Food flavor evaluation system construction based on big data and artificial intelligence

Abstract

The embodiment of the invention discloses a personalized healthy catering matching system constructed by utilizing artificial intelligence, and how to realize intelligent and dynamic menu generation and recommendation taking high-nutrition food materials (such as salmons) as a core based on multi-dimensional user data, authoritative nutrition knowledge and food material characteristics. The system not only solves the pain point of'thousands of people one face 'of traditional diet suggestions, improves user compliance and health benefits through accurate matching, but also deeply integrates nutritional science, cooking science and information technology, and provides a core tool for intelligent health management.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and big data, and builds a system for evaluating healthy food combinations by establishing a healthy food combination system. Background Art

[0002] With the booming global food industry and consumers' increasing demands for food quality, health evaluation, as a key indicator of food quality, has become increasingly important. The nutritional value of food is a crucial factor consumers consider when choosing a food. However, the limitations of traditional health-matching methods, such as manual pairing, in terms of efficiency and accuracy, are becoming increasingly apparent.

[0003] First, manual matching is inefficient. Given the scale and scale of food production, relying solely on professionals for manual evaluation is not only time-consuming and labor-intensive, but also difficult to rapidly evaluate large numbers of food samples. This not only impacts food production efficiency but also increases production costs.

[0004] Secondly, the accuracy of manual evaluation is difficult to guarantee. Manual evaluation is influenced by a variety of factors, including the evaluator's psychological and physiological state and personal preferences, and the results may be subjective and erroneous. This not only affects the objectivity and fairness of the evaluation, but can also lead to fluctuations and instability in food quality.

[0005] The rapid development of artificial intelligence (AI) technology offers a new solution to this challenge. With its powerful data processing and learning capabilities, AI can rapidly and accurately analyze the healthy components and characteristics of foods. By leveraging advanced AI technologies such as deep learning and reinforcement learning, AI systems can integrate multi-source and multi-modal data, including sensory evaluation data, instrumental analysis data, and image data, enabling comprehensive and objective evaluation of healthy food combinations.

[0006] The application of artificial intelligence in healthy food pairing not only improves the efficiency and accuracy of evaluations but also drives technological innovation and industrial upgrading within the food industry. Through the use of AI systems, food manufacturers can achieve real-time monitoring and early warning of food quality, promptly identifying and resolving potential quality issues. Furthermore, AI systems can provide strong support for food research and development, helping companies develop new products that better meet consumer needs.

[0007] In summary, applying AI technology to healthy food pairing is not only an inevitable trend driven by technological development and application needs, but also a crucial measure to enhance food quality and meet consumer demands. With the continuous advancement of AI technology and the expansion of its application areas, we believe that healthy food pairing will usher in even greater development prospects in the future. Summary of the Invention

[0008] The purpose of this invention is to build a healthy food pairing system based on big data and artificial intelligence. This system integrates multi-source data related to food health and utilizes advanced algorithms and models to achieve comprehensive and accurate evaluation of food flavor. By combining the power of big data analysis with the advantages of artificial intelligence technology, this system aims to improve the efficiency and accuracy of healthy food pairing and provide strong support for quality control and product development in the food industry.

[0009] To achieve the above objectives, the implementation scheme of the present invention is: a food health pairing system based on big data and artificial intelligence, which includes a data acquisition module, a data processing and analysis module and a food health pairing database.

[0010] The data acquisition module is responsible for collecting multi-source data related to food health, including food nutritional composition, health index, chromatographic data related to food safety and quality, mass spectrometry data, food appearance image data, and manual sensory evaluation data obtained by professional sensory evaluators. This data will serve as the basis for evaluating the healthiness of food.

[0011] The data processing and analysis module is the core of the system. It uses advanced big data analysis and artificial intelligence technologies to preprocess, extract features, and train models on collected data. First, the raw data is cleaned and standardized to eliminate noise and outliers. Then, feature extraction techniques are used to extract key features related to food flavor from the data. Finally, using deep learning and other artificial intelligence technologies, a multimodal, multi-task deep learning model is constructed to accurately evaluate food healthiness.

