Intelligent pesticide residue detection device and method

Abstract

The invention relates to the technical field of pesticide detection, and particularly discloses an intelligent pesticide residue detection method which comprises the following steps: S1, matrix interfering substance identification: weighing a preset mass of a to-be-detected sample, putting the to-be-detected sample into a spectral imaging detection instrument, collecting a diffuse reflection spectrum of the sample through a near infrared spectrometer, and determining the matrix interfering substance in the to-be-detected sample; combining a convolutional neural network model to identify crop types, and synchronously estimating relative proportions of fibers, starch and lipid to obtain matrix components; s2, a pulsed field cooperative dynamic extraction step: according to the identified matrix components, performing differential parameter regulation and control in an ultrasonic-microwave-infrared composite field extraction cavity; when the lipid proportion is larger than a first preset value, microwave and infrared synergistic heating is started to reach 50-60 DEG C, normal hexane saturated acetonitrile extraction is matched, and meanwhile the turbidity of extract liquor is monitored in real time through a camera; according to the technical scheme, aiming at the characteristics of the grain matrix, the problems of insufficient efficiency and serious matrix interference can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of pesticide detection, and in particular to an intelligent pesticide residue detection device and method. Background Art

[0002] In the field of food safety testing, succinate dehydrogenase inhibitors (SDHIs) are a new class of pesticides widely used in crops such as cereals and oilseeds. However, because these pesticides often have residues in grains at levels of micrograms per kilogram (ppb), and the grain matrix itself contains various interfering substances such as fiber, starch, and lipids, existing detection technologies face numerous challenges in practical application.

[0003] In terms of extraction efficiency, traditional detection methods mostly use liquid-liquid extraction combined with mechanical oscillation. When using oscillation extraction with an acetonitrile-acetic acid solution, the extraction rate of SDHIs in grains is generally only 70%-80%. This is particularly true for pesticides bound to starch granules, such as fluopyram, which struggle to completely release them with traditional oscillation. The cellulose and protein in grains encapsulate pesticide molecules, making it difficult to disrupt this encapsulation with conventional oscillation methods. Consequently, the detection limit for LC-QQQ testing is limited to 0.1-0.5 μg / kg, which falls short of the EU limit of 0.01 mg / kg for some SDHIs.

[0004] Regarding matrix interference, even with traditional dispersive solid-phase extraction (d-SPE) cleanup, 20%-30% of impurities such as lipids and pigments in grains will remain. Once these impurities enter the LC-TOF / MS, they can significantly suppress the detection. For example, fatty acids in wheat adsorb on the surface of the mass spectrometer ion source, reducing the ionization efficiency of flupyraclostrobin by over 40%. This shifts mass accuracy from <5ppm to over 10ppm, making it impossible to distinguish even compounds with similar masses.

[0005] Existing methods use uniform extraction parameters for different types of food, such as cereals, oilseeds, and nuts, failing to fully account for the differences between different matrices. For example, oilseeds, with their high fat content, require more robust degreasing methods. Current methods, using C18 adsorbents, achieve oil removal rates of less than 60%, significantly interfering with subsequent mass spectrometry analysis. Furthermore, reagent dosages are manually adjusted, leading to a repeatability of the relative standard deviation (RSD) exceeding 15% when dealing with large fluctuations in SDHI residue levels in food.

[0006] The traditional testing process requires vortexing, oscillation, and centrifugation, all pretreatment steps, which take a total of 1.5-2 hours. When the SDHIs content in grain approaches the limit of quantification, LC-QQQ, while accurate, can only detect known compounds. While LC-TOF / MS can screen for unknowns, its pretreatment efficiency is insufficient, hindering its high-resolution advantages and resulting in a high false-positive rate.

[0007] In general, the current detection technology has insufficient extraction efficiency and serious matrix interference when facing low-content SDHIs in grains. There is an urgent need for an intelligent pesticide residue detection device and method tailored to the characteristics of the grain matrix. Summary of the Invention

[0008] The present invention provides an intelligent pesticide residue detection device and method, which, based on the characteristics of grain matrix, can solve the problems of insufficient efficiency and severe matrix interference.

