Smell feature reasoning method, system, device and storage medium based on graph model

Abstract

This application provides a graph model-based odor feature inference method, system, device, and storage medium, relating to the field of odor recognition technology. The method includes: acquiring raw response data of a target odor based on multiple gas sensors, performing data preprocessing and feature extraction to obtain multiple odor feature data; constructing a sensor relationship graph model based on the correlation information between each gas sensor, using each gas sensor as a node; constructing an odor feature graph model based on the correlation degree between each odor feature data, using each odor feature data as a node; inputting the raw response data, sensor relationship graph model, odor feature graph model, and each odor feature data into a graph neural network for training to obtain an odor feature inference model; and inputting the odor feature graph model and each odor feature data into the odor feature inference model to obtain the feature inference result of the target odor. This application significantly improves the accuracy of odor recognition.

Description

Technical Field

[0001] This application relates to the field of odor recognition technology, specifically to an odor feature reasoning method, system, device, and storage medium based on a graph model. Background Technology

[0002] An electronic nose is an intelligent detection instrument that simulates the olfactory system of mammals. By integrating a gas sensor array and pattern recognition algorithm, it can systematically analyze and identify complex odor components. It is currently widely used in many important fields such as food industry, environmental monitoring, medical diagnosis and industrial safety.

[0003] However, current methods for processing electronic nose data, such as principal component analysis, support vector machines, and deep learning, generally treat odor features as independent variables. In reality, the complex characteristics of odor stem from the synergistic or antagonistic effects of various odor molecules on the sensor array. The neglect of this interaction in existing methods leads to insufficient information extraction, thus limiting the accuracy of odor recognition. Therefore, improving the odor recognition accuracy of electronic noses has become an urgent problem to be solved. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the prior art, this application provides a graph model-based odor feature reasoning method, system, device and storage medium, which effectively solves the problem of low odor recognition accuracy of electronic noses.

[0005] In a first aspect, the present invention provides an odor feature reasoning method based on a graph model, the method comprising: Raw response data of the target odor is acquired based on multiple gas sensors; The raw response data is preprocessed and feature extracted to obtain multiple odor feature data. Each of the gas sensors is used as a node, and a sensor relationship graph model is constructed based on the correlation information between the gas sensors. Each of the odor feature data is used as a node, and an odor feature graph model is constructed based on the correlation between the odor feature data. The original response data, the sensor relationship graph model, the odor feature graph model, and each of the odor feature data are input into a graph neural network for training to obtain an odor feature inference model. The odor feature map model and each of the odor feature data are input into the odor feature inference model to obtain the feature inference result of the target odor.

[0006] In an optional implementation, the step of inputting the original response data, the sensor relationship graph model, the odor feature graph model, and each of the odor feature data into a graph neural network for training to obtain an odor feature inference model includes: The sensor relationship graph model and the raw response data are input into a first graph neural network for processing to obtain the first high-order feature; The odor feature map model and each of the odor feature data are input into a second graph neural network for processing to obtain a second high-order feature. The first high-order feature and the second high-order feature are fused to obtain the target fused feature; The target fusion features are input into the target model for training to obtain the odor feature inference model.

[0007] In an optional implementation, the fusion of the first high-order feature and the second high-order feature is achieved by one or more of vector concatenation fusion, weighted summation fusion, and attention-based mechanism fusion, and the target model is a classifier model or a regressor model.

[0008] In an optional implementation, the step of using each of the gas sensors as nodes and constructing a sensor relationship graph model based on the correlation information between the gas sensors includes: Each of the aforementioned gas sensors is designated as a target node; The target edges between each target node are determined based on the sensing mechanism and the overlap of the sensing spectra of each gas sensor. Calculate the correlation coefficient of the original response data of the two gas sensors corresponding to the target edge to obtain the initial weight of the target edge; The initial weights are assigned to the target edge to obtain the sensor relationship graph model.

[0009] In an optional implementation, the step of using each of the odor feature data as nodes and constructing an odor feature map model based on the correlation between the odor feature data includes: Each of the aforementioned odor feature data is used as a target node; Calculate the correlation degree between each of the target nodes, wherein the correlation degree is one of the mutual information coefficient, Pearson correlation coefficient and Spearman rank correlation coefficient; If the correlation between two target nodes is greater than or equal to a preset threshold, then a target edge is constructed to obtain the odor feature map model.

