A rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion
By extracting the dynamic features of the electronic nose and electronic tongue response curves of Chinese herbal medicine slices, screening effective channels and calculating the similarity index, the problem of insufficient identification of subtle quality differences in electronic sensory fusion detection is solved, and high-precision and rapid detection of the quality of Chinese herbal medicine slices is realized.
Patent Information
- Application Number
- CN202611167957.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-03
- Publication Date
- 2026-11-10
AI Technical Summary
In existing technologies, the dynamic morphological information in the response process is not used in electronic sensor fusion detection, resulting in insufficient ability of features to represent subtle quality differences. At the same time, the lack of effective screening of multi-channel features reduces detection accuracy.
By extracting the time-domain response curves of the electronic nose and electronic tongue, the peak arrival time, rise rate, rise segment fluctuation and decay rate are determined to form a four-dimensional morphological feature vector. Based on the correlation coefficient, effective channels are selected and spliced into a fused morphological feature vector. The cosine similarity and comprehensive quality index are calculated to output the quality level.
It enhances the sensitivity and differentiation ability of Chinese herbal medicine slices, improves detection accuracy and reliability, ensures information integrity and discrimination efficiency, and improves the accuracy and stability of detection results.
Smart Images

Figure CN123066166A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of quality testing technology for traditional Chinese medicine decoction pieces, specifically to a rapid quality testing method for traditional Chinese medicine decoction pieces based on electronic sensory fusion. Background Technology
[0002] With the development of biomimetic sensing technology, electronic noses and electronic tongues have been gradually introduced into the detection of odor and taste in traditional Chinese medicine (TCM) decoction pieces. Electronic noses, simulating human olfactory perception, utilize a gas-sensitive sensor array to respond to volatile components released from the decoction piece sample, converting chemical signals into electrical signals. Electronic tongues, simulating human taste perception, detect non-volatile taste components in the aqueous extract of the decoction piece using a taste sensor. Both types of devices can rapidly acquire multidimensional sensor data related to the odor and taste of the decoction piece, and the detection process is objective, repeatable, and digital. They play an important role in rapid quality detection methods for TCM decoction pieces and provide feasible technical means for achieving objective and standardized evaluation of TCM decoction piece quality.
[0003] For example, Chinese publication CN 113705719 discloses a method for rapidly evaluating the five medicinal properties of Chinese herbal medicine slices using electronic sensory fusion. This method rapidly and accurately evaluates the four natures and five medicinal properties of Chinese herbal medicine slices using electronic sensory fusion. It can quickly identify and evaluate the five natures and four natures of Chinese herbal medicine slices, effectively solving the problems of strong subjectivity and poor repeatability of manual identification and the time-consuming, labor-intensive, and poor specificity of modern analytical methods. It is an innovation in the method of evaluating the four natures and five medicinal properties of Chinese herbal medicine slices.
[0004] However, the above method has the following technical drawbacks:
[0005] 1. Feature extraction only uses the steady-state or maximum value of the sensor response curve, neglecting dynamic morphological information such as the rise rate, peak time, fluctuation pattern, and decay rate during the response process. This results in insufficient characterization of subtle quality differences and reduced detection accuracy. In actual use, after contact with the sample, the response signal of the electronic nose and electronic tongue sensors undergoes a complete dynamic process over time, from rapid rise to saturation, and then possibly accompanied by slight decay. Different qualities of traditional Chinese medicine decoction pieces not only differ in steady-state response values but also in dynamic morphological characteristics such as the rise rate of the response curve, the time to reach the peak, the fluctuation pattern of the rise process, and the rate of decay.
[0006] 2. When the number of sensor channels of electronic nose and electronic tongue is large, the features of each channel can only be directly spliced and input into the classifier without any effective feature screening, which can easily reduce the detection accuracy. Summary of the Invention
[0007] To address this, the present invention provides a rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion, which solves the problems in the prior art where the dynamic morphological information in the response process is not used, resulting in insufficient characterization ability of features for subtle quality differences, and the lack of effective screening of multi-channel features.
[0008] To achieve the above objectives, the present invention provides the following technical solution:
[0009] A rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion includes the following steps:
[0010] S1: Take the powdered Chinese herbal medicine sample to be tested, divide it into two portions, and detect it using an electronic nose and an electronic tongue under constant temperature and humidity conditions to obtain the time domain response curve of the electronic nose and the time domain response curve of the electronic tongue.
