Chinese liquor AI evaluation system and method based on multi-modal data fusion

Abstract

The invention provides a Chinese liquor AI evaluation system and method based on multi-modal data fusion, and the system comprises a data collection module which is used for collecting multi-modal data of a liquor sample; the data preprocessing module is used for preprocessing the collected multi-modal data; the feature fusion module is used for fusing the multi-modal data and constructing feature vectors comprehensively reflecting the quality features of the white spirit; the evaluation model is used for learning a mapping relation between the multi-modal characteristics of the white spirit and the quality grade and evaluating the quality of the white spirit; and the output module is used for outputting the liquor quality identification result. Through multi-modal data fusion and a deep learning algorithm, the subjectivity of artificial evaluation is effectively reduced, the accuracy of an identification result is improved, automatic identification of the quality of the white spirit is realized, the identification efficiency is greatly improved, the identification cost is reduced, the quality of the white spirit can be comprehensively and systematically evaluated, and the method has an important application value.

Description

Technical Field

[0001] The present invention belongs to the intersection of artificial intelligence and food testing. Specifically, it relates to a Chinese liquor AI evaluation system and method based on multimodal data fusion. Background Art

[0002] As a treasure of traditional Chinese culture, the quality assessment of Chinese liquor has long relied on manual evaluation, a process characterized by subjectivity, high costs, and low efficiency. In recent years, the development of sensor technologies such as electronic noses and electronic tongues has provided new means for objectively evaluating liquor quality. However, a single sensor cannot fully capture the complex flavor profile of liquor. Electronic noses and electronic tongues can only detect physical and chemical parameters and cannot correlate with the unique stylistic characteristics of liquor (such as a rich cellar aroma and a sweet aftertaste). Foreign AI wine tasting models are designed for wine and are not suitable for the complex aroma system of Chinese liquor (12 major aromas and 53 minor aromas). Summary of the Invention

[0003] The embodiments of the present application provide a Chinese liquor AI evaluation system and method based on multimodal data fusion, which enables objective, accurate and efficient identification of liquor quality.

[0004] In a first aspect, the present application provides an AI-based Chinese liquor evaluation system based on multimodal data fusion, comprising:

[0005] Data acquisition module, used to collect multimodal data of liquor samples;

[0006] A data preprocessing module is used to preprocess the collected multimodal data;

[0007] Feature fusion module, used to fuse multimodal data and construct feature vectors that fully reflect the quality characteristics of liquor;

[0008] The evaluation model is used to learn the mapping relationship between the multimodal features of liquor and its quality grades, and evaluate the quality of liquor;

[0009] Output module, used to output the liquor quality identification results.

[0010] Among them, the multimodal data collected by the data acquisition module includes:

[0011] Volatile organic compound data of liquor samples collected using an electronic nose device;

[0012] Using a multi-channel electronic tongue device equipped with taste sensors, we collected the taste characteristics of liquor samples;

[0013] Sommeliers evaluate liquor samples, record scores across multiple dimensions, and generate structured data.

[0014] Physical and chemical index data of liquor samples obtained through laboratory testing.

[0015] Among them, the data preprocessing module is used to: perform data cleaning, feature extraction, and data standardization on the collected multimodal data; perform data cleaning through sliding average filtering or wavelet transform methods to remove noise and outliers in the sensor data; extract time domain and frequency domain features from the data collected by the electronic nose and electronic tongue, and the time domain and frequency domain features include mean, variance, peak, and spectral energy; and normalize data of different modes to eliminate dimensional differences.

[0016] Among them, the feature fusion module is used to: use a multimodal feature fusion algorithm to fuse the electronic nose collected data, electronic tongue collected data, manual evaluation data and physical and chemical index data into a high-dimensional feature vector. The multimodal feature fusion algorithm includes feature splicing, attention mechanism or graph neural network.

[0017] Among them, the evaluation model is built based on a deep learning algorithm, and the model architecture uses a multi-layer perceptron, a convolutional neural network or a Transformer model; the training data contains a multimodal data set of thousands of liquor samples; the loss function uses a mean square error or a cross-entropy loss function; the optimization algorithm uses the Adam optimizer, and the learning rate is set to 0.001.