[0012] The food health pairing database is another key component of the system. It stores processed and analyzed data and trained models. The data and models in the database can be updated and optimized based on actual needs to adapt to different food types and health assessment requirements.

[0013] As a specific embodiment, the data processing and analysis module of the present invention utilizes a multimodal, multi-task deep learning architecture. This architecture can simultaneously process information from diverse data sources, including chromatographic data, mass spectrometry data, image data, and manual sensory evaluation data. Through multi-task learning, the system can simultaneously learn multiple objectives, such as food type, nutritional intensity, and nutritional range, thereby achieving a comprehensive assessment of food healthiness.

[0014] The present invention also provides a method for constructing a healthy food pairing system based on big data and artificial intelligence, including steps such as data collection, data preprocessing, feature extraction, model training, and health evaluation. In the data collection stage, multi-source data on food health is acquired through professional equipment and sensory evaluators; in the data preprocessing stage, the raw data is cleaned and standardized; in the feature extraction stage, key features are extracted from the data using feature extraction techniques; in the model training stage, an evaluation model is constructed using artificial intelligence technologies such as deep learning; and in the healthy pairing stage, the data on the foods to be paired is input into the trained model to obtain healthy food pairing results.

[0015] The verification method of the present invention involves preprocessing and testing a series of representative food samples to obtain raw data such as chromatography and mass spectrometry, while also conducting manual sensory evaluation. This data is then input into the food health matching system of the present invention, and the system's output is compared with the actual evaluation results to verify the accuracy and reliability of the system.

[0016] This paper integrates the advantages of big data and artificial intelligence technologies to build a comprehensive and accurate food health pairing system. This system not only improves the efficiency and accuracy of healthy food pairing, but also provides strong support for quality control and product development in the food industry. Furthermore, the system is scalable and customizable, allowing it to be tailored and optimized based on different food types and nutritional requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A system structure diagram provided by the present invention; Figure 2 Schematic diagram of the research process of the present invention DETAILED DESCRIPTION

[0018] Introduction to the specific implementation method of the food health matching system based on big data and artificial intelligence Example 1: Big Data and Artificial Intelligence-Driven Healthy Food Pairing System like Figure 1As shown, this embodiment introduces a food health pairing system based on big data and artificial intelligence, and its specific implementation method. The system consists of three core components: a component sensory detection system, a computer intelligent analysis system, and a food health database. The sensory detection system consists of an LC-MS (liquid chromatography-mass spectrometry) and a GC-O (gas chromatography-olfactometer). The LC-MS captures raw data on the health characteristics of food samples, including chromatographic and mass spectrometric data; the GC-O collects manual sensory evaluation data on the food samples' odor, namely, aroma characteristics and aroma intensity. The computer intelligent analysis system performs preprocessing, feature extraction, and deep learning analysis on the data acquired by the component sensory detection system. Using the multimodal, multi-task deep learning architecture M3TDN, a comprehensive evaluation of flavor type, flavor intensity, and flavor range is achieved. The food health database utilizes computer artificial intelligence algorithms to perform fitting analysis on the data acquired by the sensory detection system, eliminate errors, identify data correlations and patterns, and establish a comprehensive food health database containing comprehensive data on food sample chromatographic information, mass spectrometric information, nutritional elements, and healthiness. The implementation includes the following steps: Step Q1: Establish a food flavor database. Collect food samples and obtain chromatographic and mass spectrometric data using LC-MS as raw data for health characteristics. Collect manual sensory evaluation data, including aroma characteristics and aroma intensity, using GC-O. Use computer software to perform fitting analysis on the raw data, eliminate errors, and establish a food health database.

[0019] Step Q2: Data preprocessing and feature extraction. The collected chromatographic and mass spectrometric data are normalized and quantified to generate high-dimensional health data. Feature extraction is performed using two methods: one is to apply classic algorithms such as LDA and PCA to extract and analyze features from the raw data; the other is to use deep restricted Boltzmann machines to extract and learn features from the high-dimensional health data.

