Smell type identification method, system and equipment for monomer perfume, and storage medium

Abstract

The invention provides a method, a system and equipment for identifying the odor type of a monomer perfume, and a storage medium, and relates to the technical field of gas detection, the method comprises the following steps: preprocessing a monomer perfume sample, obtaining a response signal of the preprocessed monomer perfume sample, and obtaining an initial response data set; performing data preprocessing on the initial response data set to obtain a sample feature vector set; a convolutional neural network model is constructed, the convolutional neural network model is trained according to the sample feature vector set and the fragrance labels of the single perfume samples, an odor and fragrance recognition model is obtained, the convolutional neural network model at least comprises a shared convolutional layer, and the shared convolutional layer is used for fusing multiple scale features; and obtaining a target feature vector of the target monomer perfume, and inputting the target feature vector into the odor type identification model for identification to obtain the odor type of the target monomer perfume. According to the method, the common characteristics of different odor types are captured by using the shared convolutional layer, so that the recognition accuracy is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of gas detection, and in particular to a method, system, device and storage medium for identifying the odor and fragrance type of a monomeric fragrance. Background Art

[0002] With the rapid development of the fragrance industry, the accurate identification of the odor and fragrance of monomeric fragrances, which are the basic raw materials for flavor blending, has become a key factor affecting the quality of flavor blending and the market competitiveness of products. At present, the industry generally adopts the traditional method of manual olfactory identification by professional perfumers for odor identification, but this method has significant limitations. First, due to individual differences in perfumers' experience accumulation and olfactory sensitivity, the judgment of the fragrance of the same monomeric fragrance is often highly subjective; secondly, perfumers are easily affected by physiological and psychological factors such as olfactory fatigue and mood swings, resulting in low detection efficiency and difficulty in meeting the requirements of modern large-scale production for detection efficiency. Although traditional instrumental analysis methods such as mass spectrometry and gas chromatography have high reliability, these technologies are difficult to directly reflect the overall characteristics of the odor, and have obvious deficiencies in real-time and portability, and cannot meet the needs of rapid and accurate identification of odor and fragrance in actual production.

[0003] In the field of olfactory detection and classification research, traditional linear methods rely primarily on manual feature extraction, which has limited adaptability and makes it difficult to effectively process complex odor data. In recent years, the rapid development of deep learning technology has brought new research opportunities to this field. Deep learning algorithms have powerful data learning and feature extraction capabilities, and can automatically learn the complex characteristics and patterns of odors from large amounts of data, thereby improving the accuracy of classification and recognition. However, existing deep learning technology research has not fully considered the correlation between multiple odor and fragrance labels, resulting in low accuracy in odor and fragrance recognition, making it difficult to meet actual production needs. This problem needs to be addressed urgently. Summary of the Invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method, system, device and storage medium for identifying the odor and fragrance of a single fragrance, which effectively solves the problem of low accuracy in identifying the odor and fragrance of a single fragrance.

[0005] In a first aspect, the present invention provides a method for identifying the odor and fragrance type of a monomeric fragrance, the method comprising:

[0006] Preprocessing the monomeric fragrance sample, obtaining a response signal of the preprocessed monomeric fragrance sample, and obtaining an initial response data set;

[0007] Performing data preprocessing on the initial response data set to obtain a sample feature vector set;

[0008] Constructing a convolutional neural network model, training the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, wherein the convolutional neural network model includes at least a shared convolution layer, and the shared convolution layer is used to fuse multiple scale features;

[0009] A target feature vector of a target monomer fragrance is obtained, and the target feature vector is input into the odor and fragrance recognition model for recognition to obtain the odor and fragrance of the target monomer fragrance.

[0010] In an optional embodiment, the shared convolutional layer includes a first branch, a second branch, a third branch, and a fourth branch, wherein:

[0011] The first branch uses deep convolution combined with maximum pooling operation to perform feature extraction to obtain first feature information;

[0012] The second branch uses point-by-point convolution combined with average pooling operation to extract features and obtain second feature information;

[0013] The third branch uses a maximum pooling operation to perform feature extraction to obtain third feature information;

[0014] The fourth branch uses an average pooling operation to perform feature extraction to obtain fourth feature information.

