System and method for configuring personalized nutritious food

Abstract

The invention relates to the technical field of nutritious food configuration, and discloses a system and a method for configuring personalized nutritious food, and the method for configuring personalized nutritious food comprises the following steps: collecting spectral characteristic data and chemical component data of a food material sample by adopting a hyperspectral imaging and mass spectrum combined technology, and obtaining holographic chemical fingerprint information; processing the collected spectral data by using a multi-scale feature extraction algorithm, extracting multi-level feature information of the food materials, and carrying out standardization and dimension reduction processing on the feature data; according to the method, high-precision prediction of the nutrient content of the food material is realized through combination of hyperspectral imaging and mass spectrometry technologies and a deep learning model, so that the nutrient content prediction accuracy and the configuration scheme accuracy are improved, the nutrient balance achievement rate is also improved, and meanwhile, the limitation of a static database is broken through; real-time nutritional ingredient analysis of any food material is achieved, the method can adapt to individual differences and freshness changes of the food materials, and accurate configuration based on actual food material characteristics is provided.

Description

Technical Field

[0001] This invention relates to the field of nutritional meal preparation technology, and more specifically, to a system and method for preparing personalized nutritional meals. Background Technology

[0002] As people increasingly value healthy eating, the demand for personalized nutritional meal planning services is growing. Existing nutritional meal planning systems primarily rely on static ingredient nutrition databases, which presents several technical problems: first, they cannot account for the impact of factors such as ingredient origin, season, storage, and cooking methods on nutrient content; second, they cannot handle individual differences in ingredients and variations in freshness; and third, they only focus on the content of major nutrients, lacking the ability to comprehensively analyze the full spectrum of bioactive substances in ingredients. These problems prevent existing systems from providing truly accurate personalized nutritional meal planning solutions. Summary of the Invention

[0003] This invention provides a system and method for configuring personalized nutritional meals, addressing the technical problem of personalized nutritional meal configuration service demand in related technologies.

[0004] This invention provides a method for configuring personalized nutritional meals, comprising the following steps:

[0005] Hyperspectral imaging and mass spectrometry were used to collect spectral feature data and chemical composition data of food samples to obtain holographic chemical fingerprint information;

[0006] The collected spectral data were processed using a multi-scale feature extraction algorithm to extract multi-level feature information of the ingredients, and the feature data were standardized and dimensionality reduced.

[0007] Based on the processed feature data, a mapping relationship between spectral features and the content of nutritional components and bioactive substances in food is established through a deep spectral analysis neural network.

[0008] For novel food ingredients, we combine transfer learning and few-shot learning methods to transfer and apply the established spectral nutrition mapping knowledge to predict the nutritional components of unknown food ingredients.

[0009] Based on the predicted nutritional characteristics of ingredients, combined with the user's health status and nutritional needs, an optimization algorithm is used to calculate the optimal ingredient combination scheme and generate a personalized nutritional meal plan.

[0010] As a further optimization of the present invention, the method of acquiring spectral characteristic data and chemical composition data of food samples using hyperspectral imaging and mass spectrometry specifically includes:

[0011] Food samples were scanned using a hyperspectral imaging device to obtain continuous spectral data in the wavelength range of 400-2500nm;

[0012] Liquid chromatography-mass spectrometry was used to analyze food samples to obtain information on the molecular weight, structure and content of bioactive substances in the food.

[0013] Hyperspectral data and mass spectrometry data are associated according to sample ID to construct a multimodal feature dataset of the samples;

[0014] Radiometric and geometric corrections were performed on the hyperspectral data, and peak alignment and normalization were performed on the mass spectrometry data.

[0015] As a further optimization of the present invention, the step of processing the acquired spectral data using a multi-scale feature extraction algorithm specifically includes:

[0016] Preprocess the hyperspectral data;

[0017] Wavelet transform combined with multi-scale analysis was used to decompose the preprocessed spectral data and extract spectral features at different scales.

[0018] Based on the results of multi-scale analysis, a three-layer feature extraction network was constructed to extract macroscopic, mesoscopic, and microscopic features of food ingredients.

[0019] An adaptive feature selection algorithm was applied to screen out a subset of spectral features that are highly correlated with the content of nutrients and bioactive substances.