[0018] Among them, the liquor quality identification results output by the output module include quality grade, flavor characteristic description, and quality defect analysis.

[0019] Among them, the output module is also used to visualize the identification results, including quality grade scores, flavor characteristic radar charts, and quality defect analysis.

[0020] Secondly, this application provides a Chinese liquor AI evaluation method based on multimodal data fusion, including:

[0021] The data acquisition module collects multimodal data of liquor samples;

[0022] The data preprocessing module preprocesses the collected multimodal data;

[0023] The feature fusion module fuses multimodal data to construct a feature vector that fully reflects the quality characteristics of liquor;

[0024] The evaluation model learns the mapping relationship between the multimodal features of liquor and its quality grades to evaluate the quality of liquor;

[0025] The output module outputs the liquor quality identification results.

[0026] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when executed by a processor.

[0027] In a fourth aspect, an embodiment of the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the program.

[0028] The Chinese liquor AI evaluation system and method based on multimodal data fusion in the embodiment of the present application has the following beneficial effects:

[0029] Objective and accurate: Through multimodal data fusion and deep learning algorithms, the subjectivity of manual evaluation is effectively reduced and the accuracy of identification results is improved.

[0030] Efficient and convenient: It realizes the automated identification of liquor quality, greatly improving the identification efficiency and reducing the identification cost.

[0031] Comprehensive and systematic: It takes into account many factors such as the aroma, taste, style, etc. of the liquor, and can comprehensively and systematically evaluate the quality of the liquor.

[0032] Strong explainability: Through methods such as feature importance analysis, the decision-making process of the AI evaluation model can be explained, thereby improving the credibility of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a schematic diagram of the structure of the Chinese liquor AI evaluation system based on multimodal data fusion in an embodiment of the present application;

[0034] Figure 2 This is a flow chart of the Chinese liquor AI evaluation method based on multimodal data fusion in an embodiment of the present application. DETAILED DESCRIPTION

[0035] The present application will be further described below with reference to the accompanying drawings and embodiments.

[0036] In the following description, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance. The following description provides multiple embodiments of the present invention, and different embodiments can be replaced or combined, so this application can also be considered to include all possible combinations of the same and / or different embodiments described. Therefore, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then this application should also be considered to include embodiments containing one or more of all other possible combinations of features A, B, C, and D, even though such embodiments may not be explicitly described in the following text.

[0037] Example 1

[0038] like Figure 1As shown, the Chinese liquor AI evaluation system based on multimodal data fusion of the present application includes: a data acquisition module 201, used to collect multimodal data of liquor samples; a data preprocessing module 202, used to preprocess the collected multimodal data; a feature fusion module 203, used to fuse multimodal data and construct a feature vector that comprehensively reflects the quality characteristics of liquor; an evaluation model 204, used to learn the mapping relationship between the multimodal features of liquor and the quality grade, and evaluate the quality of liquor; an output module 205, used to output the liquor quality identification results.

[0039] This application combines electronic sensory equipment with a manual evaluation database to construct a large model for the quality identification of Chinese liquor, achieving objective, accurate and efficient identification of liquor quality, and has important application value.

[0040] Example 2

[0041] The Chinese liquor AI evaluation system based on multimodal data fusion in this application includes: a data acquisition module: used to collect multimodal data of liquor samples, including electronic nose sensor data, electronic tongue sensor data, manual evaluation data, liquor physical and chemical index data, etc. A data preprocessing module: used to preprocess the collected multimodal data, including data cleaning, feature extraction, data standardization, etc. A feature fusion module: used to fuse data of different modalities to construct a feature vector that can comprehensively reflect the quality characteristics of liquor. An AI evaluation model: constructed based on a deep learning algorithm, used to learn the mapping relationship between the multimodal features of liquor and the quality grade, and to realize the automatic evaluation of the quality of liquor. A result output module: used to output the liquor quality identification results, including quality grade, flavor characteristic description, quality defect analysis, etc.

[0042] (1) Data acquisition module

[0043] Electronic nose sensor data acquisition:

[0044] A high-precision electronic nose equipped with a gas sensor array (such as metal oxide semiconductor sensors and electrochemical sensors) was used to collect volatile organic compound (VOC) data from liquor samples. The collection time for each liquor sample was set to 3-5 minutes to ensure data stability.