[0015] In an optional embodiment, performing data preprocessing on the initial response data set to obtain a sample feature vector set includes:

[0016] Normalizing the initial response data set to obtain a normalized data set;

[0017] The normalized data set is cleaned using a Euclidean distance method to obtain the sample feature vector set.

[0018] In an optional embodiment, the convolutional neural network model is trained according to the sample feature vector set to obtain an odor and fragrance recognition model, including:

[0019] Associating the sample feature vector set with the fragrance label, and dividing the associated sample feature vector set into a training set and a test set;

[0020] The convolutional neural network model is trained using a stochastic gradient descent algorithm according to the training set to obtain an initial odor and fragrance recognition model;

[0021] The initial odor and fragrance recognition model is tested using a test set to obtain the odor and fragrance recognition model.

[0022] In an optional embodiment, the pre-processing of the monomer fragrance sample comprises:

[0023] Selecting a plurality of monomer fragrance samples and placing them in sealed containers of the same volume;

[0024] The target temperature, target relative humidity and target equilibrium time are set so that the monomer fragrance sample can fully volatilize in the sealed container to obtain sample volatile gas.

[0025] In an optional embodiment, obtaining the response signal of the pre-processed monomer fragrance sample to obtain the initial response data set includes:

[0026] The sample volatile gas is collected by an electronic nose, and the response signal of the sensor array in the electronic nose is recorded to obtain the initial response data set.

[0027] In an optional embodiment, obtaining a target feature vector of a target monomer fragrance includes:

[0028] collecting a response signal of the target monomer fragrance through an electronic nose to obtain target response data;

[0029] performing normalization processing on the target response data to obtain target normalized data;

[0030] The target normalized data is cleaned using the Euclidean distance method to obtain the target feature vector.

[0031] In a second aspect, the present invention provides a system for identifying the odor and fragrance of a single fragrance, the system comprising:

[0032] A data acquisition module is used to pre-process the monomer fragrance sample, obtain the response signal of the monomer fragrance sample after pre-processing, and obtain an initial response data set;

[0033] A data processing module, configured to perform data preprocessing on the initial response data set to obtain a sample feature vector set;

[0034] A model training module is used to construct a convolutional neural network model, and train the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, wherein the convolutional neural network model includes at least a shared convolution layer, and the shared convolution layer is used to fuse multiple scale features;

[0035] The model recognition module is used to obtain the target feature vector of the target monomer fragrance, input the target feature vector into the odor and fragrance recognition model for recognition, and obtain the odor and fragrance of the target monomer fragrance.

[0036] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for identifying the odor and fragrance of a monomeric fragrance as described in the first aspect of the present invention.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying the odor and fragrance type of a monomeric fragrance as described in the first aspect of the present invention.

[0038] The odor and fragrance identification method, system, device and storage medium of the present invention, by considering the correlation of odor and fragrance labels, using a shared convolutional layer to capture the common features between different odor and fragrance types, combined with the powerful deep feature extraction capability of the convolutional neural network, can more accurately identify the subtle odor differences of individual spices and effectively improve the recognition accuracy. At the same time, the shared convolutional layer reduces model parameters, reduces computational complexity, accelerates training convergence, and the model structure is suitable for parallel computing, which can significantly improve work efficiency. Based on the objective data collected by the electronic nose and the odor and fragrance identification model, human interference is avoided, subjective bias is eliminated, and the results are guaranteed to be objective and repeatable. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 This is a first schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention;

[0041] Figure 2 This is a second schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention;

[0042] Figure 3 This is a third schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention;

[0043] Figure 4 is a schematic diagram of the structure of a shared convolutional layer provided by an embodiment of the present invention;

[0044] Figure 5 This is a fourth schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention;

[0045] Figure 6This is a fifth schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention;

[0046] Figure 7 Schematic diagram of the structure of the odor and fragrance identification system for a single fragrance provided by an embodiment of the present invention;

[0047] Figure 8 It is a structural diagram of an electronic device provided by an embodiment of the present invention.