[0020] As a further optimization of the present invention, the deep spectral analysis neural network comprises a hybrid neural network structure of convolutional layers, long short-term memory layers, and fully connected layers, specifically including:

[0021] The input layer receives the preprocessed spectral data;

[0022] Convolutional layers use multi-scale convolutional kernels to extract local spectral features and capture the correlation between different bands;

[0023] Long short-term memory layers process the sequential characteristics of spectral data, capturing long-term dependencies in wavelength changes;

[0024] The fully connected layer integrates the features from the previous layer and maps them to the final predicted values ​​of nutrient content.

[0025] As a further optimization of the present invention, the establishment of the mapping relationship between spectral characteristics and the content of nutritional components and bioactive substances in food specifically includes:

[0026] Introduce variables as auxiliary features to construct a multivariate influence factor model;

[0027] Train a deep spectral analysis neural network to establish a quantitative mapping relationship between spectral features and the content of nutrients and bioactive substances.

[0028] The model performance was evaluated using 10x cross-validation. By adjusting the network structure, optimizing hyperparameters, and applying regularization techniques, the model's generalization ability and prediction accuracy were improved.

[0029] As a further optimization of the present invention, the method of combining transfer learning and few-shot learning to transfer and apply the established spectral nutrient mapping knowledge to predict the nutritional components of unknown ingredients specifically includes:

[0030] Based on a pre-trained deep spectral analysis neural network, the spectral nutritional mapping knowledge of known ingredients is transferred to new ingredient types through a fine-tuning strategy.

[0031] To address the limited number of new food samples, a prototype network is used to achieve small-sample learning.

[0032] Synthetic spectral data is generated using spectral data augmentation techniques and generative adversarial networks to expand the training samples;

[0033] The Bayesian calibration method is applied to assess the prediction uncertainty and provide confidence intervals for the predicted values.

[0034] As a further optimization of the present invention, the step of combining the user's health status and nutritional needs and using an optimization algorithm to calculate the optimal food combination scheme specifically includes:

[0035] Establish a personalized nutrition needs model based on the user's health status, physiological characteristics, and health goals;

[0036] An improved mixed-integer linear programming model is used to solve for the optimal combination of ingredients that meets the nutritional needs of users.

[0037] Establish a model of the relationship between cooking techniques and nutrient loss, predict the rate of change of nutrient components under different cooking methods, and dynamically adjust the formulation scheme;

[0038] Taking into account multiple factors, a final personalized nutrition meal plan is generated.

[0039] As a further optimization of the present invention, the deep spectral analysis neural network is optimized using the following loss function:

[0040]

[0041] Where L represents the loss function; N represents the number of samples, i.e., the total number of food samples participating in the training; M represents the number of target nutrient types; u represents the sample index, from 1 to N; w represents the nutrient index, from 1 to M; y uwThis represents the measured value of the w-th nutrient component in the u-th sample; This represents the model's predicted value for the w-th nutrient component in the u-th sample; It represents the square of the difference between the predicted value and the measured value, and is used to quantify the error of a single prediction; This represents the summation over all nutrient types; This represents summing over all samples; This means averaging the total error to obtain the average error for each sample.

[0042] As a further optimization of the present invention, the step of calculating the optimal food combination scheme by combining the user's health status and nutritional needs with an optimization algorithm further includes:

[0043] Cooking Loss Compensation: A compensation coefficient is used to adjust for nutrient loss under different cooking methods.

[0044]

[0045] Among them, C comp R represents the compensation coefficient. loss This indicates the nutrient loss rate under the corresponding cooking method.

[0046] A system for configuring personalized nutritional meals, used to perform the method for configuring personalized nutritional meals described above, includes:

[0047] The hyperspectral imaging and mass spectrometry combined data acquisition module is used to acquire spectral feature data and chemical composition data of food samples to obtain holographic chemical fingerprint information;

[0048] The multi-scale feature extraction module is used to process the collected spectral data and extract multi-level feature information of the ingredients;

[0049] The deep spectral analysis module is used to establish the mapping relationship between spectral characteristics and the content of nutrients and bioactive substances in food.

[0050] The transfer learning and few-shot learning module is used to transfer and apply the spectral nutritional mapping knowledge of known ingredients to unknown ingredients.