[0010] In an optional implementation, the method further includes: New raw response data is acquired, and the odor feature inference model is incrementally trained and dynamically optimized based on the new raw response data.

[0011] In an optional implementation, the step of preprocessing and extracting features from the raw response data to obtain multiple odor feature data includes: The original response data is denoised using the wavelet threshold denoising method to obtain the initial response data. The initial response data is normalized using the minimum-maximum normalization method to obtain normalized data. Statistical features and time-frequency domain features are extracted from the normalized data to obtain multiple odor feature data.

[0012] Secondly, the present invention provides an odor feature reasoning system based on a graph model, the system comprising: The data acquisition module is used to acquire raw response data of the target odor based on multiple gas sensors; The data processing module is used to preprocess and extract features from the raw response data to obtain multiple odor feature data. The first construction module is used to construct a sensor relationship graph model based on the correlation information between each gas sensor, using each gas sensor as a node. The second construction module is used to construct an odor feature graph model by taking each of the odor feature data as nodes and based on the correlation between each of the odor feature data. The model training module is used to input the original response data, the sensor relationship graph model, the odor feature graph model, and each of the odor feature data into the graph neural network for training to obtain an odor feature inference model. The feature reasoning module is used to input the odor feature map model and each of the odor feature data into the odor feature reasoning model to obtain the feature reasoning result of the target odor.

[0013] Thirdly, the present invention provides an electronic device, including 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 graph model-based odor feature reasoning method as described in the first aspect of this application.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the odor feature reasoning method based on a graph model as described in the first aspect of this application.

[0015] The odor feature inference method, system, device, and storage medium provided in this application, based on a graph model, explicitly encodes the physical cross-sensitivity between sensors and the nonlinear dependencies between features by constructing a dual-granularity graph model consisting of a sensor relationship graph model and an odor feature graph model. This enables the graph neural network to learn more discriminative high-order joint features, significantly improving the accuracy of odor recognition, classification, and regression. Simultaneously, the sensor relationship graph model introduces hardware-level physical constraints. When individual sensors drift or malfunction, the sensor relationship graph model can rely on strongly correlated nodes in the graph for information compensation, significantly improving the system's fault tolerance and long-term stability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a first schematic diagram of the flowchart of the odor feature reasoning method based on graph model provided in the embodiments of this application; Figure 2 This is a second schematic diagram of the flowchart of the odor feature reasoning method based on graph model provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the odor feature reasoning system based on the graph model provided in this application embodiment; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0018] Explanation of key component symbols: 200. Odor feature reasoning system based on graph model; 210. Data acquisition module; 220. Data processing module; 230. First construction module; 240. Second construction module; 250. Model training module; 260. Feature reasoning module; 300. Electronic device; 310. Processor; 320. Communication interface; 330. Memory; 340. Communication bus. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be further described clearly and completely below with reference to the accompanying drawings of the embodiments. It should be noted that the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] 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 technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0022] Existing electronic nose data processing methods, such as principal component analysis, support vector machines, and deep learning, generally suffer from the following technical limitations: First, traditional methods typically treat odor features as independent variables, failing to fully consider the intrinsic correlations between different sensors and between multidimensional features. However, the complex characteristics of odor are essentially manifested as synergistic or antagonistic response effects of multi-component odor molecules on the sensor array. Ignoring this correlation leads to insufficient utilization of feature information, thus affecting the inference accuracy of odor recognition. Second, when electronic nose systems face sensor drift, performance degradation, or environmental interference, traditional models based on the independent feature assumption experience significant performance degradation and lack effective robustness mechanisms, resulting in decreased odor recognition accuracy. Therefore, improving the odor recognition accuracy of electronic noses has become an urgent problem to be solved.

[0023] Example 1 This application provides a graph model-based odor feature reasoning method, which is applied to electronic noses and effectively solves the problem of low odor recognition accuracy of electronic noses. Figure 1 This is a schematic diagram of the odor feature reasoning method based on a graph model provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes the following steps: S100: Acquire raw response data of the target odor based on multiple gas sensors.

[0024] In this embodiment, the electronic nose collects raw response data of the target odor using multiple gas sensors. Different gas sensors have broad-spectrum and cross-response characteristics to different volatile molecules in the target odor, such as resistance changes. By using multiple gas sensors, the odor can be perceived from different dimensions, forming multi-dimensional and complementary raw response data for the target odor.

[0025] S200. Perform data preprocessing and feature extraction on the raw response data to obtain multiple odor feature data.

