flavored nuts in an embodiment of the present invention;

[0062] Figure 2 This is a flowchart of an intelligent control method for freshly baked flavored nuts in an embodiment of the present invention;

[0063] Figure 3 This is a flowchart of the antioxidant process steps of an intelligent control method for baking freshly baked flavored nuts in an embodiment of the present invention;

[0064] Figure 4 This is a detection process step diagram of an intelligent control method for freshly baked flavored nuts in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0066] The preferred implementation methods of the present invention are described below with reference to the accompanying drawings. Those skilled in the art should understand that these implementation methods are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0067] As used herein, the singular forms "a," "an," and "the" may also include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "include," "comprising," "having," and the like specify the presence of stated features, integers, steps, operations, components, parts, or combinations thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof. Furthermore, the term "and / or" as used in this specification includes any and all combinations of the relevant listed items.

[0068] See also Figure 1-Figure 2 As shown, an intelligent control method for baking freshly baked flavored nuts includes:

[0069] S1, collecting reflectivity distribution data and two-dimensional temperature field matrix data of the nut surface, unifying the reflectivity distribution data and the two-dimensional temperature field matrix data into the same coordinate system to obtain spatiotemporal registration data, and preprocessing the spatiotemporal registration data to obtain preprocessed spatiotemporal registration data;

[0070] S2, extracting non-enzymatic browning reaction acceleration and surface caramelization index based on preprocessed spatiotemporal registration data;

[0071] S3, combining the extracted non-enzymatic browning reaction acceleration and surface caramelization index with the temperature field data to form a comprehensive feature vector set, performing dimensionality reduction processing on the comprehensive feature vector set, and extracting key feature vectors;

[0072] S4. Constructing a multi-physics field coupling prediction model based on the extracted key feature vectors;

[0073] S5. Inputting the real-time spatiotemporal registration data into the multi-physics field coupling prediction model to obtain the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value, and surface caramelization index prediction value of the nuts;

[0074] S6. Perform abnormality diagnosis based on the real-time detected color difference value of the roasted nut area, combined with the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value, and surface caramelization index prediction value of the nuts, and generate an adjustment control strategy based on the abnormality diagnosis result.

[0075] like Figure 2 As shown, in step S1, the reflectivity distribution data and the two-dimensional temperature field matrix data of the nut surface are collected, the reflectivity distribution data and the two-dimensional temperature field matrix data are unified into the same coordinate system to obtain spatiotemporal registration data, and the spatiotemporal registration data are preprocessed to obtain preprocessed spatiotemporal registration data.

[0076] Specifically, in this embodiment, the hyperspectral imaging system uses a hyperspectral camera with a wavelength range of 400-1000nm, a spectral resolution of 2.8nm, a spatial resolution of 1024×1024 pixels, and a sampling frequency of 10Hz. The system is installed above the baking equipment and maintains a fixed distance of 50cm from the surface of the nuts to ensure that the entire baking area is completely covered within the imaging range. Before collecting data, the imaging system is first subjected to white board correction and dark current compensation to ensure data accuracy. White board correction is to calibrate the spectral response of the imaging system by using a standard white calibration plate to eliminate systematic errors; dark current compensation is to perform dark current compensation on the imaging system. Compensation is to compensate for the noise signal of the imaging system under no-light conditions and improve the signal-to-noise ratio of the data. After the original spectral data collected by the system is converted into the reflectance distribution data of the three channels of L*, a*, and b* in the CIELAB color space through the spectral reflectance model. Among them, L* represents brightness, with a value range of 0-100; a* represents the range from green to red, with a value range of -128 to +127; b* represents the range from blue to yellow, with a value range of -128 to +127. The CIELAB color space is a color space model based on human visual perception, which can accurately describe the color characteristics of the object surface.

[0077] The infrared thermal imager uses an uncooled microbolometer with a temperature measurement range of -20°C to +650°C, a thermal sensitivity better than 0.05°C, a pixel resolution of 640×480, and a sampling frequency of 10Hz, synchronized with the hyperspectral imaging system. The thermal imager is installed on the side of the roasting equipment, maintaining a fixed distance of 60cm from the nut surface and a collection angle of 45° to ensure complete coverage of the temperature field in the roasting area. The raw thermal radiation data collected by the thermal imager is converted into two-dimensional temperature field matrix data after emissivity correction.