[0011] S2: By determining the overall curve of the electronic nose time-domain response curve and the electronic tongue time-domain response curve, the peak arrival time, rise rate, rise segment fluctuation and average decay rate from the peak point to the steady state point are determined, and a four-dimensional morphological feature vector for each channel is constructed.
[0012] S3: By calculating the electronic nose correlation coefficient and electronic tongue correlation coefficient between the rise rate of each sensor channel and the quality grade of the medicinal slices, effective channels with correlation coefficients greater than the preset threshold are selected, irrelevant channels with weak correlation are removed, and the four-dimensional morphological feature vectors of the retained effective channels are sequentially spliced into a fused morphological feature vector.
[0013] S4: Based on the fused morphological feature vectors of standard samples of each level after effective channel screening and the morphological feature templates of each level, calculate the cosine similarity and comprehensive quality index of the fused morphological feature vectors of the sample to be tested and the templates of each level.
[0014] S5: Compare the overall quality index with the preset judgment intervals for each level one by one, and output the quality level results.
[0015] Furthermore, the specific steps of S1 are as follows:
[0016] S1.1: Take the sample of Chinese herbal medicine to be tested, crush it and sieve it. Take the powder that passes through the sieve as the sample to be tested and divide it into two portions. The first portion of the sample is used for electronic nose detection; the second portion of the sample is used for electronic tongue detection.
[0017] S1.2: Acquisition of dynamic response curve of electronic nose;
[0018] S1.3: Acquisition of dynamic response curve of electronic tongue.
[0019] Furthermore, the specific steps of S2 are as follows:
[0020] S2.1: The electronic nose time-domain response curve and the electronic tongue time-domain response curve determined in S1 are collectively referred to as the total curve;
[0021] S2.2: Determine the global maximum response value of each total curve over the entire sampling time range, record the maximum value as the peak response value, and record the sampling time corresponding to the first occurrence of the maximum value as the peak arrival time;
[0022] S2.3: Calculate the rate of rise from the initial response value to the peak point based on the peak response value and the peak arrival time;
[0023] S2.4: Calculate the variability of the rising segment from the initial response value to the peak response value;
[0024] S2.5: Calculate the average decay rate from the peak response value to the steady-state point, reflecting how quickly the response signal decays to the steady state after reaching the peak value;
[0025] S2.6: The rise rate, peak arrival time, rise segment volatility, and average decay rate are combined to form the four-dimensional morphological feature vector of the channel.
[0026] Furthermore, S3 includes the following steps:
[0027] S3.1: The ascent rate in the standard sample grade of the same variety of Chinese herbal medicine decoction pieces, the four-dimensional morphological feature vector of the electronic nose, and the four-dimensional morphological feature vector of the electronic tongue. Calculate the correlation coefficient of the electronic nose respectively. Correlation coefficient with electronic tongue This is used to measure the degree of monotonic correlation between two variables;
[0028] S3.2: Combine the preset correlation screening threshold to analyze the electronic nose correlation coefficients of each channel. And the correlation coefficients of the electronic tongue for each channel. The effective channels for the electronic nose and the electronic tongue were screened to determine the effective channels for both.
[0029] S3.3: The set of effective electronic nose channels and the set of effective electronic tongue channels retained after filtering in S3.2 are then concatenated end to end according to the order of electronic nose channels first and electronic tongue channels, to obtain the fused morphological feature vector.
[0030] Furthermore, the determination rule for S3.2 is as follows:
[0031] First, the correlation coefficient of the electronic nose... Judge one by one: If Preset relevance screening threshold This indicates the current rate of ascent of the channel. There is a positive correlation between the quality grade of the medicinal slices and the determination of the grade. These channels are valid and will be reserved.
[0032] like Preset relevance screening threshold This indicates that the correlation between the dynamic response characteristics of the current channel and the quality grade of the medicinal slices is weak, and the current channel is determined to be an invalid channel and is removed. Finally, all electronic nose channels determined to be valid channels constitute the set of valid electronic nose channels.
[0033] Then, the correlation coefficient of the electronic tongue was analyzed. Perform filtering: If Preset relevance screening threshold Then determine the first These channels are valid and will be reserved.
