A gas sensitive array detection device for rapid identification of characteristic smell of Liupu tea

Abstract

The application discloses a kind of gas-sensitive array detection devices for the characteristic smell of Liupu tea fast identification, it is related to Liupu tea smell detection technical field, device includes: ontology, sampling chamber, heating cylinder, gas-sensitive sensor array, sampling pump and microprocessor.Microprocessor includes data acquisition module, theoretical delay calculation module, compensation parameter generation module, feature calibration module and classification identification module.The initial response slope of sensor is extracted in the application, and diffusion compensation time parameter is generated by mapping with theoretical delay;The time axis of original response time sequence is calibrated and amplitude weighted using the parameter, and the physical space interference of hardware is stripped, to generate pure chemical space-time characteristic matrix.The application effectively solves the signal distortion problem caused by the difference of gas path diameter, environmental temperature fluctuation and pump body mechanical aging of traditional detection equipment, which is beneficial to improve the accuracy and cross-device consistency of Liupu tea characteristic smell identification.

Description

Technical Field

[0001] This invention relates to the field of Liubao tea aroma detection technology, specifically to a gas-sensitive array detection device for rapid identification of the characteristic aroma of Liubao tea, and more particularly to a high-precision aroma detection device that eliminates physical space interference through multi-source hardware parameter fusion and signal spatiotemporal dual-domain calibration. Background Technology

[0002] Liubao tea is a historically renowned tea with unique regional characteristics in my country, a national geographical indication product, and one of the core categories of Chinese dark tea. The pile fermentation and subsequent aging are the core processes that contribute to the unique flavor and quality of modern Liubao tea, directly determining its final quality grade and market value. During the pile fermentation and subsequent aging process, Liubao tea is influenced by multiple factors, including fermentation microorganisms, temperature and humidity, storage conditions, and aging time. This results in a rich and differentiated characteristic aroma system, primarily including the pure aged aroma representing high quality, as well as unusual aromas such as musty, musty, off-flavors, and moldy smells. Accurate differentiation and quantitative identification of these aromas are crucial for controlling the stability of the fermentation process, standardizing storage and maintenance management, identifying the quality of tea, defining product grades, and achieving precise market pricing. This plays a decisive role in the standardization, regulation, and high-quality development of the Liubao tea industry.

[0003] Currently, the industry's evaluation of the aroma and quality of Liubao tea still heavily relies on traditional manual sensory assessment methods. This method suffers from numerous drawbacks, including strong subjectivity, inconsistent evaluation standards, poor repeatability and stability, and susceptibility to environmental and human interference. It is ill-suited to the precise quality inspection and standardized management requirements of a modern, large-scale tea industry. To overcome the technical bottlenecks of manual assessment and achieve objective, digital, and rapid detection of tea aroma and quality, gas-sensitive sensor array technology, with its advantages of rapid response, non-destructive testing, and simulation of human olfaction, has gradually become the mainstream technology for digital identification of tea aroma.

[0004] There is already a research foundation for tea aroma detection based on gas-sensitive sensor arrays. For example, the existing technology CN107064238A relates to a portable electronic nose system for tea aroma detection and its detection. This invention integrates a gas-sensitive sensor array, an air pump, a signal acquisition circuit, and a tablet computer into a portable case. It switches between the cleaning and detection air paths through an electromagnetic reversing valve and collects the sensor's steady-state response extreme values ​​for comparison with a preset database. However, it also has the following shortcomings: ① It always uses the rated flow rate of the sampling pump as the calculation benchmark and does not consider the influence of manufacturing tolerances in the diameter and length of the air path between different devices on the gas transmission time; ② It does not establish a compensation mechanism for the influence of ambient temperature fluctuations on gas flow rate; and it does not consider the decrease in pumping power caused by mechanical aging after long-term operation of the pump body, making it difficult to guarantee the consistency and long-term stability of the detection results. Existing technology CN103134850A relates to a rapid tea quality detection device based on characteristic aroma. This invention first establishes a database of characteristic aroma content of teas of different qualities through GC / MS analysis, and then uses a gas-sensitive sensor array for rapid detection. By comparing with the database, it outputs quality grade or characteristic aroma content information, achieving both qualitative and quantitative dual-mode detection. However, this scheme has the following shortcomings: ① GC / MS analysis requires complex pretreatment of tea samples such as crushing and extraction, which is cumbersome and cannot achieve rapid on-site detection; ② GC / MS can only analyze single odor components, making it difficult to effectively correlate with the overall gas characteristics of tea and failing to comprehensively reflect the comprehensive attributes of tea aroma; ③ The sensor array only contains 4 gas-sensitive sensors, limiting the detection dimensions, and directly uses unverified and uncalibrated raw response signals for database comparison. Existing technology CN107991444A relates to a method and device for tea quality detection. This invention involves steeping tea leaves into tea soup, heating it to produce aroma, using a sensor array for detection, and using an enzyme sensor to detect the tea polysaccharide content in the tea soup to automatically identify the tea type. However, this scheme has the following shortcomings: ① The volatile aroma components of tea leaves are significantly different when the tea is steeped into tea soup and then heated compared to when dry tea is directly heated, making it impossible to truly restore the original aroma characteristics of the tea leaves; ② The content of tea polysaccharides is affected by various factors such as tea tree variety, origin, season, and processing, and it is unreliable to rely solely on it as the only basis for variety identification; ③ It also does not take into account physical interference factors such as differences in gas path hardware, fluctuations in ambient temperature, and long-term drift of sensors.

[0005] In summary, all the above-mentioned existing technical solutions assume that the working environment of the gas sensor is ideal—that is, the gas path parameters are constant, the ambient temperature is stable, and the pump performance never degrades, so that the raw electrical signal output by the sensor can be directly used for feature extraction and pattern recognition.

[0006] However, in real-world, long-term applications, manufacturing tolerances are unavoidable in the physical dimensions of gas pipelines. Furthermore, the mechanical pumps responsible for gas delivery experience irreversible wear and power reduction over time, and fluctuations in ambient temperature trigger thermal expansion and contraction of the gas itself. When these physical variables are superimposed, the hydrodynamic transport state of volatile odor molecules within the confined space is severely altered. The delay time of molecules with different volatile concentrations reaching the sensor and the dynamic diffusion rate at the initial contact with the sensor exhibit irregular shifts. This results in a high coupling between the time axis offset and amplitude deviation caused by the hardware's physical space and mechanical aging in the final acquired response signal and the chemical concentration characteristics of the odor molecules themselves. Consequently, the subsequent algorithm-extracted spectrum suffers severe distortion and drift, making it difficult to achieve high-precision identification of the complex and overlapping odors of Liubao tea.

[0007] To address the shortcomings of the existing technologies, this invention provides a gas-sensitive array detection device for rapid identification of the characteristic aroma of Liubao tea. The device aims to eliminate hardware physical space interference by fusing multi-source hardware parameters and performing spatiotemporal dual-domain calibration of signals, thereby generating a pure chemical spatiotemporal feature matrix that can truly reflect the chemical nature of the aroma, and improving the accuracy and cross-device consistency of Liubao tea aroma identification. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a gas-sensitive array detection device for rapid identification of the characteristic aroma of Liubao tea. The core technical points of this invention are: calculating the theoretical fluid transport delay by fusing multi-source hardware parameters (gas path diameter, pipe length, heating temperature, pump rated flow rate), and generating diffusion compensation time parameters by combining the transient dynamic characteristics of the sensor's initial response slope. Finally, the original response signal is calibrated for time axis offset and weighted by amplitude to generate a purely chemical spatiotemporal feature matrix stripped of hardware physical space interference, thereby significantly improving the accuracy of aroma identification and cross-device consistency.

[0009] To achieve the above objectives, the technical solution of the present invention is as follows:

[0010] A gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea includes:

[0011] The sample chamber consists of a main body, a sampling chamber located within the main body, and a heating cylinder located within the sampling chamber. The top of the sampling chamber is provided with a sealing cover, which forms a sealing structure between the sampling chamber and the heating cylinder. The heating temperature of the heating cylinder is controlled within the range of 40~60℃.

[0012] A gas-sensitive sensor array connected to the heating cylinder and a sampling pump installed on the gas pipeline between the heating cylinder and the gas-sensitive sensor array;

[0013] The microprocessor is configured to include the following functional modules:

[0014] The data acquisition module is used to acquire the current heating temperature data of the sampling chamber, the physical diameter and physical length data of the gas pipeline located between the heating cylinder and the gas sensor array, the rated flow rate data of the sampling pump, and the raw response time series data acquired by the gas sensor array in real time after the sampling pump is started.

[0015] The theoretical delay calculation module is used to fuse and calculate the current heating temperature data, physical pipe diameter data, physical length data and rated flow rate data to generate theoretical fluid transmission delay data that characterizes the current gas path environment.

[0016] The compensation parameter generation module is used to extract the initial response slope data of the original response time series data within a preset time window, and to map the initial response slope data with the theoretical fluid transport delay data to generate diffusion compensation time parameter data that characterizes the differences in the activity of volatile molecules in different tea leaves.

[0017] The feature calibration module is used to perform time axis offset calibration and amplitude weighting on the original response time series data using diffusion compensation time parameter data, generating pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference;

[0018] The classification and recognition module is used to compare and map the pure chemical spatiotemporal feature matrix data with the preset standard odor fingerprint spectrum database, and to judge and generate characteristic odor classification result data for Liubao tea.

[0019] Preferably, the initial response slope data of the original response time series data within a preset time window is extracted, specifically including:

[0020] In the original response time series data, the voltage amplitude is continuously monitored. When the voltage amplitude deviates from the initial baseline data to the preset trigger threshold, the current moment is extracted as the response start moment data.

[0021] Starting from the data at the start of the response, a continuous data segment of a preset time length is extracted and used as the feature sequence data within the preset time window;

[0022] According to the preset sampling frequency, extract multiple discrete voltage sampling point data and the timestamp corresponding to each voltage sampling point data from the feature sequence data;

[0023] A univariate linear regression calculation was performed on all voltage sampling point data and timestamps to obtain the slope parameter of the fitted line;

[0024] The slope parameter is assigned to generate the initial response slope data.

[0025] Preferably, the initial response slope data is mapped to theoretical fluid transport delay data to generate diffusion compensation time parameter data characterizing the differences in the activity of volatile molecules in different tea leaves, specifically including:

[0026] Convert the initial response slope data into response acceleration data;

[0027] Acquire the system's preset acceleration decay constant data, pump body aging weight coefficient data, and reference offset data, and read the current cumulative running time data of the sampling pump in real time;

[0028] Calculate the product of the response acceleration data and the acceleration decay constant data to obtain the decay power variable data;

[0029] Using the natural constant as the base and the negative number of the decay power variable data as the exponent, perform exponential calculation to generate basic activity coefficient data;

[0030] The cumulative runtime data is multiplied with the pump body aging weight coefficient data, and the product result is added to the baseline offset data to calculate the mechanical attenuation correction factor data.