[0078] The spatiotemporal registration algorithm uses an affine transformation method based on feature point matching. SIFT feature points are first extracted from the hyperspectral image and thermal image. The RANSAC algorithm is then used to filter matching point pairs and establish an affine transformation matrix between the two images. This transformation matrix maps the temperature field data in the thermal image to the coordinate system of the hyperspectral image, achieving spatial alignment of the two data. In the temporal dimension, a hardware trigger signal ensures synchronization of the sampling moments of the two systems, resulting in spatiotemporal registration data.

[0079] The spatiotemporal registration data was denoised using a bidirectional filtering algorithm. Bidirectional filtering, a Bayesian-based filtering algorithm, removes noise while preserving image edge information. For hyperspectral data, bidirectional filtering was performed on each band separately; for temperature field data, bidirectional filtering was performed directly on a two-dimensional temperature matrix. The filtered data was then subjected to a consistency check to ensure that sampling errors between sensors were effectively eliminated and that the spatial distribution and temporal trends of the data remained consistent. After processing with the bidirectional filtering algorithm, preprocessed spatiotemporal registration data was generated.

[0080] like Figure 2 As shown, in step S2, the non-enzymatic browning reaction acceleration and the surface caramelization index are extracted based on the preprocessed spatiotemporal registration data.

[0081] Specifically, in this embodiment, the db4 wavelet basis function is used to perform a 4-layer multi-resolution decomposition on the time series corresponding to the preprocessed spatiotemporal registration data, such as the temperature field and the reflectivity change series, to decompose the original signal into low-frequency trend features and high-frequency detail features. The high-frequency detail features correspond to the short-term noise and instantaneous fluctuations in the data, and the low-frequency trend features retain the long-term change trend of the signal, directly reflecting the cumulative effect of the non-enzymatic browning reaction and the evolution law of the physical field. The high-frequency noise is filtered out by the soft threshold noise reduction algorithm, and the low-frequency trend component reflecting the real physical process is retained.

[0082] For the decomposed low-frequency trend features, an adaptive moving window algorithm is used to calculate the local derivative features. The window width is automatically adjusted according to the local change rate of the signal. The window is narrowed during the intense reaction stage to improve the time resolution. A cubic polynomial is fitted to the time series within each window to simultaneously calculate the first-order derivative (corresponding to the non-enzymatic browning reaction rate) and the second-order derivative (corresponding to the non-enzymatic browning reaction acceleration). The second-order derivative value at the center of the window is taken as the non-enzymatic browning reaction acceleration.

[0083] The KL divergence analysis method is introduced to compare the probability distribution of the current local derivative feature with the feature distribution of historical qualified data. If the distribution difference exceeds a preset threshold, it is determined to be an abnormal data point. The preset threshold is the KL divergence difference critical value determined based on the feature distribution of historical qualified data. It is used to judge the degree of deviation between the probability distribution of the current local derivative feature and the distribution of historical qualified data. When the calculated KL divergence value of the two exceeds the threshold, the corresponding data point is determined to be an abnormal data point. The bidirectional Kalman filter algorithm is used to perform time series smoothing correction on the abnormal point to generate a continuous and stable corrected feature sequence.

[0084] Based on the corrected characteristic sequence, the specific band (420nm) and the reference band (550nm) of the nut surface reflectance spectrum were extracted. Caramelization products showed a single peak absorption at 420nm, which is highly sensitive to the accumulation of caramelization products. The 550nm band is less affected by pigments and has no direct correlation with the caramelization reaction, making it suitable as a reference band for the visible light band. The ratio of the two was calculated and a reflectance ratio model was constructed. The larger the ratio, the more intense the caramelization reaction.

[0085] Gaussian filtering is performed on the reflectance ratio data to smooth the spatially distributed noise, and a threshold segmentation algorithm is used to identify active caramelization reaction areas, such as the darker areas on the surface of the nuts.