[0026] In this embodiment of the application, structured odor feature data can be obtained by preprocessing and extracting features from the raw response data. The data preprocessing includes the following steps: S210. The original response data is denoised using the wavelet threshold denoising method to obtain the initial response data.

[0027] As an optional implementation of this application, wavelet thresholding denoising can be used to denoise the original response data. Wavelet transform decomposes the signal in the original response data of the gas sensor into approximation coefficients and detail coefficients at different scales. The approximation coefficients are low-frequency and used to characterize trends, while the detail coefficients are high-frequency and contain noise. Since the odor response signal energy is mainly concentrated in the low-frequency and part of the mid-frequency range, while random noise is mostly distributed in the high-frequency detail coefficients, a soft thresholding process is applied to the detail coefficients: coefficients with amplitudes less than a preset threshold are set to zero, while those greater than a preset threshold are reduced to zero. This suppresses noise while preserving key abrupt changes such as the response start and peak values. The preset threshold can be determined using a general threshold or an adaptive method based on actual conditions.

[0028] Understandably, since the signals of gas sensors are susceptible to random environmental noise, wavelet transform can expand the signal in the original response data at different frequency scales, separating the main components of the signal from the noise. Simultaneously, soft thresholding can effectively remove small-amplitude noise coefficients while retaining important coefficients that characterize signal abrupt changes at the start of the response, thus achieving a faithful noise reduction effect.

[0029] S220. Normalize the initial response data based on the minimum-maximum normalization method to obtain normalized data.

[0030] Optionally, the min-max normalization method can be used to linearly map each initial response data after denoising to [0,1]. For each gas sensor, its minimum and maximum values ​​in each measurement are calculated independently. The minimum value is subtracted from the initial response data, and then divided by the difference between the maximum and minimum values ​​to obtain the normalized data. This operation eliminates the differences in dimensions and response amplitudes between sensors while preserving the relative change pattern of the original response, avoiding damage to the temporal structure information, and providing a numerically stable and physically consistent input basis for cross-sensor correlation modeling of graph neural networks.

[0031] S230. Extract statistical features and time-frequency domain features from the normalized data to obtain multiple odor feature data.

[0032] In this embodiment, statistical features and time-frequency domain features are extracted from normalized data to form structured odor feature data. Since the point information density of the original response data is low, statistical features and time-frequency domain features are extracted. Statistical features include, but are not limited to, mean, standard deviation, skewness, kurtosis, maximum value, minimum value, rise time, and recovery time, used to characterize the overall intensity, volatility, asymmetry, and dynamic characteristics of the response. Time-domain features include, but are not limited to, initial slope, steady-state slope, and area under the response curve, used to characterize adsorption and desorption rates and total amounts. Frequency-domain features include, but are not limited to, the dominant frequency amplitude, energy entropy, and spectral center obtained after FFT or wavelet packet transform, used to reflect the oscillation mode and complexity of the response.

[0033] Based on this, the original response data can be compressed into physically interpretable vectors, significantly improving the signal-to-noise ratio and discriminability, while also meeting the modeling requirements in graph models and enabling accurate modeling of mutual information correlation between features.

[0034] S300: Treat each gas sensor as a node and construct a sensor relationship graph model based on the correlation information between each gas sensor.

[0035] In this embodiment of the application, a sensor relationship graph model can be constructed based on prior knowledge and data-driven approaches, specifically including the following steps: S310, each gas sensor is used as a target node.

[0036] S320. Determine the target edges between each target node based on the sensing mechanism and sensitivity spectrum overlap of each gas sensor.

[0037] Optionally, the target edge can be determined based on the sensing mechanism and the overlap of the sensing spectra of each gas sensor. For example, based on physicochemical principles, two gas sensors that are both sensitive to alcohols have overlapping sensing spectra. Therefore, when detecting the odor of alcohols, they cooperate with each other, and the edge formed by these two gas sensors can be used as the target edge.

[0038] S330. Calculate the correlation coefficient of the raw response data of the two gas sensors corresponding to the target edge to obtain the initial weight of the target edge.

[0039] In this embodiment, the functional coupling strength between sensors is quantified by calculating the correlation coefficient of the raw response data. For each target odor sample, the steady-state response value or time-series feature vector of each gas sensor is extracted, and the correlation coefficient between two gas sensors is calculated to measure linear correlation, or mutual information is calculated to capture nonlinearity and higher-order dependencies, thereby obtaining the initial weights of each target edge.