[0031] The basic activity coefficient data is corrected based on the mechanical decay correction factor data to generate a scaling factor;

[0032] The theoretical fluid transport delay data is scaled based on a scaling factor to generate diffusion compensation time parameter data.

[0033] Preferably, the initial response slope data is converted into response acceleration data, specifically including:

[0034] The window slides at a preset time step to continuously extract the initial response slope data of at least two adjacent time windows;

[0035] Calculate the difference between the initial response slope data of the next time window and the initial response slope data of the previous time window;

[0036] The ratio of the difference to the preset time step is calculated to obtain the response acceleration data.

[0037] Preferably, the generation of pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference specifically includes:

[0038] Subtract the diffusion compensation time parameter data from the timestamps corresponding to each voltage sampling point in the original response time series data to obtain time axis aligned data that eliminates non-uniform diffusion time difference;

[0039] Input the diffusion compensation time parameter data into the preset concentration decay function to calculate and generate the dimensionless amplitude compensation weight coefficient.

[0040] The original voltage amplitude data in the original response time series data is multiplied by the amplitude compensation weighting coefficient to obtain the restored true concentration amplitude data.

[0041] The time-axis aligned data is matrix-stitched with the actual concentration amplitude data to generate a pure chemical spatiotemporal feature matrix.

[0042] Preferably, the characteristic aroma classification results data for Liubao tea are determined and generated, specifically including:

[0043] Obtain multiple standard odor template matrix data pre-stored in the standard odor fingerprint spectrum database, as well as odor category label data corresponding to each standard odor template matrix data;

[0044] The pure chemical spatiotemporal feature matrix data was subtracted element-wise from the standard odor template matrix data to obtain multiple residual matrix data.

[0045] The spatial distance data corresponding to each standard odor category is calculated by summing the squares of all elements in each residual matrix data.

[0046] Compare the numerical values ​​of all spatial distance data and select the spatial distance data with the smallest value as the target distance data.

[0047] Extract the odor category label data associated with the target distance data, and generate the odor category label data as the characteristic odor classification result data for Liubao tea.

[0048] Preferably, generating theoretical fluid transport delay data characterizing the current gas path environment specifically includes:

[0049] Obtain the initial ambient temperature data of the system's environment, and convert the current heating temperature data and the initial ambient temperature data into absolute temperatures to obtain the current absolute temperature data and the initial absolute temperature data;

[0050] Calculate the ratio of the current absolute temperature data to the initial absolute temperature data, and use the ratio as the correction variable for the actual flow velocity of the gas after thermal expansion;

[0051] The rated flow velocity data is corrected based on the actual flow velocity correction variable data to obtain the true flow velocity data inside the pipe;

[0052] The volume data of the gas pipeline located between the heating cylinder and the gas sensor array is calculated using the physical diameter and physical length data of the gas pipeline.

[0053] Calculate the ratio of pipe volume data to actual pipe velocity data to generate theoretical fluid transport delay data.

[0054] Preferably, the gas sensor array includes at least a first sensor sensitive to microbial metabolites, a second sensor sensitive to alcohols and esters, a third sensor sensitive to aldehydes and phenols, and a fourth sensor sensitive to terpenes.

[0055] Preferably, the main body is further provided with an activated carbon filter, the air inlet of the activated carbon filter is connected to the air outlet of the gas-sensitive sensor array pipeline, and the air outlet of the activated carbon filter is connected to the external environment of the main body.

[0056] Preferably, a touch screen is provided on the outer surface of the body.

[0057] The method for constructing a preset standard odor fingerprint database involves the following steps:

[0058] (1) A total of 250 standard samples of Liubao tea with 4 different aromas and 15 different storage years were collected, with each sample weighing 5.0±0.1g;

[0059] (2) Under the standard storage environment conditions of Liubao tea, the heating temperature was set to 50℃ and the heating time was 5 minutes. Pure chemical spatiotemporal feature matrix data of each standard sample were collected in sequence.

[0060] (3) Each standard sample was collected 10 times, and the average value was used as the standard odor template matrix data for that category. The corresponding odor category label data were labeled, including four categories: stale smell, musty smell, warehouse smell, and moldy smell.

[0061] (4) Store all standard template matrices and their corresponding tags into the database to complete the construction.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] 1. Eliminating cross-device detection bias through multi-source hardware parameter fusion. This invention fuses and calculates multiple hardware parameters, such as the physical dimensions of the gas pipeline, heating temperature, and pump rated flow rate, to derive the theoretical fluid transport delay. Even with manufacturing tolerances, different devices can adapt to this delay based on their actual physical parameters, ensuring that the feature matrices of the same odor sample collected from different devices are aligned to the same chemical reference on the time axis, significantly improving cross-device detection consistency.

[0064] 2. Enhancing the Distinction Between Similar Odors Through Transient Dynamics Feature Extraction. This invention uses a threshold-triggered method combined with univariate linear regression to extract the initial response slope of the sensor and further convert it into response acceleration, capturing the differences in diffusion dynamics of different volatile molecules in the initial contact with the sensor. Compared to existing technologies that only extract steady-state extreme values, this invention can effectively distinguish between similar odors with similar final concentrations but different chemical compositions (such as a slight "mustard" smell versus a "warehouse smell").

[0065] 3. Eliminating long-term operational drift through pump body aging compensation. This invention introduces a mechanical attenuation correction factor constructed from cumulative operating time and pump body aging weighting coefficient to dynamically correct the theoretical fluid transmission delay. As the equipment's service life increases, the compensation parameters automatically adapt to the pump body's aging degree, avoiding reference drift in the detection system as its service life increases.

[0066] 4. Generating a pure chemical feature matrix through spatiotemporal dual-domain calibration. This invention eliminates non-uniform diffusion time difference in the time axis dimension and compensates for concentration dissipation during transmission in the amplitude dimension, generating a pure chemical spatiotemporal feature matrix stripped of hardware physical space interference. This matrix reflects only the chemical kinetic properties of the odor molecules themselves, providing high-fidelity input data for subsequent pattern recognition. Attached Figure Description

[0067] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0068] Figure 1 This is a three-dimensional structural diagram of the present invention;

[0069] Figure 2 This is a schematic diagram of the front cross-sectional structure of the present invention;

[0070] Figure 3 This is a flowchart illustrating the microprocessor execution process of the present invention.

[0071] The diagram shows the following components: 1. Main body; 2. Touch screen; 3. Sealing cover; 4. Sampling chamber; 5. Heating cylinder; 6. Heater; 7. Microprocessor; 8. Activated carbon filter; 9. Gas sensor array; 10. Sampling pump. Detailed Implementation

[0072] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0073] Application Overview:

[0074] In the traditional field of tea quality testing, especially in the quality grading and storage management of Liubao tea, the accurate identification of characteristic aromas (such as stale, musty, and moldy aromas) is regarded as a key indicator for measuring the quality of tea transformation and market value. The generation of this high-fidelity aroma spectrum is essentially a process of stable and uniform airflow transmission at the physicochemical level. That is, through constant temperature heating and pump suction, a controlled physical gas path is used as a transmission medium to transmit volatile aroma molecules non-destructively and proportionally to the gas-sensitive sensor array at the end, thereby leaving an electrical signal trajectory with a specific chemical concentration and time correspondence at the data end.

[0075] However, existing technologies lack a verification mechanism for the consistency of fluid dynamics between the physical environment of the input hardware and the response of the output sensor. This results in an inability to accurately identify fluid transmission delays and non-uniform diffusion time differences during long-cycle detection. Fluid transmission delays manifest as a shift in the time it takes for gas to reach the sensor, even though the system has a set constant sampling time. This is due to differences in the physical diameter and length of the gas pipeline or the power attenuation caused by the aging of the mechanical pump. Non-uniform diffusion time differences manifest as changes in the diffusion acceleration of volatile molecules with different activity levels when they first come into contact with the sensor, caused by the thermal expansion of gas due to ambient temperature fluctuations. This results in the kinetic energy being limited by the dynamic changes in the physical environment. Consequently, a strict, purely chemical mapping relationship cannot be established between the fluid dynamics state within the gas path and the time-series response of the sensor. This leads to misjudgments of complex mixed odors or distortions in spectrum comparisons, thus affecting the accuracy of odor classification and the robustness of cross-device detection.

[0076] For example, in actual quality inspection scenarios at tea factories, when operators conduct continuous testing of multiple batches of Liubao tea, existing systems, using conventional algorithms, can only capture the voltage extreme value after the sensor reaches a steady state, but cannot distinguish whether the lag in response time is accompanied by a hidden flow rate decay caused by the aging of the sampling pump. Furthermore, when the temperature difference between day and night in the workshop causes the actual gas flow rate to differ from the rated flow rate, the system only records the superficial phenomenon of concentration curve distortion, failing to detect abnormal fluctuations in fluid transmission delay in the pipeline during the gas thermal expansion process and changes in the initial response slope of different molecules. Specifically, the system is prone to misjudging "aged" molecules that arrive late due to pump aging as insufficient concentration, or incorrectly classifying "storage" molecules that have transient slope distortion due to temperature interference as "moldy" molecules. This leads to the company continuously outputting incorrect quality grading results, making it impossible to form an objective evaluation benchmark that conforms to pure chemical spectrum characteristics.

[0077] If the above problems are not addressed, the detection system will continue to lose its ability to objectively distinguish the true chemical state of odors. Specifically, the failure to recognize fluid transport delays will cause the algorithm to over-rely on static peak values, leading to a deviation of the signal time axis from the true physical baseline, thus weakening the system's ability to distinguish the boundaries between similar characteristic odors (such as musty and moldy smells). Simultaneously, the failure to correct for non-uniform diffusion time differences will exacerbate the residuals between the original data and the standard fingerprint spectrum, preventing the collected feature matrix from presenting purely chemical properties stripped of hardware interference. Ultimately, this will cause the equipment to lose its long-term accuracy and consistency. Consequently, the inaccuracy of the detection results will systematically hinder tea companies from establishing standardized quality control systems, affecting the achievement of digital quality control goals in the Liubao tea industry.

[0078] like Figure 1-3 As shown, a gas-sensitive array detection device for rapid identification of the characteristic aroma of Liubao tea includes:

[0079] The main body 1, the sampling chamber 4 set inside the main body 1, and the heating cylinder 5 set inside the sampling chamber. The top of the sampling chamber 4 is provided with a sealing cover 3, which forms a sealing structure for the sampling chamber 4 and the heating cylinder 5. The heating temperature control range of the heating cylinder 5 is 40-60℃.

