Smell classification and recognition method, device and equipment and storage medium

Abstract

The invention relates to an odor classification and recognition method, device and equipment and a storage medium, and the method comprises the following steps: carrying out the odor feature extraction of preprocessed data, and obtaining an odor extraction feature vector; constructing a feature matrix based on the odor extraction feature vector, and inputting the feature matrix into a preset classification algorithm for odor classification to obtain a primary odor type and a confidence coefficient corresponding to the primary odor type; if the confidence is within a preset confidence range, taking the primary odor type as the odor type of the target odor sample; and if the confidence is not in a preset confidence range, performing clustering analysis on the target odor sample to obtain a secondary odor type, and taking the secondary odor type as the odor type of the target odor sample, the technical problem that the odor type cannot be effectively identified depending on artificial olfaction or complex chemical analysis equipment is solved.

Description

Technical Field

[0001] The present invention relates to the field of odor technology, and in particular to an odor classification and identification method, device, equipment and storage medium. Background Art

[0002] In modern society, odor classification and recognition technology has been widely used in environmental monitoring, food safety, medicine and health, and other fields. However, traditional odor classification and recognition methods rely on artificial olfaction or complex chemical analysis equipment, which cannot meet the requirements of real-time, high efficiency and high precision. In recent years, with the advancement of sensor technology and data processing technology, odor classification and recognition systems based on chemical sensors have gradually become a research hotspot. These systems can capture odor data and realize automatic odor classification through algorithmic analysis, but they still face problems such as complex odor data processing, limited accuracy of classification algorithms, and imperfect confidence processing mechanisms. Therefore, developing an effective odor classification and recognition method to improve the accuracy and reliability of odor classification is a challenge that needs to be urgently addressed in the current field of odor classification and recognition. Summary of the Invention

[0003] The main purpose of the present invention is to provide an odor classification and identification method, device, equipment and storage medium, which solves the technical problem that artificial olfaction or complex chemical analysis equipment cannot effectively identify the type of odor.

[0004] To achieve the above object, the present invention provides a method for classifying and identifying odors, comprising the following steps: The target odor sample is collected by a preset chemical sensor to obtain collected odor data; and the collected odor data is preprocessed to obtain preprocessed data; performing odor feature extraction on the preprocessed data to obtain an odor extraction feature vector; Constructing a feature matrix based on the odor extraction feature vector, and inputting the feature matrix into a preset classification algorithm to perform odor classification, thereby obtaining an odor type and a confidence level corresponding to the odor type; If the confidence level is within a preset confidence level range, the primary odor type is used as the odor type of the target odor sample; If the confidence level is not within a preset confidence level range, cluster analysis is performed on the target odor sample to obtain a secondary odor type, and the secondary odor type is used as the odor type of the target odor sample.

[0005] Furthermore, the preset chemical sensor includes a preset multi-channel metal oxide semiconductor sensor array, and the target odor sample is collected by the preset chemical sensor to obtain collected odor data; and the collected odor data is preprocessed to obtain preprocessed data, including: Dynamically sampling the target odor sample through the multi-channel metal oxide semiconductor sensor array to obtain collected odor data; wherein the collected odor data includes odor intensity information, odor component concentration information and odor molecular structure information; Performing wavelet decomposition on the collected odor data using wavelet transform technology to obtain multi-scale wavelet data; Denoising is performed on the multi-scale wavelet data to obtain denoised data, and the denoised data is used as preprocessed data.

[0006] Furthermore, odor feature extraction is performed on the pre-processed data to obtain an odor extraction feature vector, including: performing baseline drift correction on the preprocessed data to obtain baseline drift corrected data; performing high-dimensional feature encoding on the baseline drift correction data to obtain high-dimensional feature encoded data; Performing local feature extraction on the high-dimensional feature-encoded data to obtain local feature data; wherein the local feature data includes concentration changes of odor chemical components, odor reaction time constants, and spatial distribution of odor molecules; Performing structured feature analysis on the local feature data to obtain structured feature data; wherein the structured feature data includes topological structural features of the odor molecules, chemical bond type distribution of the odor molecules, composition ratio of functional groups of the odor molecules, and molecular weight distribution features of the odor molecules; Performing data enhancement on the structured feature data to obtain odor enhancement data; Performing dimensionality reduction on the odor enhancement data to obtain dimensionality reduction features; The dimensionality reduction features are constructed into a time series using a preset long-short memory algorithm to obtain a change curve of the odor at different time points; wherein the change curve includes a decay curve of the odor intensity over time, a curve of the change of the odor component ratio over time, and a curve of the diffusion rate of the odor molecules over time; An odor extraction feature vector is constructed based on the change curve.

[0007] Furthermore, the odor extraction feature vector includes an odor attenuation vector, an odor component ratio vector, and an odor molecule diffusion rate vector; constructing a feature matrix based on the odor extraction feature vector, and inputting the feature matrix into a preset classification algorithm to perform odor classification, thereby obtaining a primary odor type and a confidence level corresponding to the primary odor type, including: Arrange the odor attenuation vector, the odor component ratio vector, and the odor molecule diffusion rate vector from left to right or from top to bottom to obtain a feature matrix; Based on a preset classification algorithm, the feature matrix is ​​morphologically encoded to obtain a feature matrix; Performing molecular structure modeling on the feature matrix to obtain a molecular graph feature matrix; Classifying the molecular graph feature matrix to obtain classification labels corresponding to the molecular graph feature matrix; Calling a preset odor database and mapping the classification label with a storage label in the preset odor database to obtain a primary odor type corresponding to the classification label; wherein the storage label in the preset odor database has a corresponding primary odor type; A confidence calculation is performed on the primary odor type based on the molecular graph feature matrix to obtain a confidence level corresponding to the primary odor type.