[0009] In order to solve the above technical problems, this application provides the following technical solutions: An intelligent pesticide residue detection method, characterized by comprising: S1 Matrix Interfering Substance Identification Step: Weigh a preset mass of the test sample and place it in a spectral imaging detection instrument. Use a near-infrared spectrometer to collect the sample's diffuse reflectance spectrum. Combined with a convolutional neural network model, the crop species are identified and the relative proportions of fiber, starch, and lipids are simultaneously estimated to obtain the matrix composition. S2 pulsed field synergistic dynamic extraction step: Based on the identified matrix components, differentiated parameter regulation is performed in the ultrasound-microwave-infrared composite field extraction chamber; When the lipid ratio is greater than the first preset value, microwave and infrared synergistic heating to 50-60°C is activated, and extraction is carried out with n-hexane saturated acetonitrile. At the same time, the turbidity of the extract is monitored in real time through a camera. When the turbidity of the extract is greater than the second preset value, the ultrasonic cavitation effect is automatically enhanced until the turbidity is less than the third preset value. When the fiber content is greater than the fourth preset value, a pulsed electric field is used to destroy the plant cell wall and the conductivity change of the extract is recorded simultaneously. When the conductivity growth rate is less than the fifth preset value, the pulse frequency is automatically adjusted to 15-20Hz to increase the SDHIs release rate. During the extraction process, the temperature of the extract is collected at preset intervals, and the heating power is dynamically corrected to ensure that the sample temperature difference is less than 5°C; S3 gradient intelligent purification step: If the lipid content is greater than 30%, add 150 mg / g of sample to the magnetic bead mixing chamber of PSA / C18 magnetic microspheres, drive mixing for 2 minutes using a rotating magnetic field, and then use an acetonitrile solution containing 8% acetic acid in a microfluidic chip at a gradient flow rate of 0.5 mL / min for elution; If the starch content is greater than 40%, graphite adsorbent is added to enhance the turbulence effect through the spiral channel of the microfluidic chip to complete the purification. S4 mass spectrometry differential detection steps: First, unknown SDHIs are screened to determine whether SDHIs information exists; if not, the process ends; if present, quantitative detection is performed based on the SDHIs information.

[0010] The basic approach and its benefits are as follows: Matrix Interfering Substance Identification (S1) utilizes a near-infrared spectrometer to capture the specific spectral signals of fiber (characteristic absorption peak at 1650 cm⁻¹), starch (characteristic peak at 1450 cm⁻¹), and lipids (C-H stretching vibration peak at 930 nm) in grains. Combined with the deep learning capabilities of a convolutional neural network (CNN), this approach, using a library of spectral signatures from training samples (such as wheat, soybeans, and peanuts), enables accurate classification of crop species and rapid estimation of the relative proportions of these three interfering substances. Core to this approach is the conversion of physical spectral signals into quantifiable matrix parameters through a spectrum-component correlation model, providing a targeted basis for subsequent steps.

[0011] Pulsed-field synergistic dynamic extraction (S2) is based on the matrix components identified by S1 and utilizes different physical fields to target specific matrices. When the lipid content is high, dielectric heating from microwaves (2.45 GHz) and radiative heating from infrared (5-10 μm) work synergistically to accelerate the separation of lipids from acetonitrile, while acetonitrile saturation with n-hexane enhances lipid solubility. Turbidity (reflecting the particle suspension state) is monitored in real time by a camera, and the intensity of the ultrasonic (20-80 kHz) cavitation effect is dynamically adjusted to ensure the full release of lipid-encapsulated SDHIs.

[0012] When the fiber content is high, a pulsed electric field (5-10kV) breaks down the plant cell wall through the electroporation effect. The pulse frequency (15-20Hz) is adjusted in combination with the change in conductivity (reflecting the degree of cell fragmentation) to enhance the release efficiency of SDHIs in the fiber matrix; real-time temperature monitoring ensures uniform heating to avoid local overheating that may lead to pesticide degradation.

[0013] Gradient intelligent purification (S3) is a targeted purification mechanism selected according to the type of interfering substance. When the lipid content is high, PSA / C18 magnetic microspheres specifically adsorb lipids and polar impurities through ion exchange (PSA) and hydrophobic interaction (C18). Dynamic mixing driven by a rotating magnetic field enhances contact efficiency, and gradient elution with an acetonitrile solution containing 8% acetic acid can reduce the loss of the target. When the starch content is high, the graphite adsorbent adsorbs starch derivatives through π-π interaction, and the turbulent effect generated by the microfluidic spiral channel enhances mass transfer and improves the efficiency of impurity removal.

[0014] Mass spectrometry differential detection (S4) first screens unknown SDHIs using the full-scan mode of high-resolution time-of-flight mass spectrometry (LC-TOF / MS) (without pre-setting the target), and uses its high resolution (20,000-60,000 FWHM) to distinguish isomers. If the target is confirmed, the system switches to the multiple reaction monitoring (MRM) mode of liquid chromatography-triple quadrupole mass spectrometry (LC-QQQ) to achieve precise quantification through dual screening of parent ions and product ions, adapting to the detection needs of different SDHIs.

[0015] S1's spectral recognition replaces traditional manual classification, S2's dynamic parameter adjustment (such as ultrasonic power and pulse frequency) shortens the extraction time, and S3's targeted purification reduces invalid operations, reducing the time consumption of the entire process.

[0016] S3's differentiated purification strategy for lipids and starches improves impurity removal rates, avoiding mass spectrometry ion suppression caused by lipids in traditional methods and chromatographic column contamination by starch-derived impurities.

[0017] Able to adapt to the characteristics of grain matrices, S1-S4 forms a closed-loop adaptation mechanism of identification, extraction, purification, and detection, and dynamically adjusts parameters for different grain types such as cereals (high fiber) and oils (high lipids), solving the problems of insufficient extraction and incomplete purification caused by traditional methods.