[0051] The nutrition meal configuration module is used to calculate the optimal ingredient combination scheme and generate a personalized nutrition meal configuration scheme based on the user's health status, nutritional needs, and full spectrum of nutritional characteristics of ingredients.

[0052] The beneficial effects of this invention are as follows: This method combines hyperspectral imaging and mass spectrometry analysis with a deep learning model to achieve high-precision prediction of the nutrient content of food ingredients. This not only improves the accuracy of nutrient content prediction and the accuracy of configuration schemes, but also improves the rate of achieving nutritional balance. At the same time, it breaks through the limitations of static databases, realizes real-time nutritional component analysis of any food ingredient, and can adapt to individual differences and changes in freshness of food ingredients, providing precise configuration based on the actual characteristics of food ingredients. Attached Figure Description

[0053] Figure 1 This is a flowchart of a method for configuring personalized nutritional meals according to the present invention;

[0054] Figure 2 This is a detailed flowchart of the present invention for obtaining holographic chemical fingerprint information;

[0055] Figure 3 This is a detailed flowchart of the feature data standardization and dimensionality reduction processing of the present invention;

[0056] Figure 4 This is a detailed flowchart of the model verification and optimization of the present invention;

[0057] Figure 5 This is a detailed flowchart of the invention for predicting the nutritional components of unknown food ingredients;

[0058] Figure 6 This is a detailed flowchart of the personalized nutrition meal preparation scheme generated by the present invention. Detailed Implementation

[0059] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0060] At least one embodiment of the present invention discloses a method for configuring personalized nutritional meals, such as Figures 1 to 6 As shown, it includes the following steps:

[0061] Step 1: Use hyperspectral imaging and mass spectrometry to collect spectral feature data and chemical composition data of food samples to obtain holographic chemical fingerprint information;

[0062] The specific steps are as follows:

[0063] Step 1.1: Hyperspectral imaging data acquisition;

[0064] Specifically, the food samples are scanned using a hyperspectral imaging device to obtain continuous spectral data in the wavelength range of 400-2500nm, forming a hyperspectral data cube containing spatial and spectral dimensions, with each pixel containing complete spectral curve information.

[0065] Step 1.2: Mass spectrometry analysis data acquisition;

[0066] Specifically, high performance liquid chromatography-mass spectrometry (HPLC-MS) was used to analyze food samples to obtain information on the molecular weight, structure, and content of bioactive substances in the food.

[0067] Step 1.3: Data fusion and standardization;

[0068] Specifically, hyperspectral and mass spectrometry data are associated according to sample IDs to construct a multimodal feature dataset for the samples. Radiometric and geometric corrections are performed on the hyperspectral data to eliminate equipment noise and environmental interference; peak alignment and normalization are performed on the mass spectrometry data to ensure the comparability of different batches of data.

[0069] Step 1.4: Extraction of chemical fingerprint information;

[0070] Specifically, based on the collected multimodal data, a holographic chemical fingerprint of the food is constructed, including complete spectra of major nutrients (proteins, fats, carbohydrates, vitamins, minerals, etc.) and bioactive substances (polyphenols, flavonoids, terpenoids, alkaloids, etc.).

[0071] Step 2: Process the collected spectral data using a multi-scale feature extraction algorithm to extract multi-level feature information of the ingredients, and standardize and reduce the dimensionality of the feature data.

[0072] The specific steps are as follows:

[0073] Step 2.1: Spectral preprocessing;

[0074] Specifically, preprocessing is performed on hyperspectral data, including polynomial smoothing, standard normal variable transformation (SNV), and first derivative transformation, to eliminate scattering effects and baseline drift, and enhance spectral characteristics.

[0075] Step 2.2, Multi-scale feature extraction algorithm;

[0076] Specifically, wavelet transform (WT) combined with multi-scale analysis is used to decompose the preprocessed spectral data and extract spectral features at different scales. The wavelet transform formula is as follows:

[0077]

[0078] Among them, WT f (a, b) represents the wavelet transform result, i.e., the wavelet coefficients at scale a and position b; f(t) represents the original spectral signal, which is a function of time or wavelength; ψ * This represents the conjugate form of the wavelet basis functions; This indicates scaling and translation transformations applied to the wavelet basis functions; 'a' represents the scaling parameter, which controls the degree of scaling of the wavelet; 'b' represents the translation parameter, which controls the position of the wavelet along the signal. It is a normalization factor to ensure consistent wavelet transform energy at different scales; integral sign It represents the integration operation performed over the entire time domain or frequency domain. dt represents the derivative of the integration variable, that is, the small change in the time or wavelength variable t, and is the basic element of the integration operation.

