Smell feature recognition method

Abstract

The invention relates to the technical field of odor recognition, and discloses an odor feature recognition method, which comprises the following steps of: acquiring an odor data set by using an odor sensor array consisting of a plurality of odor sensors, generating corresponding odor feature sequences through feature extraction, and combining odor features of the odor feature sequences in the same unit time, so as to obtain an odor feature recognition result. According to the method, the dynamic change of the smell is accurately captured and described through the smell sensor array and the feature extraction technology, the dynamic change of the smell is accurately captured and described, the detection accuracy is improved, and the detection accuracy is improved. The accuracy and reliability of smell recognition are improved by combining time sequence analysis, the adaptive capacity of the system in a complex smell environment is enhanced through feature combination and time sequence correlation analysis, and the problem that in the prior art, the data analysis depth of a smell sensor array on complex smells is insufficient is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of odor recognition, and in particular to a method for recognizing odor characteristics. Background Art

[0002] In traditional odor recognition technology, a single type of odor sensor is often only sensitive to a certain type of odor or certain specific odor components, and is unable to fully and accurately identify complex odor combinations. This makes the recognition effect of a single sensor unsatisfactory in complex environments. In this case, an odor sensor array composed of multiple odor sensors can be used to replace a single sensor. However, in the prior art, the use of odor sensor arrays for complex odor detection lacks in-depth analysis of data, resulting in insufficient accuracy in odor recognition and restoration. Summary of the invention

[0003] The purpose of the present invention is to provide an odor feature recognition method, aiming to solve the problem of insufficient depth of data analysis of complex odors by odor sensor arrays in the prior art.

[0004] The present invention is implemented in this way. In a first aspect, the present invention provides a method for identifying odor characteristics, comprising: Based on an odor sensor array composed of a plurality of odor sensors, the odor data of the odor to be detected is continuously collected and recorded to obtain an odor data set corresponding to the odor to be detected by the odor sensor array; wherein the odor data set includes a plurality of odor data sequences, each of which corresponds to each of the odor sensors in the odor sensor array, and the odor data sequence includes the odor data collected by the odor sensors arranged in chronological order; Extracting odor features according to each of the odor data sequences in the odor data set to obtain an odor feature sequence corresponding to each of the odor data sequences; wherein the odor feature sequence includes odor features arranged in chronological order, and the odor features are used to describe the specific conditions of the odor fed back by the odor data; Combining the odor features of each of the odor feature sequences within the same unit time according to the time correspondence to obtain a combined feature sequence; wherein the combined feature sequence includes combined features arranged in sequence according to the time sequence, and the combined features are used to describe the specific conditions of the odors commonly fed back by each of the odor data; Based on the combined feature sequence, the correlation of the changes in the combined features of the odor to be tested in time sequence is analyzed and processed to obtain the temporal correlation relationship of the combined feature sequence, and the combined feature sequence is subjected to feedback analysis and processing of the odor to be tested according to the temporal correlation relationship to obtain the odor recognition characteristics of the odor to be tested.

[0005] The present invention provides a method for identifying odor characteristics, which has the following beneficial effects: The present invention uses an odor sensor array composed of several odor sensors to obtain an odor data set, generates a corresponding odor feature sequence through feature extraction, combines the odor features of each odor feature sequence in the same unit time, generates a combined feature sequence and analyzes the change correlation in time sequence, obtains a time series correlation relationship to perform feedback analysis on the combined feature sequence, and obtains the odor recognition feature of the odor to be tested. The method uses the odor sensor array and feature extraction technology to accurately capture and describe the dynamic changes of odor, combines time series analysis to improve the accuracy and reliability of odor recognition, and enhances the adaptability of the system in complex odor environments through feature combination and time series correlation analysis, thereby solving the problem of insufficient depth of data analysis of complex odors by the odor sensor array in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a schematic diagram of the steps of an odor feature recognition method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0007] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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.

[0008] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0009] Reference Figure 1 As shown, a preferred embodiment of the present invention is provided.

[0010] In a first aspect, the present invention provides a method for identifying odor characteristics, comprising: S1: Continuously collecting and recording odor data of an odor to be detected based on an odor sensor array composed of a plurality of odor sensors, so as to obtain an odor data set corresponding to the odor to be detected by the odor sensor array; wherein the odor data set includes a plurality of odor data sequences, each of which corresponds to each of the odor sensors in the odor sensor array, and the odor data sequence includes odor data collected by the odor sensors arranged in chronological order; S2: extracting odor features according to each of the odor data sequences in the odor data set, so as to obtain an odor feature sequence corresponding to each of the odor data sequences; wherein the odor feature sequence includes odor features arranged in chronological order, and the odor features are used to describe the specific conditions of the odor fed back by the odor data; S3: combining the odor features of each of the odor feature sequences within the same unit time according to the time correspondence to obtain a combined feature sequence; wherein the combined feature sequence includes combined features arranged in sequence according to the time sequence, and the combined features are used to describe the specific conditions of the odors commonly fed back by each of the odor data; S4: Analyze and process the correlation of changes in the time sequence of the combination characteristics of the odor to be tested based on the combination characteristic sequence to obtain the temporal correlation relationship of the combination characteristic sequence, and perform feedback analysis and processing of the odor to be tested on the combination characteristic sequence according to the temporal correlation relationship to obtain the odor recognition characteristics of the odor to be tested.