[0020] Step Q3: Deep Learning Model Construction. Based on the extracted feature data, a multimodal, multi-task deep learning architecture, M3TDN, was constructed. Training was performed on the high-dimensional, multi-source feature data, using a multi-layer LSTM and a multi-layer restricted Boltzmann machine to train the judgment model. Through model training, an AI-powered healthy food pairing system was established.

[0021] Step Q4: System Application and Evaluation. Apply the trained model to healthy food pairings to achieve intelligent judgment of food nutrient type, nutrient intensity, and nutrient range. Evaluate and optimize the system by comparing the results of traditional sensory evaluation with intelligent evaluation.

[0022] By using the method described in this example, we can establish a food flavor evaluation system based on big data and artificial intelligence, achieving comprehensive, rapid, and accurate evaluation of food flavor. This not only improves the scientific nature and accuracy of food evaluation, but also provides strong support for food research and development, quality control, and consumer selection.

[0023] Implementation Case 2: AI-driven personalized food pairing recommendation system Building on Implementation Case 1, this implementation case introduces an adaptive user profile building module and a multi-level nutritional knowledge graph module to achieve more accurate and personalized nutritional pairing recommendations. Compared to traditional rule-based recommendation systems, this solution uses deep learning technology to dynamically capture changes in user needs and utilizes graph neural networks to explore complex relationships between nutrients, improving the accuracy and personalization of recommendations.

[0024] Step Q1: Specific settings of the user portrait construction module. The user portrait construction module is set at the front end of the intelligent matching recommendation module. Its function is to build a personalized user portrait based on the user's personal information and health data, and provide a basis for accurate nutritional matching recommendations. This module adopts a distributed storage architecture, using MongoDB to store the user's structured information (such as basic physiological indicators, dietary preferences, etc.), the Redis cache system to store frequently accessed user feature vectors, and ElasticSearch to index the user's behavioral sequence data. The user information collection interface is developed based on Web technology and supports access from multiple terminal devices. The basic information that the user needs to enter includes: age, gender, height, weight, occupation type, exercise habits, past medical history, allergy information, dietary preferences, etc. For the physiological characteristic dimension, the system automatically calculates the user's body mass index (BMI); the basal metabolic rate is calculated using the Mifflin-St Jeor equation:

[0025]

[0026] Step Q2: Calculate the total daily energy expenditure and the sequential pattern of eating behavior

[0027] Where α and β are balance factors (α+β=1), is the basic activity coefficient, is the intensity coefficient of the i-th specific activity, is the corresponding time weight. Example activity coefficient settings: 1.2 for sedentary lifestyle, 1.375 for light activity, 1.55 for moderate activity, 1.725 for heavy activity, and 1.9 for very heavy activity. Regarding behavioral characteristics, a recurrent neural network is used to extract sequential patterns in user dietary behavior, converting historical dietary records into time series vectors. An LSTM network is then used to learn the user's dietary habits. Regarding preference characteristics, matrix factorization techniques are used to explore the user's potential preference factors and construct a user-food rating matrix R, employing weighted regularized matrix factorization. Finally, regarding health status, a health constraint model is constructed by combining the user's health indicators and medical history. For users with specific diseases, a disease-nutrient association matrix D is introduced.

[0028] Step Q3: Implement the Nutrition Knowledge Graph module. The Nutrition Knowledge Graph module resides within the health database and stores and manages the complex relationship network between food nutrients, supporting intelligent reasoning and knowledge discovery. This module is built using the Neo4j graph database and contains 15,000 food entity nodes, 1,200 nutrient entity nodes, and 650,000 relationship edges. The knowledge graph is constructed using a semi-automated process: Entity recognition utilizes a BERT-based named entity recognition model, pre-trained for the nutrition domain; relationship extraction utilizes a bidirectional LSTM model based on an attention mechanism, taking into account both semantic and structural relationships between entities; and knowledge fusion utilizes entity alignment and relationship alignment algorithms to eliminate duplicate and conflicting information.