[0008] Furthermore, cluster analysis is performed on the target odor samples to obtain secondary odor types, including: Performing tensor decomposition on the collected odor data of the target odor sample to obtain a tensor decomposition feature representation; Performing frequency domain analysis on the tensor decomposition feature representation using Fourier transform technology to obtain a spectrum feature distribution; Constructing an odor molecule graph based on the spectral feature distribution; Performing topological space mapping on the odor molecule graph to obtain an odor topological structure; wherein the odor topological structure is the spatial relationship and chemical bond connection between odor molecules; Performing eigenvalue decomposition on the odor topological structure based on a preset spectral graph theory technology to obtain odor spectrum features; Performing a persistence coherence analysis on the odor spectrum characteristics to obtain an odor persistence map; Sampling the odor persistence map using discrete point sets to obtain persistence barcodes, and quantifying the topological similarity between the persistence barcodes and persistence barcodes in a preset odor database to obtain an odor topological distance matrix; A hierarchical clustering analysis is performed based on the odor topological distance matrix to obtain secondary odor types.

[0009] Furthermore, the odor persistence map is sampled using discrete point sets to obtain a persistence barcode, including: Using Morse-Smale complex decomposition technology to extract critical points from the odor persistence graph to obtain a critical point set; wherein the critical point set includes minimum points, saddle points and maximum points; Extracting topological features from the critical point set to obtain a persistent homology generator; Performing multi-scale topological structure reconstruction on the persistent homology generator to obtain a multi-scale simplicial complex; Extracting topological invariants from the multi-scale simplicial complex to obtain a persistent Betti number sequence; A barcode is constructed based on the persistent Betti number sequence to obtain a persistent barcode.

[0010] Furthermore, topological invariants are extracted from the multi-scale simplicial complex to obtain a persistent Betti number sequence, including: performing discrete decomposition processing on the multi-scale simplicial complex using a discrete Morse function to obtain a discrete gradient vector field; Performing homology group calculation on the discrete gradient vector field to obtain persistent homology groups in each dimension, wherein the persistent homology groups in each dimension reflect connectivity information of the topological characteristics of the odor in different dimensions; Performing spectral sequence analysis on the persistent homology groups of each dimension to obtain a persistent spectral sequence; Extracting topological invariants from the persistent spectral sequence to obtain a DeRham cohomology group; Perform characteristic calculation on the DeRham cohomology group to obtain the twisted Betti number; The persistent random complex method is used to perform statistical analysis on the distorted Betti numbers to obtain a persistent Betti number sequence.

[0011] The present invention also provides an odor classification and identification device, comprising: The acquisition module is used to collect the target odor sample through a preset chemical sensor to obtain collected odor data; and pre-process the collected odor data to obtain pre-processed data; an extraction module, configured to extract odor features from the preprocessed data to obtain an odor extraction feature vector; A construction module, configured to construct a feature matrix based on the odor extraction feature vector, and input the feature matrix into a preset classification algorithm to perform odor classification, thereby obtaining an odor type and a confidence level corresponding to the odor type; a judgment module, configured to use the primary odor type as the odor type of the target odor sample if the confidence level is within a preset confidence level range; An analysis module is configured to perform cluster analysis on the target odor sample to obtain a secondary odor type if the confidence level is not within a preset confidence level range, and use the secondary odor type as the odor type of the target odor sample.

[0012] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0014] The odor classification and identification method provided by the present invention includes the following steps: collecting a target odor sample using a preset chemical sensor to obtain collected odor data; preprocessing the collected odor data to obtain preprocessed data; extracting odor features from the preprocessed data to obtain an extracted odor feature vector; constructing a feature matrix based on the extracted odor feature vector, and inputting the feature matrix into a preset classification algorithm to perform odor classification to obtain a primary odor type and a confidence level corresponding to the primary odor type; if the confidence level is within a preset confidence range, determining the primary odor type as the odor type of the target odor sample; if the confidence level is not within the preset confidence range, performing cluster analysis on the target odor sample to obtain a secondary odor type, and determining the secondary odor type as the odor type of the target odor sample. The above-mentioned technical solution solves the technical problem that odor types cannot be effectively identified by relying on artificial olfaction or complex chemical analysis equipment. By inputting the feature matrix into a preset classification algorithm and performing cluster analysis, complex odor data can be systematically processed, and the odor type classification results are combined with the confidence level, thereby optimizing the classification process of odor classification and identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of the steps of a method for classifying and identifying odors according to an embodiment of the present invention; Figure 2 This is a structural block diagram of an odor classification and identification device according to an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0016] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] like Figure 1 As shown, Figure 1 This is a schematic diagram of the steps of the odor classification and identification method according to one embodiment of the present invention; An embodiment of the present invention provides a method for classifying and identifying odors, comprising the following steps: Step S1: collecting a target odor sample through a preset chemical sensor to obtain collected odor data; and preprocessing the collected odor data to obtain preprocessed data.