[0048] Description of main component symbols:

[0049] 700. Smell and fragrance recognition system for a single fragrance; 710. Data acquisition module; 720. Data processing module; 730. Model training module; 740. Model recognition module; 800. Electronic device; 810. Processor; 820. Communication interface; 830. Memory; 840. Communication bus. DETAILED DESCRIPTION

[0050] To make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be further clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. It should be noted that the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present invention. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0053] The accurate identification of odor and fragrance has become a key factor affecting the quality of flavor blending and the market competitiveness of products. The traditional method of manual olfactory identification has low detection efficiency and is difficult to meet the requirements of modern large-scale production for detection efficiency. Traditional instrumental analysis methods such as mass spectrometry and gas chromatography have high reliability, but these technologies are difficult to directly reflect the overall characteristics of the odor, and have obvious deficiencies in real-time and portability. Deep learning algorithms have powerful data learning and feature extraction capabilities, and can automatically learn the complex characteristics and patterns of odors from large amounts of data, thereby improving the accuracy of classification and recognition. However, existing deep learning technology research has not fully considered the correlation between multiple odor and fragrance labels, resulting in low accuracy in odor and fragrance recognition, which is difficult to meet actual production needs. This problem needs to be solved urgently.

[0054] Example 1

[0055] The embodiment of the present invention provides a method for identifying the odor and fragrance type of a single fragrance, which effectively solves the problem of low accuracy in identifying the odor and fragrance type of a single fragrance. Figure 1 FIG. 1 is a first schematic diagram of a method for identifying the odor and fragrance of a monomeric fragrance according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0056] S100 , preprocessing the monomeric fragrance sample, obtaining the response signal of the preprocessed monomeric fragrance sample, and obtaining an initial response data set.

[0057] In this embodiment of the present invention, multiple individual fragrance samples with known odors and aromas are selected as samples. Each individual fragrance sample with a known aroma is labeled with an aroma tag. Each individual fragrance sample can include one or more aroma tags. The individual fragrance samples are placed in a sealed container and equilibrated under constant temperature and humidity conditions for a certain period of time to allow the individual fragrances to fully evaporate. Odor data of the individual fragrances is then collected using a device such as an electronic nose to obtain an initial response dataset. Figure 2 FIG. 2 is a second schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention. Figure 2 As shown in FIG, obtaining the initial response data set specifically includes the following steps:

[0058] S110. Select a plurality of monomer fragrance samples and place them in sealed containers of the same volume.

[0059] In this embodiment of the present invention, liquid fragrances are used as the monomeric fragrance samples. The more different monomeric fragrance samples there are, the more accurate the recognition model obtained through training. For each monomeric fragrance sample, a preset volume of the monomeric fragrance sample is precisely measured using a pipette. This preset volume can be set to 5uL, 10uL, or 20uL, and then placed in a 50mL beaker for sealed storage. This effectively isolates the fragrance's odor from external environmental factors such as air flow, light, and other odors.

[0060] S120 , setting a target temperature, a target relative humidity, and a target equilibrium time so that the monomer fragrance sample can fully volatilize in the sealed container to obtain sample volatile gas.

[0061] In the embodiments of the present invention, constant temperature and relative humidity are maintained. The volatilization rate and stability of fragrance molecules vary under different environmental conditions. A stable temperature and humidity environment is crucial for the volatilization and equilibrium of sample volatile gases. Temperature affects the rate of molecular movement, and relative humidity may affect the volatility characteristics of monomeric fragrance samples. Appropriate temperature and humidity ensure that the distribution of sample volatile gases between the headspace and the sample matrix reaches a stable state, minimizing experimental error and ensuring comparability between monomeric fragrance samples. Optionally, the target temperature is set to 25°C-27°C, and the target relative humidity is set to 75%-85%.

[0062] Allowing the monomeric fragrance sample to equilibrate under constant conditions for a certain period of time allows the odor molecules within the fragrance to fully diffuse and stabilize, ensuring that the collected odor information truly reflects the odor characteristics of the monomeric fragrance itself and avoiding detection bias caused by uneven odor distribution within the fragrance. Optionally, the target equilibration time is set to 20 minutes. During the target equilibration time, the monomeric fragrance sample gradually evaporates within the confined space, forming a dynamic equilibrium between the headspace region and the monomeric fragrance sample, causing the sample volatile gas concentration in the headspace region to gradually stabilize. After 20 minutes of equilibration, it is considered that the sample volatile gas has reached a relative equilibrium state between the headspace region and the monomeric fragrance sample.

[0063] S130 , collecting sample volatile gas through the electronic nose, recording the response signal of the sensor array in the electronic nose, and obtaining an initial response data set.