[0045] Electronic tongue sensor data collection:

[0046] A multi-channel electronic tongue device equipped with taste sensors (such as bitter, sweet, sour, salty, and umami) was used to collect flavor profile data for each liquor sample. The collection time for each liquor sample was set at 1-2 minutes.

[0047] Manual evaluation data collection:

[0048] A team of professional sommeliers will evaluate the liquor samples, record scores on dimensions such as aroma, taste, aftertaste, and style (e.g., a full score of 100), and generate structured data.

[0049] Physical and chemical index data collection:

[0050] Through laboratory testing, the physical and chemical index data of liquor samples such as alcohol content, total acid, total ester, and solids are obtained.

[0051] (2) Data preprocessing module

[0052] Data cleaning: removing noise and outliers from sensor data, for example through sliding average filtering or wavelet transform methods.

[0053] Feature extraction: Extract time domain and frequency domain features from electronic nose and electronic tongue data, such as mean, variance, peak value, spectral energy, etc.

[0054] Data normalization: Normalize data from different modalities to eliminate dimensional differences, such as using the Z-score normalization method.

[0055] (3) Feature fusion module

[0056] Use multimodal feature fusion algorithms (such as feature splicing, attention mechanism or graph neural network) to fuse electronic nose, electronic tongue, manual evaluation and physical and chemical index data into a high-dimensional feature vector.

[0057] For example, the 100-dimensional features of the electronic nose, the 50-dimensional features of the electronic tongue, the 10-dimensional features of manual evaluation, and the 5-dimensional features of physical and chemical indicators are spliced into a 165-dimensional feature vector.

[0058] (4) AI (Artificial Intelligence) Evaluation Model

[0059] Build models using deep learning algorithms, such as:

[0060] Model architecture: Use a multi-layer perceptron (MLP), convolutional neural network (CNN), or Transformer model.

[0061] Training Data: A multimodal dataset containing thousands of liquor samples is used for training.

[0062] Loss function: Use mean square error (MSE) or cross entropy loss function.

[0063] Optimization algorithm: Adam optimizer is used and the learning rate is set to 0.001.

[0064] The model output is the quality grade of liquor (such as superior, first-grade, second-grade) and flavor characteristics description (such as fruit aroma, cellar aroma, sauce aroma, etc.).

[0065] (5) Result output module

[0066] Visualize the AI tasting model's identification results, including: quality grade score (e.g., 95 / 100), flavor profile radar chart (displaying aroma, mouthfeel, aftertaste, etc.), and quality defect analysis (e.g., excessive acidity, insufficient aroma, etc.).

[0067] Example 3: Quality Grade Identification of Maotai-flavor Liquor (Application in the Brewing Workshop of Maotai Town Distillery)

[0068] 1. Test sample

[0069] Positive sample: 2023 Moutai core production area base wine (quality grade 1)

[0070] Defective samples: samples with "mud smell at the bottom of the pit" due to pit pollution (quality level 5), samples with "lack of caramel aroma" due to abnormal fermentation temperature (quality level 4). Data volume: a total of 120 groups of samples (including gradient dilution samples), divided into training / validation / test sets according to the ratio of 6:2:2.

[0071] 2. Sensor parameters

[0072]

[0073] 3. Feature Extraction

[0074] Electronic nose data: Extract the characteristic value at t = 30s in the dynamic response curve.

[0075] 4. Model Training

[0076] parameter set up Convolutional layer 3-layer CNN (kernel_size=5, stride=2) LSTM layer Bidirectional LSTM (hidden_size=128) Fusion Layer Fully connected layer (input dimension = 256 + 128 + 768 = 1152) Training cycle 200 epochs (batch_size=32)

[0077] 5. Verify the results

[0078] index System output Manual evaluation Quality grade accuracy 94.3％ 82.1％ Defect type recall rate 91.5% (burnt aroma defect) - Response time 8.7 seconds / sample 15 minutes / sample

[0079] 6. Industrial Verification

[0080] Application scenario: Moutai Town Distillery Brewing Workshop Base Wine Testing for July 2023 Batch

[0081] Abnormal detection: The ethyl acetate / ethyl lactate ratio was detected to be abnormal in the 37th batch (2.15 vs the standard value 2.68±0.15).