[0019] Specifically, in this step, the target odor sample is first collected using pre-set chemical sensors. This requires a set of chemical sensors that are sensitive to specific odor molecules. These sensors can be based on sensing elements made of metal oxides, conductive polymers, or other materials. They react with odor molecules to generate electrical signals. These electrical signals represent the presence and concentration of the odor molecules, which are then converted into odor data through further collection. When these sensors come into contact with the target odor sample, the signal information collected is typically multi-dimensional, including interference from the different components and concentrations of the odor sample, as well as other environmental factors. Next, the collected odor data undergoes preprocessing, a critical step to ensure data validity and reliability. This preprocessing process can include denoising, normalization, and data smoothing. Denoising eliminates noise signals introduced by external environmental influences or device errors during the collection process, ensuring the purity of the odor data. Normalization adjusts the values ​​of different odor samples to a uniform range to avoid data distortion caused by differences in odor concentration or different collection times. Data smoothing reduces random fluctuations in the data, making the collected odor signal more stable. For example, if the target odor sample is the aroma of a certain food, the chemical sensor detects multiple aroma components in the food, such as sugar, acidity, and flavors. The preprocessing process standardizes this complex multidimensional data to remove interfering odors in the environment, such as humidity or other gases in the air. This refined preprocessing step provides a clean, standardized data foundation for subsequent odor feature extraction and classification, significantly improving the overall accuracy and efficiency of odor classification and identification.

[0020] Step S2: extract odor features from the preprocessed data to obtain an odor extraction feature vector.

[0021] Specifically, when extracting odor features from preprocessed data, representative features must first be extracted from the preprocessed data. These features effectively describe the essential characteristics of the odor. This process aims to transform complex odor data into concise feature vectors that can be used for further analysis and classification. Specifically, odor feature extraction involves extracting characteristic values ​​from the preprocessed data that uniquely identify the odor sample. These characteristic values ​​can be parameters such as the concentration of odor molecules, response time, and signal intensity, which exhibit different characteristics in different odor samples. For example, odor data collected by chemical sensors, after preprocessing, contains a set of numerical values ​​representing concentration and intensity. The feature extraction step calculates and analyzes these numerical values ​​to identify the most representative features, such as changes in the concentration of specific molecules or the response patterns of multiple sensors combined. In practice, this process utilizes various mathematical and statistical methods to extract features. Commonly used techniques include principal component analysis (PCA) and linear discriminant analysis (LDA), which can transform the original preprocessed data into a smaller-dimensional feature vector while retaining key information in the odor data. Through these methods, the feature vector not only reduces the data dimensionality but also enhances its ability to distinguish odor types. For example, if the preprocessed data for an odor sample includes the response signals of multiple chemical sensors to different odor molecules, the feature extraction step will analyze the patterns of these response signals and convert them into a feature vector containing the main components of these signals. These feature vectors will reflect the intensity of the odor, the duration of the odor, and the concentration of its main components. Such feature vectors not only represent the core information of the odor sample but can also be used for subsequent odor classification and recognition tasks. Therefore, odor feature extraction is a key step in the entire odor classification and recognition system, converting the preprocessed data into a format that can be effectively processed by the classification algorithm.

[0022] Step S3: constructing a feature matrix based on the odor extraction feature vector, and inputting the feature matrix into a preset classification algorithm to perform odor classification, thereby obtaining an odor type and a confidence level corresponding to the odor type.

[0023] Specifically, in this step, a feature matrix is ​​first constructed based on the extracted odor feature vectors. This involves organizing the multiple feature vectors obtained from the odor feature extraction process into a systematic data structure, namely a feature matrix. The rows of the feature matrix typically correspond to different odor samples, while the columns represent the individual features in the odor feature vectors. By integrating the feature vectors of multiple odor samples into the feature matrix, a comprehensive dataset describing the characteristics of different odor samples can be formed. This matrix not only contains the various characteristic information of the odor samples but also provides a structured input data source for subsequent classification. Subsequently, the feature matrix is ​​input into a pre-set classification algorithm for odor classification. The classification algorithm can be a variety of machine learning models, such as support vector machines (SVMs), decision trees, random forests, or neural networks, which can classify odor samples based on the information in the feature matrix. The classification process analyzes the data patterns in the feature matrix to determine the odor type to which each odor sample belongs and calculates a corresponding confidence level for each prediction result. The confidence level reflects the reliability of the classification algorithm's prediction of a particular odor type. For example, suppose the feature matrix contains extracted feature vectors for different odor samples. These feature vectors include information such as odor concentration and component ratio. This feature matrix is ​​input into a support vector machine classifier. Using a trained model, the classifier identifies the primary type of odor samples and makes a prediction. Ultimately, the classifier outputs the odor type and its corresponding confidence score for each odor sample. If the confidence score is high enough, the odor sample is determined to belong to the primary odor type. If the confidence score is low, further processing is required to confirm the odor type.

[0024] Step S4: If the confidence level is within a preset confidence level range, the primary odor type is used as the odor type of the target odor sample.

[0025] Specifically, in this step, after the classification algorithm classifies the odor sample and obtains a primary odor type and its corresponding confidence level, it is necessary to further determine whether the confidence level is within a preset confidence range. The confidence range is a pre-set threshold interval used to measure the credibility of the classification result. If the confidence level is greater than or equal to this preset minimum confidence threshold, it means that the classification algorithm's prediction of the odor type is highly reliable. In this case, the primary odor type can be considered accurate, and the primary odor type is therefore used as the odor type of the target odor sample.

[0026] In specific implementations, when the feature matrix constructed from the odor extraction feature vectors is input into the classification algorithm, the algorithm outputs a prediction result and corresponding confidence score for each odor sample. The system then checks these confidence scores to determine whether they are within a preset confidence range. Typically, this range can be set based on experimental data and actual application requirements. For example, a confidence score range of 0.7 to 1.0 indicates that a classification result is considered reliable when the confidence score is 0.7 or greater. For example, suppose the classification algorithm predicts the odor type of a food sample and the result indicates that its primary odor type is "lemon" with a confidence score of 0.85. Since 0.85 is within the preset confidence range (assuming the range is 0.7 to 1.0), the classification algorithm's prediction of "lemon" is highly reliable. Therefore, "lemon" can be used as the odor type for the food sample without further processing. This confidence assessment mechanism ensures that the primary odor type is only adopted as the final odor type when the classification result reaches a certain confidence level. This not only improves the accuracy of odor classification and recognition but also reduces the risk of false positives. This can effectively improve the reliability and practicality of the system, enabling the odor classification and recognition system to provide more accurate and reliable results in practical applications.