[0034] like Preset relevance screening threshold If a channel is deemed invalid, it is discarded. Finally, all electronic tongue channels that are deemed valid are combined to form a set of valid electronic tongue channels.
[0035] Furthermore, the specific steps of S4 are as follows:
[0036] S4.1: Based on the fusion morphological feature vectors of all batches of standard samples, calculate the arithmetic mean element by element, and use the arithmetic mean as the first... Rank morphological feature template ;
[0037] S4.2: The fused morphological feature vector of the sample to be tested output from S3. Morphological feature templates output by S4.1 Perform cosine similarity calculation to obtain the cosine similarity. ;
[0038] S4.3: Combining cosine similarity at each level Calculate the overall quality index , The range of values is , The higher the value, the higher the morphological similarity between the sample being tested and the high-quality template.
[0039] Furthermore, the cosine similarity The calculation formula is as follows:
[0040]
[0041] in, For fusion morphological feature vectors The Euclidean norm; morphological feature template The Euclidean norm.
[0042] Furthermore, the comprehensive quality index The calculation formula is as follows:
[0043]
[0044] in, For the preset first Grade weighting coefficient; for .
[0045] The present invention has the following advantages: By extracting a four-dimensional morphological feature vector for each response curve, including the rise rate, peak arrival time, rise segment fluctuation, and average decay rate, the present invention can effectively identify subtle differences in the speed of response rise, the timing of peak appearance, the degree of fluctuation during the rise process, and the decay relaxation rate of different quality Chinese herbal medicine slices. This enhances the sensitivity and distinguishing ability of the features to changes in quality grade, thereby effectively solving the problem of insufficient steady-state value features in characterizing subtle quality differences and improving detection accuracy and reliability.
[0046] Meanwhile, by calculating the correlation coefficients of the electronic nose and electronic tongue based on the monotonic correlation between the quality grade of the standard sample and the rate of increase of each channel, and combining the preset correlation screening threshold, the channels are judged one by one, invalid channels with weak correlation with the quality grade are eliminated, and the set of valid channels is retained. This avoids the interference of irrelevant features on subsequent similarity matching, thereby improving the discrimination efficiency of feature input while ensuring the integrity of information.
[0047] In addition, by calculating the cosine similarity between the fused morphological feature vector of the sample to be tested and each template, and then combining it with the preset weights to calculate the comprehensive quality index, the grade results are output. Moreover, the judgment interval of each grade is determined based on the actual distribution statistics of the standard sample. Therefore, the final output quality grade improves the accuracy and stability of the judgment results while ensuring the detection speed.
[0048] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. Attached Figure Description
[0049] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).
[0050] Figure 1 This is a flowchart illustrating the implementation of the rapid quality detection method for traditional Chinese medicine decoction pieces based on electronic sensory fusion according to the present invention.
[0051] Figure 2 This is a diagram illustrating the generation architecture of the four-dimensional morphological feature vector in the rapid quality detection method for traditional Chinese medicine decoction pieces based on electronic sensory fusion of this invention. Detailed Implementation
[0052] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Technical engineers in the field can make some non-essential improvements and adjustments to the present invention based on the above-described content. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0053] Please see Figure 1 A rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion includes the following steps:
[0054] S1: Take the powdered Chinese herbal medicine sample to be tested, divide it into two portions, and detect it using an electronic nose and an electronic tongue under constant temperature and humidity conditions. Obtain the time-domain response curve of the electronic nose and the time-domain response curve of the electronic tongue. This allows for complete recording of the entire dynamic process information of the sensor from contact with the sample to response equilibrium, including the signal rise phase, response peak, and possible decay phase. This provides raw data for subsequent extraction of morphological parameters that reflect the dynamic behavior of the response, such as rise rate, peak arrival time, rise segment fluctuation, and decay rate.
[0055] The specific steps of S1 are as follows:
[0056] S1.1: Take the sample of Chinese herbal medicine to be tested, crush it and sieve it. Take the powder that passes through the sieve as the sample to be tested and divide it into two portions. Place the first portion of the sample in the headspace sampling bottle of the electronic nose and seal it at a temperature of 20-25℃ and a relative humidity of 45-55% to allow the volatile odor substances to reach gas-solid equilibrium in the bottle;
[0057] The second sample is placed in the sample cup of the electronic tongue, and deionized water is added at a material-to-liquid ratio of 1:10-1:60. The sample is then sealed and allowed to stand at a temperature of 20-25℃ and a relative humidity of 45-55% to allow the soluble flavor substances to fully dissolve and form the test solution.