[0064] In an embodiment of the present invention, an electronic nose is used to detect pretreated monomer fragrance samples. The electronic nose is equipped with a sensor array consisting of 10 different types of gas sensors. The sensor array collects volatile gases from the samples and records the response signals of each sensor in real time to form an initial response data set.

[0065] Optionally, the sampling time of the electronic nose is set to 120 seconds, the sampling interval is set to 10 seconds, and the washing time is set to 120 seconds. Thus, the response values ​​of the individual fragrance samples collected by the electronic nose form a 10×120-dimensional sample matrix, where 120 represents the sampling time of each sensor in the data collection phase, and 10 represents the number of sensors in the electronic nose. The sample matrix is ​​represented by X, and the i-th sample matrix is ​​represented as follows:

[0066]

[0067] In the above formula, X i represents the sample matrix of the i-th monomer flavor sample, a im,n It represents the response value of the i-th monomer fragrance sample collected at the n-th second in the m-th sensor.

[0068] The sample volatile gases of all monomeric fragrance samples are collected by an electronic nose to obtain sample matrices of all monomeric fragrance samples, and all sample matrices are integrated to obtain an initial response data set.

[0069] S200: Preprocess the initial response data set to obtain a sample feature vector set.

[0070] In an embodiment of the present invention, the collected initial response data set is normalized and cleaned, the normalization method is used to extract features from the initial response data set, the Euclidean distance method is selected for data cleaning, and the remaining sample data is used as a feature vector. Figure 3 FIG3 is a third schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention. Figure 3 As shown in Figure 2, the specific steps of data preprocessing are as follows:

[0071] S210 , normalizing the initial response data set to obtain a normalized data set.

[0072] In the embodiment of the present invention, Z-score normalization is used to normalize each sample matrix X i Perform feature extraction to obtain a normalized data set. The response values ​​of each sensor collected by the electronic nose may have different dimensions and value ranges. Z-score normalization effectively eliminates the impact of differences in the scale of the original data between different features by converting feature data of different dimensions and value ranges into a standard normal distribution with a mean of 0 and a standard deviation of 1. The normalized data range is around -1 to 1, which is more in line with the common assumptions of machine learning models on input data. This standardization process allows the model parameters to be updated within a reasonable range at the beginning of training, reducing the oscillation in the gradient descent process caused by different data scales, thereby accelerating the convergence of the model and improving training efficiency. The Z-score normalization formula is as follows:

[0073]

[0074] In the above formula, x ′ Represents the normalized feature data value, x represents the sample matrix X i The original response value in x max Represents the maximum value of the 120 original response values, x mim Represents the minimum value of the 120 original response values, x mean Represents the average of 120 raw response values.

[0075] Through Z-score normalization, not only the data range is unified, but also the distribution characteristics of the original data are retained. This allows subsequent models to focus more on learning the characteristic pattern differences between different odor types, rather than being disturbed by the absolute values ​​of the data. For example, when distinguishing between floral and fruity single fragrances, the model can more accurately capture the relative change pattern of each sensor response rather than the absolute intensity difference. At the same time, the normalized feature data is more suitable for model structures such as convolutional neural networks based on shared convolutional layers, which helps the model extract deep features and consider the correlation between odor type labels, thereby improving the accuracy of odor type prediction for unknown samples.

[0076] S220. Use the Euclidean distance method to clean the normalized data set and obtain a sample feature vector set.

[0077] Data cleaning is to compare and eliminate duplicate or invalid samples. In this embodiment of the present invention, the Euclidean distance method is used to calculate the similarity of samples in the normalized data set. The Euclidean distance is actually the linear distance between two points in two-dimensional space. Its calculation formula is as follows:

[0078]

[0079] In the above formula, d(X,Y) represents the Euclidean distance between sample X and sample Y, n represents the number of feature data in the sample, and x i Represents the feature data in sample X, y i Represents the feature data in sample Y.

[0080] Euclidean distance is an effective measure of the similarity between individual fragrance samples collected by the electronic nose. The smaller the Euclidean distance, the higher the similarity of the individual fragrance samples. By calculating the Euclidean distance between each individual fragrance sample, duplicate or highly similar individual fragrance samples can be accurately identified. These redundant samples often carry similar odor characteristic information. Retaining them will not only fail to provide new and useful knowledge to the model, but may also interfere with the model's learning of different odor types. Removing them can greatly improve the quality of the sample feature vector set. At the same time, after removing similar samples, the feature differences between individual fragrance samples of different odor types become more obvious. This allows the key features of different fragrance types to be captured more accurately during the subsequent feature extraction and analysis process, helping to build more effective feature representations and improve the accuracy of odor and fragrance perception.