[0040] S340. Assign initial weights to the target edge to obtain the sensor relationship graph model.

[0041] In this embodiment of the application, the initial weights of all target edges are assigned to the target edges formed by the corresponding target nodes to obtain the sensor relationship graph model.

[0042] Based on this, the sensor relationship graph model understands the inherent connections between the sensor hardware itself. When a gas sensor drifts or fails, the model can compensate for the information through other gas sensors that are strongly correlated with that gas sensor, which greatly improves the robustness of the electronic nose system.

[0043] S400. Using each odor feature data as a node, construct an odor feature map model based on the correlation between each odor feature data.

[0044] In this embodiment of the application, constructing the odor feature map model includes the following steps: S410, Use each odor feature data as the target node.

[0045] In this embodiment, structured odor feature data, such as response peak, recovery time and frequency domain energy ratio, can be used as target nodes, and a knowledge graph can be constructed based on their inherent statistical dependencies.

[0046] S420. Calculate the correlation degree between each target node. The correlation degree is one of the mutual information coefficient, Pearson correlation coefficient and Spearman rank correlation coefficient.

[0047] S430. If the correlation between two target nodes is greater than or equal to a preset threshold, then construct the target edge and obtain the odor feature map model.

[0048] Based on this, the odor feature map model aggregates neighborhood features by utilizing the complex dependencies between features, learns joint pattern representations, breaks through the erroneous assumption of feature independence in traditional methods, and significantly improves the discriminative power and generalization ability for subtle odor differences.

[0049] S500: Input the raw response data, sensor relationship graph model, odor feature graph model and various odor feature data into the graph neural network for training to obtain the odor feature inference model.

[0050] In this embodiment, a dual-channel heterogeneous graph neural network collaborative architecture is adopted to achieve joint modeling of sensor physical constraints and odor semantic association. The sensor relationship graph model and odor feature graph model, along with the corresponding original response data and odor feature data, are input into the graph neural network for training to obtain an odor feature inference model. The specific steps include the following: S510. Input the sensor relationship graph model and the raw response data into the first graph neural network for processing to obtain the first high-order features.

[0051] In this embodiment, the first graph neural network uses a sensor relationship graph as its topology and takes the original response data of each gas sensor as the initial feature of the node. Through multi-layer message passing, it aggregates the response information of neighboring sensors to generate a first high-order feature for hardware perception enhancement, representing an implicit compensation mechanism. For example, when a gas sensor drifts, its feature is corrected by the robust response of a strongly correlated gas sensor.

[0052] S520. Input the odor feature map model and each odor feature data into the second graph neural network for processing to obtain the second high-order features.

[0053] In this embodiment, the second graph neural network is based on an odor feature graph model, with structured odor feature data as node attributes, learns nonlinear collaborative patterns between features, and outputs a second higher-order feature with enhanced semantic discrimination.

[0054] S530, fuse the first high-order feature and the second high-order feature to obtain the target fused feature.

[0055] As an optional implementation method of this application, one or more of the following methods can be used to dynamically fuse the first high-order feature and the second high-order feature to obtain the target fused feature, thereby avoiding dimensional redundancy and semantic conflicts caused by simple concatenation.

[0056] S540. Input the target fusion features into the target model for training to obtain the odor feature inference model.

[0057] As an optional implementation of this application, the target model is a classifier model or a regressor model. The target fusion features are input into the classifier or regressor for training to obtain an odor feature inference model.

[0058] Based on this, the odor feature reasoning model uses dual-granularity graph neural network collaborative learning, which combines physical interpretability and data-driven discriminative power. The dual-path representation is fused with attention to avoid information redundancy and effectively improve the accuracy of odor recognition.

[0059] S600. Input the odor feature map model and each odor feature data into the odor feature inference model to obtain the feature inference result of the target odor.

[0060] In this embodiment, the odor feature graph model and corresponding odor feature data are input into the odor feature inference model. Multi-hop message passing is performed based on graph topology. For example, if the peak response node and the dominant frequency node in a target odor sample are strongly correlated and have high edge weights, their features will mutually reinforce each other, forming a high-order specific representation of aldehyde odors. The model outputs the feature inference result of the target odor. This result can be a discriminative embedding vector, decoded by a lightweight classifier, providing not only category or concentration regression values ​​but also inverting key contributing features and highly activated subgraphs in the graph, enabling interpretable inference.