[0058] S1.2: Electronic Nose Dynamic Response Curve Acquisition: The electronic nose device performs continuous headspace sampling of the sample under test. Starting from the moment the sensor contacts the headspace gas of the sample, this contact moment is recorded as the initial sampling moment. The response signals of each sensor are continuously acquired and recorded at fixed sampling intervals, resulting in a sequence of response values for each channel over time, which is recorded as the electronic nose time-domain response curve. The sampling process continues until the signals of all channels reach a stable state, i.e., the change amplitude of the response values at multiple consecutive sampling points is less than a preset threshold (e.g., 0.5-0.8% of full scale). At this point, acquisition stops, thus obtaining the complete electronic nose time-domain response curve.
[0059] S1.3: Acquisition of Dynamic Response Curve of Electronic Tongue: An electronic tongue device is used to continuously detect the taste of the sample. The sensor array of the electronic tongue includes n taste sensors. The moment the sensor is immersed in the test solution is recorded as the initial sampling time. The membrane potential of each sensor selectively responds to the taste substances in the solution, and the potential value changes dynamically over time. The potential response signals of each sensor are continuously acquired and recorded at the same fixed sampling interval, resulting in a sequence of response values for each channel over time, which is recorded as the time-domain response curve of the electronic tongue. Sampling continues until all channel signals reach a stable state, i.e., the change amplitude of the response value at multiple consecutive sampling points is less than a preset threshold. Acquisition is then stopped, thus obtaining the complete time-domain response curve of the electronic tongue.
[0060] S2: By determining the overall curve of the electronic nose time-domain response curve and the electronic tongue time-domain response curve, the peak arrival time, rise rate, rise segment fluctuation, and average decay rate from the peak point to the steady state point are determined, and a four-dimensional morphological feature vector for each channel is constructed. This facilitates the transformation of the dynamic response process into a unified four-dimensional feature vector, thereby making it easier to determine the difference information from the time-domain waveform and providing feature data for subsequent channel screening and similarity matching.
[0061] The specific steps of S2 are as follows:
[0062] S2.1: The time-domain response curves of the electronic nose and electronic tongue determined in S1 are collectively denoted as the overall curve. , , This represents the total number of sampling points.
[0063] S2.2: Determine each total curve The global maximum response value over the entire sampling time range is denoted as the peak response value. The sampling time corresponding to the first occurrence of this maximum value is recorded as the peak arrival time. , representing the time required for the sensor response signal to reach its peak value from its initial state. The starting sampling time The initial response value corresponding to 0. If multiple sampling times correspond to the same global maximum response value during the scan, the time when that value is first reached is taken as the peak arrival time. .
[0064] S2.3: Based on peak response value and peak arrival time Calculate the initial response value to peak point The rate of increase between This reflects how quickly the sensor's response signal reaches its maximum value after contacting the sample. The calculation formula is as follows:
[0065]
[0066] in, This is the starting sampling time.
[0067] S2.4: Calculate the initial response value Peak response value The volatility of the rising segment within the rising segment This reflects the degree to which the actual response curve deviates from the ideal linear growth during its ascent. The closer the actual curve is to a straight upward line, the better. The smaller the value; if the curve shows obvious fluctuations or phased increases, then... The larger it is. The calculation formula is as follows:
[0068]
[0069] S2.5: Calculate the peak response value Average decay rate between the steady state and the steady state This reflects how quickly the response signal decays towards a steady state after reaching its peak. If the curve shows no significant decay after reaching its peak, that is... and If they are basically equal, then It approaches zero. The calculation formula is as follows:
[0070]
[0071] in, This represents the average response of the sampling points in this segment after the curve enters a steady state. This represents the initial moment of steady state.
[0072] S2.6: The rate of increase Peak arrival time Volatility during the upward phase and average decay rate The combination constitutes the four-dimensional morphological feature vector of this channel.
[0073] For electronic noses The four-dimensional morphological feature vectors extracted from each channel are denoted as follows: , ; For electronic tongue The four-dimensional morphological feature vectors extracted from each channel are denoted as follows: , .