[0027] Step S5: If the confidence level is not within a preset confidence level range, cluster analysis is performed on the target odor sample to obtain a secondary odor type, and the secondary odor type is used as the odor type of the target odor sample.

[0028] Specifically, in this step, if the confidence level falls outside the preset confidence range, it indicates that the classification algorithm's prediction of the primary odor type is not certain or reliable, and further action is required. Specifically, cluster analysis is performed on the target odor sample to obtain a secondary odor type, further confirming the accurate classification of the odor sample. Cluster analysis is an unsupervised learning method that aims to group samples based on data similarity. In this step, the feature matrix of the target odor sample is compared with the data of other known odor samples, using a clustering algorithm (such as K-means clustering, hierarchical clustering, or DBSCAN) to discover the inherent structure and similarities between samples. This analysis can reveal other odor categories to which the target odor sample belongs, especially when the initial classification confidence level is low, providing a more detailed basis for classification. For example, if the initial classification result for a food odor sample is "orange scent," but the confidence level is 0.6, which is below the preset minimum confidence threshold of 0.7, cluster analysis is required. By comparing the food sample with other similar odor samples through cluster analysis, it can be found that it is actually closer to the "grapefruit scent" odor type. The secondary odor type provided by the cluster analysis, namely "grapefruit scent," is then used as the final odor type, replacing the initial classification result. This process ensures that even when the classification confidence is low, a more accurate odor type can be found through deeper data analysis, thereby improving the classification accuracy and reliability of the system.

[0029] In a specific embodiment, the preset chemical sensor includes a preset multi-channel metal oxide semiconductor sensor array, and the target odor sample is collected by the preset chemical sensor to obtain collected odor data; and the collected odor data is preprocessed to obtain preprocessed data, including: Dynamically sampling the target odor sample through the multi-channel metal oxide semiconductor sensor array to obtain collected odor data; wherein the collected odor data includes odor intensity information, odor component concentration information and odor molecular structure information; Performing wavelet decomposition on the collected odor data using wavelet transform technology to obtain multi-scale wavelet data; Denoising is performed on the multi-scale wavelet data to obtain denoised data, and the denoised data is used as preprocessed data.

[0030] Specifically, a target odor sample is first collected using a pre-set chemical sensor, including a pre-set multi-channel metal oxide semiconductor (MOS) sensor array. This array can simultaneously detect different aspects of the odor sample through multiple channels. The collected odor data includes odor intensity, odor component concentrations, and odor molecular structure information. This information comprehensively covers the essential characteristics of the odor sample, providing a rich data source for subsequent processing. Next, the collected odor data is pre-processed. The first step in pre-processing is dynamic sampling of the target odor sample using the multi-channel MOS sensor array. This means that the sensor array repeatedly detects the odor sample at different time points or under different conditions, thereby obtaining more comprehensive and accurate odor data. The collected data includes odor intensity, odor component concentrations, and odor molecule structure information. After initial collection, this data requires further processing to improve its quality. The collected odor data is then subjected to wavelet decomposition using wavelet transform technology. Wavelet transform is an effective signal processing method that converts time-domain signals into the frequency domain, facilitating data analysis at different scales. Wavelet decomposition of the odor data produces multi-scale wavelet data that reflects the characteristics of the odor data at different frequencies, revealing more subtle variations. After obtaining multi-scale wavelet data, denoising is required. This process improves data quality by removing unnecessary noise, making the final data more accurate and reliable. Denoising is typically achieved by setting a threshold or other denoising algorithm to remove noise signals from the data while retaining the valid information. This processed data is called preprocessed data and serves as the basis for further analysis and classification.

[0031] In a specific embodiment, performing odor feature extraction on the preprocessed data to obtain an odor extraction feature vector includes: performing baseline drift correction on the preprocessed data to obtain baseline drift corrected data; performing high-dimensional feature encoding on the baseline drift correction data to obtain high-dimensional feature encoded data; Performing local feature extraction on the high-dimensional feature-encoded data to obtain local feature data; wherein the local feature data includes concentration changes of odor chemical components, odor reaction time constants, and spatial distribution of odor molecules; Performing structured feature analysis on the local feature data to obtain structured feature data; wherein the structured feature data includes topological structural features of the odor molecules, chemical bond type distribution of the odor molecules, composition ratio of functional groups of the odor molecules, and molecular weight distribution features of the odor molecules; Performing data enhancement on the structured feature data to obtain odor enhancement data; Performing dimensionality reduction on the odor enhancement data to obtain dimensionality reduction features; The dimensionality reduction features are constructed into a time series using a preset long-short memory algorithm to obtain a change curve of the odor at different time points; wherein the change curve includes a decay curve of the odor intensity over time, a curve of the change of the odor component ratio over time, and a curve of the diffusion rate of the odor molecules over time; An odor extraction feature vector is constructed based on the change curve.