[0080] A gas-sensitive sensor array 9 connected to the heating cylinder 5 and a sampling pump 10 installed on the gas pipeline between the heating cylinder 5 and the gas-sensitive sensor array 9;

[0081] The microprocessor is configured to include the following functional modules:

[0082] The data acquisition module is used to acquire the current heating temperature data of the sampling chamber 4, the physical diameter and physical length data of the gas pipeline located between the heating cylinder 5 and the gas sensor array 9, the rated flow rate data of the sampling pump 10, and the raw response time series data acquired by the gas sensor array 9 in real time after the sampling pump 10 is started.

[0083] The theoretical delay calculation module is used to fuse and calculate the current heating temperature data, physical pipe diameter data, physical length data and rated flow rate data to generate theoretical fluid transmission delay data that characterizes the current gas path environment.

[0084] The compensation parameter generation module is used to extract the initial response slope data of the original response time series data within a preset time window, and to map the initial response slope data with the theoretical fluid transport delay data to generate diffusion compensation time parameter data that characterizes the differences in the activity of volatile molecules in different tea leaves.

[0085] The feature calibration module is used to perform time axis offset calibration and amplitude weighting on the original response time series data using diffusion compensation time parameter data, generating pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference;

[0086] The classification and recognition module is used to compare and map the pure chemical spatiotemporal feature matrix data with the preset standard odor fingerprint spectrum database, and to judge and generate characteristic odor classification result data for Liubao tea.

[0087] Specifically, a heater 6 is installed at the bottom of the sampling chamber 4, and the output end of the heater 6 is in contact with the bottom of the heating cylinder 5. A sealing cover 3 is installed on the top of the main body 1, which seals both the sampling chamber 4 and the heating cylinder 5.

[0088] Furthermore, the gas-sensitive sensor array 9 includes at least a first sensor sensitive to microbial metabolites, a second sensor sensitive to alcohols and esters, a third sensor sensitive to aldehydes and phenols, and a fourth sensor sensitive to terpenes. Wherein:

[0089] The first sensor is a metal oxide semiconductor sensor, which is highly sensitive to microbial metabolites such as 2-methylisoborneol, and is specifically designed to capture the precursor molecules of "musty smell" that Liubao tea produces under poor storage conditions.

[0090] The second sensor is an electrochemical sensor that is sensitive to volatile alcohols and esters, and is used to characterize the "stuffy smell" produced by the damp heat during the fermentation of Liubao tea.

[0091] The third sensor is an organic field-effect transistor sensor, which is sensitive to aldehydes and phenols and is used to capture the "storage smell" markers formed in tea leaves during long-term environmental adsorption.

[0092] The fourth sensor is a piezoelectric quartz crystal sensor, which is sensitive to macromolecular terpenes, the core source of the high-quality "aged flavor" that Liubao tea develops through normal transformation.

[0093] Furthermore, a touch screen display 2 is provided on the outer surface of the main body 1. The heater 6, sampling pump 10, gas sensor array 9, and touch screen display 2 are all electrically connected to the microprocessor 7. During detection, a quantitative amount of loose Liubao tea (or a piece of compressed tea) is taken, the sealing cover 3 on the top of the main body 1 is opened, the tea leaves are placed in the heating cylinder 5, and then the sealing cover 3 is closed and pressed tightly to ensure airtightness. The detection command is issued through the touch screen display 2. After receiving the command, the microprocessor 7 controls the heater 6 to heat the heating cylinder 5 to the set temperature and maintain the temperature for a specific time to allow the characteristic aroma molecules (i.e., volatile substances) of the Liubao tea sample to fully escape and reach a concentration equilibrium state in the sealed space.

[0094] Microprocessor 7 execution flow:

[0095] After quantitative sampling and array response enrichment are completed, the microprocessor 7 starts the sampling pump 10 located in the gas pipeline between the heating cylinder 5 and the gas sensor array 9, extracts the mixed gas in the heating cylinder 5 at a preset rated flow rate, and guides it into the gas sensor array 9. When the first to fourth sensors in the array come into contact with the gas containing different odor molecules, they generate their own independent electrical signal responses. The microprocessor 7 is configured to perform the following steps during the detection process:

[0096] Step 1: Data Collection

[0097] The microprocessor acquires the current heating temperature data of sampling chamber 4, the physical diameter and length data of the gas pipeline between heating cylinder 5 and gas sensor array 9, the rated flow rate data of sampling pump 10, and the raw response time series data collected by gas sensor array 9 in real time after sampling pump 10 is started. The microprocessor first collects and summarizes the underlying data. On the one hand, it calls hardware status parameters through sensors, including acquiring the current heating temperature data of sampling chamber 4, the physical diameter and length data of the gas pipeline between heating cylinder 5 and gas sensor array 9, and the rated flow rate data of sampling pump 10 built into the system. On the other hand, after sampling pump 10 is started, microprocessor 7 collects the electrical signals output by gas sensor array 9 in real time at a preset frequency. When the first to fourth sensors in the array come into contact with gases containing different odor molecules, they generate their own independent electrical signal responses. Microprocessor 7 synchronously collects these continuous electrical signals in real time, continuously setting the response time, to form raw response time series data representing changes in gas concentration. The data dimensions of space mechanical properties, thermodynamic state, and chemical sensing response were unified, providing complete physical boundary conditions for subsequent calculations.

[0098] Step 2: Generate theoretical fluid transport delay data:

[0099] The system integrates current heating temperature data, physical pipe diameter data, physical length data, and rated flow rate data to generate theoretical fluid transport delay data characterizing the current gas path environment. After acquiring the aforementioned basic data, the microprocessor 7 performs a fusion calculation based on the principles of fluid mechanics and thermodynamics. The system first considers that the actual flow rate changes due to gas thermal expansion; therefore, it incorporates the current heating temperature data into the flow rate evaluation system to dynamically compensate for the rated flow rate of the sampling pump 10, thus deriving the true flow rate within the pipe. Subsequently, the system calculates the physical volume space of the pipe by combining the physical pipe diameter and physical length data. By mapping this physical volume space to the true flow rate, the microprocessor generates theoretical fluid transport delay data characterizing the current gas path environment. This data objectively quantifies the purely mechanical and physical time consumed by the gas from its escape to its arrival at the sensor, eliminating interference from manufacturing tolerances and environmental temperature differences.

[0100] Step 3: Generate diffusion compensation time parameter data:

[0101] The system extracts the initial response slope data from the original response time series data within a preset time window, and maps this initial response slope data to theoretical fluid transport delay data to generate diffusion compensation time parameter data characterizing the differences in the activity of different volatile molecules in tea. Since the diffusion activity of volatile substances in tea with different molecular weights (such as alcohols and terpenes) in air varies, the microprocessor 7 further extracts transient features from the original response time series data. In the initial stage of gas contact with the sensor (i.e., within the preset time window), the system extracts the initial response slope data indicating the increase in concentration. This slope data directly reflects the diffusion acceleration of a specific volatile molecule. The microprocessor 7 dynamically maps and weights this slope data, characterizing the activity of chemical molecules, with the aforementioned theoretical fluid transport delay data characterizing the physical hardware state, ultimately generating dedicated diffusion compensation time parameter data. This parameter achieves dual adaptive compensation for the diffusion characteristics of different odor molecules and the potential mechanical attenuation of the pump body.

[0102] Step 4: Generate pure chemical spatiotemporal feature matrix data:

[0103] The system utilizes diffusion-compensated time parameter data to perform time axis offset calibration and amplitude weighting on the original response time series data, generating a pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference. After calculating high-precision compensation parameters, the microprocessor 7 performs interference-free reconstruction of the original electrical signal sequence. The system uses this diffusion-compensated time parameter data to perform reverse time offset calibration on the original response time series data in the time axis dimension, eliminating time lag caused by fluid delay; simultaneously, it uses this parameter for weighted scaling in the amplitude dimension to restore the concentration dissipation of gas molecules during transmission. Through this joint calibration of spatiotemporal axes, the system splices and combines the corrected time node sequence with the true concentration amplitude sequence to generate a pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference. This matrix data truthfully and purely restores the complete picture of the chemical reaction between the volatiles of Liubao tea and the sensor.

[0104] Step 5: Generate characteristic odor classification result data:

[0105] The pure chemical spatiotemporal feature matrix data is compared and mapped with a preset standard odor fingerprint spectrum database to determine and generate characteristic odor classification results for Liubao tea. Finally, the microprocessor 7 enters the pattern recognition and decision stage. The system calls the preset standard odor fingerprint spectrum database, which contains template matrices and labels for moldy, musty, stored, and aged odors that have been standardized. The microprocessor 7 performs matrix-level similarity comparison between the real-time extracted pure chemical spatiotemporal feature matrix data and the various standard odor template matrix data in the database. By calculating their residual distance in multidimensional space, the system automatically selects the target spectrum with the smallest error distance and the highest matching degree, extracts its corresponding category label, and finally generates and outputs characteristic odor classification results for the current Liubao tea sample, and sends the results to the display screen 2 for display.

[0106] The core innovation of the aforementioned technology lies in the fusion calculation of multiple hardware parameters, such as the physical dimensions (diameter and length) of the gas pipeline, heating temperature, and pump flow rate, to derive the theoretical fluid transport delay. This delay is then mapped to the initial response slope of volatile molecules to generate diffusion compensation time parameters. By performing time axis offset calibration and amplitude weighting on the original time series data, physical spatial interference caused by hardware structure and fluid dynamics is effectively eliminated. The resulting pure chemical spatiotemporal feature matrix data can truly reflect the chemical essence of tea aroma, solving the signal distortion problem caused by slight differences in gas pipeline diameter, ambient temperature fluctuations, and pump mechanical aging in traditional detection equipment. This ensures that the extracted feature matrix has high cross-device consistency and anti-interference capability, thereby effectively improving the recognition accuracy and long-term robustness of the complex and overlapping aromas of Liubao tea.

[0107] Specific implementation of each step:

[0108] (I) Multi-dimensional underlying data collection:

[0109] Conventional odor detection equipment typically focuses solely on the electrical signal sequence output from the sensor during data acquisition, neglecting the objective constraints on gas fluid dynamics imposed by physical hardware conditions such as the front-end heating device, transmission pipe dimensions, and pump flow rate. This single-dimensional data acquisition method results in a lack of boundary parameters necessary for evaluating fluid transmission delay and calculating diffusion compensation. Consequently, subsequent signal analysis, lacking a benchmark of physical variables, is prone to judgment biases caused by hardware differences. Therefore, this paper proposes acquiring the current heating temperature data of sampling chamber 4, the physical diameter and length data of the gas path pipe located between heating cylinder 5 and gas sensor array 9, the rated flow rate data of sampling pump 10, and the raw response time series data acquired in real-time by gas sensor array 9 after sampling pump 10 starts, as detailed below:

[0110] Current heating temperature data: refers to the Celsius temperature value of the air inside the sampling chamber 4 that the microprocessor 7 collects in real time through the temperature sensor, which is used to characterize the basic physical state of gas thermal expansion.