[0086] The average reflectance ratio within the active area is calculated and weighted based on the distance from the pixel to the edge of the nut to generate a surface caramelization index to quantify the degree and spatial uniformity of the caramelization reaction.

[0087] like Figure 2 As shown, in step S3, the extracted non-enzymatic browning reaction acceleration and surface caramelization index are combined with the temperature field data to form a comprehensive feature vector set, and the comprehensive feature vector set is subjected to dimensionality reduction processing to extract key feature vectors.

[0088] Specifically, in this embodiment, the entire baking area is first divided into 10×10 grids, and the average temperature, temperature gradient, non-enzymatic browning reaction acceleration and surface caramelization index are calculated in each grid to form a 400-dimensional original feature vector (10×10×4). The global statistical features are calculated, including the average value, standard deviation, maximum value, minimum value and median of the entire area, for a total of 20 dimensions (4×5). Then, the time series features are extracted, including the rate of change, fluctuation amplitude and trend coefficient of each parameter in the past 30 seconds, for a total of 12 dimensions (4×3). Finally, the spatial distribution features are calculated, including the spatial autocorrelation coefficient, entropy value and heterogeneity index of each parameter, for a total of 12 dimensions (4×3). These features are combined to form a 444-dimensional comprehensive feature vector set.

[0089] Principal component analysis (PCA) is a commonly used dimensionality reduction algorithm. By projecting the original data onto the principal component direction, the key eigenvectors in the data can be extracted, the dimension of the data can be reduced, and the main information of the data can be retained. The principal component analysis dimensionality reduction process first standardizes the eigenvectors so that each feature has zero mean and unit variance, then calculates the characteristic covariance matrix, and solves its eigenvalues ​​and eigenvectors. The eigenvalues ​​are sorted and the top K principal components with a cumulative contribution rate of 95% are selected as the key eigenvector set. The value of K is usually between 15 and 25, depending on the complexity of the data. The dimension of the final key eigenvector set is greatly reduced, but the main information of the original data is retained, which can effectively characterize the key characteristics of the nut roasting process.

[0090] like Figure 2 As shown, in step S4, a multi-physics field coupling prediction model is constructed based on the extracted key feature vectors.

[0091] Specifically, in this embodiment, a convolutional neural network is used to process historical baking data to establish a multi-physics field coupling prediction model. A convolutional neural network (CNN) is a deep learning algorithm with powerful feature extraction and classification capabilities. By inputting historical baking data into a convolutional neural network, a multi-physics field coupling prediction model capable of predicting future baking states can be trained. During the model construction process, 100,000 sets of historical nut baking data are preprocessed according to the key feature vectors extracted in step S3. The historical baking includes multi-dimensional data of temperature field, reflectivity, and reaction rate of historically completed nut baking. The Z-score algorithm is used to identify and eliminate outliers such as data points that deviate from the mean by ±3σ. The K-nearest neighbor interpolation method (KNN) is used to fill in the missing data based on spatiotemporal correlation to ensure data continuity, and multiple target labels are annotated for each set of data. The data include chromaticity values ​​(L*, a*, b* values ​​in CIELAB space), measured values ​​of non-enzymatic browning reaction acceleration, and measured values ​​of surface caramelization index, forming a training data set containing multi-physics field information.

[0092] A network architecture consisting of five layers of residual convolutional blocks was constructed. Each residual block consists of two 3×3 convolutional layers, a batch normalization layer, and a ReLU activation function. Cross-layer skip connections are used to mitigate gradient vanishing, enabling the network to capture the long-range dependency between the temperature field and the browning reaction. A spatial attention module (SAM) and a channel attention module (CAM) were embedded after the third and fourth residual blocks. The SAM generates a spatial weight matrix through two-dimensional convolution to focus on high-temperature or high-browning areas. The CAM generates channel weights through global average pooling and fully connected layers to enhance key feature channels such as temperature and acceleration.