[0032] Specifically, in this step, odor feature extraction is first performed on the preprocessed data to obtain an odor extraction feature vector. The odor feature extraction process begins with baseline drift correction to ensure that baseline drift in the data is corrected, thereby obtaining baseline drift-corrected data. The purpose of baseline drift correction is to eliminate systematic errors caused by the measurement environment or sensor drift, making the data more accurate and reliable. Next, high-dimensional feature encoding is performed on the baseline drift-corrected data. The purpose of high-dimensional feature encoding is to convert the data into a higher-dimensional representation to capture the more complex and subtle features in the odor data. This high-dimensional feature encoding data integrates multiple odor feature information, providing a rich foundation for subsequent local feature extraction. Local feature extraction is then performed on the high-dimensional feature encoding data to obtain local feature data. Local feature data includes concentration changes of odor chemical components, odor reaction time constants, and the spatial distribution of odor molecules. These features can reflect the specific behavior and changing patterns of the odor sample in the local environment. Structured feature analysis is then performed on the local feature data to obtain structured feature data. This step further extracts structured information from the odor sample by analyzing the topological structure of the odor molecules, the distribution of chemical bond types, the proportion of functional groups, and the molecular weight distribution. This information provides a more comprehensive perspective for accurately describing odor characteristics. Next, data enhancement is performed on this structured feature data to generate enhanced odor data. Data enhancement techniques increase sample diversity by generating more variant data, thereby improving the robustness and generalization of the model. Subsequently, dimensionality reduction is performed on the enhanced odor data to generate reduced-dimensionality features. Dimensionality reduction aims to simplify high-dimensional data into low-dimensional data for easier processing and analysis while preserving important feature information. Using a pre-defined long-short memory algorithm, the reduced-dimensionality features are constructed into time series, generating odor change curves at different time points. These curves include the decay curve of odor intensity over time, the change curve of odor component ratio over time, and the diffusion rate curve of odor molecules over time. These curves detail the dynamic changes of the odor sample over time. Finally, odor extraction feature vectors are constructed based on these change curves. These feature vectors integrate all important information of the odor sample in the time series, providing a complete and accurate feature description for final odor classification and analysis. Through this series of processing steps, the extraction of odor characteristics can be ensured to be not only comprehensive but also accurate, providing a solid data foundation for subsequent analysis and decision-making.

[0033] In a specific embodiment, the odor extraction feature vector includes an odor attenuation vector, an odor component ratio vector, and an odor molecule diffusion rate vector; constructing a feature matrix based on the odor extraction feature vector, and inputting the feature matrix into a preset classification algorithm to perform odor classification, thereby obtaining a primary odor type and a confidence level corresponding to the primary odor type, includes: Arrange the odor attenuation vector, the odor component ratio vector, and the odor molecule diffusion rate vector from left to right or from top to bottom to obtain a feature matrix; Based on a preset classification algorithm, the feature matrix is ​​morphologically encoded to obtain a feature matrix; Performing molecular structure modeling on the feature matrix to obtain a molecular graph feature matrix; Classifying the molecular graph feature matrix to obtain classification labels corresponding to the molecular graph feature matrix; Calling a preset odor database and mapping the classification label with a storage label in the preset odor database to obtain a primary odor type corresponding to the classification label; wherein the storage label in the preset odor database has a corresponding primary odor type; A confidence calculation is performed on the primary odor type based on the molecular graph feature matrix to obtain a confidence level corresponding to the primary odor type.

[0034] Specifically, the odor extraction feature vector includes an odor attenuation vector, an odor component ratio vector, and an odor molecule diffusion rate vector. First, these vectors are arranged from left to right or top to bottom to form a feature matrix. The purpose of this step is to integrate odor features of different dimensions into a unified matrix representation to facilitate subsequent processing and analysis. Next, based on a preset classification algorithm, the feature matrix is ​​state-encoded to obtain a state-encoded feature matrix. State encoding is a method of converting a feature matrix into another encoding format, making it more adaptable to the requirements of the classification algorithm. This step ensures that the information in the feature matrix is ​​effectively preserved and fully utilized during the classification process. Then, molecular structure modeling is performed on the state-encoded feature matrix to obtain a molecular graph feature matrix. Molecular structure modeling models the specific structural features of odor molecules and converts them into a feature matrix in the form of a graph structure, which can more intuitively reflect the complex relationships and characteristics of odor molecules. Next, the molecular graph feature matrix is ​​classified to obtain the corresponding classification labels. This step uses a classification algorithm to process the molecular graph feature matrix and classify it into specific labels based on its characteristics. These labels represent different primary odor types. Then, the preset odor database is called, and the classification label is mapped with the stored labels in the preset odor database to obtain the primary odor type corresponding to the classification label. The stored labels in the preset odor database have corresponding primary odor types. Through the mapping process, the classification label can be converted into a specific odor type, such as "orange scent" or "grapefruit scent." Finally, the confidence level of the primary odor type is calculated based on the molecular graph feature matrix to obtain the confidence level corresponding to the primary odor type. The confidence level calculation is intended to assess the reliability of the classification results. By analyzing the molecular graph feature matrix, the confidence level of each primary odor type is calculated to facilitate subsequent decision-making and processing.

[0035] In a specific embodiment, performing cluster analysis on the target odor sample to obtain the secondary odor type includes: Performing tensor decomposition on the collected odor data of the target odor sample to obtain a tensor decomposition feature representation; Performing frequency domain analysis on the tensor decomposition feature representation using Fourier transform technology to obtain a spectrum feature distribution; Constructing an odor molecule graph based on the spectral feature distribution; Performing topological space mapping on the odor molecule graph to obtain an odor topological structure; wherein the odor topological structure is the spatial relationship and chemical bond connection between odor molecules; Performing eigenvalue decomposition on the odor topological structure based on a preset spectral graph theory technology to obtain odor spectrum features; Performing a persistence coherence analysis on the odor spectrum characteristics to obtain an odor persistence map; Sampling the odor persistence map using discrete point sets to obtain persistence barcodes, and quantifying the topological similarity between the persistence barcodes and persistence barcodes in a preset odor database to obtain an odor topological distance matrix; A hierarchical clustering analysis is performed based on the odor topological distance matrix to obtain secondary odor types.