[0111] Physical pipe diameter data: refers to the internal cylinder diameter of the gas passage pipe located between the outlet of heating cylinder 5 and the inlet of gas sensor array 9.

[0112] Physical length data: refers to the actual axial transmission straight distance of the airflow in the gas path between the outlet of the heating cylinder 5 and the inlet of the gas sensor array 9.

[0113] Rated flow rate data: refers to the theoretical gas volume extracted per unit time by sampling pump 10 under standard operating voltage and initial conditions.

[0114] Raw response time series data: refers to the data set formed by arranging the voltage amplitude signals obtained by the microprocessor 7 in chronological order after the sampling pump 10 starts and the gas sensor array 9 comes into contact with the volatile gas of Liubao tea.

[0115] Step 1: Acquire current heating temperature data. The microprocessor 7 acquires an analog voltage signal through a thermocouple temperature sensor located at the bottom of the heating cylinder 5, and converts it into a digital temperature value through an analog-to-digital converter circuit. The preset range for this data is typically configured to be 40 to 60 degrees Celsius.

[0116] Step 2: Obtain the physical dimensions of the gas pipeline. Microprocessor 7 reads the hardware parameters pre-written during the device's factory calibration phase from the system motherboard's non-volatile memory (EEPROM) via the I2C bus. This reading step does not require real-time measurement. The preset range for the physical pipe diameter data is configured to be 1.0 to 5.0 mm; the preset range for the physical length data is configured to be 50.0 to 200.0 mm.

[0117] Step 3: Acquire the rated flow rate data of sampling pump 10. Microprocessor 7 also reads the calibrated pump parameters from non-volatile memory. The preset range for this rated flow rate data is configured to be 100 to 500 ml per minute.

[0118] Step 4: Acquire raw response time series data. After the sampling pump 10 is started, the timer module inside the microprocessor 7 is triggered, controlling the analog-to-digital conversion module (A / D module) to synchronously acquire the output voltage of all sensors in the gas sensor array 9 according to the preset sampling frequency.

[0119] In this process, microprocessor 7 first calculates the timestamp of each discrete sampling point based on the sampling frequency. The specific formula is as follows:

[0120] ;

[0121] in, For the first The relative timestamps of each sampling point are in seconds; This is the sampling sequence number, which takes the value of a positive integer greater than or equal to 1; The system's preset sampling frequency is configured to range from 10 to 100 Hz. By establishing a mapping between discrete time and physical time, an accurate time reference is provided for subsequent signal time axis offset calibration.

[0122] Subsequently, the microprocessor 7 aggregates and concatenates the voltage amplitude data corresponding to each timestamp to generate the raw response time series data for each sensor. The specific formula is as follows:

[0123] ;

[0124] in, The first in the gas-sensitive sensor array Raw response time series data from each sensor; The sensor number is 1, 2, 3, or 4, corresponding to the first through fourth sensors, respectively. For the first One sensor in The voltage amplitude data output by the analog-to-digital converter module at any time, in volts; The total number of sampling points within a preset time window is determined by the product of the total detection time and the sampling frequency. By structuring the scattered voltage points, a standard mathematical vector structure is generated, facilitating subsequent matrix mapping calculations.

[0125] Example 1: Assume a gas-sensitive array detection device for detecting Liubao tea is in operation:

[0126] The microprocessor 7 obtains the current heating temperature data of 50°C from the temperature sensor.

[0127] The microprocessor 7 reads the physical pipe diameter data as 2.0 mm and the physical length data as 100.0 mm from the EEPROM.

[0128] The microprocessor 7 reads the rated flow rate data of the sampling pump from the EEPROM as 200 mL / min.

[0129] Microprocessor 7 configuration sampling frequency The frequency is 10 Hz, meaning 10 points are collected per second. We select the first 3 sampling points ( To demonstrate the first sensor ( Timestamp calculation and time series data generation.

[0130] Calculate timestamps :

[0131] when hour, Second;

[0132] when hour, Second;

[0133] when hour, Second;

[0134] Assume the voltage amplitude data of the first sensor collected by the analog-to-digital converter at the above times are 1.25 volts, 1.30 volts, and 1.42 volts, respectively. Substituting these data into the sequence formula, the original response time series data of the first sensor is obtained as follows:

[0135] ;

[0136] The data relationship mapping obtained from the above process is shown in the table below:

[0137]

[0138] In the aforementioned technology, by establishing a multi-dimensional underlying data architecture that covers space mechanical properties, thermodynamic state, and chemical sensing response, abstract electrical response signals are associated with specific physical hardware parameters. This provides sufficient calculation basis for subsequent algorithms to correct for gas thermal expansion, calculate pipeline volume, and compensate for fluid dynamic delays. It ensures that subsequent data processing can be based on the actual physical environment and improves the objectivity and comprehensiveness of basic data input.

[0139] (II) Generation of theoretical fluid transport delay data:

[0140] When gas-sensitive array detection equipment operates under different seasons or ambient temperatures, the degree of thermal expansion of volatile gases from Liubao tea varies, causing the actual gas dynamic velocity flowing through the pipeline to deviate from the pump's factory-set rated flow rate. Simultaneously, during assembly, different batches of equipment inevitably exhibit mechanical tolerances in the physical diameter and length of their internal gas path pipelines. The superposition of these temperature and spatial variables leads to an uncontrollable shift in the actual transmission time of odor molecules from escape to contact with the sensor surface, causing misalignment in the original response time series data collected by the sensor on the time axis and reducing the accuracy of subsequent odor waveform feature extraction. Therefore, it is proposed to fuse current heating temperature data, physical pipe diameter data, physical length data, and rated flow rate data to generate theoretical fluid transport delay data characterizing the current gas path environment, as detailed below:

[0141] Generate theoretical fluid transport delay data characterizing the current gas path environment, specifically including:

[0142] Obtain the initial ambient temperature data of the system's environment, and convert the current heating temperature data and the initial ambient temperature data into absolute temperatures to obtain the current absolute temperature data and the initial absolute temperature data;

[0143] Calculate the ratio of the current absolute temperature data to the initial absolute temperature data, and use the ratio as the correction variable for the actual flow velocity of the gas after thermal expansion;

[0144] The rated flow velocity data is corrected based on the actual flow velocity correction variable data to obtain the true flow velocity data inside the pipe;

[0145] The volume data of the gas pipeline is calculated using the physical diameter and physical length data of the gas pipeline located between the heating cylinder 5 and the gas sensor array 9.

[0146] Calculate the ratio of pipe volume data to actual pipe velocity data to generate theoretical fluid transport delay data.

[0147] Initial ambient temperature data: refers to the current space temperature value in Celsius collected by the microprocessor 7 through an ambient temperature sensor located outside the housing or sampling chamber 4 before the detection begins, which is used as a thermodynamic benchmark for evaluating the degree of thermal expansion of the gas.

[0148] Actual flow rate correction variable data: refers to the proportionality coefficient of gas volume expansion due to temperature increase, which is a dimensionless constant used to compensate for the hydrodynamic changes of heated gas.

[0149] Actual flow velocity data: refers to the actual volumetric flow rate of the volatile gases from Liubao tea that are heated and expanded within the gas pipeline.

[0150] Pipeline volume data: refers to the objective three-dimensional spatial volume inside the physical gas pipeline section located between the heating cylinder 5 and the gas-sensitive sensor array 9.

[0151] Theoretical fluid transport delay data: refers to the theoretical transport time required for gas molecules to travel from the outlet of the heating cylinder to the inlet of the gas sensor array 9, calculated based on the pure physical space dimensions and the thermodynamically corrected flow velocity.

[0152] Step 1: Absolute Temperature Conversion: Microprocessor 7 acquires the initial ambient temperature data of the system's environment through an external ambient temperature sensor. The preset range for this data is typically between 10 and 40 degrees Celsius. Subsequently, microprocessor 7 converts the previously acquired current heating temperature data and the initial ambient temperature data into absolute thermodynamic temperatures. By uniformly converting degrees Celsius to the Kelvin scale, a standard thermodynamic calculation benchmark is provided for subsequent calculations of the isobaric thermal expansion of gases based on Charles's Law. The specific formula is as follows:

[0153] ;

[0154] in, These are the initial absolute temperature data, in Kelvin. This is the initial ambient temperature data, in degrees Celsius.

[0155] ;

[0156] in, This is the current absolute temperature data, in Kelvin. This refers to the current heating temperature data obtained in the previous step, in degrees Celsius.

[0157] Step 2: Calculate the actual flow rate correction variable data: Microprocessor 7 calculates the ratio of the current absolute temperature data to the initial absolute temperature data. The specific formula is as follows:

[0158] ;

[0159] in, Correct variable data for actual flow rate. By quantifying the volume expansion rate of gas after heating, it objectively reflects the degree of physical expansion of gas after it changes from a cold state to a hot state.

[0160] Step 3: Calculate the actual flow velocity data in the pipe: The microprocessor 7 performs a product operation based on the actual flow velocity correction variable data and the rated flow velocity data of the sampling pump 10 obtained in the previous step. The specific formula is as follows:

[0161] ;

[0162] in, The data represents actual flow velocity within the pipe, expressed in milliliters per minute. The rated flow rate data obtained in the preliminary steps is expressed in milliliters per minute. By thermodynamically compensating the nominal mechanical parameters of the pump body, the flow rate error caused by gas expansion is eliminated, resulting in the actual gas flow rate that conforms to the current physical environment.

[0163] Step 4: Calculate pipe volume data: The microprocessor 7 uses the physical diameter and length data of the gas pipeline obtained in the previous step to calculate the volume of the cylinder. The specific formula is as follows:

[0164] ;

[0165] in, This is pipeline volume data, in cubic millimeters. Pi is a constant, with a value of approximately 3.14159. The physical pipe diameter data obtained in the previous steps is in millimeters. The physical length data, in millimeters, is obtained from the preliminary steps. By establishing a spatial distance barrier model for fluid transport, the physical volume that volatile gases must traverse is precisely quantified.

[0166] Step 5: Generate theoretical fluid transport delay data: The microprocessor 7 calculates the ratio between the calculated pipe volume data and the actual flow velocity data within the pipe. A unit conversion constant is introduced during the calculation to unify the dimensions. The specific formula is as follows:

[0167] ;

[0168] in, This is theoretical fluid transport delay data, in seconds; A preset unit conversion constant is used to convert milliliters per minute to cubic millimeters per second, and its constant value is 1000 / 60 (approximately 16.667). By mapping the three-dimensional spatial variable to the one-dimensional velocity variable in the time dimension, a purely physical delay time parameter is generated that excludes interference from environmental temperature differences and hardware tolerances.