[0081] S300, constructing a convolutional neural network model, training the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, wherein the convolutional neural network model includes at least a shared convolution layer, and the shared convolution layer is used to fuse multiple scale features.

[0082] In an embodiment of the present invention, a convolutional neural network model is used as a pre-trained model to identify the odor and aroma of individual fragrances. This convolutional neural network model incorporates a shared convolutional layer before the first convolution operation. This shared convolutional layer employs matrix addition within the convolutional layer, effectively summing multiple feature extraction methods. Figure 4 Schematic diagram of the structure of the shared convolutional layer provided by the embodiment of the present invention, such as Figure 4 As shown, the shared convolutional layer includes a first branch, a second branch, a third branch, and a fourth branch.

[0083] The first branch uses depthwise convolution combined with maximum pooling for feature extraction. By using 3×3 convolution kernels, deep convolution deeply mines the local detail features of the sample feature vector set, strengthening the capture of key features, and then outputting the specified size through maximum pooling to obtain the first feature information. The second branch uses point-by-point convolution combined with average pooling for feature extraction. Point-by-point convolution performs dimensionality reduction and reduces the matrix depth, thereby alleviating overfitting pressure. Then, average pooling is used to obtain useful feature information and obtain the second feature information. The third branch uses maximum pooling for feature extraction. Maximum pooling is used to capture the strongest partial feature information in the sample feature vector set to obtain the third feature information. The fourth branch uses average pooling for feature extraction. Average pooling is used to capture the comprehensive feature information of the sample feature vector set to obtain the fourth feature information.

[0084] Due to the adjacent reinforcement characteristics between the flavor labels, the design concats the four branches in the form of a parallel multi-scale structure, thereby introducing information flow between the multiple flavor labels, making the convolutional neural network model pay more attention to the dependency between different odor flavors, capturing the comprehensive information of the odor flavor characteristics, learning more effective features, and thus improving the accuracy of model detection and classification effect. Specifically, the convolution kernels in the shared convolution layer share weights on the data of different flavor labels, which means that no matter which flavor label, the same convolution kernel performs convolution operations on the input data in the same way. For example, in the feature vector of a single fragrance sample, the same convolution kernel will identify the same or similar local feature patterns that may exist in data with different flavor labels. Therefore, when the data of different flavor labels pass through the shared convolution layer, they will be affected by the same convolution kernel, thereby introducing information flow in the feature extraction process, which provides a basis for information interaction between different flavor labels.

[0085] Furthermore, the shared convolutional layer maps the input data of different aroma labels into the same feature space. Through the convolution operation, the raw aroma data is converted into a set of feature representations. These feature representations summarize the data of different aroma labels, integrating the information of different aroma labels into this feature space. For example, after passing the shared convolutional layer through aroma data of different odor types, the molecular structure characteristics and odor intensity characteristics may be represented in the same feature space, thereby enabling information flow and fusion between multiple aroma labels.

[0086] The four branches are fused to form a shared convolutional layer, capturing comprehensive information from the sample feature vector set. Through depthwise and pointwise convolution, the convolution kernel and dilation rate are adjusted to achieve a larger receptive field. The design of depthwise and pointwise convolution in the shared convolutional layer significantly reduces the number of neurons. Both maximum pooling and average pooling also maximize the capture of feature information from the sample feature vector set. This shared convolutional layer integrates features at different scales, effectively avoiding the incomplete representation of features at a single scale. It captures local features within the receptive fields of different scales, thereby mining more comprehensive feature information.

[0087] Figure 5 FIG4 is a fourth schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention. Figure 5 As shown in FIG, the training of the odor and fragrance recognition model specifically includes the following steps:

[0088] S310, associating the sample feature vector set with the fragrance label, and dividing the associated sample feature vector set into a training set and a test set;

[0089] In an embodiment of the present invention, the feature vector of each individual fragrance sample in the sample feature vector set is associated with the corresponding fragrance label, and the associated sample feature vector set is divided into a training set and a test set, wherein the training set accounts for 75% and the test set accounts for 25%.