[0074] S3: Channel Correlation Screening and Feature Fusion: This involves calculating the electronic nose correlation coefficient between the rise rate of each sensor channel and the quality grade of the medicinal slices. Correlation coefficient with electronic tongue The process involves selecting effective channels with correlation coefficients greater than a preset threshold, discarding irrelevant channels with weak correlations, and then sequentially concatenating the four-dimensional morphological feature vectors of the remaining effective channels into a fused morphological feature vector. This facilitates the selection of effective channels with significant correlations to the quality grade of medicinal slices from all sensor channels of the electronic nose and electronic tongue, while discarding irrelevant channels with weak correlations to quality. The denoised fused morphological feature vector is formed by concatenating channel features, providing effective feature input for subsequent morphological similarity matching.
[0075] S3 includes the following steps:
[0076] S3.1: The ascent rate in the standard sample grade of the same variety of Chinese herbal medicine decoction pieces, the four-dimensional morphological feature vector of the electronic nose, and the four-dimensional morphological feature vector of the electronic tongue. Calculate the correlation coefficient of the electronic nose respectively. Correlation coefficient with electronic tongue This metric measures the degree of monotonic correlation between two variables, and its value ranges from -1 to 1. The closer the absolute value is to 1, the stronger the correlation. The calculation formula is as follows:
[0077]
[0078]
[0079] in, For the electronic nose The rate of rise of each channel on all standard samples; For electronic tongue The rate of rise of each channel on all standard samples; A vector is formed by assigning values to the quality grades of each standard sample, with the first grade assigned the value K and the Kth grade assigned the value 1. This is a rank transformation function, used to convert a raw data sequence into a sorted rank. Let covariance function be used. The standard deviation is denoted as .
[0080] The above standard sample grades are determined by traditional experience and classified according to their quality. Levels (e.g., Level 1, Level 2, ..., Level 3) Quality levels, where level 1 is the highest quality level. Convert quality levels to numerical values: assign level 1 a value... The second level is assigned the value -1, and so on, the th The level is assigned a value of 1.
[0081] S3.2: Combine with preset relevance screening thresholds The correlation coefficients of the electronic nose for each channel were analyzed separately. And the correlation coefficients of the electronic tongue for each channel. The effective channels for the electronic nose and the electronic tongue were screened to determine the effective channels.
[0082] The range of values for the relevance screening threshold is: The basis for setting this threshold is that when the absolute value of the correlation coefficient is greater than 0.3, it is generally considered that there is a moderate or higher degree of monotonic correlation between the two variables. Setting the upper limit of the correlation screening threshold to 0.6 is to retain a sufficient number of effective channels and avoid excessive loss of effective information due to an excessively high threshold.
[0083] Correlation coefficient of electronic nose Judge one by one: If This indicates the rate of ascent of the channel. There is a positive correlation between the response dynamic characteristics of this channel and the quality grade of the medicinal slices, meaning that the dynamic characteristics of this channel are sensitive to changes in the quality of the medicinal slices. These channels are valid and will be reserved.
[0084] like This indicates that the correlation between the dynamic response characteristics of this channel and the quality grade of the medicinal slices is weak, therefore this channel is deemed invalid and discarded. The morphological feature vector of this channel... It does not participate in subsequent processing steps. All electronic nose channels determined to be valid channels constitute the set of valid electronic nose channels.
[0085] Using the exact same judgment rules, the correlation coefficient of the electronic tongue was evaluated. Perform filtering: If Then determine the first One channel is a valid channel and will be reserved; if If the morphological feature vector of the channel is invalid, it will be discarded. It does not participate in subsequent processing steps. All electronic tongue channels determined to be valid channels constitute the set of valid electronic tongue channels.
[0086] S3.3: After filtering in S3.2, the effective channels of the electronic nose and the effective channels of the electronic tongue are concatenated end-to-end, following the order of electronic nose channels first and then electronic tongue channels, to obtain a fused morphological feature vector. .
[0087] S4: Based on the fused morphological feature vectors of standard samples of each grade after effective channel screening and the morphological feature templates of each grade, calculate the cosine similarity and comprehensive quality index of the fused morphological feature vectors of the sample to be tested and the templates of each grade, so as to reflect the overall matching of the Chinese medicine sample to be tested with each grade.