[0169] Example 2: Continuing from the previous example, the data obtained beforehand is: current heating temperature data. Physical pipe diameter data Physical length data Rated flow rate data .

[0170] Absolute temperature conversion: Microprocessor 7 acquires initial ambient temperature data through an ambient temperature sensor. .

[0171] Initial absolute temperature: .

[0172] Current absolute temperature: .

[0173] Calculate the actual flow rate correction variable: .

[0174] Calculate the actual flow velocity inside the pipe: .

[0175] Calculate the pipe volume: .

[0176] Generate theoretical fluid transport delay data:

[0177] constant .

[0178] Second.

[0179] In the aforementioned technology, static mechanical dimensional parameters are numerically fused with dynamic thermodynamic environmental variables. By introducing the temperature expansion coefficient to correct the rated flow rate, the interference of external environmental temperature fluctuations on the gas flow rate inside the pipe is eliminated. By calculating the pipe volume, the influence of hardware manufacturing tolerances between equipment is eliminated. The generated theoretical fluid transport delay data provides an objective and accurate time offset benchmark, laying the physical and fluid-level data foundation for reconstructing the true chemical reaction sequence of odor molecules in subsequent steps.

[0180] (III) Generation of diffusion compensation time parameter data:

[0181] The volatile aroma of Liubao tea is composed of various chemical substances with different molecular weights (such as small-molecule alcohols and large-molecule terpenes), and these molecules exhibit objective differences in diffusion rates and kinetic activity in the air. Meanwhile, the pump body undergoes mechanical aging after long-term operation, leading to a decrease in airflow driving capability. If a uniform theoretical fluid transport delay is directly used for calibration, ignoring the differences in diffusion kinetics between different molecules and the mechanical decay of the hardware over time, the time axis of the sensor response data cannot be accurately aligned, thus affecting the accuracy of the classification model in resolving overlapping odor features. Therefore, this paper proposes: extracting the initial response slope data of the original response time series data within a preset time window, and mapping the initial response slope data with the theoretical fluid transport delay data to generate diffusion compensation time parameter data characterizing the differences in the activity of volatile molecules in different teas, as detailed below:

[0182] Step 1: Extract initial response slope data:

[0183] Conventional electronic nose algorithms typically extract the maximum voltage value of the sensor after it reaches steady state as the comparison feature. However, the physical diffusion rates of different odor molecules in Liubao tea (such as small molecules corresponding to musty smells and large terpenoid molecules corresponding to aged smells) in the air vary objectively, resulting in different steepness of the initial rise of the sensor output curve. If only static extrema are used, the system will lose the transient dynamics information of the odor molecules in the early stage of volatilization, making it difficult to distinguish similar odors with similar final concentrations but different chemical compositions. In addition, the method of directly subtracting adjacent points to obtain the slope is highly susceptible to high-frequency hardware noise interference, leading to distortion of feature parameters. Therefore, this paper proposes to extract the initial response slope data of the original response time series data within a preset time window, specifically including:

[0184] In the original response time series data, the voltage amplitude is continuously monitored. When the voltage amplitude deviates from the initial baseline data to the preset trigger threshold, the current moment is extracted as the response start moment data.

[0185] Starting from the data at the start of the response, a continuous data segment of a preset time length is extracted and used as the feature sequence data within the preset time window;

[0186] According to the preset sampling frequency, extract multiple discrete voltage sampling point data and the timestamp corresponding to each voltage sampling point data from the feature sequence data;

[0187] A univariate linear regression calculation was performed on all voltage sampling point data and timestamps to obtain the slope parameter of the fitted line;

[0188] The slope parameter is assigned to generate the initial response slope data.

[0189] Initial baseline data: refers to the average background output voltage of the sensor array 9 collected by the microprocessor 7 in a clean air environment before the introduction of volatile gases from Liubao tea. It is used as a numerical reference to determine whether the sensor has a chemical response.

[0190] Preset trigger threshold: refers to the minimum voltage change set by the system to determine the effective contact of odor molecules with the sensor. It is a physical boundary parameter to prevent false judgments caused by background electrical noise.

[0191] Response start time data: refers to the time point when the sensor's output voltage deviates from the initial baseline data and the deviation reaches the preset trigger threshold, representing the zero point time when odor molecules actually cause a chemical reaction in the sensor.

[0192] Preset time length: refers to the time span set by the system to capture the early changes in the sensor signal, forming a time window to lock in the initial dynamic diffusion stage of volatile gases.

[0193] Feature sequence data: refers to a local continuous data subset extracted from the original response time series data based on the response start time data and a preset time length.

[0194] Initial response slope data: refers to the overall rate of change of voltage in the characteristic sequence data calculated by the linear regression algorithm within a preset time window, used to quantify the transient activity of diffusion of a specific volatile molecule.

[0195] Step 1: Identify the response start time data: During system initialization, microprocessor 7 reads the initial baseline data under clean air conditions. Then, it calculates the absolute difference between the current voltage amplitude and the initial baseline data point-by-point in the original response time series data. When this difference first exceeds or equals a preset trigger threshold, microprocessor 7 records the current timestamp. The specific formula is as follows:

[0196] ;

[0197] in, For the aforementioned obtained number One sensor in The raw voltage amplitude data at each moment, in volts; The initial baseline data, in volts, is collected and recorded by the microprocessor 7 before measurement. The preset trigger threshold is configured with a preset value range of 0.02 to 0.10 volts. The condition that this inequality is met... That is, the data assigned to the start time of the response is denoted as . By establishing an anti-interference signal triggering mechanism, fluctuations in electrical background noise are effectively filtered out, and the true starting point of the substantial chemical reaction between volatile molecules and the sensor is accurately located.

[0198] Step 2: Extract Feature Sequence Data: The microprocessor 7 uses the identified response start time as the starting point and a preset time length fixed by the system as the span to establish an observation window on the time axis. It then independently extracts the original response time sequence data segments falling within this window. The preset time length is typically configured to range from 1.0 to 5.0 seconds. The specific formula is as follows:

[0199] ;

[0200] in, Define the domain of the preset time window; The data is in response to the start time, and the unit is seconds; The preset time duration is in seconds. (Falling in) The voltage sampling data within the interval constitutes the feature sequence data. By defining the time boundary for feature extraction, the steady-state or saturated data from the later stages of sensor processing are discarded, and computational resources are concentrated on the initial diffusion stage, which best reflects the differences in molecular activity.

[0201] Step 3: Extract Sampling Points and Timestamps: Following the preset sampling frequency set in the previous steps, the microprocessor 7 sequentially reads the valid data points within the feature sequence data. Each discrete voltage value within this interval is defined as a voltage sampling point data point, and its corresponding relative time is defined as the corresponding timestamp. Assume a total of [number missing] sampling points are extracted within this window. One data pair.

[0202] Step 4: Univariate linear regression calculation: Microprocessor 7 performs the extracted... A univariate linear regression calculation using least squares was performed on the voltage sampling data and their corresponding timestamps to derive the slope parameter of the fitted line. The specific formula is as follows:

[0203] ;

[0204] in, The slope parameter; This represents the total number of sampling points included within the preset time window. This is the index number of the local sampling point within the window, with a value ranging from 1 to... Positive integers; For the first characteristic sequence data Each timestamp is in seconds; For the first characteristic sequence data Data from each voltage sampling point, in volts; This is the arithmetic mean of all timestamps within the window; This is the arithmetic mean of all voltage sampling points within the window. By finding the globally optimal linear trend among multiple sampling points containing discrete noise, the problem of traditional two-point differentiation being easily affected by a single noise point is effectively avoided, generating a transient rate of change feature with high robustness.

[0205] Step 5: Generate initial response slope data: Microprocessor 7 will calculate the slope parameters. The value is directly assigned and defined as the initial response slope data for subsequent diffusion parameter mapping.

[0206] By converting static amplitude extraction into dynamic rate of change extraction, the activity differences (i.e., transient dynamic characteristics) of different volatile molecules in tea leaves during the initial contact with the sensor were successfully captured. By employing a combination of threshold triggering and interval linear regression, the interference of sudden electrical noise changes at single sampling points was effectively suppressed, generating high signal-to-noise ratio and robust initial response slope data, providing reliable chemical transient input for subsequent fluid delay mapping and feature matrix construction.

[0207] Example 3: Continuing from the previous example, the sampling frequency configured for microprocessor 7... Hertz, the system sets the initial baseline data for the first sensor. Volts, preset trigger threshold Volts, preset time length Seconds. The first four sampling points acquired by microprocessor 7 are:

[0208] ; ; ; .

[0209] Determine the start time of the response: Calculate the difference:

[0210] time: Volt, less than Volts, not triggered.

[0211] time: Volts, greater than the preset trigger threshold Volt, trigger.

[0212] Obtain response start time data Second.

[0213] Extracting feature sequence data: the time window interval is Seconds. The sampling points falling within this interval are , , Total points .

[0214] Calculate the average:

[0215] Time average Second.

[0216] Average voltage volt.

[0217] Calculating the slope of a univariate linear regression:

[0218] Molecular part:

[0219] ;

[0220] Denominator:

[0221] ;

[0222] Initial response slope data Volts per second.

[0223] Step 2: Generate diffusion compensation time parameter data:

[0224] Theoretical fluid transport delay data only considers the static space of the physical pipeline and the thermal expansion and contraction of the gas, but ignores two key dynamic variables: first, the diffusion acceleration of different molecules in Liubao tea (such as large molecules of aged flavor substances and small molecules of musty flavor substances) in the air varies objectively; second, after long-term operation, mechanical wear of the pump body will cause its actual traction force to decrease month by month. If the theoretical delay is directly used for time axis calibration, residual time differences will exist when aligning different molecules and equipment with different service cycles, reducing the accuracy of characteristic odor identification. To address this, we propose mapping the initial response slope data with the theoretical fluid transport delay data to generate diffusion compensation time parameter data characterizing the differences in the activity of volatile molecules in different teas, specifically including:

[0225] Convert the initial response slope data into response acceleration data;

[0226] Acquire the system's preset acceleration decay constant data, pump body aging weight coefficient data, and reference offset data, and read the current cumulative running time data of sampling pump 10 in real time;

[0227] Calculate the product of the response acceleration data and the acceleration decay constant data to obtain the decay power variable data;

[0228] Using the natural constant as the base and the negative number of the decay power variable data as the exponent, perform exponential calculation to generate basic activity coefficient data;

[0229] The cumulative runtime data is multiplied with the pump body aging weight coefficient data, and the product result is added to the baseline offset data to calculate the mechanical attenuation correction factor data.

[0230] The basic activity coefficient data is corrected based on the mechanical decay correction factor data to generate a scaling factor;

[0231] The theoretical fluid transport delay data is scaled based on a scaling factor to generate diffusion compensation time parameter data.

