Fatty acid detection method, fatty acid detection device, and computer storage medium

Abstract

The application provides a fatty acid detection method, a fatty acid detection device and a computer storage medium, and belongs to the technical field of spectral data processing. The fatty acid detection method comprises the following steps: acquiring near-infrared spectral data of a biological tissue to be detected in an original physical form; mapping the near-infrared spectral data to matched equivalent spectral data through a spectral transmission model, wherein the equivalent spectral data are spectral data of the biological tissue to be detected in a reference physical form; and inputting the equivalent spectral data into a trained quantitative detection model to determine the fatty acid content in the biological tissue to be detected, wherein the quantitative detection model is established based on spectral data in the reference physical form and corresponding true values of the fatty acid content. In this way, the fatty acid content in the biological tissue to be detected is non-destructively and accurately predicted, and the problem of meeting the non-invasive detection requirement of clinical or in-vivo research is solved.

Description

Technical Field

[0001] This application relates to the field of spectral data processing technology, and in particular to a fatty acid detection method, a fatty acid detection device, and a computer storage medium, belonging to the field of spectral data processing technology. Background Technology

[0002] Omega-3 polyunsaturated fatty acids (mainly including alpha-linolenic acid, eicosapentaenoic acid, and docosahexaenoic acid) and Omega-6 polyunsaturated fatty acids (mainly including linoleic acid and arachidonic acid) are two essential classes of fatty acids for the human body. The human body cannot synthesize these unsaturated fatty acids on its own and must obtain them through diet. Major food sources of Omega-3 and Omega-6 fatty acids include deep-sea fish (such as salmon, sardines, and mackerel), flaxseed and flaxseed oil, algae and algal oil, and egg yolks. As important regulators of human physiological functions, Omega-3 and Omega-6 have significant benefits for cardiovascular health. EPA (eicosapentaenoic acid) and DHA (docosahexaenoic acid) can effectively reduce triglycerides and inhibit platelet aggregation. In the nervous and visual systems, DHA, as a key component of nerve cell membranes, can promote neural plasticity, and long-term intake helps maintain cognitive function. Simultaneously, Omega-3 and Omega-6 can also exert anti-inflammatory and immunomodulatory functions by inhibiting the release of inflammatory factors, which is of positive significance for improving chronic inflammatory diseases.

[0003] Currently, the conventional methods for quantitative analysis of Omega-3 and Omega-6 fatty acid composition in biological samples are gas chromatography or gas chromatography-mass spectrometry (GC-MS). While these methods offer high sensitivity and good separation, they suffer from drawbacks such as complex sample pretreatment, time-consuming analysis, and high cost. In recent years, near-infrared spectroscopy has become increasingly popular as a rapid and non-destructive alternative to traditional gas chromatography for the detection of fatty acid content in single rapeseed seeds, fatty acids in peanut quality breeding, and unsaturated fatty acids in fresh meat.

[0004] However, the detection targets of near-infrared spectroscopy in related technologies are mostly liquids or ground homogenates, and there is no non-invasive detection for specific Omega-3 and Omega-6 fatty acids in intact biological tissues. As a result, quantitative models based on liquids cannot be directly applied to tissue samples, resulting in poor prediction accuracy of fatty acids, which cannot meet the needs of non-invasive detection in clinical or in vivo studies. Summary of the Invention

[0005] This application provides a fatty acid detection method, a fatty acid detection device, and a computer storage medium, aiming to solve the problem that quantitative models based on liquids cannot be directly applied to tissue samples, resulting in poor prediction accuracy of fatty acids and failing to meet the needs of non-invasive detection in clinical or in vivo studies.

[0006] To achieve the above objectives, this application provides the following technical solution: In a first aspect, this application provides a method for detecting fatty acids, comprising: acquiring near-infrared spectral data of a biological tissue under its original physical morphology; mapping the near-infrared spectral data into matching equivalent spectral data through a spectral transfer model, wherein the equivalent spectral data is the spectral data of the biological tissue under its baseline physical morphology; and inputting the equivalent spectral data into a trained quantitative detection model to determine the fatty acid content in the biological tissue under its baseline physical morphology, wherein the quantitative detection model is established based on the spectral data under its baseline physical morphology and the corresponding true values ​​of fatty acid content.

[0007] In some embodiments of this application, acquiring near-infrared spectral data of the biological tissue under test in its original physical morphology includes: measuring the original near-infrared spectral data of the biological tissue under test according to preset acquisition parameters based on near-infrared spectral acquisition technology; and truncating the original near-infrared spectral data according to a preset characteristic wavelength window to obtain target band near-infrared spectral data, which is then used as near-infrared spectral data.

[0008] In some embodiments of this application, the determination of the characteristic wavelength window includes: acquiring calibration sample data corresponding to the target fatty acid, and dividing it into training data and test data, wherein the calibration sample data includes near-infrared spectral data of the standard solution and the corresponding target fatty acid concentration information; establishing a quantitative prediction model for the target fatty acid based on the training data, and determining the contribution of each wavelength variable to the quantitative prediction model, wherein the contribution of each wavelength variable to the quantitative prediction model is characterized by the variable importance projection value; screening spectral variables according to the contribution of each wavelength variable to obtain the key characteristic wavelengths corresponding to the target fatty acid, and determining the characteristic wavelength window based on the key characteristic wavelengths.