[0079] Step 2.3: Hierarchical feature extraction;

[0080] Specifically, based on the results of multi-scale analysis, a three-layer feature extraction network is constructed:

[0081] Macro layer: Extracting visible features such as surface texture, shape, and color of food ingredients;

[0082] Mesoscopic layer: Extracting characteristics such as the internal structure, tissue density, and moisture distribution of food ingredients;

[0083] Microscopic layer: Extracting molecular-level characteristics such as molecular composition, chemical bond structure, and functional group distribution.

[0084] Step 2.4, Feature Optimization and Filtering;

[0085] Specifically, an adaptive feature selection algorithm is applied, based on information gain and mutual information criteria, to screen out a subset of spectral features that are highly correlated with the content of nutrients and bioactive substances, thereby reducing data redundancy and improving the accuracy of subsequent modeling.

[0086] Step 3: Based on the processed feature data, establish a mapping relationship between spectral features and the content of nutritional components and bioactive substances in food through a deep spectral analysis neural network.

[0087] The specific steps are as follows:

[0088] Step 3.1: Construction of a deep spectral analysis neural network;

[0089] Specifically, a hybrid neural network structure comprising convolutional layers, long short-term memory layers, and fully connected layers is designed to process high-dimensional spectral data. The network structure is as follows:

[0090] Input layer: Receives preprocessed spectral data, with dimensions of [number of samples × number of spectral bands];

[0091] Convolutional layers: use multi-scale convolutional kernels to extract local spectral features and capture the correlation between different bands;

[0092] Long Short-Term Memory (LSTM) layer: processes the sequential characteristics of spectral data and captures long-term dependencies in wavelength changes;

[0093] Fully connected layer: Integrates features from previous layers and maps them to the final predicted nutrient content.

[0094] Step 3.2: Multivariate influencing factor modeling;

[0095] Specifically, variables such as food origin, season, and storage time are introduced as auxiliary features to construct a multivariate influencing factor model. The model adopts an additive structure:

[0096] Y = f1(X1) + f2(X2) + ... + f n1 (X n1 )+ε;

[0097] Where Y represents the content of the target nutrient, and X1, X2, X... n1 f1, f2, and fn1 represent the 1st, 2nd, and n1th influencing factors, respectively. n1 Let represent the nonlinear mapping functions corresponding to the 1st, 2nd, and n1th influencing factors, respectively, where n1 represents the total number of influencing factors and ε represents the residual term.

[0098] Step 3.3: Establish the mapping relationship;

[0099] Specifically, a deep spectral analysis neural network is trained to establish a quantitative mapping relationship between spectral features and the content of nutrients and bioactive substances. The network is optimized by minimizing the mean square error between predicted and measured values.

[0100]

[0101] Where L represents the loss function; N represents the number of samples, i.e., the total number of food samples participating in the training; M represents the number of target nutrient types; u represents the sample index, from 1 to N; w represents the nutrient index, from 1 to M; y uw This represents the measured value of the w-th nutrient component in the u-th sample; This represents the model's predicted value for the w-th nutrient component in the u-th sample; It represents the square of the difference between the predicted value and the measured value, and is used to quantify the error of a single prediction; This represents the summation over all nutrient types; This represents summing over all samples; This means that the average error of each sample is obtained by averaging the total error.

[0102] Step 3.4: Model Validation and Optimization;

[0103] Specifically, 10x cross-validation was used to evaluate model performance, and the generalization ability and prediction accuracy of the model were improved by adjusting the network structure, optimizing hyperparameters, and applying regularization techniques.

[0104] Step 4: For novel food ingredients, combine transfer learning and few-shot learning methods to transfer and apply the established spectral nutrition mapping knowledge to predict the nutritional components of unknown food ingredients.