[0018] After training on samples, the CNN model achieved high accuracy in identifying crop species and accurately estimated the fiber and lipid content, providing reliable parameters for subsequent steps. Matrix interference was reduced after purification, improving the LC-QQQ detection capability for SDHIs. This system can cover grain matrices such as wheat, soybeans, and peanuts, eliminating the need for manual process adjustments and increasing automation to reduce human error. The high resolution of LC-TOF / MS, combined with the deep cleanup of S3, reduces the false positive rate in screening for novel SDHI derivatives and expands the detection range.

[0019] In summary, the present invention targets the characteristics of grain matrices, improves extraction efficiency, and reduces matrix interference, thereby achieving rapid and accurate detection of pesticide residues in grains.

[0020] Furthermore, the S4 includes: screening for unknown SDHIs using an LC-TOF / MS coupling instrument equipped with a biphenyl column, dynamically controlling the temperature of the column oven according to the lipid content, and in full-scan mode of the mass spectrometer, automatically triggering the source to increase the collision-induced dissociation energy by 10 eV if the mass accuracy is greater than 5 ppm to obtain SDHIs information, and then performing quantitative detection using an LC-QQQ coupling instrument with a C18 column and a mobile phase gradient of acetonitrile from 20% to 80% in 0-3 min. The dynamic MRM mode is enabled on the mass spectrometer end, and the collision energy is automatically adjusted according to the real-time signal intensity so that the quantitative RSD at a concentration of 0.01 mg / kg is less than 5%.

[0021] Furthermore, in the S1 matrix interfering substance identification step: The near-infrared spectrometer has a wavelength range of 900-2500 nm, a spectral resolution of ≤8 cm⁻¹, and a scanning number of ≥32 times to obtain fine spectral characteristics of the sample; The convolutional neural network model uses a transfer learning strategy, with a ResNet-50 architecture as its foundation. It is pre-trained on a public dataset containing spectra of 500 crop species and fine-tuned on a private dataset containing at least 1,000 samples covering five major food crops: wheat, corn, soybeans, peanuts, and rapeseed, as well as their common hybrids. During the process of identifying crop types, the model output includes the confidence level of the crop category, the predicted value of the proportion of interfering substances, and the matrix complexity index. The matrix complexity index is calculated based on the weighted sum of the proportions of fiber, starch, and lipids, and is used to set the initial parameter values ​​in subsequent steps; When the model's confidence in the crop category is less than 90%, the hyperspectral imaging module is automatically triggered to collect the sample surface texture features, and then reconfirmed through a multimodal fusion algorithm, that is, a weighted fusion of spectral features and texture features. The weight coefficient of the fusion algorithm is obtained through genetic algorithm optimization.

[0022] Furthermore, the process of simultaneously estimating the relative proportions of fiber, starch, and lipids includes: The collected near-infrared diffuse reflectance spectra were preprocessed by using Savitzky-Golay smoothing and standard normal variate transformation to eliminate baseline drift, and then the spectral features were enhanced by first-order derivative. An improved partial least squares regression model was used to estimate the proportion of each component. The improved partial least squares regression model optimized band selection through a genetic algorithm to screen out 15 characteristic bands with the strongest correlation with fiber, starch, and lipids. The matrix interaction compensation factor is introduced to correct the traditional least squares regression prediction results by the following formula:

[0023] After the correction The relative proportion of each component, that is, the actual mass proportion of fiber, starch or lipid in the grain after correction by the matrix interaction compensation factor, where Represents fiber, Represents starch, represents lipids; The traditional partial least squares regression model predicts The relative proportion of the different components is not considered. For the The first ingredient and the The interaction coefficient of the two components reflects the mutual interference intensity of the two components in the spectral response; for example: when the fiber-lipid interaction coefficient =0.03, indicating that the spectral interference coefficient of lipid on fiber is 0.03; obtained by fitting the spectrum-component data of 100 sets of mixed standard substances to form an interaction coefficient matrix; is the summation term, indicating that All ingredients except , the total amount of interference caused to it; Output the component proportion prediction value including 95% confidence interval and calculate the uncertainty index ,when The spectrum re-acquisition process is triggered when For the The half-width of the 95% confidence interval of the predicted value of the species proportion; The step of obtaining matrix components further includes generating a three-dimensional matrix feature vector ,in: : The relative proportion of fiber after correction, that is, the mass proportion of fiber components in grain after correction by the matrix interaction compensation factor; : The relative proportion of starch after correction, that is, the mass proportion of starch components in the grain after correction by the matrix interaction compensation factor; : The relative proportion of corrected lipids, that is, the mass proportion of lipid components in grains after correction by the matrix interaction compensation factor; : The 95% confidence interval half width of the predicted value of fiber proportion indicates that the true value of fiber proportion falls within The probability of being within the range is 95%; : The 95% confidence interval half width of the predicted value of starch proportion indicates that the true value of starch proportion falls within The probability of being within the range is 95%; : The 95% confidence interval half width of the predicted value of lipid proportion indicates that the true value of lipid proportion falls within The probability of being within the range is 95%; The three-dimensional matrix feature vector is then directly mapped to the extraction parameter space of step S2.