[0061] As a further implementation of the embodiments of this application, Figure 2 This is a second schematic diagram of the flowchart of the odor feature reasoning method based on graph model provided in this application embodiment, as shown below. Figure 2 As shown, the method also includes the following steps: S700: Obtain new raw response data, and perform incremental training and dynamic optimization of the odor feature inference model based on the new raw response data.

[0062] For example, when new raw response data flows in, the correlation coefficients or mutual information are recalculated using a sliding window on the sensor relationship graph, and the weights of target edges are fine-tuned to maintain hardware topology stability. On the odor feature graph, the mutual information between features is re-estimated based on the mixed datasets of new and old data, and target edges are dynamically pruned or added, allowing the graph structure to evolve with the odor distribution. Finally, an elastic weight solidification or memory replay strategy is adopted to fine-tune only the last two layers and the fusion module while freezing the parameters of the backbone graph neural network, completing lightweight incremental training.

[0063] Based on this, when changes in ambient temperature and humidity cause sensor correlation shifts, the weights are recalculated to achieve online calibration of the graph structure. The dynamic model update mechanism ensures long-term deployment reliability through incremental dual-graph adaptive reconstruction.

[0064] To verify the effectiveness of the graph model-based odor feature reasoning method provided in this application, the method provided in this application is used to identify the quality grade of tobacco raw materials, as follows: First, an electronic nose device is used. Its sensor array consists of six different metal oxide semiconductor gas sensors, with sensitive materials including tungsten trioxide, tin dioxide, and zinc oxide, to cover the detection of key volatile components in tobacco such as nitrogen oxides, hydrocarbons, alcohols, and aldehydes. Three known quality grades of tobacco samples are prepared: Grade A, Grade B, and Grade C, with one hundred samples prepared for each grade, totaling three hundred samples. The tobacco samples are placed in a sealed gas chamber and headspace equilibrated for one hour. Then, the odor of the tobacco samples is steadily delivered to the electronic nose sensor array through a gas path system. The resistance response curves of each gas sensor are collected at a set frequency to obtain raw response data.

[0065] The collected raw data was then preprocessed using wavelet thresholding to remove noise. High-frequency noise coefficients were filtered using a soft threshold and normalized to map the values ​​of each gas sensor response curve to the [0,1] interval. Eight time-domain features, including steady-state response value, maximum response value, and recovery time, were extracted from each preprocessed curve to form a 48-dimensional feature vector, thus obtaining odor feature data.

[0066] A sensor relationship graph model is then constructed, with six physical sensors as target nodes. Edge connections are built by calculating the Pearson correlation coefficients between sensor pairs in all sample data. The weight of each target edge is the absolute value of the correlation coefficient. The sensor relationship graph serves to establish a collaborative sensing model within the sensor array, thereby enabling an understanding of the inherent correlation patterns between sensors. By explicitly modeling the relationships between sensors, the sensor relationship graph allows the electronic nose to compensate and correct for performance drift or temporary malfunctions of individual gas sensors based on information from other strongly correlated gas sensors in the graph. This significantly improves the stability and fault tolerance of the entire electronic nose system.

[0067] After constructing the sensor relationship graph, an odor feature graph model was built. This model uses 48 extracted feature vectors as nodes, and edge connections are constructed by calculating the mutual information values ​​between feature pairs, forming a knowledge network representing the complex relationships between features. The purpose of building the odor feature graph model is to overcome the limitations of the feature independence assumption at the beginning of a line. The odor feature graph enables the model to identify and learn joint patterns between features, thereby more comprehensively capturing the essential characteristics of tobacco odor and enhancing the electronic nose system's ability to discriminate subtle quality differences.

[0068] Finally, the two graphical models are input into two independent graphical attention networks for processing, outputting 32-dimensional and 64-dimensional feature vectors. These are then concatenated to form a 96-dimensional fused feature vector, which is finally used for classification decision through a fully connected network to obtain the feature inference result. Experimental analysis and demonstration show that the accuracy of identifying the quality grade of tobacco raw materials reaches approximately 97%, significantly improving the identification accuracy and robustness of the electronic nose, and outperforming traditional methods and single-view graphical models.