[0011] Specifically, in step S1 of the embodiment provided by the present invention, the odor to be tested is collected in real time through an odor sensor array, the odor data of each sensor is recorded, and each odor data is arranged in chronological order to generate an odor data sequence, and the odor data sequence containing each sensor data is collectively regarded as an odor data set.

[0012] It can be understood that the use of multiple sensors to provide comprehensive collection of odors increases the accuracy and coverage of the data. Continuous odor data collection can capture the dynamic information of odor changes over time and ensure real-time monitoring of odors. The odor data sequence arranged in chronological order helps to analyze the changes in odors over time, thereby improving the reliability of the data and the effectiveness of the analysis.

[0013] Specifically, in step S2 of the embodiment provided by the present invention, the odor data is preprocessed, and odor features, such as peak intensity, mean, variance, spectral features, etc., are extracted from the preprocessed data. Algorithms such as Fourier transform or principal component analysis are used to extract features, and key parameters related to the odor are extracted to obtain an odor feature sequence containing the odor features.

[0014] More specifically, the extracted odor features are sorted in chronological order to form an odor feature sequence, ensuring that the data in the feature sequence is arranged in chronological order to obtain an ordered odor feature sequence that can reflect the temporal variation characteristics of the odor data.

[0015] It can be understood that odor feature extraction can provide a detailed description of odor data and reveal the specific conditions of the odor. The odor feature sequence arranged in chronological order facilitates time series analysis and identifies odor change trends and patterns. Preprocessing and feature extraction improve data quality and make odor identification and classification more accurate.

[0016] Specifically, in step S3 of the embodiment provided by the present invention, each odor feature sequence is synchronized on the time axis to ensure that all feature data points are aligned on the same time unit, the timestamps of each odor feature sequence are aligned, and missing values ​​and interpolation are processed to ensure the consistency of the time axis, thereby obtaining each odor feature sequence after time axis synchronization.

[0017] More specifically, within the same unit time, the odor features in each odor feature sequence are combined and processed, and aggregation operations such as vectorization, concatenation or weighted averaging are performed on the odor features on each time unit to obtain the combined features under each time unit and form a combined feature sequence.

[0018] More specifically, the combined features are sorted in chronological order to form a new combined feature sequence, ensuring that the combined feature sequence is arranged in chronological order and the features at each time point are complete, resulting in an ordered combined feature sequence that contains comprehensive information of all odor features at each time unit.

[0019] It can be understood that combining feature sequences provides a more comprehensive and detailed description of odors by integrating multiple odor features. By combining features on a unified time unit, it can better perform consistency analysis and identify the global characteristics and changing trends of odors. Combining features can reduce the errors caused by a single feature and improve the accuracy of odor identification and classification. Combining feature sequences provides a high-quality data foundation for subsequent pattern recognition and machine learning, enhancing the depth and breadth of analysis.

[0020] Specifically, in step S4 of the embodiment provided by the present invention, a time series analysis is performed on the combined feature sequence to find the regularity and pattern of the change of the odor feature over time, and the change of the combined feature sequence over time is expressed through a time correlation relationship.

[0021] More specifically, the combined feature sequence is based on the odor characteristics of the odor data collected by the odor sensor array, and the odor sensor array includes several different odor sensors. The performance of these odor sensors has cross-selectivity, that is, different odor sensors have different sensitivities and adaptabilities to different gas components. By combining and analyzing the odor characteristics of the odor data collected by each odor sensor in the odor sensor array, the gas components of the odor to be tested can be evaluated and analyzed more clearly.

[0022] More specifically, different gas components have different characteristics changing over time in space, which results in different changes in the odor data collected by the gas sensor array over time when the odor data of different gas components are continuously monitored by the gas sensor array.

[0023] Therefore, based on the above technical principles, the combined feature sequence can be analyzed from at least two analysis dimensions. The first analysis dimension is the array dimension, that is, multiple cross-selective odor sensors in the odor sensor array analyze the odor to be tested from different sensitive angles within unit time to obtain a comprehensive and detailed analysis of the odor to be tested. The second analysis dimension is the time dimension, that is, reverse analysis and processing is performed on the changing characteristics of various gas components in the odor to be tested within unit time, so as to analyze the mapping relationship between the actual changes in the odor to be tested and the theoretical changing characteristics of various gas components within unit time, so as to determine the gas components in the odor to be tested.

[0024] More specifically, starting from the above-mentioned at least two analysis dimensions, the analysis results are combined, mutually verified and jointly inferred to obtain the odor recognition characteristics of the odor to be tested, thereby forming a complete odor characteristic description.

[0025] It can be understood that by conducting feature analysis on multiple analytical dimensions of the odor to be tested, the odor to be tested can be analyzed more comprehensively and accurately, thereby obtaining the specific form of the gas components of the odor to be tested, and at the same time, the temporal change pattern of the odor to be tested can be accurately captured, providing strong support for subsequent odor restoration work.