[0036] Specifically, in this step, cluster analysis is performed on the target odor sample to obtain secondary odor types. First, tensor decomposition is performed on the collected odor data of the target odor sample to obtain a tensor decomposition feature representation. The purpose of tensor decomposition is to decompose high-dimensional odor data into multiple low-dimensional subspaces, thereby extracting useful feature information. Next, the tensor decomposition feature representation is subjected to frequency domain analysis using Fourier transform technology to obtain a spectral feature distribution. Fourier transform can convert time domain signals into frequency domain signals, revealing the distribution characteristics of frequency components in the odor data. Based on this spectral feature distribution, an odor molecule graph is constructed. This graph represents the relationships and characteristics of different odor molecules and reflects their interactions. Next, a topological spatial mapping is performed on the odor molecule graph to obtain the odor topological structure. The odor topological structure describes the spatial relationships and chemical bond connections between odor molecules, providing detailed spatial and chemical information between odor molecules. Subsequently, eigenvalue decomposition is performed on the odor topological structure based on a pre-defined spectral graph theory technique to obtain odor spectral features. Spectral graph theory mathematically analyzes the topological structure to extract eigenvalues ​​that describe the characteristics of odor molecules. Persistent coherence analysis is performed on the odor spectrum features to obtain an odor persistence map. Persistent coherence analysis helps capture long-term structural features in odor data and presents them in the form of a persistence map. The odor persistence map is sampled as a discrete point set to obtain persistence barcodes, and the persistence barcodes are topologically quantified with the persistence barcodes in a preset odor database to obtain an odor topological distance matrix. Topological similarity quantification evaluates the similarity between the sample and known odors in the database by calculating the distance between them. This step converts the topological features of the odor sample into a distance matrix for further analysis. Finally, hierarchical clustering analysis is performed based on the odor topological distance matrix to obtain secondary odor types. Hierarchical clustering analysis processes the similarities in the distance matrix to group the odor samples into different odor types. These secondary odor types provide further classification results for the target odor sample, enabling more accurate identification and description of the odor characteristics. For example, suppose the target odor sample is a new type of fragrance. First, its odor data is tensor decomposition to obtain a feature representation, and then Fourier transform is used to perform frequency domain analysis to obtain its spectral feature distribution. Based on these spectral features, a molecular graph of the fragrance's odor is constructed and topologically mapped to obtain its odor topological structure. Spectral graph theory techniques are then used to perform eigenvalue decomposition of the odor topological structure, yielding odor spectrum features. Persistence homology analysis is then performed to generate an odor persistence graph. The persistence graph is sampled to generate persistence barcodes, which are then quantified for similarity with barcodes in a database. Finally, hierarchical clustering analysis is used to determine the fragrance's secondary odor type, such as "floral" or "woody," enabling detailed classification and identification of target odor samples.

[0037] In a specific embodiment, discrete point set sampling is performed on the odor persistence map to obtain a persistence barcode, including: Using Morse-Smale complex decomposition technology to extract critical points from the odor persistence graph to obtain a critical point set; wherein the critical point set includes minimum points, saddle points and maximum points; Extracting topological features from the critical point set to obtain a persistent homology generator; Performing multi-scale topological structure reconstruction on the persistent homology generator to obtain a multi-scale simplicial complex; Extracting topological invariants from the multi-scale simplicial complex to obtain a persistent Betti number sequence; A barcode is constructed based on the persistent Betti number sequence to obtain a persistent barcode.

[0038] Specifically, in this step, the odor persistence graph is sampled using a discrete point set to obtain a persistence barcode. First, critical points are extracted from the odor persistence graph using Morse-Smale complex decomposition technology to obtain a critical point set. Morse-Smale complex decomposition technology is used to identify important points in a topological space, including minima, saddle points, and maxima, which reflect important structural features in the odor data. Next, topological feature extraction is performed on the critical point set to obtain persistent homology generators. The topological feature extraction process involves identifying structural features at different scales, which are used to describe the topological structure of the odor persistence graph. Subsequently, the persistent homology generators are reconstructed at multiple scales to obtain a multiscale simplicial complex. A multiscale simplicial complex is a tool for representing topological structures at different scales, which can effectively capture the different levels and complexities of odor data. Topological invariants are extracted from the multiscale simplicial complex to obtain a sequence of persistent Betti numbers. Topological invariants, such as Betti numbers, are used to describe the topological properties of a space, such as the number of connected branches or the number of holes. The persistent Betti number sequence provides a detailed description of the topological properties of the odor persistence graph. Finally, a barcode is constructed based on the persistent Betti number sequence, resulting in a persistent barcode. A persistent barcode visualizes the Betti number sequence as a barcode graph, representing the persistence and variability of topological features at different scales. For example, suppose we have a persistence graph for an odor sample. Using the Morse-Smale complex decomposition technique, we can extract minima, saddle points, and maxima, which represent distinct topological features in the odor sample. Next, through topological feature extraction and multi-scale reconstruction, we obtain a multiscale simplicial complex representing the structure at different scales, from which we extract the persistent Betti number sequence. Finally, by constructing a persistent barcode, these features are represented as barcodes. This barcode can be compared and matched with barcodes in a pre-set odor database for further odor classification and analysis.