[0082] Treatment suggestion: The system automatically prompts "adjust the stacking fermentation time +24 hours", and it has been verified that the defect rate has decreased by 67%.

[0083] Example 4: Diagnosis of Luzhou-flavor Defects (Monitoring of Luzhou-flavor Liquor Cellars in Yibin, Sichuan)

[0084] 1. Data characteristics

[0085] Key indicator: Ethyl hexanoate content (standard range: 2.1-2.8g / L)

[0086] Defect Type:

[0087] Type A: Insufficient ethyl hexanoate (<1.9g / L) → "weak cellar aroma"

[0088] Type B: Ethyl butyrate exceeds the standard (>0.15g / L) → "unpleasant taste".

[0089] 2. Multimodal Data Association

[0090] Physical signal Human reviews Semantic association weight Electronic nose-ethyl hexanoate "Rich cellar aroma" 0.87 (Pearson correlation) Electronic tongue-bitterness value "Bitter aftertaste" 0.79

[0091] 3. Real-time monitoring of data flow

[0092] 4. Defect traceability analysis

[0093] Abnormal event: On September 5, 2023, three consecutive batches of ethyl butyrate were detected to exceed the standard.

[0094] Root cause analysis: The system called historical data and found that the pH value of the cellar mud decreased by 0.3 (consistent with the manual detection records); the model traceability module located the northwest area of the cellar number CT-17.

[0095] 5. Economic benefits

[0096] index Traditional method This system Single batch testing cost ￥1,200 ￥280 Defect missed detection rate 15.2％ 3.8％ Evaluation report generation time 2 hours Automatically generated

[0097] Example 5

[0098] 1. Test sample library

[0099] Fragrance brand Defect Type Number of samples Light fragrance Fenjiu Ethyl acetate exceeds the standard 50 Rice-flavored Guilin Three Flowers β-phenylethanol deficiency 30 Soybean flavor Shiwan Jade Ice BBQ Excessive fatty acids 20

[0100] 2. Cross-fragrance detection performance

[0101] Fragrance Accuracy Key Features Sauce-flavored 93.2％ 4-vinylguaiacol / phenylethanol ratio Special fragrance 88.7％ Slope of dynamic response curve of ethyl propionate Mixed fragrance 90.1％ Synergistic effect of ethyl hexanoate and ethyl heptanoate

[0102] 3. Comparison with traditional methods

[0103] Detection items Manual evaluation This system Fragrance classification Requires confirmation by 3 experts Automatic identification (confidence> 95%) Quality Grade The difference between the two groups was ±8.7 points Standard deviation ±2.3 points Defect location Relying on experience Provide quantitative indicators (e.g., "ethyl hexanoate gap = 23%)"

[0104] Key features of this application include: Database Uniqueness: Establishes a standardized library of poor-quality samples (including defective samples) covering 19 mainstream fragrance brands. Multimodal Fusion Architecture: Cross-modally aligns electronic sensory signals (physical layer) with human review semantics (cognitive layer). Fragrance Adaptive Mechanism: Designs differentiated feature extraction modules for different fragrance types, such as Strong, Sauce, and Light.

[0105] This application has the following beneficial effects:

[0106] Objective and accurate: Through multimodal data fusion and deep learning algorithms, the subjectivity of manual evaluation is effectively reduced and the accuracy of identification results is improved.

[0107] Efficient and convenient: It realizes the automated identification of liquor quality, greatly improving the identification efficiency and reducing the identification cost.

[0108] Comprehensive and systematic: It takes into account many factors such as the aroma, taste, style, etc. of the liquor, and can comprehensively and systematically evaluate the quality of the liquor.

[0109] Strong explainability: Through methods such as feature importance analysis, the decision-making process of the AI evaluation model can be explained, thereby improving the credibility of the model.