[0093] The model training uses the Adaptive Moment Estimation (Adam) optimizer. The Adaptive Moment Estimation (Adam) optimizer is a commonly used deep learning optimization algorithm that can automatically adjust the learning rate according to the gradient information during training to improve the convergence speed of the model. The initial learning rate is set to 0.001, and it decays by 10% every 50 epochs. The batch size is 64, and the number of training rounds is 300. The training dataset is divided into a training set (for model learning), a validation set (for monitoring overfitting), and a test set (for evaluating generalization ability). The ratio is usually 7:2:1. The data is input into the network in batches, and the predicted value is calculated through forward propagation. The difference between the predicted value and the true label is measured using the mean squared error function. The network parameters are then updated through the backpropagation algorithm. The loss value of the validation set is monitored in real time during training. When the loss of the validation set no longer decreases, the early stopping mechanism is triggered to avoid model overfitting. At the same time, the learning rate is dynamically adjusted. For example, the initial learning rate is set to 0.001 and gradually decayed in the later stage to improve training efficiency and model convergence speed.

[0094] The model outputs a 30-second prediction of the Maillard reaction completion rate, along with a confidence interval and uncertainty indicator. The Maillard reaction is a crucial chemical reaction in baking, and predicting its completion allows for real-time monitoring of the baking process. Confidence intervals and uncertainty indicators measure the reliability of the prediction results, providing a basis for subsequent control decisions.

[0095] Multiple sampling points were selected within a temperature range (e.g., 80°C to 160°C). A piecewise polynomial function was constructed based on the cubic spline interpolation principle, ensuring that adjacent segment functions met continuity and smoothness conditions at the connection points. This generated a response surface that intuitively reflects the effect of temperature changes on nut color.

[0096] By combining the nonlinear prediction ability of neural networks with the interpretability of response surface models, a multi-physics field coupling prediction model is formed. This model can not only accurately predict complex coupling states, but also analyze the quantitative relationship between temperature, browning, and caramelization through response surface analysis, providing a theoretical basis for baking control.

[0097] like Figure 2 As shown, in step S5, the spatiotemporal registration data collected in real time is input into the multi-physics field coupling prediction model to obtain the color prediction value, non-enzymatic browning reaction acceleration prediction value and surface caramelization index prediction value of the nuts.

[0098] Specifically, in this embodiment, the real-time data processing process includes: using a hyperspectral imaging system and an infrared thermal imager to collect the reflectance distribution data and two-dimensional temperature field matrix data of the nut surface in real time at a frequency of 10 Hz, and after preprocessing and spatiotemporal registration, obtaining real-time spatiotemporal registration data. Then, according to the processing flow in step S3, the key feature vector at the current moment is extracted;

[0099] The key feature vector sequence at the current moment is input into the trained multi-physics field coupling prediction model, and the model outputs the color prediction value (L*, a*, b*) within the next 30 seconds. For example, when the current temperature is 110°C, the model predicts that L*=48.5 (light brown) after 10 seconds and L*=45.2 (dark brown) after 20 seconds, reflecting the progress of the Maillard reaction;

[0100] The predicted chromaticity value is compared with a response surface model. The response surface model stores the nonlinear relationship between temperature and color. For example, the color changes slowly at low temperatures and browns rapidly at high temperatures. If the color L value is predicted to be 45 at a certain temperature of 130°C, and the surface model shows that the theoretical value of L at this temperature should be 43-44, then it is determined that the prediction result deviates from the physical law.

[0101] Based on the deviation direction, such as if the predicted L* value is too high or too low, combined with the temperature-color mapping relationship, the non-enzymatic browning reaction acceleration and the caramelization index prediction value are adjusted synchronously. For example, if the measured temperature is normal but the color prediction is too light, it may indicate insufficient browning reaction. The browning acceleration prediction value needs to be adjusted downward, and the caramelization index should be checked to see if the low value is caused by local temperature unevenness.

[0102] After the predicted values ​​are adjusted, adjusted multi-dimensional prediction data including chromaticity, non-enzymatic browning reaction acceleration and surface caramelization index are generated, which improves the reliability of the prediction results.