[0105] The specific steps are as follows:

[0106] Step 4.1: Construction of the transfer learning model;

[0107] Specifically, based on a pre-trained deep spectral analysis neural network, a fine-tuning strategy is used to transfer the spectral nutritional mapping knowledge of known food ingredients to new food types. Transfer learning is described by the following formula:

[0108] J transfer (θ t ) = J target (θ t )+λ·R(θ t θ s );

[0109] Among them, J transfer (θ t J represents the total loss function for transfer learning, used to measure the overall performance of the model on the target task; target (θ t ) represents the loss function for the target task, used to evaluate the model's prediction error on the target data; θ t θ represents the set of parameters of the target model, i.e., the model weights that need to be optimized; s R(θ) represents the set of parameters of the source model, i.e., the weights of the pre-trained model. t θ s ) represents the regularization term, used to measure the difference between the parameters of the target model and the parameters of the source model; λ represents the tradeoff coefficient, which controls the influence of the regularization term and is used to balance the relationship between target task learning and knowledge transfer.

[0110] Step 4.2: Application of few-shot learning algorithm;

[0111] Specifically, to address the limited number of samples for new food ingredients, prototypical networks are used to achieve few-shot learning. For each food ingredient category c, its prototypical representation is calculated using a small number of samples:

[0112]

[0113] Where, p c S represents the prototype vector of category c, which is the average of the features of all samples in that category; c S represents the support set for category c, i.e., the set of samples used for training in that category; c | Indicates support for set S c The number of samples in the middle; (x k y k ) represents a sample and its label in the support set; x k y represents the k-th input sample; k This represents the label corresponding to the k-th input sample; This represents the feature extraction function, used to extract the original input x. k Convert to feature representation; This represents the summation of all samples in the support set for category c.

[0114] Step 4.3, Data Augmentation and Compositing;

[0115] Specifically, for food types with scarce samples, spectral data augmentation techniques are used to expand the training samples. Augmentation methods include additive Gaussian noise, spectral shifting, band mixing, and spectral interpolation. Simultaneously, a Generative Adversarial Network (GAN) is used to generate synthetic spectral data to further expand the training set size.

[0116] Step 4.4: Prediction result calibration;

[0117] Specifically, for predictions of unknown ingredients, a Bayesian calibration method is applied to assess prediction uncertainty and provide confidence intervals for the predicted values. The calibration formula is:

[0118] p(y|x, D)=∫p(y|x, θ)p(θ|D)dθ;

[0119] Where p(y|x,D) represents the posterior distribution of the target value y given the training data D and the input x; p(y|x,θ) represents the prediction distribution when the model parameter is θ; p(θ|D) represents the posterior distribution of the parameter, that is, the probability distribution of the model parameter θ given the training data D; ∫ represents the integral operation over all possible parameters θ, used to calculate the marginal probability distribution; dθ represents the infinitesimal element of the parameter space, which is part of the integral operation.

[0120] Step 5: Based on the predicted nutritional characteristics of the ingredients, combined with the user's health status and nutritional needs, use an optimization algorithm to calculate the optimal ingredient combination scheme and generate a personalized nutritional meal plan.

[0121] The specific steps are as follows:

[0122] Step 5.1: Modeling user nutritional needs;

[0123] Specifically, based on the user's health status, physiological characteristics, and health goals, a personalized nutritional needs model is established to determine the target intake of various nutrients and bioactive substances and their allowable fluctuation range.

[0124] Step 5.2, Ingredient combination optimization algorithm;

[0125] Specifically, an improved Mixed Integer Linear Programming (MILP) model is used to solve for the optimal food combination that meets the user's nutritional needs. The optimization objective function is set as follows:

[0126]

[0127] Where minJ represents minimizing the objective function J, that is, finding the solution that minimizes the objective function; This represents the total cost of all ingredients; i represents the index of the ingredient, from 1 to n; n represents the total number of available ingredients; c i x represents the unit cost or weighting coefficient of the i-th ingredient; i represents the selection quantity of the i-th ingredient; w represents the deviation penalty coefficient, used to balance the weight between cost and nutritional requirement deviation. This represents the sum of all nutrient deviations; j represents the nutrient index, from 1 to m; m represents the total number of nutrients considered. This represents the deviation of the j-th nutrient from the target amount; This represents the deviation value of the j-th nutrient from the target amount.