[0232] Response acceleration data: refers to the rate of change calculated by comparing the initial response slope under different time windows, used to quantify the dynamic acceleration of the diffusion motion of volatile gas molecules of Liubao tea on the sensor surface.

[0233] Acceleration decay constant data: refers to a dimensionless constant preset by the system, used to map linearly changing acceleration to a nonlinear exponential decay model, which characterizes the diffusion damping of gas molecules in a specific fluid environment.

[0234] Pump body aging weight coefficient data: refers to the proportional coefficient preset by the system to quantify the mechanical wear degree of sampling pump 10, representing the contribution of unit running time to the decrease in pump body flow rate.

[0235] Reference offset data: refers to the system preset constant, representing the initial mechanical health reference value of sampling pump 10 in its brand new factory condition (i.e., when the cumulative running time is zero).

[0236] Cumulative runtime data: refers to the total operating time of sampling pump 10 since its first activation, recorded by the non-volatile memory inside microprocessor 7.

[0237] Decaying power variable data: refers to the transitional parameter that serves as an intermediate step in exponential operations, representing the product effect of fluid dynamics and molecular diffusion characteristics.

[0238] Basic activity coefficient data: refers to the dimensionless proportionality coefficient that reflects the diffusion and penetration ability of a specific gas molecule in the current environment after removing mechanical factors.

[0239] Mechanical attenuation correction factor data: refers to the numerical variable used to compensate for the decrease in pumping power of sampling pump 10 due to long-term operation.

[0240] Scaling factor: refers to a combined multiplier that integrates chemical diffusion properties and hardware mechanical properties, used for bidirectional scaling of purely theoretical delay time.

[0241] Diffusion compensation time parameter data: refers to the final calculated actual compensation time value used to directly align the time axis offset of the original sensor waveform.

[0242] Step 1: Convert the initial response slope data into response acceleration data:

[0243] A single initial response slope (first derivative) only reflects the average rate of change of gas concentration within a local time period. However, volatile components of different molecular weights in Liubao tea (such as light alcohols and heavy terpenes) not only have different initial diffusion rates in a closed pipe, but also exhibit significantly different degrees of rate decay or acceleration. If the algorithm only focuses on the slope dimension, the system will be unable to distinguish between different chemical substances with the same initial velocity but insufficient subsequent momentum and those with the same initial velocity but sustained subsequent momentum, thus reducing the resolution of odor identification. Therefore, it is proposed to convert the initial response slope data into response acceleration data, specifically including:

[0244] The window slides at a preset time step to continuously extract the initial response slope data of at least two adjacent time windows;

[0245] Calculate the difference between the initial response slope data of the next time window and the initial response slope data of the previous time window;

[0246] The ratio of the difference to the preset time step is calculated to obtain the response acceleration data.

[0247] Preset sampling period: refers to the time step between two consecutive triggering of feature sequence data extraction and linear regression calculation within a preset time window by the microprocessor 7, used to standardize the time interval limit for dynamic feature extraction.

[0248] Preset time step: A fixed time span constant for the microprocessor 7 when moving forward the truncation window on the time series data. Its value is preset to be between 0.1s and 1.0s, and the specific value is determined by the physical sampling frequency of the sensor during system calibration and stored in read-only memory.

[0249] Adjacent time windows: In the same segment of feature sequence data, two data segments of the same length are extracted, and their starting time points differ by a preset time step.

[0250] Time window sliding and slope extraction: On the feature sequence data, the microprocessor 7 locates the data by delaying the start of the next time window by a preset time step relative to the start of the previous time window. The initial response slope data of the previous time window is generated by the univariate linear regression calculation in the preceding step and is temporarily stored in a register. Based on the location, the microprocessor obtains the data for the next time window and calls the aforementioned univariate linear regression model to calculate the initial response slope data for the next time window. The calculation formula for the initial time relationship is as follows:

[0251] ;

[0252] in, This is the start time data for the next time window, in seconds (s). This is the start time data for the previous time window, in seconds (s). The preset time step is in seconds (s). By establishing two observation intervals with a defined time interval on the time axis, a data extraction benchmark that strictly corresponds in the time dimension is provided for subsequent difference calculations.

[0253] Calculating the difference: Microprocessor 7 subtracts the initial response slope data of the next time window from the initial response slope data of the previous time window to obtain the difference between the two. The specific formula is as follows:

[0254] ;

[0255] in, The difference is calculated in volts per second. This is the initial response slope data for the next time window, in volts per second. The initial response slope data for the previous time window is expressed in volts per second. By quantifying the fluctuations in the rate of change of gas concentration at adjacent time points, the static velocity dimension is removed, and the pure trend increment is obtained.

[0256] Calculating response acceleration data: Microprocessor 7 extracts the absolute value of the difference and calculates its ratio with the time interval data corresponding to the preset sampling period. The specific formula is as follows:

[0257] ;

[0258] in, For response acceleration data, the unit is volts per square second; This is the absolute value of the difference calculated in the preceding steps, in volts per second. The preset time step is in seconds. Discrete slope increments are converted into continuous rates of change over time, generating standard acceleration parameters characterizing the diffusion dynamics of volatile molecules, providing scientific input for subsequent characterization of material activity.

[0259] By quantifying the dynamic rate of change of the slope, the algorithm can keenly capture the microscopic dynamic differences of different odor molecules in the initial contact with the sensor. Through feature extraction methods that delve into the essence of physicochemical diffusion, the system's ability to identify highly similar and easily confused odor components in Liubao tea is effectively improved.

[0260] Step 2: Obtain preset data and cumulative running time: The microprocessor 7 retrieves a set of pre-set constants from the system memory: acceleration decay constant data, pump body aging weight coefficient data, and reference offset data. Simultaneously, the microprocessor 7 reads the real-time updated cumulative running time data of the sampling pump 10. The acceleration decay constant data is typically configured with a value range of 0.1 to 1.0; the pump body aging weight coefficient data is typically configured with a value range of 0.0001 to 0.005; and the reference offset data is fixed at a value of 1.0.

[0261] Step 3: Calculate the decay power variable data: The microprocessor 7 multiplies the response acceleration data with the acceleration decay constant data. The specific formula is as follows:

[0262] ;

[0263] in, Data for decaying power variables; The system is pre-set with acceleration decay constant data. By associating the extracted purely electrical characteristic dimensions with the fluid damping model, the linear acceleration characteristics are transformed into input parameters suitable for nonlinear mapping.

[0264] Step 4: Generate basic activity coefficient data: Microprocessor 7 uses the natural constant as the base and the negative number of the decaying power variable data as the exponent to perform exponential evaluation. The specific formula is as follows:

[0265] ;

[0266] in, Based on basic activity coefficient data; The constant is a natural constant with a value of approximately 2.71828. By simulating the natural physical laws of volatile gases diffusing rapidly towards saturation in a confined space, a dimensionless coefficient between 0 and 1 is generated, objectively quantifying the differences in chemical activity among different tea molecules.

[0267] Step 5: Calculate the mechanical attenuation correction factor data: The microprocessor 7 multiplies the acquired cumulative runtime data with the pump body aging weight coefficient data, and then adds the baseline offset data to the product. The specific formula is as follows:

[0268] ;

[0269] in, Data for mechanical attenuation correction factor; This is the current cumulative runtime data of the sampling pump, in hours; This is the data for the pump body aging weighting coefficient; This serves as the baseline offset data. By establishing a mechanical compensation model for the linear aging of hardware over time, the reduction in pumping capacity caused by long-term service of the equipment is offset, thus preventing baseline drift of the detection system as its service life increases.

[0270] Step 6: Generate scaling factor: Microprocessor 7 multiplies the base activity coefficient data with the mechanical decay correction factor data. The specific formula is as follows:

[0271] ;

[0272] in, This is the scaling factor. By mathematically fusing the purely chemical molecular diffusion characteristics with the purely mechanical equipment aging status, a comprehensive calibration multiplier that takes into account multiple variables of both hardware and software is output.

[0273] Step 7: Generate diffusion compensation time parameter data: Finally, microprocessor 7 uses a scaling factor to perform a scaling product operation on the theoretical fluid transport delay data calculated in the previous step. The specific formula is as follows:

[0274] ;

[0275] in, This is the diffusion compensation time parameter data, in seconds; The theoretical fluid transport delay data, in seconds, is generated from the calculations in the preceding steps. By dynamically calibrating the static time delay derived from pure physical space, a true time offset absolutely bound to the specific odor component being measured and the current specific hardware state is obtained, which is used for subsequent high-precision time alignment.

[0276] An exponential model realistically recreates the nonlinear diffusion process of molecules, while a linear compensation model offsets hardware wear variables. The generated diffusion compensation time parameter data is adaptable to various volatile components and detection devices with different levels of aging, ensuring that subsequent time axis calibration has high-fidelity physicochemical significance.

[0277] Example 4: Following the calculation example above, microprocessor 7 has already... Instant The initial response slope data of the previous time window is calculated within a window of seconds. and theoretical fluid transport delay data .

[0278] Extracting continuous slope data: The system's preset time step is 0.2 seconds. Microprocessor 7 performs the same linear regression calculation in the next step window (e.g., 0.4 seconds to 0.6 seconds) to obtain the initial response slope data for the subsequent time window. .

[0279] Calculate the difference: .

[0280] Calculate the response acceleration data: .

[0281] Obtaining preset parameters and runtime: Microprocessor 7 reads the acceleration decay constant data set by the system. Pump body aging weighting coefficient data Baseline offset data The microprocessor 7 reads the system's non-volatile memory to obtain the cumulative runtime data of the current sampling pump. Hour.

[0282] Calculate the decay power variable data: .

[0283] Generate basic activity coefficient data: .

[0284] Calculate the mechanical attenuation correction factor data: .

[0285] Generate scaling factor: .

[0286] Generate diffusion compensation time parameter data: .

[0287] In the aforementioned technology, the differences in activity of different tea volatile molecules in the initial contact with the sensor are objectively quantified by extracting the initial response slope and acceleration. Mechanical attenuation correction is introduced to offset the dynamic inconsistencies caused by long-term hardware operation. The generated diffusion compensation time parameter data not only includes the static physical delay of the pipeline space but also takes into account the chemical diffusion characteristics of gas molecules and the dynamic wear state of the pump body, providing an accurate time offset calibration basis for the subsequent generation of a high-precision pure chemical feature matrix.