[0024] Furthermore, in the S2 pulsed field synergistic dynamic extraction step: The first preset value is 25%, that is, , the second preset value is 10NTU, and the third preset value is 5NTU; After enhancing the ultrasonic cavitation effect, if the turbidity is still greater than the third preset value after three consecutive monitorings, it is judged as an extraction abnormality, and a real-time guidance message is output on the display screen: Please check whether the amount of n-hexane saturated acetonitrile is insufficient or confirm whether the sample is sufficiently homogenized; The real-time guidance information is based on the three-dimensional matrix feature vector in The UI value is dynamically generated. When UI>0.1, it indicates that the matrix is ​​complex and the extraction time is extended by 2 minutes.

[0025] Furthermore, in the S2 pulsed field synergistic dynamic extraction step: The fourth preset value is 15%, that is, , the fifth preset value is 0.1mS / cm·min; When using a pulsed electric field, if the conductivity growth rate is still less than 0.05mS / cm·min after adjusting the frequency to 15-20Hz, a real-time guidance message will be triggered: the fiber is not fully broken, extend the pulsed electric field time by 30 seconds or increase the pulse voltage; Real-time guidance information and 3D matrix feature vectors in and Related, when When the fiber ratio estimation fluctuates greatly, high-frequency pulses are used.

[0026] Furthermore, in the S2 pulsed field synergistic dynamic extraction step: The preset time is 30 seconds, and the temperature of multiple points of the extract is collected by an infrared temperature sensor; When the temperature difference between any two points is ≥5°C, in addition to dynamically correcting the heating power, it also outputs real-time guidance information: uneven temperature distribution, turn on the cavity stirring function; If the temperature difference after correction is still ≥3℃, based on Generate real-time guidance information: Starch gelatinization leads to impeded heat transfer and lower infrared temperature.

[0027] Furthermore, in the S3 gradient intelligent purification step: When the lipid content is greater than 30%, The particle size of the PSA / C18 magnetic microspheres is 50-100 μm, the specific surface area is ≥500 m² / g, the rotating magnetic field strength is 0.5-1.5 T, and the rotation speed is 300-600 r / min; if the absorbance of the purified solution at a wavelength of 450 nm is greater than 0.2, 50 mg / g of the sample of GCB microspheres is added, and the mixing time is extended to 3 minutes; When the starch content is greater than 40%, The addition amount of the graphite adsorbent is 80-120 mg / g sample, the inner diameter of the spiral channel of the microfluidic chip is 1 mm, the radius of curvature is 5 mm, and the turbulence intensity is controlled by the Reynolds number within the range of 2000-4000; if the conductivity change rate of the eluent is less than 0.05 mS / cm·min, ultrasonic assisted elution is triggered, and the ultrasonic parameters are a frequency of 40 kHz and a power of 50 W; During the purification process, the UV-visible spectrum of the eluate is collected in real time, and the purification effect is identified by the convolutional neural network model. When the convolutional neural network model predicts that the purification efficiency is less than 90%, the three-dimensional matrix feature vector is used to identify the purification effect. in and Dynamically adjust purification strategies: like , which indicated that the estimation error of lipid proportion was large, so step elution was used, first with 5% acetic acid and acetonitrile, then with 10% acetic acid and acetonitrile; like , PSA / C18 magnetic microspheres and graphite adsorbent are activated simultaneously, and a pulsed electric field is introduced in the spiral channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The present invention is a flow chart of Example 1 of an intelligent pesticide residue detection method. DETAILED DESCRIPTION

[0029] The following is further described in detail through specific implementation methods: Example 1

[0030] An intelligent pesticide residue detection method (such as Figure 1 shown), including: S1 Matrix Interfering Substance Identification Step: Weigh a preset mass of the test sample and place it in a spectral imaging detection instrument. Use a near-infrared spectrometer to collect the sample's diffuse reflectance spectrum. Combined with a convolutional neural network model, the crop species are identified and the relative proportions of fiber, starch, and lipids are simultaneously estimated to obtain the matrix composition. S2 pulsed field synergistic dynamic extraction step: Based on the identified matrix components, differentiated parameter regulation is performed in the ultrasound-microwave-infrared composite field extraction chamber; When the lipid ratio is greater than the first preset value, microwave and infrared synergistic heating to 50-60°C is activated, and extraction is carried out with n-hexane saturated acetonitrile. At the same time, the turbidity of the extract is monitored in real time through a camera. When the turbidity of the extract is greater than the second preset value, the ultrasonic cavitation effect is automatically enhanced until the turbidity is less than the third preset value. When the fiber content is greater than the fourth preset value, a pulsed electric field is used to destroy the plant cell wall and the conductivity change of the extract is recorded simultaneously. When the conductivity growth rate is less than the fifth preset value, the pulse frequency is automatically adjusted to 15-20Hz to increase the SDHIs release rate. During the extraction process, the temperature of the extract is collected at preset intervals, and the heating power is dynamically corrected to ensure that the sample temperature difference is less than 5°C; S3 gradient intelligent purification step: If the lipid content is greater than 30%, add 150 mg / g of sample to the magnetic bead mixing chamber of PSA / C18 magnetic microspheres, drive mixing for 2 minutes using a rotating magnetic field, and then use an acetonitrile solution containing 8% acetic acid in a microfluidic chip at a gradient flow rate of 0.5 mL / min for elution; If the starch content is greater than 40%, graphite adsorbent is added to enhance the turbulence effect through the spiral channel of the microfluidic chip to complete the purification. S4 mass spectrometry differential detection steps: First, unknown SDHIs are screened to determine whether SDHIs information exists; if not, the process ends; if present, quantitative detection is performed based on the SDHIs information.