[0103] like Figure 2 As shown, in step S6, an abnormality diagnosis is performed based on the real-time detected color difference value of the roasted nut area, combined with the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value and surface caramelization index prediction value of the nuts, and an adjustment control strategy is generated based on the abnormality diagnosis result.

[0104] Specifically, in this embodiment, a hyperspectral imaging system installed on the top of the baking equipment is used to collect CIELAB chromaticity data of the nut surface in real time, and calculate the color difference value of the local area of ​​the real-time data. When it is detected that the color difference value of the local area of ​​the real-time data exceeds a preset threshold, the system performs abnormal diagnosis to determine the cause of the deviation. The preset threshold is set to 1.5. The abnormal diagnosis analyzes the deviation between the predicted value and the preset standard value (such as the Hunter chromaticity coordinate gold standard), combines the change trend of the non-enzymatic browning reaction acceleration and the surface caramelization index, and determines the specific cause of the deviation, such as too high temperature, too long baking time or insufficient cold air injection;

[0105] If the real-time local area color difference value is detected to be no more than 1.5, the system determines that the current baking state is good, continues to monitor subsequent data, and records the current non-enzymatic browning reaction acceleration and surface caramelization index. The browning acceleration and caramelization index characteristic values ​​at the current moment are stored in a historical database for regular updating of the training data set of the physical field coupling prediction model to achieve online adaptive optimization of the model;

[0106] Based on the abnormal diagnosis results, the system matches the control scheme from the preset rule and strategy library, which stores a large number of predefined control rules and strategies. For example, when the acceleration of the non-enzymatic browning reaction is detected to exceed the standard value by 20% and is globally distributed, it is judged as "over-browning risk" and the "step-by-step cooling + pulse cooling air" combination strategy is triggered. First, the power of the heating module is reduced by 15%-20%, and the cold air system is simultaneously started to perform pulse cooling on the baking area to avoid the imbalance of moisture inside the nuts caused by a sudden drop in temperature. If the surface caramelization index is lower than the standard value by 10% and concentrated in the edge area, it is judged as "inadequate local reaction" and the "regional heating + extended baking time" strategy is executed. The power of the edge heating unit is increased by 10%, and the current baking stage is extended by 1-2 minutes to promote the sugar pyrolysis reaction and ensure the uniform generation of caramelized products.

[0107] Convert the selected regulation strategy into executable instructions for the equipment, and drive the actuator to adjust the parameters through the PLC programmable logic controller, for example:

[0108] Adjust the heating module power by zones, such as reducing the power of the left zone by 15%;

[0109] Adjust the conveying speed of the mesh belt to extend the time the nuts stay in the high temperature area;

[0110] Start the nitrogen protection system to inhibit oxidation reaction and cooperate with temperature regulation;

[0111] After the adjustment is executed, the system continuously monitors the changes in real-time data and predicted values. If the color difference value does not converge significantly within 30 seconds, the strategy optimization mechanism is automatically triggered, such as increasing the power adjustment range or switching to the backup strategy to ensure the stability of baking quality.

[0112] like Figure 3 As shown, the degree of oxidation during the roasting process of freshly roasted flavored nuts is reduced through antioxidant technology.

[0113] Specifically, in this embodiment, before the raw nuts are baked, a glucose oxidase solution is added to soak the raw nuts. The glucose oxidase solution has an enzyme activity of 50U / g-100U / g and the soaking time is 30 minutes. The glucose oxidase can catalyze the oxidation reaction of glucose to produce gluconic acid and hydrogen peroxide. Before the baking process, glucose oxidase is added to the soaking liquid of the nuts. The glucose oxidase can effectively consume oxygen in the nuts. Trehalose can also be added to the soaking liquid. Combined with the glucose oxidase, it can maintain the activity of the enzyme during the baking process, consume excess oxygen in the baking environment, and reduce the oxidation reaction of the nut oil during baking.