[0088] The specific steps for S4 are as follows:
[0089] S4.1: Based on the batch standard samples used in S3. After completing the effective channel screening described in step 3.3, a unified set of effective channels for the electronic nose and the electronic tongue is determined. For each batch of standard samples within each level, the effective channel morphological feature vectors are concatenated to form a fused morphological feature vector, using the same method as in S3.3. For the first... grade Take the fusion morphological feature vectors of all batches of standard samples within that grade, calculate the arithmetic mean of each element, and use this mean vector as the first... Rank morphological feature template .
[0090] S4.2: The fused morphological feature vector of the sample to be tested output from S3. Morphological feature templates output by S4.1 Perform cosine similarity calculation to obtain the cosine similarity. The range of values for cosine similarity is: The closer the value is to 1, the better the fusion morphological feature vector of the sample under test is to the first digit. The more consistent the orientation of the morphological feature templates in the vector space, the higher the degree of morphological similarity. Cosine similarity. The calculation formula is as follows:
[0091]
[0092] in, For fusion morphological feature vectors The Euclidean norm; morphological feature template The Euclidean norm.
[0093] S4.3: Combining cosine similarity at each level Calculate the overall quality index , The range of values is , The higher the value, the greater the morphological similarity between the sample and the high-quality template, indicating better overall intrinsic quality of the sample. The calculation formula is as follows:
[0094]
[0095] in, For the preset first The grade weighting coefficient corresponds to a higher grade and a larger weight.
[0096] S5: Incorporate quality index It compares the results against preset level ranges one by one and outputs the quality level results. This allows for the determination of the overall quality index. It transforms the results into intuitive quality grade judgments and outputs them, enabling rapid detection and grade determination of the quality of Chinese herbal medicine slices.
[0097] The method for establishing the preset quality grade judgment interval is as follows: based on the N batches of standard samples with known quality grades used in S3, calculate the comprehensive quality index of each batch of standard samples one by one according to the method described in S4. Values, statistics corresponding to each quality level The range of value distribution is used to determine the relationship between adjacent levels. Value boundary threshold.
[0098] For example, Angelica sinensis slices are divided into three grades: first-class, second-class, and third-class (K=3). Statistical analysis determined the judgment interval as follows: It was determined to be a first-class product. It was determined to be a second-class product. It was determined to be a third-class product.
[0099] A specific embodiment is also given below.
[0100] This embodiment uses Astragalus membranaceus (Huang Qi) as the detection object, and employs an electronic nose device (sensor channels m=10) and an electronic tongue device (sensor channels n=8) for rapid quality grade detection. The quality grade of Astragalus membranaceus is divided into three levels, K=3: Level 1 is Grade 1, Level 2 is Grade 2, and Level 3 is Grade 3. Quality grade assignments are: Grade 1 = 3, Grade 2 = 2, Grade 3 = 1.
[0101] First, a total of 30 batches of Astragalus membranaceus standard samples were collected, which were identified by traditional experience and confirmed by high performance liquid chromatography fingerprinting. Among them, there were 10 batches each of first-class, second-class and third-class products.
[0102] Each batch of Astragalus membranaceus standard sample was pulverized and passed through an 80-mesh sieve. Approximately 5.0 grams of the sieved powder was divided into two portions. The first portion was placed in an electronic nose headspace sampling bottle and sealed and left to stand for 30 minutes at 25°C and 50% relative humidity. Starting from the moment the sensor contacts the headspace gas, the electronic nose continuously collected response signals from each channel at 0.1-second intervals until the change in response value at 10 consecutive sampling points was less than 0.5% of the full scale. At this point, the collection was stopped, resulting in 10 electronic nose time-domain response curves.
[0103] The second sample was placed in the electronic tongue sample cup, and deionized water was added at a material-to-liquid ratio of 1:50. The sample was then shaken and extracted for 5 minutes under the same temperature and humidity conditions. The electronic tongue began immersing the eight taste sensors in the test solution the instant the sensor was placed in the solution. Potential response signals from each channel were continuously acquired at 0.1-second intervals. Stability was determined by the variation amplitude at 10 consecutive sampling points being less than 0.5% of the full scale, resulting in eight electronic tongue time-domain response curves.