[0029] Step Q4: Apply personalized recommendation algorithm. Based on user profile and nutrition knowledge graph, a multi-objective optimization recommendation algorithm is used, taking into account four goals: nutritional balance, user preference matching, cost-effectiveness, and food availability:

[0030] Constraints include nutritional needs constraints, health status constraints, and budget constraints:

[0031] Step Q5: System Performance Verification. On a test set consisting of 5,000 users and 12 months of historical data, the recommendation system in this implementation case achieved significant performance improvements compared to implementation case 1. The nutritional goal achievement rate increased from 76.8% to 89.2%, user satisfaction scores increased from 3.6 to 4.3 (on a 5-point scale), and the recommendation diversity index increased from 0.42 to 0.67.

[0032] The accuracy of personalized recommendations was evaluated by consistency with the recommendations of expert nutritionists, with a similarity of 84.7%, indicating that the system can well simulate the recommendation logic of professional nutritionists.

[0033] Case Study 3: Using AI to Build a Personalized Healthy Dining System – Intelligent Menu Recommendations with Salmon as the Core Ingredient With rising health awareness, consumers are increasingly demanding personalized, scientifically-based dining. As a high-nutrient-density ingredient rich in omega-3 fatty acids, high-quality protein, and a variety of vitamins and minerals, salmon's cooking methods and combinations directly impact nutrient absorption and health benefits. To accurately meet the health goals and nutritional needs of different individuals, we have developed and implemented an intelligent healthy dining pairing system based on big data and artificial intelligence. This example details how the system recommends healthy menus centered around salmon for individual users.

[0034] Step Q1: Data Collection. Nutritional and characteristic data collection for ingredients: Liquid chromatography-mass spectrometry (LC-MS) is used to separate and detect the components in salmon samples, obtaining chromatographic and mass spectrometric data. This data contains information on the various flavor compounds in the salmon. Appearance image data collection: A high-definition camera is used to capture appearance image data of the salmon, including characteristics such as color and texture. Characteristics that influence nutrition and cooking choices, such as the salmon's freshness, location (belly, back), and origin (farmed / wild), are recorded. Manual sensory evaluation data collection: Professional tasters are organized to conduct sensory evaluations of salmon cooked in different ways, recording information such as flavor type, flavor intensity, and flavor amplitude. This evaluation data will serve as an important reference for system training.

[0035] Step Q2: Data Preprocessing. The chromatographic and mass spectrometric data obtained by LC-MS are cleaned, standardized, and normalized to remove noise and dimensionality. Image data is preprocessed, including enhancement and cropping, to extract key features. Manual sensory evaluation data and free-text feedback are encoded and subjected to natural language processing (NLP) for subsequent analysis. User health goals are quantified into metrics that the model can understand (e.g., target daily calories, protein grams, omega-3 intake).

[0036] Step Q3: Feature Extraction. Deep learning techniques, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), are used to extract key features from the chromatographic, mass spectrometric, and image data. These features can reflect the cooking flavor characteristics and nutritional value of the salmon. The extracted features are then fused and dimensionality reduced to form a unified feature vector.

[0037] Step Q4: Model training. Build a multimodal, multi-task deep learning model. This model can simultaneously process chromatography, mass spectrometry, image, and manual sensory evaluation data. Learn the complex mapping relationship between user characteristics and ingredient / cooking method / menu characteristics to predict the user's acceptance, compliance, and potential health benefit probability for a specific menu. Use collaborative filtering, content filtering, deep neural networks (DNN), or graph neural networks (GNN). Divide the preprocessed data into training and test sets. Train the model using the training set and adjust model parameters to optimize performance. During training, use appropriate loss functions and optimization algorithms to guide the model training process. Based on the matching results and nutritional rules (such as the dietary pyramid and nutritional guidelines for specific diseases), use sequence generation models (such as Transformer) or constrained optimization algorithms to dynamically generate personalized menu plans centered on salmon.