[0128] Specifically, a model is established to show the relationship between cooking techniques and nutrient loss, predicting the rate of change in nutrient composition under different cooking methods. Based on this, the formulation is dynamically adjusted to compensate for nutrient loss during cooking. The cooking loss compensation coefficient is calculated as follows:

[0129]

[0130] Among them, C comp R represents the compensation coefficient. loss This indicates the nutrient loss rate under the corresponding cooking method.

[0131] Step 5.4: Generation and optimization of nutritional meal plans;

[0132] Specifically, taking into account multiple factors such as nutritional balance, taste preferences, ingredient freshness, and cost-effectiveness, a final personalized nutritional meal plan is generated, and a detailed ingredient list, dosage guidelines, and cooking suggestions are provided.

[0133] The personalized nutritional meal preparation method provided in this embodiment has the following significant technical advantages compared with the prior art:

[0134] High-precision nutrient prediction: By combining hyperspectral imaging and mass spectrometry analysis with deep learning models, high-precision prediction of nutrient content in food can be achieved, improving the accuracy of nutrient content prediction, enhancing the accuracy of formulation plans, and increasing the rate of achieving nutritional balance.

[0135] Real-time dynamic analysis capability: Breaking through the limitations of static databases, it enables real-time nutritional analysis of any ingredient, adapting to individual differences and changes in freshness, and providing precise configuration based on the actual characteristics of the ingredients.

[0136] Full-spectrum nutritional analysis: Expanding from traditional major nutrient analysis to comprehensive assessment of bioactive substances, providing more comprehensive health regulation functions to meet users' diverse health needs.

[0137] High adaptability and scalability: Through transfer learning and few-shot learning techniques, the system can continuously learn and adapt to new food types, achieving high-precision prediction without a large amount of sample data, and significantly reducing model update and maintenance costs.

[0138] Real-world application examples of this implementation method

[0139] Application Scenario: This implementation method has been applied in the personalized nutrition meal planning system of a large health management center. The center needs to provide personalized nutrition meal plans for 500 members with different health conditions and nutritional needs daily. Before applying this method, the center used a traditional nutrition meal planning system based on a static database, which faced problems such as insufficient accuracy in nutrient calculations, inability to handle differences in ingredients, and a limited variety of nutrients.

[0140] Implementation process example:

[0141] Hyperspectral imaging and mass spectrometry data acquisition: Hyperspectral imaging equipment and liquid chromatography-mass spectrometry were used to analyze 200 commonly used food ingredients at the center. Table 1 shows the hyperspectral data characteristics of some of the food ingredients.

[0142] Table 1: Hyperspectral Data Characteristics of Some Food Ingredients

[0143] Ingredient Name Effective band range (nm) Characteristic peak position (nm) Data Dimensions Number of characteristic bands broccoli 450-2350 675,980,1450 1024×1024×240 120 salmon 420-2400 550,1210,1720 1024×1024×256 135 Quinoa 440-2300 1120,1450,1950 1024×1024×235 110 beef 430-2380 540,1180,1680 1024×1024×250 140 spinach 455-2340 680,990,1460 1024×1024×238 118

[0144] Multi-scale feature extraction and accurate modeling: Multi-scale feature extraction algorithms were applied to the collected hyperspectral data to extract food features at different levels. Table 2 shows the results of multi-scale feature extraction.

[0145] Table 2: Examples of multi-scale feature extraction results (taking broccoli as an example)

[0146]

[0147] A mapping relationship between the spectral characteristics of broccoli and its nutrient content was established using a deep spectral analysis neural network. Table 3 shows a comparison of the prediction results with those of traditional methods.

[0148] Table 3: Comparison of Predicted Nutritional Components of Broccoli

[0149]

[0150] Application of Transfer Learning and Few-Shot Learning: For the newly introduced purple cabbage (a non-pre-trained ingredient), transfer learning and few-shot learning methods were used to predict its nutritional components. Table 4 shows the prediction accuracy under different training sample numbers.