[0026] The present invention provides a method for identifying odor characteristics, which has the following beneficial effects: The present invention uses an odor sensor array composed of several odor sensors to obtain an odor data set, generates a corresponding odor feature sequence through feature extraction, combines the odor features of each odor feature sequence in the same unit time, generates a combined feature sequence and analyzes the change correlation in time sequence, obtains a time series correlation relationship to perform feedback analysis on the combined feature sequence, and obtains the odor recognition feature of the odor to be tested. The method uses the odor sensor array and feature extraction technology to accurately capture and describe the dynamic changes of odor, combines time series analysis to improve the accuracy and reliability of odor recognition, and enhances the adaptability of the system in complex odor environments through feature combination and time series correlation analysis, thereby solving the problem of insufficient depth of data analysis of complex odors by the odor sensor array in the prior art.

[0027] Preferably, the step of continuously collecting and recording odor data of the odor to be detected based on an odor sensor array composed of a plurality of odor sensors to obtain an odor data set corresponding to the odor to be detected by the odor sensor array includes: S11: analyzing and processing the odor signal of the odor to be detected by each of the odor sensors in the preset odor sensor array respectively, so as to obtain odor data corresponding to the odor to be detected; S12: generating a time tag corresponding to the odor data according to the time node corresponding to the odor data, and binding the time tag to the odor data; S13: Arranging and processing each of the odor data collected by the odor sensor according to the time tag bound to the odor data to obtain an odor data sequence of the odor sensor; S14: taking the odor data sequences of the odor sensors together as the odor data set of the odor sensor array.

[0028] Specifically, each odor sensor in the odor sensor array is used to collect and analyze the odor signal of the odor to be tested. Each odor sensor preprocesses the collected odor signal (such as signal filtering and denoising) and extracts features to generate preliminary odor data, thereby obtaining preliminary odor data of the odor to be tested by each odor sensor.

[0029] More specifically, according to the time node of odor data collection, a corresponding time tag is generated for each odor data, and the time tag is bound to the odor data, the timestamp when each odor data is collected is obtained, a time tag is generated, and it is bound to the corresponding odor data (such as through the time field in the data structure) to obtain the odor data with a time tag.

[0030] More specifically, according to the time label, the odor data collected by each odor sensor is arranged in time sequence to form an odor data sequence. The odor data of each odor sensor is sorted according to the time label to deal with the problem of time interval consistency. If necessary, interpolation processing is performed to ensure the consistency of the time axis to obtain the odor data sequence of each odor sensor.

[0031] More specifically, the odor data sequences of each odor sensor are combined to form a complete odor data set, the odor data sequences of all odor sensors are aligned along the time axis, and all odor data sequences are combined to form an odor data set of the odor sensor array, thereby obtaining the odor data set of the odor sensor array.

[0032] It can be understood that by generating and binding time tags, the precise synchronization of odor data in time is ensured, which is crucial for time series analysis and identification of dynamic characteristics of odors. The arrangement processing and time consistency processing of odor data ensure the integrity and continuity of the data, reduce the impact of data missing and inconsistency on subsequent analysis, and combine the data sequences of various odor sensors into an odor data set, which provides a comprehensive description of the odor to be tested and improves the accuracy of odor feature identification. The systematic steps make the collection, processing and organization of odor data more efficient, providing a solid data foundation for subsequent odor identification and analysis.

[0033] Preferably, the step of performing odor feature extraction processing on each of the odor data sequences in the odor data set to obtain an odor feature sequence corresponding to each of the odor data sequences comprises: S21: Obtaining the specification parameters of the odor sensor corresponding to the odor data sequence, and performing simulation analysis on the odor signal acquisition performance of the odor sensor according to the specification parameters to obtain the performance characteristic distribution of the odor sensor; wherein the performance characteristic distribution is used to describe the performance status of the odor sensor when collecting odor signals of different forms; S22: performing feature extraction processing on each odor data in the odor data sequence in turn according to the performance feature distribution, so as to obtain the odor feature corresponding to each odor data in the odor data sequence; S23: Arranging the odor features corresponding to the individual odor data according to the arrangement order of the individual odor data in the odor data sequence to obtain the odor feature sequence.

[0034] Specifically, the specification parameters of the odor sensor, including its sensitivity, response time, detection range, noise level, stability, etc., are collected, and relevant parameters are obtained from the technical manual or product manual of the sensor manufacturer, or these parameters are measured through experimental calibration, so as to obtain the detailed specification parameters of the odor sensor.

[0035] More specifically, based on the acquired specifications of the odor sensor, the odor signal acquisition performance of the odor sensor is simulated and analyzed to obtain the performance characteristic distribution. A sensor model or simulation software is used to simulate the response of the sensor in different odor environments according to its specifications. The performance of the sensor under different odor concentrations, temperature and humidity conditions is analyzed, and a performance characteristic distribution graph is generated to describe the performance characteristic distribution of the odor sensor under different conditions.

[0036] More specifically, according to the performance characteristic distribution, feature extraction processing is performed on each odor data in the odor data sequence. According to the performance characteristic distribution, a suitable feature extraction algorithm (such as principal component analysis, linear discriminant analysis, wavelet transform, etc.) is selected to process each odor data and extract feature vectors, such as mean, variance, peak, spectral characteristics, etc., to obtain the odor characteristics corresponding to each odor data point.