[0114] During the baking process, nitrogen is used as the protective gas in the gradient baking process, and a natural antioxidant is sprayed; the natural antioxidant contains 0.1%-0.3% rosemary extract and 0.05% vitamin E. The rosemary extract mainly contains phenolic compounds, which can provide phenolic hydroxyl groups to scavenge free radicals and interrupt the lipid peroxidation chain reaction. When used in combination with vitamin E, it can regenerate oxidized vitamin E and improve the overall antioxidant effect. In addition, both rosemary extract and vitamin E have high thermal stability and can continue to be effective during the baking process of nuts.

[0115] During the packaging process, the packaging of freshly roasted flavored nuts is filled with nitrogen, a common deoxidizer is added, and the packaging is sealed. This can reduce oxidation reactions that occur during the transportation and storage of nut products, greatly improving the retention time of the freshly roasted flavor of the nut products.

[0116] In this embodiment, the antioxidant process achieves antioxidant effect throughout the entire process of nut handling, processing, transportation and storage through the triple mechanisms of enzymatic synergy, nitrogen isolation and chemical blocking, thereby avoiding the oxidation of unsaturated fatty acids in nuts to produce odorous substances, ensuring the freshly roasted flavor of the nuts and prolonging the retention time of the freshly roasted flavor.

[0117] like Figure 4 As shown, after completing the baking process of step S1 to step S6, quality inspection is performed on the baked batch of freshly baked flavored nut samples.

[0118] Specifically, in this embodiment, the crispness of the nut samples was measured using a texture analyzer, the acetaldehyde content in the nut samples was measured using a gas chromatography-mass spectrometer (GC-MS), and the sensory scores of the nut samples were manually evaluated. The sensory scores were obtained by a review team based on the appearance, smell, taste, and aftertaste of the nut samples, and specifically included:

[0119] The judging team consists of 10 people, and the judging environment is the same. Each round of evaluation is for at least 5 nuts, and the judges rinse their mouths with water between each round. The sensory score is determined by observing the appearance of the nuts, smelling the nut odor, tasting the nut texture, and feeling the nut aftertaste. The sensory score is the average of multiple scores, and only one decimal place is retained.

[0120] Among them, when observing the appearance of nuts, if the surface of the nuts is evenly glossy, the appearance score is 0.8-1 points; if the surface is partially dark, the appearance score is 0.5-0.7 points; if there are burnt spots or obvious discoloration on the surface of the nuts, the appearance score is 0-0.4 points;

[0121] In the nose sniffing of nutty odor, if a strong nutty aroma can be smelled, the odor score is 1.3-1.5 points; if there is a slight oxidation smell in the odor, the odor score is 0.8-1.2 points; if a clear rancid smell can be smelled, the odor score is 0-0.7 points;

[0122] When tasting the taste of nuts, if the nut tastes crispy and has no residue, the taste score is 1.8-2.0 points; if the taste is hard or soft in the middle, the taste score is 1.0-1.7 points; if the taste is sticky or powdery, the taste score is 0-0.9 points;

[0123] In the perception of nutty aftertaste, if there is no bitterness in the aftertaste, the aftertaste score is 0.5 points; if there is a slight bitterness in the aftertaste, the aftertaste score is 0.3 points; if there is a clear sour and bitter taste, the aftertaste score is 0 points;

[0124] Perform quality inspection on the baked batch of freshly baked flavored nut samples. The standards for qualified nut samples include: the crispness of the freshly baked flavor is not less than 800gf, the acetaldehyde content is not more than 0.5mg / kg, and the sensory score is not less than 4.0 points.

[0125] If only the crispness or acetaldehyde content or both are unqualified, reprocessing will be carried out; among them, if the crispness is within 600gf-800gf, the baking temperature will be increased by 5°C based on the temperature during the baking process, and the baking time will be extended by 8 minutes, and the freshly baked flavored nuts in this batch will be re-baked; if the acetaldehyde content is within 0.5mg / kg-1.2mg / kg, the aldehyde substances in the freshly baked flavored nuts in this batch will be removed by vacuum negative pressure adsorption; if the crispness is lower than 600gf or the acetaldehyde content is higher than 1.2mg / kg, the freshly baked flavored nuts in this batch will be discarded and destroyed;

[0126] If the crispness is lower than 600gf, it is necessary to inspect and repair the relevant equipment in the baking process, including the temperature control equipment and the humidity control equipment, to prevent equipment failure from causing crispness problems in multiple batches of nuts. If the acetaldehyde content is higher than 1.2mg / kg, it is necessary to inspect the relevant steps of the antioxidant process, including the enzyme activity test of the enzyme solution and the sealing test of the baking equipment, to prevent excessive acetaldehyde content due to failure of the antioxidant process.