[0104] For each response curve, the peak response value, peak arrival time, initial response value, and steady-state average response were determined by scanning. Taking a batch of first-grade samples from the third channel of the electronic nose as an example, the initial response value was 0.03, the peak response value was 0.78, the peak arrival time was at the 380th sampling point, the steady-state start time was at the 2200th sampling point, and the steady-state average response was 0.65. Therefore, the following were calculated: the rise rate was 0.00197 per sampling point; the peak arrival time was taken as 380; the average absolute value of the rise segment fluctuation, calculated point-by-point, was 0.028; and the average decay rate was 0.0000714. The four-dimensional morphological feature vector of this channel was obtained as [0.00197, 380, 0.028, 0.0000714]. The above morphological feature extraction was completed for all 18 channels of all 30 batches of standard samples.
[0105] The rise rate of each channel in each batch was extracted, and the Spearman rank correlation coefficient between the rate and the quality grade assignment was calculated. A correlation screening threshold of 0.4 was set. After screening, 6 out of 10 channels of the electronic nose had correlation coefficients greater than 0.4, constituting the effective channel set for the electronic nose; similarly, 4 out of 8 channels of the electronic tongue had correlation coefficients greater than 0.4, constituting the effective channel set for the electronic tongue. The four-dimensional morphological feature vectors of the effective channels from each batch were concatenated in the order of electronic nose first, then electronic tongue, resulting in a fused morphological feature vector with a dimension of 40.
[0106] The morphological feature vectors of 10 batches of samples within each grade were arithmetically averaged element by element to obtain the morphological feature templates for first-grade, second-grade, and third-grade products. Weighting coefficients were set as follows: 0.5 for first-grade, 0.3 for second-grade, and 0.2 for third-grade.
[0107] The cosine similarity of 30 batches of standard samples with three grade templates was calculated and then weighted and summed to obtain the comprehensive quality index. The statistical determination intervals were as follows: a comprehensive quality index greater than or equal to 0.48 was classified as first-grade, greater than or equal to 0.22 and less than 0.48 as second-grade, and less than 0.22 as third-grade.
[0108] Take the Astragalus membranaceus slices to be tested, pulverize them through an 80-mesh sieve, take 5.0 grams of the sieved powder and divide them into two equal portions. Collect the time-domain response curves of the electronic nose and the electronic tongue under the same conditions as above, extract the four-dimensional morphological feature vectors of each channel, and use the determined effective channel set of the electronic nose and the effective channel set of the electronic tongue to screen and splice them to obtain the fused morphological feature vector of the sample to be tested.
[0109] The cosine similarity between the fused morphological feature vector of the sample to be tested and the templates of the three grades is calculated: the cosine similarity with the morphological feature template of the first grade is 0.85, the cosine similarity with the morphological feature template of the second grade is 0.12, and the cosine similarity with the morphological feature template of the third grade is 0.03.
[0110] The calculated comprehensive quality index is 0.467. Since 0.467 falls within the range of 0.22 or greater and less than 0.48, the tested Astragalus membranaceus slices are classified as Grade II. The output comprehensive quality index is 0.467, and the quality grade is Grade II.
[0111] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion, characterized in that, Includes the following steps: S1: Take the powdered Chinese herbal medicine sample to be tested, divide it into two portions, and detect it using an electronic nose and an electronic tongue under constant temperature and humidity conditions to obtain the time domain response curve of the electronic nose and the time domain response curve of the electronic tongue. S2: By determining the overall curve of the electronic nose time-domain response curve and the electronic tongue time-domain response curve, the peak arrival time, rise rate, rise segment fluctuation and average decay rate from the peak point to the steady state point are determined, and a four-dimensional morphological feature vector for each channel is constructed. S3: By calculating the electronic nose correlation coefficient and electronic tongue correlation coefficient between the rise rate of each sensor channel and the quality grade of the medicinal slices, effective channels with correlation coefficients greater than the preset threshold are selected, irrelevant channels with weak correlation are removed, and the four-dimensional morphological feature vectors of the retained effective channels are sequentially spliced into a fused morphological feature vector. S4: Based on the fused morphological feature vectors of standard samples of each level after effective channel screening and the morphological feature templates of each level, calculate the cosine similarity and comprehensive quality index of the fused morphological feature vectors of the sample to be tested and the templates of each level. S5: Compare the overall quality index with the preset judgment intervals for each level one by one, and output the quality level results.