[0031] Specific usage: Take rapeseed (lipid content of about 40%, fiber content of about 12%) as an example for detailed description. In the S1 matrix interfering substance identification step, 5.00g of rapeseed crushed sample was weighed and placed in the detection cavity of the near-infrared spectrometer, the wavelength range was set to 900-2500nm, the spectral resolution was 8cm⁻¹, and 64 scans were performed. The convolutional neural network model (pre-trained on a public dataset containing spectra of 500 crops, and fine-tuned on a private dataset containing 1000 samples) output: Crop category: rapeseed (confidence 98%); Percentage of interfering substances: lipid 39.8%±1.2%, fiber 11.5%±0.8%, starch 5.2%±0.5%; Matrix complexity index: 0.72 (through the formula calculate).

[0032] During the S2 pulsed-field synergistic dynamic extraction step, because the lipid content was >25%, microwave (300W) and infrared (55°C) heating was activated, and 20 mL of n-hexane-saturated acetonitrile was added. After the extraction began, a camera monitored the turbidity in real time. When the turbidity reached 12 NTU (>10 NTU), the ultrasonic power was automatically increased to 500W. After 150 seconds, when the turbidity dropped to 4.8 NTU (<5 NTU), the ultrasonic enhancement was stopped. Temperature measurements were taken every 30 seconds. When the temperature difference between the top and bottom of the chamber reached 6°C, the microwave power was automatically reduced by 50W, and stirring (200 rpm) was initiated. After 3 minutes, the temperature difference stabilized at 2°C.

[0033] During the S3 gradient intelligent cleanup step, because the lipid content was >30%, 750 mg of PSA / C18 magnetic microspheres (80 μm, surface area 550 m² / g) were added, and mixing was carried out for 2 minutes in a rotating magnetic field of 1.0 T at 400 rpm. Elution was performed on the microfluidic chip using a gradient of acetonitrile containing 8% acetic acid at a flow rate of 0.5 mL / min. The absorbance of the purified solution at 450 nm was 0.15 (<0.2), eliminating the need for additional GCB microspheres.

[0034] During the S4 mass spectrometry differential detection step, an ion peak suspected to be fluopyram (m / z 397.0845, 2.3 ppm from the theoretical value) was detected using the LC-TOF / MS in full scan mode. The analysis was then switched to the LC-QQQ instrument, monitoring the ion transition m / z 397.0→173.3 in dynamic MRM mode at a collision energy of 25 eV. The fluopyram content was determined to be 0.085 mg / kg with an RSD of 3.2%. Example 2

[0035] Detailed description of some of the procedures was performed using wheat flour (approximately 70% starch content, approximately 2% lipid content) as the test sample. The only difference compared to Example 1 was that the LC-TOF / MS instrument was equipped with a biphenyl column and a column temperature of 40°C during the S4 mass spectrometry differential detection step (a conventional column temperature was used due to the low lipid content). In full scan mode, an ion peak suspected of being pyraclostrobin (m / z 387.1352) was detected with a mass deviation of 4.8 ppm. The CID energy within the trigger source was increased from 30 eV to 40 eV, yielding a characteristic fragment ion at m / z 141.05.

[0036] The LC-QQQ instrument used a C18 column and the mobile phase gradient was linearly increased from 20% to 80% acetonitrile in 0-3 min.

[0037] In dynamic MRM mode, the monitoring ion transition m / z 387.1→141.1 was used with an initial collision energy of 22 eV.

[0038] When the injection concentration was 0.01 mg / kg, the system automatically adjusted the collision energy to 25 eV according to the signal intensity, and the RSD of three repeated injections was 2.8%. Example 3

[0039] Taking hybrid soybeans (lipid content of about 22%, fiber content of about 8%, and starch content of about 30%) as the test sample for illustration, the only difference compared to Example 1 is that in the S1 matrix interfering substance identification step, the near-infrared spectrometer has a wavelength range of 900-2500nm, a resolution of 8cm⁻¹, and scans 32 times to obtain the original spectral data.

[0040] The convolutional neural network model (ResNet-50 architecture) initially outputs the crop classification as soybeans (confidence level 88% to 90%), triggering the hyperspectral imaging module. After fusing spectral features (weight 0.6) with texture features (weight 0.4), the hyperspectral imaging module reconfirms the crop classification as hybrid soybeans (confidence level 95%).

[0041] The proportion of output interfering substances: lipid 21.8%±1.5%, fiber 8.2%±0.7%, starch 29.5%±1.2%.