[0090] S320: training the convolutional neural network model using a stochastic gradient descent algorithm based on the training set to obtain an initial odor and fragrance recognition model.

[0091] In an embodiment of the present invention, to enhance the training process of the convolutional neural network model, the model's hyperparameters are appropriately set. Optionally, the initial parameter learning rate is set to 0.01, which enables rapid updates of model parameters in the early stages of training and accelerates model convergence. Momentum decay is set to 0.85, which helps maintain a certain level of inertia during training, making parameter updates more stable and avoiding oscillations around local optimal solutions. Weight decay is set to 0.001 to prevent model parameters from being too large, effectively alleviating overfitting problems and improving the model's generalization ability.

[0092] Optionally, a stochastic gradient descent algorithm is used to train the convolutional neural network model. This algorithm uses only one sample or a small batch of samples to update the model parameters at each iteration. This algorithm requires little computation and is fast to train, making it suitable for processing large sets of sample feature vectors of individual fragrances. Furthermore, the algorithm can escape from some local optimal solutions, helping to find the global optimal solution. A weighted cross-entropy loss function is used as the loss function. This weighted cross-entropy loss function takes into account the importance of different odor types or the imbalance of sample distribution. By setting different weights for different odor types, the model pays more attention to those important odor types or those with fewer samples during training, thereby improving the model's classification accuracy for each odor type.

[0093] S330: Using the test set to test the initial odor and fragrance recognition model to obtain the odor and fragrance recognition model.

[0094] In an embodiment of the present invention, a test set is used to evaluate the initial odor and fragrance recognition model. Optionally, by calculating indicators such as the accuracy, recall rate, F1 value, etc. of the initial odor and fragrance recognition model on the test set, a comprehensive and objective understanding of the initial odor and fragrance recognition model's ability to classify the odors and fragrances of individual spices can be obtained, providing a reliable basis for the optimization and improvement of the model, and obtaining the odor and fragrance recognition model.

[0095] S400: Obtain a target feature vector of a target monomer fragrance, input the target feature vector into an odor and fragrance type recognition model for recognition, and obtain the odor and fragrance type of the target monomer fragrance.

[0096] Figure 6 FIG5 is a fifth schematic diagram of the process of the method for identifying the odor and fragrance type of a monomeric fragrance provided by an embodiment of the present invention. Figure 6As shown in FIG, the acquisition of the target feature vector specifically includes the following steps:

[0097] S410 , collecting a response signal of a target monomer fragrance through an electronic nose to obtain target response data.

[0098] In the embodiment of the present invention, an electronic nose is used to detect a target monomer fragrance, volatile gases of the target monomer fragrance are collected through a sensor array, and response signals of each sensor are recorded in real time to obtain target response data.

[0099] S420: Perform normalization processing on the target response data to obtain target normalized data.

[0100] In an embodiment of the present invention, Z-score normalization is used to perform feature extraction on target response data to obtain target normalized data.

[0101] S430. Use the Euclidean distance method to clean the target normalized data to obtain a target feature vector.

[0102] In an embodiment of the present invention, a Euclidean distance method is used to calculate the similarity of data in the target normalized data, and data cleaning is performed based on the similarity to obtain a target feature vector.

[0103] The target feature vector of the unknown target monomer vector is input into the odor and fragrance recognition model for recognition, so as to obtain the odor and fragrance of the target monomer fragrance.

[0104] The method, system, device, and storage medium for identifying the odor and fragrance types of individual fragrances provided by the embodiments of the present invention consider the correlation between odor and fragrance labels and utilize shared convolutional layers to capture common features between different odor and fragrance types. Combined with the powerful deep feature extraction capabilities of convolutional neural networks, these methods can more accurately identify subtle odor differences among individual fragrances, effectively improving recognition accuracy. Furthermore, the shared convolutional layers reduce model parameters, lower computational complexity, and accelerate training convergence. Furthermore, the model structure is suitable for parallel computing, significantly improving work efficiency.