[0037] More specifically, the odor features corresponding to each odor data are arranged in the arrangement order of the odor data sequence to generate an odor feature sequence. The extracted odor features are arranged in chronological order or data collection order to form a complete odor feature sequence, which reflects the characteristics of the odor changing over time and obtains a complete odor feature sequence.

[0038] It can be understood that by analyzing the performance characteristic distribution of the sensor, adaptive feature extraction can be performed for different forms of odor signals, thereby improving the accuracy and reliability of feature extraction. The simulation analysis and processing of sensor performance can comprehensively describe the performance of the odor sensor in different environments, providing an important reference for feature extraction. The extracted odor features are arranged in chronological order, and the odor feature sequence formed can accurately reflect the dynamic changes of the odor signal, which is helpful for subsequent odor identification and analysis. Through simulation analysis and feature extraction processing, the system can adapt to different types and brands of odor sensors, and has strong adaptability and versatility.

[0039] Preferably, the step of sequentially performing feature extraction processing on each odor data in the odor data sequence according to the performance feature distribution to obtain the odor feature corresponding to each odor data in the odor data sequence includes: S221: performing a graph conversion process on the odor data to obtain an odor graph for graphically expressing the odor data; S222: performing an analysis process on the spectrum representative vector based on the odor spectrum to determine a plurality of spectrum representative vectors in the odor spectrum, and taking the spectrum representative vectors as a representative vector set; wherein the spectrum representative vectors are used to represent the odor spectrum; S223: performing vector weight distribution processing on the representative vector set according to the performance characteristic distribution to generate vector weight values ​​of each of the graph representative vectors in the representative vector set; wherein the vector weight values ​​are used to describe the degree of confidence of the graph representative vector in providing feedback on the situation of the odor to be tested; S224: taking the representative vector set in which each of the graph representative vectors has the vector weight value as the odor feature corresponding to the odor data.

[0040] Specifically, the odor data is subjected to a graph conversion process to obtain an odor graph, and the data output by the odor sensor is preprocessed, such as denoising and smoothing, and an appropriate conversion method (such as Fourier transform, short-time Fourier transform, wavelet transform, etc.) is used to convert the odor data in the time domain into the frequency domain or other graph domains to form an odor graph, thereby obtaining an odor graph for graphically expressing the odor data.

[0041] More specifically, based on the odor map, the representative vector of the map is analyzed and processed to determine a number of representative vectors of the map, form a representative vector set, analyze the odor map, and extract the main feature vectors. For example, principal component analysis (PCA) can be used to extract the main feature components, or cluster analysis can be used to find representative data points, and select a number of representative feature vectors that can effectively describe the main features in the odor map.

[0042] More specifically, according to the performance characteristic distribution, the vector weights of the various graph representative vectors in the representative vector set are assigned, and the weight value of each graph representative vector is calculated using the performance characteristic distribution of the odor sensor. The weight value reflects the degree of confidence of each representative vector in the current odor detection result. The methods that can be used include regression analysis, Bayesian method or other statistical analysis methods to assign weight values ​​based on the performance characteristic distribution.

[0043] More specifically, a set of graph representative vectors with vector weight values ​​is used as the odor features corresponding to the odor data, and the graph representative vectors and their corresponding weight values ​​are combined to form a complete odor feature. These odor features can accurately describe the main characteristic information of the odor data, which is convenient for further analysis and identification to obtain odor features containing vector weight values.

[0044] It can be understood that through graph conversion processing, complex odor data can be converted into a graph form that is easy to analyze, which is helpful to extract the main features. By analyzing the graph representative vector, the main features in the odor graph can be effectively captured, reducing the data dimension while ensuring the representativeness of the features. Assigning weights to each graph representative vector according to the performance characteristic distribution can accurately reflect the contribution of each vector to the odor feature and improve the accuracy of feature expression. By combining graph representative vectors with weight values, the generated odor features can comprehensively reflect the main information of the odor data, which is convenient for subsequent odor analysis and identification. This method can adapt to different odor sensors and odor data types, and improves the robustness of feature extraction through weight allocation, which helps to maintain high performance in complex environments.

[0045] Preferably, the step of combining the odor features of each odor feature sequence within the same unit time according to the time correspondence to obtain a combined feature sequence includes: S31: generating a time stamp for each odor feature according to the odor data corresponding to each odor feature in the odor feature sequence; S32: combining the odor features with the same time stamp in each of the odor feature sequences to obtain the combined feature; S33: sorting the respective combination features according to the time stamps corresponding to the respective combination features to obtain the combination feature sequence; The step of combining the odor features with the same time stamp in each of the odor feature sequences to obtain the combined feature includes: S321: performing query processing and comparison processing on the time tags of each odor feature in each of the odor feature sequences, so as to summarize the odor features with the same time tags in each of the odor feature sequences into a pre-combined feature group; S322: Perform comprehensive analysis based on each of the odor features in the pre-combined feature group to weight-combine each of the graph representative vectors with the vector weight value in each of the odor features to obtain a plurality of graph representative vectors with combined weight values, and use each of the graph representative vectors with combined weight values ​​as the combined feature.

[0046] Specifically, based on the odor data corresponding to each odor feature in the odor feature sequence, a time stamp for each odor feature is generated, and a timestamp is attached to each odor feature in the odor feature sequence to ensure that each feature corresponds to a clear time point. The sensor sampling time or data recording time is used as the time stamp to obtain an odor feature sequence with a time stamp.