[0038] Step Q5: Personalized Menu Recommendation and Interaction. The test set data is fed into the trained model to generate nutritional and flavor evaluation results for salmon cooked in different ways. Based on the evaluation results, the flavor characteristics and nutritional value of the salmon cooked in different ways are analyzed. The system then calls the trained model to generate and present multiple personalized salmon menu options in real time. The evaluation results are displayed to the user in a visual format, such as a flavor radar chart or flavor wheel.

[0039] Step Q6: Result Verification and Optimization. Compare and verify the results with those of professional nutritionists to ensure the accuracy and reliability of the system's evaluation results. Fine-tune and optimize the model based on the verification results to improve system performance. Continuously collect new data and add it to the system. Use user feedback, performance data, and health tracking results as new training data to update the model regularly or in real time (online learning).

[0040] This example demonstrates how a personalized healthy dining pairing system built using AI can intelligently and dynamically generate and recommend menus centered around highly nutritious ingredients (such as salmon) based on multi-dimensional user data, authoritative nutritional knowledge, and ingredient characteristics. This system not only addresses the pain point of traditional, one-size-fits-all dietary recommendations, improving user compliance and health benefits through precise matching, but also deeply integrates nutrition science, culinary science, and information technology, providing a core tool for intelligent health management.

Claims

1. A food health matching system based on big data and artificial intelligence, characterized by: The system includes: a data acquisition module for collecting multi-source data related to food health, including nutritional composition data, health index data, food quality and safety data, food appearance image data and artificial sensory evaluation data; a data processing and analysis module, including a component sensory detection system and a computer intelligent analysis system, wherein: the component sensory detection system is composed of LC-MS and GC-O, LC-MS is used to obtain raw data such as nutritional composition and harmful components of food, GC-O is used to obtain artificial sensory evaluation data of food odor, namely odor intensity and odor characteristics; the computer intelligent analysis system is used to analyze the nutritional composition and harmful components obtained by the sensory detection system The raw data and artificial sensory evaluation data are preprocessed and feature extracted, the raw data are subjected to dimensionality reduction and cluster analysis, and the multimodal, multi-task deep learning architecture M3TDN is used for model training; the multimodality includes nutritional component data, health index data, food quality and safety data and food appearance image data, and the multitask includes simultaneous learning of food type, nutritional intensity and nutritional range; a food health matching database is used to store processed data and trained models. The data in the database is fitted and analyzed by a computer artificial intelligence algorithm on the data obtained by the sensory detection system, error points are eliminated, and the correlation and rules of the data are found.

2. The food health matching system based on big data and artificial intelligence according to claim 1 is characterized in that: The specific steps include: First, data collection is carried out to collect multi-source data related to food health; The second step is data preprocessing. Based on the total amount of sample substances and relevant instrument parameters during instrument analysis, the collected data is normalized and quantified to form high-dimensional health data. Next, feature extraction is performed on high-dimensional health data using two methods: one is to use classic algorithms such as LDA and PCA to perform feature extraction and analysis on data obtained by machine judgment and manual interpretation; the other is to use deep restricted Boltzmann machines to extract and learn features from high-dimensional flavor data; Then, we conduct model training. Using the deep learning architecture M3TDN, combined with multi-layer LSTM and multi-layer restricted Boltzmann machines, we train the feature-extracted data to establish an AI-powered healthy food pairing system.

3. Finally, the database is established. Based on the results of the fitting analysis, the error points are eliminated, the correlation and pattern of the data are found, and a food health combination database is established.

4. The method for constructing a healthy food pairing system based on big data and artificial intelligence according to claim 3, characterized in that: During the data collection process, food samples are separated by GC, one path enters the chemical detector FID or MS to obtain the original data of nutritional characteristics, and the other path enters the sniffer port O to obtain artificial sensory evaluation data. The artificial sensory evaluation data is formed by human nose smelling and manual knob electronic signal recording and then processed by computer.

Images