[0105] Example 2

[0106] Based on the same technical concept, the embodiment of the present invention provides a system for identifying the odor and fragrance of a single fragrance. Figure 7 FIG. 1 is a schematic diagram of the structure of the odor and fragrance identification system for a monomeric fragrance provided by an embodiment of the present invention. Figure 7 As shown, the odor and fragrance recognition system 700 of the monomer fragrance includes:

[0107] The data acquisition module 710 is used to pre-process the individual fragrance samples, obtain the response signals of the pre-processed individual fragrance samples, and obtain an initial response data set.

[0108] The data processing module 720 is used to perform data preprocessing on the initial response data set to obtain a sample feature vector set.

[0109] The model training module 730 is used to construct a convolutional neural network model, train the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample, and obtain an odor and fragrance recognition model. The convolutional neural network model includes at least a shared convolution layer, which is used to fuse multiple scale features.

[0110] The model recognition module 740 is used to obtain a target feature vector of a target monomer fragrance, input the target feature vector into an odor and fragrance recognition model for recognition, and obtain the odor and fragrance of the target monomer fragrance.

[0111] The odor and fragrance recognition system for a single fragrance provided by an embodiment of the present invention is based on objective data collected by an electronic nose and an odor and fragrance recognition model, thereby avoiding interference from human factors, eliminating subjective bias, and ensuring that the results are objective and repeatable.

[0112] It can be understood that the implementation method of the odor and fragrance identification method of the monomer fragrance described in the above embodiment 1 is also applicable to this embodiment and can achieve the same technical effect, so it will not be repeated here.

[0113] Example 3

[0114] Based on the same concept, an embodiment of the present invention further provides an electronic device, Figure 8 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention, such as Figure 8 As shown, the electronic device 800 may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. The processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the steps of the method for identifying the odor and fragrance type of a single fragrance as described in the above embodiments. For example, the steps include:

[0115] S100 , preprocessing the monomeric fragrance sample, obtaining the response signal of the preprocessed monomeric fragrance sample, and obtaining an initial response data set.

[0116] S200, performing data preprocessing on the initial response data set to obtain a sample feature vector set;

[0117] S300, constructing a convolutional neural network model, training the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, wherein the convolutional neural network model includes at least a shared convolution layer, and the shared convolution layer is used to fuse multiple scale features;

[0118] S400: Obtain a target feature vector of a target monomer fragrance, input the target feature vector into an odor and fragrance type recognition model for recognition, and obtain the odor and fragrance type of the target monomer fragrance.

[0119] The processor 810 may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0120] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0121] The memory 830 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely located relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0122] Example 4

[0123] Based on the same concept, an embodiment of the present invention further provides a computer-readable storage medium storing a computer program. The computer program includes at least one code segment that can be executed by a main control device to control the main control device to implement the steps of the method for identifying the odor and fragrance type of a single fragrance as described in the above embodiments. For example, the steps include:

[0124] S100 , preprocessing the monomeric fragrance sample, obtaining the response signal of the preprocessed monomeric fragrance sample, and obtaining an initial response data set.

[0125] S200, performing data preprocessing on the initial response data set to obtain a sample feature vector set;

[0126] S300, constructing a convolutional neural network model, training the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, wherein the convolutional neural network model includes at least a shared convolution layer, and the shared convolution layer is used to fuse multiple scale features;

[0127] S400: Obtain a target feature vector of a target monomer fragrance, input the target feature vector into an odor and fragrance type recognition model for recognition, and obtain the odor and fragrance type of the target monomer fragrance.

[0128] Based on the same technical concept, an embodiment of the present invention further provides a computer program, which, when executed by a main control device, is used to implement the above method embodiment.

[0129] The computer program may be stored in whole or in part on a computer-readable storage medium packaged with the processor, or may be stored in whole or in part on a memory not packaged with the processor.

[0130] Based on the same technical concept, an embodiment of the present invention further provides a processor for implementing the above method embodiment. The above processor may be a chip.

[0131] In summary, the odor and fragrance identification method, system, device and storage medium of the monomeric fragrance provided by the present invention, by considering the correlation of odor and fragrance labels, using a shared convolutional layer to capture the common features between different odor and fragrance types, combined with the powerful deep feature extraction capability of the convolutional neural network, can more accurately identify the subtle odor differences of monomeric fragrances and effectively improve the recognition accuracy. At the same time, the shared convolutional layer is used to reduce model parameters, reduce computational complexity, accelerate training convergence, and the model structure is suitable for parallel computing, which can significantly improve work efficiency. Based on the objective data collected by the electronic nose and the odor and fragrance identification model, interference from human factors is avoided, subjective bias is eliminated, and the results are guaranteed to be objective and repeatable.