[0042] Matrix complexity index: 0.55 (by formula calculation), used for setting the initial values ​​of the parameters in step S2. Example 4

[0043] For this example, using corn flour (approximately 4% lipid content, approximately 10% fiber content, and approximately 75% starch content) as the test sample, the only difference from Example 1 is the simultaneous estimation of the relative proportions of fiber, starch, and lipid. The original spectrum is smoothed using Savitzky-Golay technology (window width 7, polynomial order 2). A standard normal variate (SNV) transformation is used to eliminate the effects of particle size, and the first-order derivative is used to enhance spectral features.

[0044] The improved partial least squares regression model (using genetic algorithm to screen 15 characteristic bands: 1100-1120nm, 1420-1440nm, etc.) predicted: starch content 74.8%, fiber content 9.5%, and lipid content 4.2%.

[0045] Introducing the interaction coefficient matrix Corrected forecast results:

[0046] Output 95% confidence interval: starch 73.98% ± 1.8%, fiber 9.43% ± 0.9%, lipid 4.17% ± 0.6%.

[0047] Uncertainty assessment: Calculating uncertainty index , trigger the spectrum re-acquisition program, after the second acquisition . Example 5

[0048] For this example, peanut kernels (approximately 48% lipid content, approximately 6% fiber content) were tested. The only difference compared to Example 1 was that, in the S2 pulsed field synergistic dynamic extraction step, microwave (400W) and infrared (58°C) heating was used because the lipid content was greater than 25%, and 30 mL of n-hexane-saturated acetonitrile was added. 100 seconds after the extraction began, the turbidity reached 15 NTU (>10 NTU), and the ultrasonic power was increased to 600 W. After 200 seconds, the turbidity dropped to 8 NTU (>5 NTU), and ultrasonic power was increased. After 250 seconds, the turbidity was still 7.5 NTU, and the system detected a turbidity greater than 5 NTU for three consecutive times, indicating an extraction anomaly.

[0049] The real-time guidance message is: Please check whether the amount of n-hexane-saturated acetonitrile is insufficient (currently 30 mL, add 5 mL).

[0050] Cause , Additional tips: If the matrix is ​​complex, extend the extraction time by 2 minutes.

[0051] Effect after adjustment: After adding 5 mL of extractant and extending the extraction time for 2 minutes, the turbidity dropped to 4.5 NTU, and the extraction was completed. Example 6

[0052] For example, the test sample was oatmeal (fiber content approximately 18%, lipid content approximately 7%). Compared to Example 1, the only difference was that in the S2 pulsed field synergistic dynamic extraction step, because the fiber content was greater than 15%, a pulsed electric field (initial voltage 6kV, frequency 10Hz) was applied. The conductivity growth rate was monitored. Within the first two minutes, the conductivity growth rate was 0.08mS / cm·min (<0.1mS / cm·min), and the frequency was automatically adjusted to 18Hz. After the adjustment, the conductivity growth rate remained at 0.04mS / cm·min (<0.05mS / cm·min) within one minute, triggering real-time guidance.

[0053] The real-time guidance information is: the fiber is not broken sufficiently, and the pulse electric field action time is extended by 30 seconds.

[0054] Cause , Additional tip: The fiber ratio estimation fluctuates greatly, so a 20Hz high-frequency pulse is used.

[0055] After the pulse frequency was adjusted to 20 Hz and extended for 30 seconds, the conductivity growth rate increased to 0.12 mS / cm·min, reaching the preset threshold. Example 7

[0056] Taking the test sample of sweet potato puree (starch content of about 55%, fiber content of about 3%) as an example, compared with Example 1, the only difference is that in the S2 pulse field synergistic dynamic extraction step, the infrared temperature sensor collects multiple temperature points every 30 seconds, and it is found that the upper temperature is 62°C and the bottom temperature is 57°C (temperature difference of 5°C). The infrared power is automatically reduced by 10%, the microwave power is reduced by 50W, and stirring is turned on (150r / min). One minute after the adjustment, the temperature difference is still 3.5°C (≥3°C), based on , Tip: Starch gelatinization causes heat transfer to be blocked, lower the infrared temperature to 50℃ (currently 57℃).

[0057] After the infrared temperature dropped to 50°C, the temperature difference stabilized at 1.8°C within 30 seconds, and the extraction was completed. Example 8

[0058] For example, the test sample was sesame (approximately 58% lipid content, approximately 12% starch content). The only difference compared to Example 1 was that in the S3 gradient intelligent purification step, because the lipid content was greater than 30%, 870 mg of PSA / C18 magnetic microspheres (75 μm particle size, 580 m² / g specific surface area) were added, and the mixture was mixed for 2 minutes at a rotating magnetic field strength of 1.2 T and a speed of 500 r / min. After gradient elution on the microfluidic chip, when the absorbance at 450 nm was 0.25 (>0.2), an additional 290 mg of GCB microspheres were added, and the mixing time was extended to 3 minutes.

[0059] UV-visible spectrum showed that the purification efficiency was 88% (<90%), because> , Tip: "The estimation error of lipid ratio is large, so step elution is used (first 5% acetic acid acetonitrile, then 10% acetic acid acetonitrile)".