[0047] More specifically, the odor features with the same time stamp in each odor feature sequence are queried and compared to summarize the pre-combined feature group, all odor feature sequences are traversed, the time stamp of each feature is checked, and the odor features with the same time stamp are summarized into the same pre-combined feature group to form multiple pre-combined feature groups, each of which contains odor features collected at the same time point.

[0048] More specifically, based on the comprehensive analysis of each odor feature in the pre-combined feature group, the spectrum representative vectors with vector weight values ​​in each odor feature are weighted combined, weight analysis is performed on the odor features in each pre-combined feature group, the combination weight value of each spectrum representative vector is calculated, and the spectrum representative vectors with vector weight values ​​are combined using weighted average or other appropriate methods to generate several spectrum representative vectors with combination weight values ​​to form a combined feature.

[0049] More specifically, each combination feature is sorted according to the time stamp corresponding to each combination feature to obtain a combination feature sequence, and the combination features are sorted according to the time stamp to ensure that the combination feature sequence is arranged in chronological order and that the position of each combination feature in the combination feature sequence corresponds to its time stamp.

[0050] It can be understood that by generating time tags and comparing them, it can ensure that the odor features are combined and processed at the same time point, thereby achieving time synchronization of different odor feature sequences. By comprehensively analyzing and processing each odor feature, it is possible to effectively integrate odor features from different sensors or different sampling times, thereby improving the accuracy and completeness of feature expression. Through weighted combination processing, feature fusion can be performed based on the weight values ​​of each odor feature, thereby ensuring that the contribution of each graph representative vector is reasonably reflected, thereby enhancing the reliability of feature expression. By sorting the combined features, the resulting combined feature sequence can maintain temporal order, thereby facilitating subsequent analysis and application.

[0051] Preferably, the steps of analyzing and processing the correlation of changes in the time sequence of the combined features of the odor to be detected based on the combined feature sequence to obtain the time sequence correlation relationship of the combined feature sequence, and performing feedback analysis and processing of the odor to be detected on the combined feature sequence according to the time sequence correlation relationship to obtain the odor recognition feature of the odor to be detected include: S41: Perform time-sequential difference analysis on each representative vector set in the combined feature sequence to obtain a vector difference distribution between each representative vector set in the combined feature sequence, and use the vector difference distribution between each representative vector set in the combined feature sequence as a temporal association relationship of the combined feature sequence; wherein the vector difference distribution is used to describe the vector difference and weight difference between each of the graph representative vectors having a combined weight value in adjacent representative vector sets, wherein the vector difference is the difference between the graph representative vectors, and the weight difference is the difference between the combined weight values ​​corresponding to the same graph representative vectors; S42: performing a simulation process on the odor to be tested according to each of the combined features in the combined feature sequence, so as to obtain a predicted odor spectrum to be tested corresponding to each of the combined features; wherein the predicted odor spectrum to be tested is a graphical display of the gas component prediction of the odor to be tested based on the specific conditions of the odor fed back by the combined features; S43: According to the relationship between the property changes of various gas compositions over time, each of the vector difference distributions in the time series association relationship is subjected to reverse inference processing of the odor to be tested, so as to obtain a reverse inference map of the odor to be tested corresponding to each of the vector difference distributions; wherein the reverse inference map of the odor to be tested is a graphical display of reverse inference of the gas composition of the odor to be tested on the vector difference distribution based on the relationship between the changes of different gas compositions over time; S44: taking the combined feature, the inferred map of the odor to be tested and the reverse inferred map of the odor to be tested together as the odor recognition feature.

[0052] Specifically, a time-sequential difference analysis process is performed on each representative vector set in the combined feature sequence to obtain the vector difference distribution between each representative vector set in the combined feature sequence, and each representative vector set in the combined feature sequence is analyzed one by one, and the vector difference and weight difference between adjacent representative vector sets are calculated, which refers to the difference between the atlas representative vectors and the difference between the combination weight values ​​of the same atlas representative vectors, to obtain the vector difference distribution of the combined feature sequence.

[0053] More specifically, the vector difference distributions between the representative vector sets in the combined feature sequence are taken together as the temporal correlation relationship of the combined feature sequence, the vector differences and weight differences between all adjacent representative vector sets are sorted and recorded, and these difference distributions are integrated into an overall temporal correlation relationship to combine the temporal correlation relationship of the feature sequence.

[0054] More specifically, the odor to be tested is simulated and processed according to each combined feature in the combined feature sequence to obtain an inferred spectrum of the odor to be tested corresponding to each combined feature. The specific conditions of the odor fed back by the combined features are used to simulate the gas components, generate an inferred spectrum of the odor to be tested, and display the inferred results of the gas components.

[0055] More specifically, based on the relationship between the changing properties of various gas compositions over time, each vector difference distribution in the time series correlation relationship is reversely inferred to obtain a reverse inference map of the odor to be tested corresponding to each vector difference distribution. By utilizing the changing pattern of gas composition over time, the vector difference distribution is reversely analyzed to generate a reverse inference map of the odor to be tested, which displays the reverse inference results of the gas composition.