2. The rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion according to claim 1, characterized in that, The specific steps of S1 are as follows: S1.1: Take the sample of Chinese herbal medicine to be tested, crush it and sieve it. Take the powder that passes through the sieve as the sample to be tested and divide it into two portions. The first portion of the sample is used for electronic nose detection; the second portion of the sample is used for electronic tongue detection. S1.2: Acquisition of dynamic response curve of electronic nose; S1.3: Acquisition of dynamic response curve of electronic tongue.
3. The rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion according to claim 1, characterized in that, The specific steps of S2 are as follows: S2.1: The electronic nose time-domain response curve and the electronic tongue time-domain response curve determined in S1 are collectively referred to as the total curve; S2.2: Determine the global maximum response value of each total curve over the entire sampling time range, record the maximum value as the peak response value, and record the sampling time corresponding to the first occurrence of the maximum value as the peak arrival time; S2.3: Calculate the rate of rise from the initial response value to the peak point based on the peak response value and the peak arrival time; S2.4: Calculate the variability of the rising segment from the initial response value to the peak response value; S2.5: Calculate the average decay rate from the peak response value to the steady-state point, reflecting how quickly the response signal decays to the steady state after reaching the peak value; S2.6: The rise rate, peak arrival time, rise segment volatility, and average decay rate are combined to form the four-dimensional morphological feature vector of the channel.
4. The rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion according to claim 1, characterized in that, S3 includes the following steps: S3.1: The ascent rate in the standard sample grade of the same variety of Chinese herbal medicine decoction pieces, the four-dimensional morphological feature vector of the electronic nose, and the four-dimensional morphological feature vector of the electronic tongue. Calculate the correlation coefficient of the electronic nose respectively. Correlation coefficient with electronic tongue This is used to measure the degree of monotonic correlation between two variables; S3.2: Combine the preset correlation screening threshold to analyze the electronic nose correlation coefficients of each channel. And the correlation coefficients of the electronic tongue for each channel. The effective channels for the electronic nose and the electronic tongue were screened to determine the effective channels for both. S3.3: The set of effective electronic nose channels and the set of effective electronic tongue channels retained after filtering in S3.2 are then concatenated end to end according to the order of electronic nose channels first and electronic tongue channels, to obtain the fused morphological feature vector.
5. The rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion according to claim 4, characterized in that, The determination rule for S3.2 is as follows: First, the correlation coefficient of the electronic nose... Judge one by one: if Preset relevance screening threshold This indicates the current rate of ascent of the channel. There is a positive correlation between the quality grade of the medicinal slices and the determination of the grade. These channels are valid and will be reserved. like Preset relevance screening threshold This indicates that the correlation between the dynamic response characteristics of the current channel and the quality grade of the medicinal slices is weak, and the current channel is determined to be an invalid channel and is removed. Finally, all electronic nose channels determined to be valid channels constitute the set of valid electronic nose channels. Then, the correlation coefficient of the electronic tongue was analyzed. Perform filtering: If Preset relevance screening threshold Then determine the first These channels are valid and will be reserved. like Preset relevance screening threshold If a channel is deemed invalid, it is discarded. Finally, all electronic tongue channels that are deemed valid are combined to form a set of valid electronic tongue channels.
6. The rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion according to claim 1, characterized in that, The specific steps of S4 are as follows: S4.1: Based on the fusion morphological feature vectors of all batches of standard samples, calculate the arithmetic mean element by element, and use the arithmetic mean as the first... Rank morphological feature template ; S4.2: The fused morphological feature vector of the sample to be tested output from S3. Morphological feature templates output by S4.1 Perform cosine similarity calculation to obtain the cosine similarity. ; S4.3: Combining cosine similarity at each level Calculate the overall quality index , The range of values is , The higher the value, the higher the morphological similarity between the sample being tested and the high-quality template.
7. The rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion according to claim 1, characterized in that, The cosine similarity The calculation formula is as follows: in, For fusion morphological feature vectors The Euclidean norm; morphological feature template The Euclidean norm.
8. The rapid detection method for the quality of traditional Chinese medicine decoction pieces based on electronic sensory fusion according to claim 1, characterized in that, The comprehensive quality index The calculation formula is as follows: in, For the preset first Grade weighting coefficient; for .