[0151] Table 4: Prediction accuracy of nutritional components in purple cabbage (different sample sizes)

[0152]

[0153] Personalized meal planning application: A personalized nutritional lunch was planned for a 45-year-old male member (mild hypertension, slightly hyperlipidemia). Table 5 shows the nutritional needs and planning results.

[0154] Table 5: Results of Personalized Nutritional Lunch Configuration

[0155]

[0156] Technical effectiveness verification:

[0157] Verification of high-precision nutrient prediction results: The accuracy of nutrient prediction using this method was compared with that of traditional methods on 200 commonly used food ingredients. The results are shown in Table 6.

[0158] Table 6: Comparison of Nutrient Prediction Accuracy

[0159] index This method Traditional methods Improvement rate (%) Prediction accuracy rate of major nutrients (%) 96.85 84.26 14.94 Micronutrient prediction accuracy (%) 95.42 78.62 21.37 Accuracy rate (%) in predicting bioactive substances 94.75 72.48 30.73 Overall prediction accuracy (%) 95.67 78.45 21.95 Consistency of prediction results (%) 97.82 83.56 17.07

[0160] Verification of the effect of personalized nutrition meal preparation: The effect of this method on the nutrition meal preparation of 100 members with different health conditions was compared with that of the traditional method. The results are shown in Table 7.

[0161] Table 7: Comparison of the effects of personalized nutrition meal preparation

[0162] index This method Traditional methods Improvement rate (%) Nutritional balance achievement rate (%) 96.35 82.18 17.24 Full spectrum nutritional coverage (%) 92.84 68.45 35.63 User satisfaction rating (out of 5) 4.62 3.85 20.00 Health indicator improvement rate (%) 78.65 56.32 39.65 Program compliance rate (%) 86.45 64.28 34.49

[0163] The above application examples and effect verification fully demonstrate the innovation and practicality of this method in the field of personalized nutrition meal preparation, and achieve a significant improvement in the accuracy of nutrient prediction and the precision of preparation scheme.

[0164] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for preparing personalized nutritional meals, characterized in that, Includes the following steps: Hyperspectral imaging and mass spectrometry were used to collect spectral feature data and chemical composition data of food samples to obtain holographic chemical fingerprint information; The collected spectral data were processed using a multi-scale feature extraction algorithm to extract multi-level feature information of the ingredients, and the feature data were standardized and dimensionality reduced. Based on the processed feature data, a mapping relationship between spectral features and the content of nutritional components and bioactive substances in food is established through a deep spectral analysis neural network. For novel food ingredients, we combine transfer learning and few-shot learning methods to transfer and apply the established spectral nutrition mapping knowledge to predict the nutritional components of unknown food ingredients. Based on the predicted nutritional characteristics of ingredients, combined with the user's health status and nutritional needs, an optimization algorithm is used to calculate the optimal ingredient combination scheme and generate a personalized nutritional meal plan.

2. The method for configuring personalized nutritional meals according to claim 1, characterized in that, The acquisition of spectral characteristic data and chemical composition data of food samples using hyperspectral imaging and mass spectrometry specifically includes: Food samples were scanned using a hyperspectral imaging device to obtain continuous spectral data in the wavelength range of 400-2500nm; Liquid chromatography-mass spectrometry was used to analyze food samples to obtain information on the molecular weight, structure and content of bioactive substances in the food. Hyperspectral data and mass spectrometry data are associated according to sample ID to construct a multimodal feature dataset of the samples; Radiometric and geometric corrections were performed on the hyperspectral data, and peak alignment and normalization were performed on the mass spectrometry data.

3. The method for configuring personalized nutritional meals according to claim 1, characterized in that, The process of processing the acquired spectral data using a multi-scale feature extraction algorithm specifically includes: Preprocess the hyperspectral data; Wavelet transform combined with multi-scale analysis was used to decompose the preprocessed spectral data and extract spectral features at different scales. Based on the results of multi-scale analysis, a three-layer feature extraction network was constructed to extract macroscopic, mesoscopic, and microscopic features of food ingredients. An adaptive feature selection algorithm was applied to screen out a subset of spectral features that are highly correlated with the content of nutrients and bioactive substances.