[0039] In a specific embodiment, topological invariants are extracted from the multi-scale simplicial complex to obtain a persistent Betti number sequence, including: performing discrete decomposition processing on the multi-scale simplicial complex using a discrete Morse function to obtain a discrete gradient vector field; Performing homology group calculation on the discrete gradient vector field to obtain persistent homology groups in each dimension, wherein the persistent homology groups in each dimension reflect connectivity information of the topological characteristics of the odor in different dimensions; Performing spectral sequence analysis on the persistent homology groups of each dimension to obtain a persistent spectral sequence; Extracting topological invariants from the persistent spectral sequence to obtain a DeRham cohomology group; Perform characteristic calculation on the DeRham cohomology group to obtain the twisted Betti number; The persistent random complex method is used to perform statistical analysis on the distorted Betti numbers to obtain a persistent Betti number sequence.

[0040] Specifically, in this step, topological invariants are extracted from the multiscale simplicial complex to obtain a persistent Betti number sequence. First, the multiscale simplicial complex is discretely decomposed using discrete Morse functions to obtain a discrete gradient vector field. Discrete Morse functions decompose the complex into a series of simple gradient steps, which capture the key topological features of the multiscale simplicial complex. Next, homology groups are calculated on the discrete gradient vector field to obtain persistent homology groups in each dimension. These persistent homology groups in each dimension reflect the connectivity information of the topological features of the odor in different dimensions. For example, in one dimension, this involves rings in the topological structure; in two dimensions, it involves holes in the topological structure. Homology group calculations enable us to identify these different topological features. Spectral sequence analysis is then performed on the persistent homology groups in each dimension to obtain persistent spectral sequences. Spectral sequence analysis is a mathematical tool used to describe the changes and persistence in various dimensions of a topological space. Based on the persistent spectral sequences, we further extract topological invariants to obtain DeRham cohomology groups. DeRham cohomology groups are used to describe different homology types in a topological space, providing topological information at different levels of the space. Subsequently, characteristic computations were performed on the DeRham cohomology groups to obtain twisted Betti numbers. Twisted Betti numbers are an extension of traditional Betti numbers, capable of capturing more subtle topological features. Finally, the persistent random complex method was used to statistically analyze the twisted Betti numbers to obtain a persistent Betti number sequence. The persistent random complex method uses randomized statistics on Betti numbers to reveal their stability and variability under different random environments. For example, for the multiscale simplicial complexes of odor samples, discrete Morse functions and gradient vector field analysis can identify key topological features in the odor data, such as distinct rings or holes. Subsequently, through homology group computations and spectral sequence analysis, the persistent Betti number sequence of the odor can be extracted. These values ​​accurately describe the topological characteristics of the odor sample at different scales, enabling detailed analysis and classification of the odor.

[0041] The above describes the odor classification and identification method in the embodiment of the present invention. The following describes the odor classification and identification device in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an odor classification and identification device includes: The acquisition module 21 is configured to acquire a target odor sample through a preset chemical sensor to obtain acquired odor data; and to pre-process the acquired odor data to obtain pre-processed data; An extraction module 22 is configured to extract odor features from the preprocessed data to obtain an odor extraction feature vector; A construction module 23 is configured to construct a feature matrix based on the odor extraction feature vector, and input the feature matrix into a preset classification algorithm to perform odor classification to obtain an odor type and a confidence level corresponding to the odor type; a judgment module 24, configured to determine the primary odor type as the odor type of the target odor sample if the confidence level is within a preset confidence level range; The analysis module 25 is configured to perform cluster analysis on the target odor sample to obtain a secondary odor type if the confidence level is not within a preset confidence level range, and use the secondary odor type as the odor type of the target odor sample.

[0042] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0043] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided, wherein the internal structure of the computer device can be as follows: Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0044] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0045] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0046] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0047] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0048] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for classifying and identifying odors, characterized in that: The following steps are involved: The target odor sample is collected through a preset chemical sensor to obtain collected odor data; and preprocessing the collected odor data to obtain preprocessed data; performing odor feature extraction on the preprocessed data to obtain an odor extraction feature vector; Constructing a feature matrix based on the odor extraction feature vector, and inputting the feature matrix into a preset classification algorithm to perform odor classification, thereby obtaining an odor type and a confidence level corresponding to the odor type; If the confidence level is within a preset confidence level range, the primary odor type is used as the odor type of the target odor sample; If the confidence level is not within a preset confidence level range, cluster analysis is performed on the target odor sample to obtain a secondary odor type, and the secondary odor type is used as the odor type of the target odor sample.

2. The odor classification and identification method according to claim 1, characterized in that: The preset chemical sensor includes a preset multi-channel metal oxide semiconductor sensor array, and the target odor sample is collected by the preset chemical sensor to obtain collected odor data; The collected odor data is preprocessed to obtain preprocessed data, including: Dynamically sampling the target odor sample through the multi-channel metal oxide semiconductor sensor array to obtain collected odor data; wherein the collected odor data includes odor intensity information, odor component concentration information and odor molecular structure information; Performing wavelet decomposition on the collected odor data using wavelet transform technology to obtain multi-scale wavelet data; Denoising is performed on the multi-scale wavelet data to obtain denoised data, and the denoised data is used as preprocessed data.