[0060] After step elution, the purification efficiency was improved and the matrix effect was improved from 82% to 96%.

[0061] The above are only embodiments of the present invention. The invention is not limited to the fields involved in this implementation case. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the relevant field are aware of all common technical knowledge in the technical field to which the invention belongs before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. An intelligent pesticide residue detection method, characterized in that: include: S1 Matrix Interfering Substance Identification Step: Weigh a preset mass of the test sample and place it in a spectral imaging detection instrument. Use a near-infrared spectrometer to collect the sample's diffuse reflectance spectrum. Combined with a convolutional neural network model, the crop species are identified and the relative proportions of fiber, starch, and lipids are simultaneously estimated to obtain the matrix composition. S2 pulsed field synergistic dynamic extraction step: Based on the identified matrix components, differentiated parameter regulation is performed in the ultrasound-microwave-infrared composite field extraction chamber; When the lipid ratio is greater than the first preset value, microwave and infrared synergistic heating to 50-60°C is activated, and extraction is carried out with n-hexane saturated acetonitrile. At the same time, the turbidity of the extract is monitored in real time through a camera. When the turbidity of the extract is greater than the second preset value, the ultrasonic cavitation effect is automatically enhanced until the turbidity is less than the third preset value. When the fiber content is greater than the fourth preset value, a pulsed electric field is used to destroy the plant cell wall and the conductivity change of the extract is recorded simultaneously. When the conductivity growth rate is less than the fifth preset value, the pulse frequency is automatically adjusted to 15-20Hz to increase the SDHIs release rate. During the extraction process, the temperature of the extract is collected at preset intervals, and the heating power is dynamically corrected to ensure that the sample temperature difference is less than 5°C; S3 gradient intelligent purification step: If the lipid content is greater than 30%, add 150 mg / g of sample to the magnetic bead mixing chamber of PSA / C18 magnetic microspheres, drive mixing for 2 minutes using a rotating magnetic field, and then use an acetonitrile solution containing 8% acetic acid in a microfluidic chip at a gradient flow rate of 0.5 mL / min for elution; If the starch content is greater than 40%, graphite adsorbent is added to enhance the turbulence effect through the spiral channel of the microfluidic chip to complete the purification. S4 mass spectrometry differential detection steps: First, unknown SDHIs are screened to determine whether SDHIs information exists; if not, the process ends; if present, quantitative detection is performed based on the SDHIs information.

2. An intelligent pesticide residue detection method according to claim 1, characterized in that: The S4 includes: screening for unknown SDHIs using an LC-TOF / MS coupling instrument equipped with a biphenyl column, dynamically controlling the temperature of the column oven according to the lipid content, and automatically triggering the source to increase the collision-induced dissociation energy by 10 eV in the full-scan mode of the mass spectrometer if the mass accuracy is greater than 5 ppm to obtain SDHIs information, and then performing quantitative detection using an LC-QQQ coupling instrument with a C18 column and a mobile phase gradient of acetonitrile from 20% to 80% in 0-3 min. The dynamic MRM mode is enabled on the mass spectrometer end, and the collision energy is automatically adjusted according to the real-time signal intensity so that the quantitative RSD at a concentration of 0.01 mg / kg is less than 5%.

3. An intelligent pesticide residue detection method according to claim 2, characterized in that: In the S1 matrix interfering substance identification step: The near-infrared spectrometer has a wavelength range of 900-2500 nm, a spectral resolution of ≤8 cm⁻¹, and a scanning number of ≥32 times to obtain fine spectral characteristics of the sample; The convolutional neural network model uses a transfer learning strategy, with a ResNet-50 architecture as its foundation. It is pre-trained on a public dataset containing spectra of 500 crop species and fine-tuned on a private dataset containing at least 1,000 samples covering five major food crops: wheat, corn, soybeans, peanuts, and rapeseed, as well as their common hybrids. During the process of identifying crop types, the model output includes the confidence level of the crop category, the predicted value of the proportion of interfering substances, and the matrix complexity index. The matrix complexity index is calculated based on the weighted sum of the proportions of fiber, starch, and lipids, and is used to set the initial parameter values ​​in subsequent steps; When the model's confidence in the crop category is less than 90%, the hyperspectral imaging module is automatically triggered to collect the sample surface texture features, and then reconfirmed through a multimodal fusion algorithm, that is, a weighted fusion of spectral features and texture features. The weight coefficient of the fusion algorithm is obtained through genetic algorithm optimization.