[0288] (iv) Generation of pure chemical spatiotemporal feature matrix data:

[0289] The raw response time series data consists of physical electrical signals collected under specific pipe lengths, diameters, and flow rates. This data includes time delays caused by fluid dynamics and concentration dissipation due to diffusion and adsorption during gas transport. If this data, containing physical spatial interference, is directly input into a classification model, the system may mistakenly identify waveform lags or amplitude attenuations caused by non-chemical factors as characteristics of the tea aroma itself. This leads to inconsistencies in features across different devices, reducing the accuracy of aroma classification. Therefore, this paper proposes using diffusion-compensated time parameter data to perform time axis offset calibration and amplitude weighting on the raw response time series data, generating a pure chemical spatiotemporal feature matrix data stripped of physical spatial interference, as detailed below:

[0290] Generating pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference, specifically including:

[0291] Subtract the diffusion compensation time parameter data from the timestamps corresponding to each voltage sampling point in the original response time series data to obtain time axis aligned data that eliminates non-uniform diffusion time difference;

[0292] Input the diffusion compensation time parameter data into the preset concentration decay function to calculate and generate the dimensionless amplitude compensation weight coefficient.

[0293] The original voltage amplitude data in the original response time series data is multiplied by the amplitude compensation weighting coefficient to obtain the restored true concentration amplitude data.

[0294] The time-axis aligned data is matrix-stitched with the actual concentration amplitude data to generate a pure chemical spatiotemporal feature matrix.

[0295] Concentration decay function: refers to a mathematical model preset by the system to quantify the law of decrease in actual concentration of gas molecules over time due to fluid dynamic dissipation and physical adsorption of the container wall during pipeline transmission.

[0296] Amplitude compensation weighting coefficient: refers to the dimensionless proportional multiplier calculated based on the concentration attenuation function, used to amplify voltage signals affected by physical transmission losses by a corresponding proportion.

[0297] Time-aligned data: refers to the relative time node sequence obtained by uniformly subtracting the physical transmission and diffusion delay from the timestamps of the original response signal, and mapping it to the zero point of the actual chemical reaction.

[0298] True concentration amplitude data: refers to the actual concentration mapping voltage of odor molecules after scaling the original voltage amplitude collected by the sensor using the above weighting coefficients, thus removing the physical spatial loss.

[0299] Pure chemical spatiotemporal characteristic matrix data: refers to a two-dimensional data structure composed of a calibrated time axis sequence and a reduced concentration amplitude sequence in a corresponding relationship. This matrix only characterizes the chemical kinetic properties of the characteristic aroma of tea and is not affected by the mechanical or spatial variables of the detection device.

[0300] Step 1: Obtain Time Axis Alignment Data: The microprocessor 7 extracts the timestamps corresponding to each discrete voltage sampling point in the original response time series data, and uniformly subtracts the diffusion compensation time parameter data calculated in the previous step from each timestamp value. The specific formula is as follows:

[0301] ;

[0302] in, Align the data to the timeline, in seconds; The first in the original response time series data The relative timestamps corresponding to each voltage sampling point are in seconds; This is the diffusion compensation time parameter data generated previously, in seconds. By performing a translation operation on the time axis in the mathematical coordinate system, the non-uniform diffusion time difference caused by pipeline length, pump aging, and differences in gas diffusion is eliminated, so that the response signals collected by different devices are aligned to the same physical reference on the time axis.

[0303] Step 2: Calculate the amplitude compensation weighting coefficient: The microprocessor calls a preset concentration decay function to transform the time-dimensional delay variable into a concentration-dimensional compensation variable. In the embodiments of this application, a compensation model based on time-linear decay is used to establish the correlation. The specific formula is as follows:

[0304] ;

[0305] in, The amplitude compensation weighting coefficient is a dimensionless parameter. The pre-calibrated and stored pipe wall adsorption dissipation constant is used to characterize the concentration loss rate per unit time for a specific pipe diameter, in reciprocal seconds. The preset value range is configured to be 0.1 to 3.0; The data represents the diffusion compensation time parameter, in seconds. By establishing a physically consistent positive correlation between time delay and concentration dissipation, a dimensionless gain multiplier greater than 1 is generated to eliminate the dimension mismatch problem and quantify the signal loss caused by physical transmission.

[0306] Step 3: Obtain the true concentration amplitude data: The microprocessor 7 uses the calculated amplitude compensation weighting coefficient as a multiplier and performs a point-by-point multiplication operation with each voltage amplitude data point in the original response time series data. The specific formula is as follows:

[0307] ;

[0308] in, The values ​​are the restored true concentration amplitude data, in volts; For the first One sensor in The raw voltage amplitude data collected at all times, in volts; The amplitude compensation weighting coefficient is calculated for the preceding steps. By scaling the amplitude of the raw electrical signal output by the sensor, the signal amplitude, which has been weakened by physical space loss, is restored to the actual chemical reaction concentration level.

[0309] Step 4: Generate pure chemical spatiotemporal feature matrix data: Microprocessor 7 constructs a column vector from the time-axis aligned data corresponding to all sampling points, and constructs another column vector from all the corresponding true concentration amplitude data, then concatenates the two into a matrix. The specific formula is as follows:

[0310] ;

[0311] in, The generated purely chemical spatiotemporal feature matrix data; For all of them arranged in sampling order The constructed time column vector; For those arranged in sampling order, and One-to-one correspondence of all The resulting amplitude column vector. By integrating the spatiotemporal calibration results, a structured and standardized two-dimensional feature matrix is ​​constructed, which serves as the standard input for subsequent fingerprint comparison.

[0312] Example 5: Continuing from the previous example, the microprocessor has obtained the original timestamps of the first 3 sampling points. The original voltage amplitude data of the first sensor for 0.1s, 0.2s, and 0.3s are respectively. The voltage levels are 1.25V, 1.30V, and 1.42V. This is based on the diffusion compensation time parameter data generated from previous calculations. Second.

[0313] Calculate timeline aligned data:

[0314] .

[0315] .

[0316] .

[0317] Calculate the amplitude compensation weighting coefficient: .

[0318] Calculate the actual concentration amplitude data:

[0319] .

[0320] .

[0321] .

[0322] Generate matrix data:

[0323] By concatenating the two columns of data above, a pure chemical spatiotemporal feature matrix is ​​generated. .

[0324] In the aforementioned technology, by eliminating the time difference caused by non-uniform diffusion and physical pipelines in the time dimension, and compensating for signal attenuation during transmission in the spatial (concentration) dimension, the electrical signal, which is dependent on specific hardware conditions, is transformed into standardized data that is independent of the physical structure of the detection device through this joint calibration and matrix reconstruction of spatiotemporal axes. The generated pure chemical spatiotemporal feature matrix data retains only the essential characteristics of the chemical reaction between the volatile odor molecules of Liubao tea and the sensor, thereby improving the consistency of the data and the reliability of subsequent pattern recognition.

[0325] (v) Generation of characteristic odor classification results data:

[0326] The characteristic aroma of Liubao tea is composed of a mixture of various complex volatile components. Traditional electronic nose algorithms typically extract only the maximum steady-state voltage as a single feature for threshold comparison. This single-dimensional comparison method ignores the dynamic response process of aroma molecules over time throughout the detection period, making it difficult for the system to distinguish between similar aromas with similar components but different concentrations or release rates (e.g., a slight "musty" smell versus the "stuffy" smell of early fermentation), resulting in low aroma classification accuracy. To address this, we propose comparing and mapping pure chemical spatiotemporal feature matrix data with a pre-defined standard aroma fingerprint database to determine and generate characteristic aroma classification results for Liubao tea, as detailed below:

[0327] The system identifies and generates characteristic aroma classification results for Liubao tea, specifically including:

[0328] Obtain multiple standard odor template matrix data pre-stored in the standard odor fingerprint spectrum database, as well as odor category label data corresponding to each standard odor template matrix data;

[0329] The pure chemical spatiotemporal feature matrix data was subtracted element-wise from the standard odor template matrix data to obtain multiple residual matrix data.

[0330] The spatial distance data corresponding to each standard odor category is calculated by summing the squares of all elements in each residual matrix data.

[0331] Compare the numerical values ​​of all spatial distance data and select the spatial distance data with the smallest value as the target distance data.

[0332] Extract the odor category label data associated with the target distance data, and generate the odor category label data as the characteristic odor classification result data for Liubao tea.

[0333] Standard odor fingerprint database: refers to a set of data pre-stored in the non-volatile memory of the microprocessor 7, which contains benchmark data generated by calibrating Liubao tea samples with known characteristic odors multiple times under standard conditions.

[0334] Standard odor template matrix data: refers to a two-dimensional mathematical matrix structure pre-established for the known typical odor types of Liubao tea. This structure is completely consistent with the number of rows and columns of the pure chemical spatiotemporal feature matrix data, and is used as a benchmark template for comparison.

[0335] Odor category label data: refers to text classification identifiers that establish a one-to-one mapping relationship with each standard odor template matrix data. Their preset values ​​include "mold", "stuffy", "warehouse", and "stale".

[0336] Residual matrix data: refers to a new matrix that records the numerical deviation of each coordinate point after comparing and subtracting the pure chemical spatiotemporal characteristic matrix data of the sample to be tested with the standard odor template matrix data of a specific category item by item.

[0337] Spatial distance data: refers to a scalar value used to objectively measure the overall similarity between two matrices of the same dimension. The smaller the value, the higher the degree of matching between the data to be tested and the standard spectrum in terms of temporal dynamics and concentration amplitude characteristics.

[0338] Target distance data: refers to the data with the smallest value among all spatial distance data obtained after global numerical comparison and filtering.

[0339] Step 1: Obtain standard odor template matrix data and tag data: Microprocessor 7 accesses non-volatile memory via the bus interface and calls the pre-stored data in the standard odor fingerprint database. A standard odor template matrix and corresponding odor category label data. The default value is a positive integer greater than or equal to 4. Microprocessor 7 establishes a mapping read list based on the index number. The specific formula is as follows:

[0340] ;

[0341] in, A set of standards built inside the microprocessor 7; The total number of standard odor categories; This is the category index number within the set, with a value ranging from 1 to... Positive integers; For the corresponding index number The A standard odor template matrix data; For the corresponding index number The Odor category label data. By constructing a standard comparison target set within the microprocessor 7, a clear reference origin is provided for subsequent similarity calculations.

[0342] Step 2: Calculate the residual matrix data: Microprocessor 7 performs matrix subtraction on the pure chemical spatiotemporal feature matrix data generated in the previous step and on each standard odor template matrix data in the list. Specifically, it subtracts elements in the same row and column from each other in the two matrices. The specific formula is as follows:

[0343] ;

[0344] in, For the corresponding to the first Residual matrix data for each standard odor category; This is the pure chemical spatiotemporal feature matrix data generated by the preceding steps; For the first A standard odor template matrix data is generated. By calculating the direct differences between the test sample features and the standard template features at each discrete time point and concentration point, a deviation distribution matrix that preserves the spatiotemporal topology is generated, thereby capturing local mismatch phenomena.