[0069] The odor feature reasoning method based on graph models provided in this application constructs a dual-granularity graph model consisting of a sensor relationship graph model and an odor feature graph model. This explicitly encodes the physical cross-sensitivity between sensors and the nonlinear dependency between features, enabling the graph neural network to learn more discriminative high-order joint features and significantly improving the accuracy of odor recognition, classification, and regression.

[0070] Example 2 Based on the same technical concept as Embodiment 1 above, this application provides a graph model-based odor feature reasoning system applied to electronic nose devices. Figure 3 This is a schematic diagram of the structure of the odor feature reasoning system based on a graph model provided in this application embodiment, as shown below. Figure 3 As shown, the graph model-based odor feature inference system 200 includes: The data acquisition module 210 is used to acquire raw response data of the target odor based on multiple gas sensors.

[0071] The data processing module 220 is used to preprocess the raw response data and extract features to obtain multiple odor feature data.

[0072] The first construction module 230 is used to construct a sensor relationship graph model by taking each gas sensor as a node and based on the correlation information between each gas sensor.

[0073] The second construction module 240 is used to construct an odor feature graph model by using each odor feature data as a node and based on the correlation between each odor feature data.

[0074] The model training module 250 is used to input the raw response data, sensor relationship graph model, odor feature graph model and various odor feature data into the graph neural network for training to obtain the odor feature inference model.

[0075] The feature reasoning module 260 is used to input the odor feature map model and various odor feature data into the odor feature reasoning model to obtain the feature reasoning result of the target odor.

[0076] The odor feature reasoning system based on graph models provided in this application enables graph neural networks to learn more essential and discriminative odor feature representations, thereby significantly improving the accuracy of odor classification and regression, and enhancing the stability and fault tolerance of electronic nose devices.

[0077] It is understood that the implementation method of the odor feature reasoning method based on graph model in Embodiment 1 above is also applicable to this embodiment and can achieve the same technical effect, so it will not be described again here.

[0078] Example 3 Based on the same concept, this application also provides an electronic device. Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 4As shown, the electronic device 300 may include a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute the steps of the graph model-based odor feature reasoning method as described in the above embodiments. For example, this includes: S100: Acquire raw response data of the target odor based on multiple gas sensors; S200. Perform data preprocessing and feature extraction on the raw response data to obtain multiple odor feature data; S300: Using each gas sensor as a node, construct a sensor relationship graph model based on the correlation information between each gas sensor; S400. Using each odor feature data as a node, construct an odor feature map model based on the correlation between each odor feature data. S500: Input the raw response data, sensor relationship graph model, odor feature graph model and various odor feature data into the graph neural network for training to obtain the odor feature inference model; S600. Input the odor feature map model and each odor feature data into the odor feature inference model to obtain the feature inference result of the target odor.

[0079] The processor 310 can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0080] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] The memory 330 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0082] Example 4 Based on the same concept, embodiments of this application also provide a computer-readable storage medium storing a computer program. This computer program includes at least one piece of code executable by a master control device to control the master control device to implement the steps of the graph model-based odor feature reasoning method as described in the above embodiments. For example, it includes: S100: Acquire raw response data of the target odor based on multiple gas sensors; S200. Perform data preprocessing and feature extraction on the raw response data to obtain multiple odor feature data; S300: Using each gas sensor as a node, construct a sensor relationship graph model based on the correlation information between each gas sensor; S400. Using each odor feature data as a node, construct an odor feature map model based on the correlation between each odor feature data. S500: Input the raw response data, sensor relationship graph model, odor feature graph model and various odor feature data into the graph neural network for training to obtain the odor feature inference model; S600. Input the odor feature map model and each odor feature data into the odor feature inference model to obtain the feature inference result of the target odor.

[0083] Based on the same technical concept, this application also provides a computer program, which, when executed by a main control device, is used to implement the above-described method embodiments.

[0084] The computer program may be stored, in whole or in part, on a computer-readable storage medium packaged with the processor, or in part or in whole on a memory not packaged with the processor.

[0085] Based on the same technical concept, this application also provides a processor for implementing the above-described method embodiments. The processor can be a chip.

[0086] In summary, the odor feature inference method, system, device, and storage medium based on graph models provided in this application explicitly encode the physical cross-sensitivity between sensors and the nonlinear dependencies between features by constructing a dual-granularity graph model consisting of a sensor relationship graph model and an odor feature graph model. This enables the graph neural network to learn more discriminative high-order joint features, significantly improving the accuracy of odor recognition, classification, and regression. Simultaneously, the sensor relationship graph model introduces hardware-level physical constraints. When individual sensors drift or malfunction, the sensor relationship graph model can rely on strongly correlated nodes in the graph for information compensation, significantly improving the system's fault tolerance and long-term stability.