4. An intelligent pesticide residue detection method according to claim 3, characterized in that: The process of simultaneously estimating the relative proportions of fiber, starch, and lipids includes: The collected near-infrared diffuse reflectance spectra were preprocessed by using Savitzky-Golay smoothing and standard normal variate transformation to eliminate baseline drift, and then the spectral features were enhanced by first-order derivative. An improved partial least squares regression model was used to estimate the proportion of each component. The improved partial least squares regression model optimized band selection through a genetic algorithm to screen out 15 characteristic bands with the strongest correlation with fiber, starch, and lipids. The matrix interaction compensation factor is introduced to correct the traditional least squares regression prediction results by the following formula: in, is the corrected component ratio, is the traditional least squares regression prediction value, is the interaction coefficient matrix, which is obtained by fitting the spectrum-composition data of 100 sets of mixed standard substances; Output the component proportion prediction value including 95% confidence interval and calculate the uncertainty index ,when The spectrum re-acquisition process is triggered when For the The half-width of the 95% confidence interval of the predicted value of the species proportion; The step of obtaining matrix components further includes generating a three-dimensional matrix feature vector ,in: : The relative proportion of fiber after correction, that is, the mass proportion of fiber components in grain after correction by the matrix interaction compensation factor; : The relative proportion of starch after correction, that is, the mass proportion of starch components in the grain after correction by the matrix interaction compensation factor; : The relative proportion of corrected lipids, that is, the mass proportion of lipid components in grains after correction by the matrix interaction compensation factor; : The 95% confidence interval half width of the predicted value of fiber proportion indicates that the true value of fiber proportion falls within The probability of being within the range is 95%; : The 95% confidence interval half width of the predicted value of starch proportion indicates that the true value of starch proportion falls within The probability of being within the range is 95%; : The 95% confidence interval half width of the predicted value of lipid proportion indicates that the true value of lipid proportion falls within The probability of being within the range is 95%; The three-dimensional matrix feature vector is then directly mapped to the extraction parameter space of step S2.

5. An intelligent pesticide residue detection method according to claim 4, characterized in that: In the S2 pulsed field synergistic dynamic extraction step: The first preset value is 25%, that is, , the second preset value is 10NTU, and the third preset value is 5NTU; After enhancing the ultrasonic cavitation effect, if the turbidity is still greater than the third preset value after three consecutive monitorings, it is judged as an extraction abnormality, and a real-time guidance message is output on the display screen: Please check whether the amount of n-hexane saturated acetonitrile is insufficient or confirm whether the sample is sufficiently homogenized; The real-time guidance information is based on the three-dimensional matrix feature vector in The UI value is dynamically generated. When UI>0.1, it indicates that the matrix is ​​complex and the extraction time is extended by 2 minutes.

6. An intelligent pesticide residue detection method according to claim 5, characterized in that: In the S2 pulsed field synergistic dynamic extraction step: The fourth preset value is 15%, that is, , the fifth preset value is 0.1mS / cm·min; When using a pulsed electric field, if the conductivity growth rate is still less than 0.05mS / cm·min after adjusting the frequency to 15-20Hz, a real-time guidance message will be triggered: the fiber is not fully broken, extend the pulsed electric field time by 30 seconds or increase the pulse voltage; Real-time guidance information and 3D matrix feature vectors in and Related, when When the fiber ratio estimation fluctuates greatly, high-frequency pulses are used.

7. An intelligent pesticide residue detection method according to claim 6, characterized in that: In the S2 pulsed field synergistic dynamic extraction step: The preset time is 30 seconds, and the temperature of multiple points of the extract is collected by an infrared temperature sensor; When the temperature difference between any two points is ≥5°C, in addition to dynamically correcting the heating power, it also outputs real-time guidance information: uneven temperature distribution, turn on the cavity stirring function; If the temperature difference after correction is still ≥3℃, based on Generate real-time guidance information: Starch gelatinization leads to impeded heat transfer and lower infrared temperature.

8. An intelligent pesticide residue detection method according to claim 7, characterized in that: In the S3 gradient intelligent purification step: When the lipid content is greater than 30%, The particle size of the PSA / C18 magnetic microspheres is 50-100 μm, the specific surface area is ≥500 m² / g, the rotating magnetic field strength is 0.5-1.5 T, and the rotation speed is 300-600 r / min; if the absorbance of the purified solution at a wavelength of 450 nm is greater than 0.2, 50 mg / g of the sample of GCB microspheres is added, and the mixing time is extended to 3 minutes; When the starch content is greater than 40%, The addition amount of the graphite adsorbent is 80-120 mg / g sample, the inner diameter of the spiral channel of the microfluidic chip is 1 mm, the radius of curvature is 5 mm, and the turbulence intensity is controlled by the Reynolds number within the range of 2000-4000; if the conductivity change rate of the eluent is less than 0.05 mS / cm·min, ultrasonic assisted elution is triggered, and the ultrasonic parameters are a frequency of 40 kHz and a power of 50 W; During the purification process, the UV-visible spectrum of the eluate is collected in real time, and the purification effect is identified by the convolutional neural network model. When the convolutional neural network model predicts that the purification efficiency is less than 90%, the three-dimensional matrix feature vector is used to identify the purification effect. in and Dynamically adjust purification strategies: like , which indicated that the estimation error of lipid proportion was large, so step elution was used, first with 5% acetic acid and acetonitrile, then with 10% acetic acid and acetonitrile; like , PSA / C18 magnetic microspheres and graphite adsorbent are activated simultaneously, and a pulsed electric field is introduced in the spiral channel.

9. An intelligent pesticide residue detection system, characterized in that: The method according to any one of claims 1 to 8 is adopted.

Images