3. The odor classification and identification method according to claim 1, characterized in that: Performing odor feature extraction on the preprocessed data to obtain an odor extraction feature vector includes: performing baseline drift correction on the preprocessed data to obtain baseline drift corrected data; performing high-dimensional feature encoding on the baseline drift correction data to obtain high-dimensional feature encoded data; Performing local feature extraction on the high-dimensional feature-encoded data to obtain local feature data; wherein the local feature data includes concentration changes of odor chemical components, odor reaction time constants, and spatial distribution of odor molecules; Performing structured feature analysis on the local feature data to obtain structured feature data; wherein the structured feature data includes topological structural features of the odor molecules, chemical bond type distribution of the odor molecules, composition ratio of functional groups of the odor molecules, and molecular weight distribution features of the odor molecules; Performing data enhancement on the structured feature data to obtain odor enhancement data; Performing dimensionality reduction on the odor enhancement data to obtain dimensionality reduction features; The dimensionality reduction features are constructed into a time series using a preset long-short memory algorithm to obtain a change curve of the odor at different time points; wherein the change curve includes a decay curve of the odor intensity over time, a curve of the change of the odor component ratio over time, and a curve of the diffusion rate of the odor molecules over time; An odor extraction feature vector is constructed based on the change curve.

4. The odor classification and identification method according to claim 1, characterized in that: The odor extraction feature vector includes an odor attenuation vector, an odor component ratio vector, and an odor molecule diffusion rate vector. The feature matrix is ​​constructed based on the odor extraction feature vector, and the feature matrix is ​​input into a preset classification algorithm to perform odor classification to obtain an odor type and a confidence level corresponding to the odor type, including: Arrange the odor attenuation vector, the odor component ratio vector, and the odor molecule diffusion rate vector from left to right or from top to bottom to obtain a feature matrix; Based on a preset classification algorithm, the feature matrix is ​​morphologically encoded to obtain a feature matrix; Performing molecular structure modeling on the feature matrix to obtain a molecular graph feature matrix; Classifying the molecular graph feature matrix to obtain classification labels corresponding to the molecular graph feature matrix; Calling a preset odor database and mapping the classification label with a storage label in the preset odor database to obtain a primary odor type corresponding to the classification label; wherein the storage label in the preset odor database has a corresponding primary odor type; A confidence calculation is performed on the primary odor type based on the molecular graph feature matrix to obtain a confidence level corresponding to the primary odor type.

5. The odor classification and identification method according to claim 1, characterized in that: The cluster analysis of the target odor sample is performed to obtain secondary odor types, including: Performing tensor decomposition on the collected odor data of the target odor sample to obtain a tensor decomposition feature representation; Performing frequency domain analysis on the tensor decomposition feature representation using Fourier transform technology to obtain a spectrum feature distribution; Constructing an odor molecule graph based on the spectral feature distribution; Performing topological space mapping on the odor molecule graph to obtain an odor topological structure; wherein the odor topological structure is the spatial relationship and chemical bond connection between odor molecules; Performing eigenvalue decomposition on the odor topological structure based on a preset spectral graph theory technology to obtain odor spectrum features; Performing a persistence coherence analysis on the odor spectrum characteristics to obtain an odor persistence map; Sampling the odor persistence map using discrete point sets to obtain persistence barcodes, and quantifying the topological similarity between the persistence barcodes and persistence barcodes in a preset odor database to obtain an odor topological distance matrix; A hierarchical clustering analysis is performed based on the odor topological distance matrix to obtain secondary odor types.

6. The odor classification and identification method according to claim 5, characterized in that: The odor persistence map is sampled by discrete point sets to obtain a persistence barcode, including: Using Morse-Smale complex decomposition technology to extract critical points from the odor persistence graph to obtain a critical point set; wherein the critical point set includes minimum points, saddle points and maximum points; Extracting topological features from the critical point set to obtain a persistent homology generator; Performing multi-scale topological structure reconstruction on the persistent homology generator to obtain a multi-scale simplicial complex; Extracting topological invariants from the multi-scale simplicial complex to obtain a persistent Betti number sequence; A barcode is constructed based on the persistent Betti number sequence to obtain a persistent barcode.

7. The odor classification and identification method according to claim 6, characterized in that: The topological invariants of the multi-scale simplicial complex are extracted to obtain a persistent Betti number sequence, including: performing discrete decomposition processing on the multi-scale simplicial complex using a discrete Morse function to obtain a discrete gradient vector field; Performing homology group calculation on the discrete gradient vector field to obtain persistent homology groups in each dimension, wherein the persistent homology groups in each dimension reflect connectivity information of the topological characteristics of the odor in different dimensions; Performing spectral sequence analysis on the persistent homology groups of each dimension to obtain a persistent spectral sequence; Extracting topological invariants from the persistent spectral sequence to obtain a DeRham cohomology group; Perform characteristic calculation on the DeRham cohomology group to obtain the twisted Betti number; The persistent random complex method is used to perform statistical analysis on the distorted Betti numbers to obtain a persistent Betti number sequence.

8. An odor classification and identification device, characterized in that: include: The collection module is used to collect the target odor sample through a preset chemical sensor to obtain the collected odor data; and preprocessing the collected odor data to obtain preprocessed data; An extraction module, configured to extract odor features from the preprocessed data to obtain an odor extraction feature vector; A construction module, configured to construct a feature matrix based on the odor extraction feature vector, and input the feature matrix into a preset classification algorithm to perform odor classification, thereby obtaining an odor type and a confidence level corresponding to the odor type; a judgment module, configured to use the primary odor type as the odor type of the target odor sample if the confidence level is within a preset confidence level range; An analysis module is configured to perform cluster analysis on the target odor sample to obtain a secondary odor type if the confidence level is not within a preset confidence level range, and use the secondary odor type as the odor type of the target odor sample.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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 steps of the method according to any one of claims 1 to 7 are implemented.

Images