[0132] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0133] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying the odor and fragrance of a monomeric fragrance, characterized in that: The method comprises: Preprocessing the monomeric fragrance sample, obtaining a response signal of the preprocessed monomeric fragrance sample, and obtaining an initial response data set; Performing data preprocessing on the initial response data set to obtain a sample feature vector set; Constructing a convolutional neural network model, training the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, wherein the convolutional neural network model includes at least a shared convolution layer, and the shared convolution layer is used to fuse multiple scale features; A target feature vector of a target monomer fragrance is obtained, and the target feature vector is input into the odor and fragrance recognition model for recognition to obtain the odor and fragrance of the target monomer fragrance.

2. The method for identifying the odor and fragrance type of a monomeric fragrance according to claim 1, wherein: The shared convolutional layer includes a first branch, a second branch, a third branch, and a fourth branch, wherein: The first branch uses deep convolution combined with maximum pooling operation to perform feature extraction to obtain first feature information; The second branch uses point-by-point convolution combined with average pooling operation to extract features and obtain second feature information; The third branch uses a maximum pooling operation to perform feature extraction to obtain third feature information; The fourth branch uses an average pooling operation to perform feature extraction to obtain fourth feature information.

3. The method for identifying the odor and fragrance type of a monomeric fragrance according to claim 1, wherein: The performing data preprocessing on the initial response data set to obtain a sample feature vector set includes: Normalizing the initial response data set to obtain a normalized data set; The normalized data set is cleaned using a Euclidean distance method to obtain the sample feature vector set.

4. The method for identifying the odor and fragrance type of a monomeric fragrance according to claim 3, wherein: The convolutional neural network model is trained according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, including: Associating the sample feature vector set with the fragrance label, and dividing the associated sample feature vector set into a training set and a test set; The convolutional neural network model is trained using a stochastic gradient descent algorithm according to the training set to obtain an initial odor and fragrance recognition model; The initial odor and fragrance recognition model is tested using a test set to obtain the odor and fragrance recognition model.

5. The method for identifying the odor and fragrance type of a monomeric fragrance according to claim 1, wherein: The pre-processing of the monomer fragrance sample comprises: Selecting a plurality of monomer fragrance samples and placing them in sealed containers of the same volume; The target temperature, target relative humidity and target equilibrium time are set so that the monomer fragrance sample can fully volatilize in the sealed container to obtain sample volatile gas.

6. The method for identifying the odor and fragrance type of a monomeric fragrance according to claim 5, wherein: The step of obtaining the response signal of the pre-processed monomer fragrance sample to obtain an initial response data set includes: The sample volatile gas is collected by an electronic nose, and the response signal of the sensor array in the electronic nose is recorded to obtain the initial response data set.

7. The method for identifying the odor and fragrance type of a monomeric fragrance according to claim 1, wherein: The step of obtaining a target feature vector of a target monomer fragrance includes: collecting a response signal of the target monomer fragrance through an electronic nose to obtain target response data; performing normalization processing on the target response data to obtain target normalized data; The target normalized data is cleaned using the Euclidean distance method to obtain the target feature vector.

8. A system for identifying the odor and fragrance of a single fragrance, characterized in that: The system comprises: A data acquisition module is used to pre-process the monomer fragrance sample, obtain the response signal of the monomer fragrance sample after pre-processing, and obtain an initial response data set; A data processing module, configured to perform data preprocessing on the initial response data set to obtain a sample feature vector set; A model training module is used to construct a convolutional neural network model, and train the convolutional neural network model according to the sample feature vector set and the fragrance label of the monomer fragrance sample to obtain an odor and fragrance recognition model, wherein the convolutional neural network model includes at least a shared convolution layer, and the shared convolution layer is used to fuse multiple scale features; The model recognition module is used to obtain the target feature vector of the target monomer fragrance, input the target feature vector into the odor and fragrance recognition model for recognition, and obtain the odor and fragrance of the target monomer fragrance.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor executes the computer program to implement the method for identifying the odor and fragrance type of a monomeric fragrance as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for identifying the odor and fragrance type of a monomeric fragrance according to any one of claims 1 to 7 is implemented.

Images