[0127] If only the sensory score is lower than 4.0, the batch of freshly roasted flavored nuts will be crushed and produced as a by-product and used as a baking ingredient; if the sensory score and other items are unqualified at the same time, the batch of freshly roasted flavored nuts will be discarded and destroyed;

[0128] In this embodiment, the above method determines whether the freshly roasted flavor of the finished nuts meets the qualified standards through flavor indicators, and reprocesses unqualified finished nuts, thereby improving the production quality and product qualification rate of freshly roasted flavored nuts.

[0129] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0130] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An intelligent control method for baking freshly baked flavored nuts, characterized in that: The following steps are involved: S1, collecting reflectivity distribution data and two-dimensional temperature field matrix data of the nut surface, unifying the reflectivity distribution data and the two-dimensional temperature field matrix data into the same coordinate system to obtain spatiotemporal registration data, and preprocessing the spatiotemporal registration data to obtain preprocessed spatiotemporal registration data; S2, extracting non-enzymatic browning reaction acceleration and surface caramelization index based on preprocessed spatiotemporal registration data; S3, combining the extracted non-enzymatic browning reaction acceleration and surface caramelization index with the temperature field data to form a comprehensive feature vector set, performing dimensionality reduction processing on the comprehensive feature vector set, and extracting key feature vectors; S4. Constructing a multi-physics field coupling prediction model based on the extracted key feature vectors; S5. Inputting the real-time spatiotemporal registration data into the multi-physics field coupling prediction model to obtain the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value, and surface caramelization index prediction value of the nuts; S6. Perform abnormality diagnosis based on the real-time detected color difference value of the roasted nut area, combined with the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value, and surface caramelization index prediction value of the nuts, and generate an adjustment control strategy based on the abnormality diagnosis result.

2. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The reflectivity distribution data and the two-dimensional temperature field matrix data of the nut surface are collected, the reflectivity distribution data and the two-dimensional temperature field matrix data are unified into the same coordinate system to obtain spatiotemporal registration data, and the spatiotemporal registration data are preprocessed to obtain the preprocessed spatiotemporal registration data: Real-time collection of nut surface reflectance distribution data in CIELAB color space and surface two-dimensional temperature field matrix data; Using a spatiotemporal registration algorithm to coordinately align and time-synchronize the reflectivity distribution data and the temperature field matrix data to generate spatiotemporal registration data; A bidirectional filtering algorithm is used to perform denoising on the spatiotemporal registration data to generate preprocessed spatiotemporal registration data.

3. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The non-enzymatic browning reaction acceleration and surface caramelization index extracted based on the pre-processed spatiotemporal registration data are: Based on the pre-processed spatiotemporal registration data, a time series corresponding to the pre-processed spatiotemporal registration data is decomposed into high-frequency and low-frequency features by using wavelet transform; Aiming at low-frequency features, a moving window algorithm is used to calculate local derivative features and extract the acceleration of non-enzymatic browning reaction; Based on the local derivative features and combined with historical data distribution, abnormal data points are detected by KL divergence analysis method to generate a corrected feature sequence; The caramelization index is calculated based on the modified characteristic sequence and the change in the surface reflectance of the nuts.

4. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The extracted non-enzymatic browning reaction acceleration and surface caramelization index are combined with the temperature field data to form a comprehensive feature vector set, and the comprehensive feature vector set is subjected to dimensionality reduction processing to extract the key feature vector: The baking area is divided into several grids, and the average temperature, temperature gradient, non-enzymatic browning reaction acceleration and caramelization index in each grid are calculated to form the original feature vector; Calculating regional global statistical features of the original feature vectors, extracting temporal features from the original feature vectors, and calculating spatial distribution features, and combining them to form a comprehensive feature vector set; For the comprehensive feature vector set, principal component analysis is used to perform dimensionality reduction processing to obtain the feature variance contribution rate after dimensionality reduction; According to the feature variance contribution rate after the dimensionality reduction, the principal component whose cumulative contribution reaches a preset threshold is screened out as the key feature vector.

5. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The multi-physics field coupling prediction model is constructed based on the extracted key feature vectors: Based on the extracted key feature vectors, cleaning and annotating the historical baking data to construct a training data set containing multi-physics field information; Based on the characteristics of the training dataset, a convolutional neural network architecture including residual connections and attention mechanism is designed; According to the convolutional neural network architecture, an adaptive moment estimation optimizer is used for model training; For the trained model, the temperature-color nonlinear relationship is fitted by a cubic spline interpolation algorithm to generate a response surface model for multi-physics field coupling.

6. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The real-time spatiotemporal registration data is input into the multi-physics field coupling prediction model to obtain the chromaticity prediction value, non-enzymatic browning reaction acceleration prediction value and surface caramelization index prediction value of nuts: Perform feature standardization and missing value interpolation on the spatiotemporal registration data collected in real time to obtain real-time preprocessed spatiotemporal registration data; Based on the feature distribution of the real-time preprocessed spatiotemporal registration data, a convolutional neural network architecture including residual connections and an attention mechanism is called to perform prediction to obtain a chromaticity prediction value for a period of time in the future; According to the prediction results of the convolutional neural network, the temperature-color nonlinear response surface model fitted by cubic spline interpolation is synchronously combined to adjust the prediction values ​​of the non-enzymatic browning reaction acceleration and the surface caramelization index; Based on the prediction results after calibration of the response surface model, multi-dimensional prediction data including chromaticity, non-enzymatic browning reaction acceleration and surface caramelization index are generated.

7. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The method comprises performing abnormal diagnosis based on the color difference value of the roasted nut area detected in real time, combining the color prediction value of the nut, the non-enzymatic browning reaction acceleration prediction value, and the surface caramelization index prediction value, and generating an adjustment control strategy based on the abnormal diagnosis result: Comparing the color difference value of the roasted nut area detected in real time with a preset threshold value, if the color difference value of the roasted nut area detected in real time exceeds the preset threshold value, combining the deviation analysis of the comparison between the predicted value and the preset standard value, determining the cause of the deviation in the roasting process, and generating an abnormality diagnosis result; According to the abnormal diagnosis results, refer to the pre-set rules and strategy library to formulate targeted adjustment and control strategies; For the regulation control strategy, parameter adjustments are performed via a closed-loop control system.

8. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The method further includes reducing the degree of oxidation during the baking process of the freshly baked flavored nuts through an antioxidant process, wherein the antioxidant process is as follows: Before baking the raw nuts, soaking the raw nuts in a glucose oxidase solution; During the baking process, nitrogen is used as the protective gas in the gradient baking process, and a natural antioxidant is sprayed.

9. The intelligent control method for baking freshly baked flavored nuts according to claim 1, characterized in that: The method further includes testing the roasted nuts, specifically: Sampling the freshly roasted flavored nuts to obtain nut samples; determining the flavor index of the nut sample; If the flavor index is qualified, the batch of freshly roasted flavored nuts will be packaged; if the flavor index is unqualified, the batch of freshly roasted flavored nuts will be re-roasted or subjected to negative pressure formaldehyde removal operation.

10. The intelligent control method for baking freshly baked flavored nuts according to claim 9, characterized in that: The re-baking or negative pressure formaldehyde removal operation is as follows: If the crispness is within 600gf-800gf, the batch of freshly roasted flavored nuts is re-roasted by regularly raising the temperature; If the acetaldehyde content is within 0.5 mg / kg-1.2 mg / kg, the aldehyde substances in the batch of freshly roasted flavored nuts are removed by vacuum negative pressure adsorption.

Images