[0345] Step 3: Calculate spatial distance data: For each newly generated residual matrix, the microprocessor 7 extracts all element values, squares them, and then sums all the squared values. The specific formula is as follows:

[0346] ;

[0347] in, For the corresponding to the first Spatial distance data for each standard odor category; This represents the total number of rows in the matrix, which is the total number of sampling points extracted within the preceding time window. The row coordinates of the matrix are variables, taking values ​​from 1 to... ; This is the total number of columns in the matrix. The fixed value is 2 (corresponding to the time column and amplitude column); The column coordinates of the matrix take values ​​from 1 to... ; For residual matrix data The middle is located in the first line, number The deviation element values ​​of the column. The overall similarity between matrices is measured by the sum of squares of the Euclidean distance. Positive and negative deviations in the matrix are uniformly converted into positive penalty values ​​and globally accumulated, thereby objectively quantifying the comprehensive spatiotemporal difference level between the test signal sequence and the standard signal sequence.

[0348] Step 4: Filter target distance data: Microprocessor 7 traverses all We take several spatial distance data points, sort them by value, compare them, and then select the one with the smallest value. The specific formula is as follows:

[0349] ;

[0350] in, For target distance data; to These are the spatial distance data corresponding to each classification index number. By performing the least-squares optimal matching selection, the known standard sample category that is closest to the comprehensive characteristics of the sample to be tested is located.

[0351] Step 5: Generate characteristic odor classification result data:

[0352] Microprocessor 7 based on the filtered target distance data Index number From the standard set Extract the corresponding odor category label data and output the results via touch screen 2. The specific formula is as follows:

[0353] ;

[0354] in, This is data on the classification of characteristic aromas of Liubao tea; For target distance data The corresponding index number in the set; This is the odor category label data corresponding to the index number. The result of the underlying mathematical operations is mapped back to specific business semantics and then sent to the touch screen 2 for display.

[0355] Example 6: Continuing from the previous example, microprocessor 7 has generated data containing 3 time points (number of rows). ), 2 dimensions (number of columns) Pure chemical spatiotemporal feature matrix data The system database is configured to store two types of data (…). Standard template: Index Corresponding tags It smells "musty"; Index Corresponding tags It has a "aged flavor".

[0356] Previous output The data is as follows:

[0357] Line 1: ;

[0358] Line 2: ;

[0359] Line 3: .

[0360] Retrieve standard odor template matrix data:

[0361] Assuming the "mold" template is retrieved from the database for:

[0362] Line 1: ;

[0363] Line 2: ;

[0364] Line 3: .

[0365] Assuming the "aged flavor" template is retrieved from the database for:

[0366] Line 1: ;

[0367] Line 2: ;

[0368] Line 3: .

[0369] Calculate the residual matrix data:

[0370] Regarding "musty smell" ( ), :

[0371] First row difference: ;

[0372] Second row difference: ;

[0373] Third row difference: .

[0374] Regarding "aged flavor" ( ), :

[0375] First row difference: ;

[0376] Second row difference: ;

[0377] Third row difference: .

[0378] Calculate spatial distance data:

[0379] ;

[0380] .

[0381] Filter target distance data:

[0382] Compare and ,Discover .

[0383] Select the minimum value to obtain the target distance data. .

[0384] Generate classification result data:

[0385] Due to target distance data Source: Index number Microprocessor 7 extracts the corresponding odor category label data. (i.e., "musty smell").

[0386] The final characteristic odor classification results data The value is assigned as "musty smell", and the classification result is displayed on the touch screen 2.

[0387] The aforementioned technology preserves the global response topology of the gas-sensitive sensor array 9 for Liubao tea volatiles at the matrix level. By summing the residual squares element-by-element, it nonlinearly amplifies and globally accumulates the local minute differences in the time and concentration dimensions, providing an objective and rigorous similarity measurement mechanism. Finding the matching method with the minimum spatial distance ensures that the classification result is based on a comprehensive judgment of the entire odor reaction dynamics process, improving the system's ability to analyze complex and overlapping odor characteristics and its classification accuracy.

[0388] Gas filtration and emission:

[0389] An activated carbon filter 8 is also provided inside the main body 1. The air inlet of the activated carbon filter 8 is connected to the air outlet of the gas sensor array 9 pipeline, and the air outlet of the activated carbon filter 8 is connected to the external environment of the main body 1.

[0390] Specifically, after the gas is detected by the gas-sensitive sensor array 9, it is filtered through the activated carbon filter 8 and then discharged outside the main body 1 to avoid secondary pollution to the environment.

[0391] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea, characterized in that, include: The sample chamber consists of a main body, a sampling chamber located within the main body, and a heating cylinder located within the sampling chamber. The top of the sampling chamber is provided with a sealing cover, which forms a sealing structure between the sampling chamber and the heating cylinder. The heating temperature of the heating cylinder is controlled within the range of 40-60℃. A gas-sensitive sensor array connected to the heating cylinder and a sampling pump installed on the gas pipeline between the heating cylinder and the gas-sensitive sensor array; The microprocessor is configured to include the following functional modules: The data acquisition module is used to acquire the current heating temperature data of the sampling chamber, the physical diameter and physical length data of the gas pipeline located between the heating cylinder and the gas sensor array, the rated flow rate data of the sampling pump, and the raw response time series data acquired by the gas sensor array in real time after the sampling pump is started. The theoretical delay calculation module is used to fuse and calculate the current heating temperature data, physical pipe diameter data, physical length data and rated flow rate data to generate theoretical fluid transmission delay data that characterizes the current gas path environment. The compensation parameter generation module is used to extract the initial response slope data of the original response time series data within a preset time window, and to map the initial response slope data with the theoretical fluid transport delay data to generate diffusion compensation time parameter data that characterizes the differences in the activity of volatile molecules in different tea leaves. The feature calibration module is used to perform time axis offset calibration and amplitude weighting on the original response time series data using diffusion compensation time parameter data, generating pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference; The classification and recognition module is used to compare and map the pure chemical spatiotemporal feature matrix data with the preset standard odor fingerprint spectrum database, and to judge and generate characteristic odor classification result data for Liubao tea.

2. The gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: Extract the initial response slope data of the original response time series data within a preset time window, specifically including: In the original response time series data, the voltage amplitude is continuously monitored. When the voltage amplitude deviates from the initial baseline data to the preset trigger threshold, the current moment is extracted as the response start moment data. Starting from the data at the start of the response, a continuous data segment of a preset time length is extracted and used as the feature sequence data within the preset time window; According to the preset sampling frequency, extract multiple discrete voltage sampling point data and the timestamp corresponding to each voltage sampling point data from the feature sequence data; A univariate linear regression calculation was performed on all voltage sampling point data and timestamps to obtain the slope parameter of the fitted line; The slope parameter is assigned to generate the initial response slope data.

3. The gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: The initial response slope data is mapped to theoretical fluid transport delay data to generate diffusion compensation time parameter data characterizing the differences in the activity of volatile molecules in different tea leaves. Specifically, this includes: Convert the initial response slope data into response acceleration data; Acquire the system's preset acceleration decay constant data, pump body aging weight coefficient data, and reference offset data, and read the current cumulative running time data of the sampling pump in real time; Calculate the product of the response acceleration data and the acceleration decay constant data to obtain the decay power variable data; Using the natural constant as the base and the negative number of the decay power variable data as the exponent, perform exponential calculation to generate basic activity coefficient data; The cumulative runtime data is multiplied with the pump body aging weight coefficient data, and the product result is added to the baseline offset data to calculate the mechanical attenuation correction factor data. The basic activity coefficient data is corrected based on the mechanical decay correction factor data to generate a scaling factor; The theoretical fluid transport delay data is scaled based on a scaling factor to generate diffusion compensation time parameter data.

4. The gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 3, characterized in that: The initial response slope data is converted into response acceleration data, specifically including: The window slides at a preset time step to continuously extract the initial response slope data of at least two adjacent time windows; Calculate the difference between the initial response slope data of the next time window and the initial response slope data of the previous time window; The ratio of the difference to the preset time step is calculated to obtain the response acceleration data.

5. The gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: Generating pure chemical spatiotemporal feature matrix data stripped of hardware physical space interference, specifically including: Subtract the diffusion compensation time parameter data from the timestamps corresponding to each voltage sampling point in the original response time series data to obtain time axis aligned data that eliminates non-uniform diffusion time difference; Input the diffusion compensation time parameter data into the preset concentration decay function to calculate and generate the dimensionless amplitude compensation weight coefficient. The original voltage amplitude data in the original response time series data is multiplied by the amplitude compensation weighting coefficient to obtain the restored true concentration amplitude data. The time-axis aligned data is matrix-stitched with the actual concentration amplitude data to generate a pure chemical spatiotemporal feature matrix.

6. The gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: The system identifies and generates characteristic aroma classification results for Liubao tea, specifically including: Obtain multiple standard odor template matrix data pre-stored in the standard odor fingerprint spectrum database, as well as odor category label data corresponding to each standard odor template matrix data; The pure chemical spatiotemporal feature matrix data was subtracted element-wise from the standard odor template matrix data to obtain multiple residual matrix data. The spatial distance data corresponding to each standard odor category is calculated by summing the squares of all elements in each residual matrix data. Compare the numerical values ​​of all spatial distance data and select the spatial distance data with the smallest value as the target distance data. Extract the odor category label data associated with the target distance data, and generate the odor category label data as the characteristic odor classification result data for Liubao tea.

7. A gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: Generate theoretical fluid transport delay data characterizing the current gas path environment, specifically including: Obtain the initial ambient temperature data of the system's environment, and convert the current heating temperature data and the initial ambient temperature data into absolute temperatures to obtain the current absolute temperature data and the initial absolute temperature data; Calculate the ratio of the current absolute temperature data to the initial absolute temperature data, and use the ratio as the correction variable for the actual flow velocity of the gas after thermal expansion; The rated flow velocity data is corrected based on the actual flow velocity correction variable data to obtain the true flow velocity data inside the pipe; The volume data of the gas pipeline located between the heating cylinder and the gas sensor array is calculated using the physical diameter and physical length data of the gas pipeline. Calculate the ratio of pipe volume data to actual pipe velocity data to generate theoretical fluid transport delay data.

8. The gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: The gas sensor array includes at least a first sensor sensitive to microbial metabolites, a second sensor sensitive to alcohols and esters, a third sensor sensitive to aldehydes and phenols, and a fourth sensor sensitive to terpenes.

9. A gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: The main body is also equipped with an activated carbon filter. The air inlet of the activated carbon filter is connected to the air outlet of the gas-sensitive sensor array pipeline, and the air outlet of the activated carbon filter is connected to the external environment of the main body.

10. A gas-sensitive array detection device for rapid identification of characteristic aromas of Liubao tea according to claim 1, characterized in that: The outer surface of the body is provided with a touch screen display.

Images