[0056] More specifically, different gas components have different changing properties in the air. Therefore, when different gas components are used for data collection by the odor sensor array, the resulting combination features are also different. In other words, the change differences between each combination feature are different. Therefore, by reverse analyzing the change characteristics between each combination feature in the combination feature sequence, it can be inferred that the odor to be tested changes in unit time, and then the gas component of the odor to be tested can be analyzed based on this change.

[0057] More specifically, the combined features, the inferred map of the odor to be tested, and the reverse inferred map of the odor to be tested are used together as odor recognition features, and the combined features, the inferred map, and the reverse inferred map are integrated to generate a unified odor recognition feature for final odor recognition.

[0058] It can be understood that by performing time-series difference analysis and time-series correlation generation on the combined feature sequence, the changes in odor characteristics can be captured more accurately and the accuracy of odor recognition can be improved. The combination of test odor simulation processing and reverse inference processing can provide a comprehensive odor simulation map based on actual characteristics and time-varying laws, and enhance the recognition ability of complex odors. By analyzing the vector difference distribution in time sequence, the regularity of odor characteristics changing over time can be identified, providing important time-series information support for odor recognition. The combined features, inference maps and reverse inference maps are combined to form multi-dimensional odor recognition features, provide rich odor information, and improve the accuracy and reliability of recognition. By utilizing the time-varying nature of gas composition, dynamic odor reverse inference processing can be performed, which can identify the dynamic changes of odor characteristics over time and adapt to the needs of odor recognition in complex environments. This method is suitable for a variety of odor detection and analysis scenarios, can provide efficient and accurate odor recognition solutions in practical applications, and has broad application prospects.

[0059] Preferably, the method further comprises: S51: performing gas disturbance processing on the odor to be detected by a gas disturbance device provided simultaneously with the odor sensor array, and continuously recording the gas disturbance parameters of the gas disturbance device; wherein the gas disturbance device is used to expand the gas component change of the odor to be detected per unit time, and the gas disturbance parameters are used to describe the expansion degree of the gas component change of the odor to be detected per unit time; S52: After the time series correlation of the combined feature sequence is obtained, the time series correlation is subjected to disturbance correction processing according to the gas disturbance parameter to obtain a corrected version of the time series correlation, and the corrected version of the time series correlation is used for subsequent feedback analysis processing of the odor to be tested.

[0060] Specifically, the odor to be tested is subjected to gas disturbance processing by a gas disturbance device which is arranged simultaneously with the odor sensor array, and the gas disturbance parameters of the gas disturbance device are continuously recorded. The gas disturbance device is started to perform disturbance processing on the odor to be tested, the change of gas components of the odor to be tested per unit time is expanded, and the gas disturbance parameters of the gas disturbance device are continuously recorded, including disturbance intensity, frequency and duration.

[0061] More specifically, after obtaining the vector difference distribution of the combined feature sequence, the temporal association relationship of the combined feature sequence is generated according to these difference distributions. Based on the vector difference distribution obtained in the previous step, the temporal association relationship is generated, and the vector differences and weight differences between adjacent representative vector sets are integrated into an overall temporal association relationship.

[0062] More specifically, the time series correlation is disturbed and corrected according to the gas disturbance parameters to obtain a corrected version of the time series correlation. The recorded gas disturbance parameters are used to analyze the expansion degree of the change of gas components per unit time, and the initial time series correlation is corrected. The influence of the disturbance equipment on the odor characteristics is considered, and the gas disturbance parameters are combined with the vector difference distribution to correct the time series correlation.

[0063] More specifically, a revised version of the timing correlation relationship is used to perform feedback analysis and processing of the odor to be tested. Based on the revised timing correlation relationship, the characteristics of the odor to be tested are re-analyzed to provide more accurate odor identification results. The impact of gas disturbance on odor characteristics is taken into account to obtain more precise feedback analysis results of the odor to be tested.

[0064] It is understandable that through the use of gas disturbance equipment, the changes in gas components of the odor to be tested per unit time can be expanded, thereby capturing more odor characteristic details and improving the accuracy of odor identification. The time series correlation relationship can be corrected according to the gas disturbance parameters to compensate for the impact of gas disturbance on odor characteristics, making the identification results more reliable and accurate. The disturbance correction processing can better reflect the actual situation of odor characteristics changing over time, enhance the effectiveness of time series correlation analysis, and comprehensively consider the combined features, inferred maps, reverse inferred maps and corrected time series correlation relationships to provide multi-dimensional odor identification features and improve the overall performance of the identification system.

[0065] Preferably, the step of performing disturbance correction processing on the time sequence association relationship according to the gas disturbance parameter to obtain a corrected version corresponding to the time sequence association relationship includes: S521: taking the property change relationship of various gas components over time in a non-disturbance state as the first analysis object; S522: performing corresponding difference simulation processing on the first analysis object according to the difference between the disturbance state fed back by the gas disturbance parameter and the non-disturbance state, so as to obtain the property change relationship of various gas components under the disturbance state fed back by the gas disturbance parameter, and taking the property change relationship of the gas components under the gas disturbance as the second analysis object; S523: Perform analysis processing according to the difference between the second analysis object and the first analysis object to obtain correction parameters, and perform correction processing on the timing association relationship according to the correction parameters to obtain a corrected version of the timing association relationship.