[0110] like Figure 2 As shown, the present application also provides an AI evaluation method for Chinese liquor based on multimodal data fusion, including: S101, the data acquisition module collects multimodal data of liquor samples; S103, the data preprocessing module preprocesses the collected multimodal data; S105, the feature fusion module fuses the multimodal data to construct a feature vector that comprehensively reflects the quality characteristics of the liquor; S107, the evaluation model learns the mapping relationship between the multimodal features of the liquor and the quality grade, and evaluates the quality of the liquor; S109, the output module outputs the liquor quality identification result.

[0111] In this application, the embodiment of the Chinese liquor AI evaluation method based on multimodal data fusion is basically similar to the embodiment of the Chinese liquor AI evaluation system based on multimodal data fusion. For relevant matters, please refer to the introduction of the embodiment of the Chinese liquor AI evaluation system based on multimodal data fusion.

[0112] The present application also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0113] The present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above method steps. The computer-readable storage medium may include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a microdrive, a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0114] The above 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 are intended to be within the scope of protection of the present invention.

Claims

1. A Chinese liquor AI evaluation system based on multimodal data fusion, characterized by: include: Data acquisition module, used to collect multimodal data of liquor samples; A data preprocessing module is used to preprocess the collected multimodal data; Feature fusion module, used to fuse multimodal data and construct feature vectors that fully reflect the quality characteristics of liquor; The evaluation model is used to learn the mapping relationship between the multimodal features of liquor and its quality grades, and evaluate the quality of liquor; Output module, used to output the liquor quality identification results.

2. The Chinese liquor AI evaluation system based on multimodal data fusion according to claim 1 is characterized in that: The multimodal data collected by the data acquisition module includes: Volatile organic compound data of liquor samples collected using an electronic nose device; Using a multi-channel electronic tongue device equipped with taste sensors, we collected the taste characteristics of liquor samples; Sommeliers evaluate liquor samples, record scores across multiple dimensions, and generate structured data. Physical and chemical index data of liquor samples obtained through laboratory testing.

3. The Chinese liquor AI evaluation system based on multimodal data fusion according to claim 1 or 2 is characterized in that: The data preprocessing module is used to: perform data cleaning, feature extraction, and data standardization on the collected multimodal data; perform data cleaning through sliding average filtering or wavelet transform methods to remove noise and outliers in the sensor data; extract time domain and frequency domain features from the data collected by the electronic nose and electronic tongue, including mean, variance, peak value, and spectral energy; and normalize data of different modes to eliminate dimensional differences.

4. The Chinese liquor AI evaluation system based on multimodal data fusion according to claim 1 or 2 is characterized in that: The feature fusion module is used to fuse the electronic nose collected data, electronic tongue collected data, manual evaluation data and physical and chemical index data into a high-dimensional feature vector using a multimodal feature fusion algorithm. The multimodal feature fusion algorithm includes feature splicing, attention mechanism or graph neural network.

5. The Chinese liquor AI evaluation system based on multimodal data fusion according to claim 1 or 2 is characterized in that: The evaluation model is built based on a deep learning algorithm, and the model architecture uses a multi-layer perceptron, a convolutional neural network, or a Transformer model; the training data includes a multimodal dataset of thousands of liquor samples; the loss function uses a mean square error or a cross-entropy loss function; the optimization algorithm uses the Adam optimizer, and the learning rate is set to 0.

001.

6. The Chinese liquor AI evaluation system based on multimodal data fusion according to claim 1 or 2 is characterized in that: The output module outputs the liquor quality identification results, including quality grade, flavor characteristic description, and quality defect analysis.

7. The Chinese liquor AI evaluation system based on multimodal data fusion according to claim 1 or 2 is characterized in that: The output module is also used to visualize the identification results, including quality grade scores, flavor profile radar charts, and quality defect analysis.

8. A Chinese liquor AI evaluation method based on multimodal data fusion, characterized by: include: The data acquisition module collects multimodal data of liquor samples; The data preprocessing module preprocesses the collected multimodal data; The feature fusion module fuses multimodal data to construct a feature vector that fully reflects the quality characteristics of liquor; The evaluation model learns the mapping relationship between the multimodal features of liquor and its quality grades to evaluate the quality of liquor; The output module outputs the liquor quality identification results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to claim 8 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to claim 8 are implemented.

Images