4. The method for configuring personalized nutritional meals according to claim 1, characterized in that, The deep spectral analysis neural network comprises a hybrid neural network structure of convolutional layers, long short-term memory layers, and fully connected layers, specifically including: The input layer receives the preprocessed spectral data; Convolutional layers use multi-scale convolutional kernels to extract local spectral features and capture the correlation between different bands; Long short-term memory layers process the sequential characteristics of spectral data, capturing long-term dependencies in wavelength changes; The fully connected layer integrates the features from the previous layer and maps them to the final predicted values ​​of nutrient content.

5. The method for configuring personalized nutritional meals according to claim 1, characterized in that, The establishment of the mapping relationship between spectral characteristics and the content of nutritional components and bioactive substances in food ingredients specifically includes: Introduce variables as auxiliary features to construct a multivariate influence factor model; Train a deep spectral analysis neural network to establish a quantitative mapping relationship between spectral features and the content of nutrients and bioactive substances. The model performance was evaluated using 10x cross-validation. By adjusting the network structure, optimizing hyperparameters, and applying regularization techniques, the model's generalization ability and prediction accuracy were improved.

6. The method for configuring personalized nutritional meals according to claim 1, characterized in that, The method combining transfer learning and few-shot learning applies established spectral nutrient mapping knowledge to predict the nutritional components of unknown ingredients, specifically including: Based on a pre-trained deep spectral analysis neural network, the spectral nutritional mapping knowledge of known ingredients is transferred to new ingredient types through a fine-tuning strategy. To address the limited number of new food samples, a prototype network is used to achieve small-sample learning. Synthetic spectral data is generated using spectral data augmentation techniques and generative adversarial networks to expand the training samples; The Bayesian calibration method is applied to assess the prediction uncertainty and provide confidence intervals for the predicted values.

7. The method for configuring a personalized nutritional meal according to claim 1, characterized in that, The process of combining the user's health status and nutritional needs, and using an optimization algorithm to calculate the optimal food combination scheme, specifically includes: Establish a personalized nutrition needs model based on the user's health status, physiological characteristics, and health goals; An improved mixed-integer linear programming model is used to solve for the optimal combination of ingredients that meets the nutritional needs of users. Establish a model of the relationship between cooking techniques and nutrient loss, predict the rate of change of nutrient components under different cooking methods, and dynamically adjust the formulation scheme; Taking into account multiple factors, a final personalized nutrition meal plan is generated.

8. A method for configuring personalized nutritional meals according to claim 1, characterized in that, The deep spectral analysis neural network is optimized using the following loss function: Where L represents the loss function; N represents the number of samples, i.e., the total number of food samples participating in the training; M represents the number of target nutrient types; u represents the sample index, from 1 to N; W represents the nutrient index, from 1 to M; y uw This represents the measured value of the W-th nutrient component in the u-th sample; This represents the model's predicted value for the w-th nutrient component in the u-th sample; It represents the square of the difference between the predicted value and the measured value, and is used to quantify the error of a single prediction; This represents the summation over all nutrient types; This represents summing over all samples; This means averaging the total error to obtain the average error for each sample.

9. A method for configuring personalized nutritional meals according to claim 7, characterized in that, The method of calculating the optimal food combination scheme by combining the user's health status and nutritional needs with an optimization algorithm also includes: Cooking Loss Compensation: A compensation coefficient is used to adjust for nutrient loss under different cooking methods. Among them, C comp R represents the compensation coefficient. loss This indicates the nutrient loss rate under the corresponding cooking method.

10. A system for configuring personalized nutritional meals, used to perform the method for configuring personalized nutritional meals according to any one of claims 1-9, characterized in that, include: The hyperspectral imaging and mass spectrometry combined data acquisition module is used to acquire spectral feature data and chemical composition data of food samples to obtain holographic chemical fingerprint information; The multi-scale feature extraction module is used to process the collected spectral data and extract multi-level feature information of the ingredients; The deep spectral analysis module is used to establish the mapping relationship between spectral characteristics and the content of nutrients and bioactive substances in food. The transfer learning and few-shot learning module is used to transfer and apply the spectral nutritional mapping knowledge of known ingredients to unknown ingredients. The nutrition meal configuration module is used to calculate the optimal ingredient combination scheme and generate a personalized nutrition meal configuration scheme based on the user's health status, nutritional needs and full spectrum of nutritional characteristics of ingredients.

Images