[0066] Specifically, the relationship between the property changes of various gas components over time under non-perturbation conditions is taken as the first analysis object, and the original data of the changes of various gas components over time without the influence of gas disturbance equipment are collected. These data are analyzed to establish a time change model of gas components under non-perturbation conditions.

[0067] More specifically, according to the difference between the disturbance state and the non-disturbance state fed back by the gas disturbance parameters, the first analysis object is subjected to corresponding difference simulation processing to obtain the gas disturbance parameters when the gas disturbance equipment is working, including the disturbance intensity, frequency, duration, etc. According to these disturbance parameters, the gas component change model under the non-disturbance state is simulated to predict the influence of the gas disturbance on the gas component change, and generate the property change relationship of various gas components under the disturbance state as the second analysis object.

[0068] More specifically, the difference between the second analysis object and the first analysis object is analyzed and processed to obtain correction parameters, the difference between the second analysis object and the first analysis object is compared, the influence of the disturbance on the change of gas components is determined, the influence is quantified, and the correction parameters are generated. These parameters are used to adjust the time series correlation relationship to compensate for the influence of gas disturbance.

[0069] More specifically, the timing correlation relationship is corrected according to the correction parameters to obtain a corrected version of the timing correlation relationship. The correction parameters are used to adjust the initial timing correlation relationship to correct the error caused by gas disturbance, ensuring that the corrected timing correlation relationship can accurately reflect the actual situation of the odor characteristics changing over time.

[0070] It can be understood that by conducting detailed analysis and simulation of the changes in gas components under non-perturbation and disturbance states, the generated correction parameters can accurately adjust the timing correlation relationship and improve the overall accuracy of the model. This method can dynamically adapt to the changes in odor under different disturbance parameters, so that the odor recognition system can still maintain high accuracy under various disturbance environments. By compensating for the errors caused by gas disturbances, the corrected timing correlation relationship can more accurately reflect the true characteristics of the odor and reduce errors in the recognition process. The system is compatible with multiple disturbance states, adapting to the needs of odor recognition in different environments, and has a wider range of application scenarios.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying odor characteristics, characterized in that: include: Based on an odor sensor array composed of a plurality of odor sensors, the odor data of the odor to be detected is continuously collected and recorded to obtain an odor data set corresponding to the odor to be detected by the odor sensor array; wherein the odor data set includes a plurality of odor data sequences, each of which corresponds to each of the odor sensors in the odor sensor array, and the odor data sequence includes the odor data collected by the odor sensors arranged in chronological order; Extracting odor features according to each of the odor data sequences in the odor data set to obtain an odor feature sequence corresponding to each of the odor data sequences; wherein the odor feature sequence includes odor features arranged in chronological order, and the odor features are used to describe the specific conditions of the odor fed back by the odor data; Combining the odor features of each of the odor feature sequences within the same unit time according to the time correspondence to obtain a combined feature sequence; wherein the combined feature sequence includes combined features arranged in sequence according to the time sequence, and the combined features are used to describe the specific conditions of the odors commonly fed back by each of the odor data; Based on the combined feature sequence, the correlation of the changes in the combined features of the odor to be tested in time sequence is analyzed and processed to obtain the temporal correlation relationship of the combined feature sequence, and the combined feature sequence is subjected to feedback analysis and processing of the odor to be tested according to the temporal correlation relationship to obtain the odor recognition characteristics of the odor to be tested.

2. The method for identifying odor characteristics according to claim 1, characterized in that: The steps of continuously collecting and recording odor data of the odor to be detected based on an odor sensor array composed of a plurality of odor sensors to obtain an odor data set corresponding to the odor to be detected by the odor sensor array include: Analyzing and processing the odor signal of the odor to be detected by each of the odor sensors in the pre-set odor sensor array, so as to obtain odor data corresponding to the odor to be detected; generating a time tag corresponding to the odor data according to a time node corresponding to the odor data, and binding the time tag to the odor data; Arrange and process each of the odor data collected by the odor sensor according to the time tag bound to the odor data to obtain an odor data sequence of the odor sensor; The odor data sequences of the odor sensors are collectively regarded as the odor data set of the odor sensor array.

3. The method for identifying odor characteristics according to claim 1, characterized in that: The steps of respectively performing odor feature extraction processing on each of the odor data sequences in the odor data set to obtain an odor feature sequence corresponding to each of the odor data sequences include: Obtaining the specification parameters of the odor sensor corresponding to the odor data sequence, and performing simulation analysis on the odor signal acquisition performance of the odor sensor according to the specification parameters to obtain the performance characteristic distribution of the odor sensor; wherein the performance characteristic distribution is used to describe the performance status of the odor sensor when collecting odor signals of different forms; performing feature extraction processing on each odor data in the odor data sequence in sequence according to the performance feature distribution, so as to obtain the odor feature corresponding to each odor data in the odor data sequence; The odor features corresponding to the individual odor data are arranged according to the arrangement order of the individual odor data in the odor data sequence to obtain the odor feature sequence.