[0087] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0088] The embodiments described above are merely examples of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A graph model-based odor feature reasoning method, characterized in that, The method includes: Raw response data of the target odor is acquired based on multiple gas sensors; The raw response data is preprocessed and feature extracted to obtain multiple odor feature data. Each of the gas sensors is used as a node, and a sensor relationship graph model is constructed based on the correlation information between the gas sensors. Each of the odor feature data is used as a node, and an odor feature graph model is constructed based on the correlation between the odor feature data. The original response data, the sensor relationship graph model, the odor feature graph model, and each of the odor feature data are input into a graph neural network for training to obtain an odor feature inference model. The odor feature map model and each of the odor feature data are input into the odor feature inference model to obtain the feature inference result of the target odor.

2. The odor feature reasoning method based on graph model according to claim 1, characterized in that, The step of inputting the original response data, the sensor relationship graph model, the odor feature graph model, and each of the odor feature data into a graph neural network for training to obtain an odor feature inference model includes: The sensor relationship graph model and the raw response data are input into a first graph neural network for processing to obtain the first high-order feature; The odor feature map model and each of the odor feature data are input into a second graph neural network for processing to obtain a second high-order feature. The first high-order feature and the second high-order feature are fused to obtain the target fused feature; The target fusion features are input into the target model for training to obtain the odor feature inference model.

3. The odor feature reasoning method based on graph model according to claim 2, characterized in that, The fusion of the first high-order feature and the second high-order feature is achieved by one or more of the following methods: vector concatenation fusion, weighted summation fusion, and attention-based fusion. The target model is a classifier model or a regressor model.

4. The odor feature reasoning method based on graph model according to claim 1, characterized in that, The step of using each of the gas sensors as nodes and constructing a sensor relationship graph model based on the correlation information between the gas sensors includes: Each of the aforementioned gas sensors is designated as a target node; The target edges between each target node are determined based on the sensing mechanism and the overlap of the sensing spectra of each gas sensor. Calculate the correlation coefficient of the original response data of the two gas sensors corresponding to the target edge to obtain the initial weight of the target edge; The initial weights are assigned to the target edge to obtain the sensor relationship graph model.

5. The odor feature reasoning method based on graph model according to claim 1, characterized in that, The step of using each of the odor feature data as nodes and constructing an odor feature map model based on the correlation between the odor feature data includes: Each of the aforementioned odor feature data is used as a target node; Calculate the correlation degree between each of the target nodes, wherein the correlation degree is one of the mutual information coefficient, Pearson correlation coefficient and Spearman rank correlation coefficient; If the correlation between two target nodes is greater than or equal to a preset threshold, then a target edge is constructed to obtain the odor feature map model.

6. The odor feature reasoning method based on graph model according to claim 1, characterized in that, The method further includes: New raw response data is obtained, and the odor feature inference model is incrementally trained and dynamically optimized based on the new raw response data.

7. The odor feature reasoning method based on graph model according to claim 1, characterized in that, The raw response data is preprocessed and feature extracted to obtain multiple odor feature data, including: The original response data is denoised using the wavelet threshold denoising method to obtain the initial response data. The initial response data is normalized using the minimum-maximum normalization method to obtain normalized data. Statistical features and time-frequency domain features are extracted from the normalized data to obtain multiple odor feature data.

8. An odor feature reasoning system based on a graph model, characterized in that, The system includes: The data acquisition module is used to acquire raw response data of the target odor based on multiple gas sensors; The data processing module is used to preprocess and extract features from the raw response data to obtain multiple odor feature data. The first construction module is used to construct a sensor relationship graph model based on the correlation information between each gas sensor, using each gas sensor as a node. The second construction module is used to construct an odor feature graph model by taking each of the odor feature data as nodes and based on the correlation between each of the odor feature data. The model training module is used to input the original response data, the sensor relationship graph model, the odor feature graph model, and each of the odor feature data into the graph neural network for training to obtain an odor feature inference model. The feature reasoning module is used to input the odor feature map model and each of the odor feature data into the odor feature reasoning model to obtain the feature reasoning result of the target odor.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the odor feature reasoning method based on the graph model as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the odor feature reasoning method based on the graph model as described in any one of claims 1-7.

Images