4. The method for identifying odor characteristics according to claim 3, characterized in that: The step of sequentially performing feature extraction processing on each odor data in the odor data sequence according to the performance feature distribution to obtain the odor feature corresponding to each odor data in the odor data sequence comprises: Performing a graph conversion process on the odor data to obtain an odor graph for graphically expressing the odor data; Performing an analysis and processing of the spectrum representative vector based on the odor spectrum to determine a plurality of spectrum representative vectors in the odor spectrum, and taking the spectrum representative vectors as a representative vector set; wherein the spectrum representative vector is used to represent the odor spectrum; Performing vector weight distribution processing on the representative vector set according to the performance characteristic distribution to generate vector weight values ​​of each of the graph representative vectors in the representative vector set; wherein the vector weight values ​​are used to describe the degree of confidence of the graph representative vector in providing feedback on the situation of the odor to be tested; The representative vector set in which each of the graph representative vectors has the vector weight value is used as the odor feature corresponding to the odor data.

5. The method for identifying odor characteristics according to claim 4, characterized in that: The step of combining the odor features of each odor feature sequence within the same unit time according to the time correspondence to obtain a combined feature sequence includes: generating a time stamp for each of the odor features according to the odor data corresponding to each of the odor features in the odor feature sequence; Combining the odor features with the same time stamp in each of the odor feature sequences to obtain the combined feature; According to the time stamp corresponding to each of the combination features, sorting the combination features to obtain the combination feature sequence; The step of combining the odor features with the same time stamp in each of the odor feature sequences to obtain the combined feature includes: Performing query processing and comparison processing on the time stamp of each odor feature in each of the odor feature sequences, so as to summarize the odor features with the same time stamp in each of the odor feature sequences into a pre-combined feature group; Based on the comprehensive analysis processing of each of the odor features in the pre-combined feature group, each of the graph representative vectors with the vector weight value in each of the odor features is weighted combined to obtain a number of graph representative vectors with combined weight values, and each of the graph representative vectors with combined weight values ​​is collectively used as the combined feature.

6. The method for identifying odor characteristics according to claim 5, characterized in that: The steps of analyzing and processing the correlation of changes in the combined features of the odor to be detected in time sequence based on the combined feature sequence to obtain the time sequence correlation relationship of the combined feature sequence, and performing feedback analysis and processing of the odor to be detected on the combined feature sequence according to the time sequence correlation relationship to obtain the odor recognition feature of the odor to be detected include: Performing time-sequential difference analysis processing on each representative vector set in the combined feature sequence to obtain vector difference distribution between each representative vector set in the combined feature sequence, and taking the vector difference distribution between each representative vector set in the combined feature sequence as the temporal association relationship of the combined feature sequence; wherein the vector difference distribution is used to describe the vector difference and weight difference between each of the graph representative vectors having a combined weight value in adjacent representative vector sets, the vector difference is the difference between the graph representative vectors, and the weight difference is the difference between the combined weight values ​​corresponding to the same graph representative vector; According to each of the combined features in the combined feature sequence, a simulation process of the odor to be tested is performed respectively to obtain a predicted odor spectrum to be tested corresponding to each of the combined features; wherein the predicted odor spectrum to be tested is a graphical display of the gas component prediction of the odor to be tested based on the specific conditions of the odor fed back by the combined features; According to the relationship between the property changes of various gas compositions over time, each of the vector difference distributions in the time series association relationship is subjected to reverse inference processing of the odor to be tested, so as to obtain a reverse inference spectrum of the odor to be tested corresponding to each of the vector difference distributions; wherein the reverse inference spectrum of the odor to be tested is a graphical display of reverse inference of the gas composition of the odor to be tested on the vector difference distribution based on the relationship between the changes of different gas compositions over time; The combined feature, the inferred map of the odor to be tested and the reverse inferred map of the odor to be tested are collectively used as the odor recognition feature.

7. The method for identifying odor characteristics according to claim 1, characterized in that: Also includes: The odor to be detected is subjected to gas disturbance processing by a gas disturbance device provided simultaneously with the odor sensor array, and the gas disturbance parameters of the gas disturbance device are continuously recorded; wherein the gas disturbance device is used to amplify the change of the gas component of the odor to be detected per unit time, and the gas disturbance parameters are used to describe the degree of amplification of the change of the gas component of the odor to be detected per unit time; After the time series correlation of the combined feature sequence is obtained, the time series correlation is subjected to disturbance correction processing according to the gas disturbance parameter to obtain a corrected version of the time series correlation, and the corrected version of the time series correlation is used for subsequent feedback analysis processing of the odor to be tested.

8. The method for identifying odor characteristics according to claim 7, characterized in that: The step of performing disturbance correction processing on the time sequence correlation according to the gas disturbance parameter to obtain a corrected version corresponding to the time sequence correlation comprises: The relationship between the property changes of various gas components over time under non-perturbation conditions is taken as the first analysis object; According to the difference between the disturbance state fed back by the gas disturbance parameter and the non-disturbance state, the first analysis object is subjected to corresponding difference simulation processing to obtain the property change relationship of various gas components under the disturbance state fed back by the gas disturbance parameter, and the property change relationship of the gas components under the gas disturbance is used as the second analysis object; An analysis process is performed based on the difference between the second analysis object and the first analysis object to obtain a correction parameter, and a correction process is performed on the time series association relationship based on the correction parameter to obtain a corrected version of the time series association relationship.

Images