[0009] In some embodiments of this application, obtaining calibration sample data corresponding to the target fatty acid includes: preparing a standard solution and measuring the original spectral data of the standard solution according to the acquisition parameters based on near-infrared spectroscopy acquisition technology, wherein the standard solution is a pre-prepared solution containing the target fatty acid; performing at least one preprocessing step of denoising, smoothing, and scattering correction on the original spectral data to obtain preprocessed spectral data; and constructing calibration sample data based on the preprocessed spectral data and the target fatty acid concentration information corresponding to each standard solution.

[0010] In some embodiments of this application, the preparation of standard solutions includes: preparing at least one standard solution of eicosapentaenoic acid, docosahexaenoic acid, alpha-linolenic acid, gamma-linolenic acid and arachidonic acid in different concentration ratios using n-heptane as a solvent, wherein the concentration range of the standard solutions covers the concentration range of each target fatty acid in the biological tissue to be tested, and multiple standard solutions with different concentrations are prepared for each target fatty acid.

[0011] In some embodiments of this application, near-infrared spectral data is mapped to matching equivalent spectral data through a spectral transfer model, including: determining a spectral transfer model, and based on the spectral transfer model, determining a reference physical form corresponding to the original physical form, wherein the reference physical form is adapted to the quantitative detection model; and performing inter-morphological spectral correction on the near-infrared spectral data according to the spectral transfer model to obtain equivalent spectral data under the reference physical form.

[0012] In some embodiments of this application, determining a spectral transfer model includes: acquiring near-infrared spectral data of sample biological tissue under various physical morphologies; establishing spectral transfer functions between adjacent physical morphologies according to the spectral scattering differences corresponding to various physical morphologies; and constructing a spectral transfer model based on each spectral transfer function.

[0013] In some embodiments of this application, before establishing the spectral transfer function between adjacent physical morphologies, the method further includes: performing spectral preprocessing on the near-infrared spectral data of the sample biological tissue under various physical morphologies within a preset characteristic wavelength window to obtain preprocessed near-infrared spectral data, wherein the spectral preprocessing includes at least one of smoothing and denoising, baseline correction and scattering correction, and the spectral correction between morphologies is based on the preprocessed near-infrared spectral data.

[0014] In some embodiments of this application, the method further includes: acquiring near-infrared spectral data of validation set samples in their original physical form and determining the true content of target fatty acids in the validation set samples; inputting the near-infrared spectral data into a quantitative detection model to obtain direct prediction results; and converting the near-infrared spectral data into equivalent spectral data through a spectral transfer model and inputting the equivalent spectral data into a quantitative detection model to obtain migration prediction results; and evaluating the migration effect of the spectral transfer model based on the differences between the direct prediction results and the migration prediction results and the true content values.

[0015] In some embodiments of this application, the training process of the quantitative detection model specifically includes: determining sample data and performing at least one preprocessing step, such as smoothing and denoising, derivative transformation, baseline correction, and background interference correction, on the near-infrared spectral data in the sample data. The sample data includes the near-infrared spectral data of each sample and the true value of the target fatty acid content. From the preprocessed near-infrared spectral data, the spectral data within the characteristic wavelength window corresponding to the target fatty acid is extracted as the input variable of the quantitative detection model. The sample data is divided into a training set and a prediction set. Based on the input variables in the training set and the corresponding true value of the target fatty acid content, the quantitative detection model is trained. The prediction performance of the quantitative detection model is evaluated based on the prediction set.

[0016] In some embodiments of this application, the quantitative detection model is established based on the partial least squares regression algorithm, and the performance of the quantitative detection model is evaluated based on at least one of the coefficient of determination, root mean square error, and relative analysis error.

[0017] In some embodiments of this application, determining sample data includes: pre-treating symmetrically taken biological tissue samples to obtain fatty acid extract; preparing fatty acid extract samples of different concentrations based on the fatty acid extract; and performing near-infrared spectroscopy acquisition and gas chromatography analysis on the fatty acid extract samples to obtain near-infrared spectral data and the true value of the target fatty acid content, respectively, as sample data.

[0018] Secondly, this application provides a fatty acid detection device, comprising: an acquisition module for acquiring near-infrared spectral data of a biological tissue under its original physical morphology; a mapping module for mapping the near-infrared spectral data into matching equivalent spectral data through a spectral transfer model, wherein the equivalent spectral data is the spectral data of the biological tissue under its baseline physical morphology; and a determination module for inputting the equivalent spectral data into a trained quantitative detection model to determine the fatty acid content in the biological tissue under its baseline physical morphology, wherein the quantitative detection model is established based on the spectral data under its baseline physical morphology and the corresponding true values ​​of fatty acid content.

[0019] Thirdly, this application provides a computer storage medium storing at least one instruction, which is executed by a processor to implement the fatty acid detection method of any of the above embodiments.

[0020] According to the fatty acid detection method, fatty acid detection device, and computer storage medium provided in this application, the near-infrared spectral data of the biological tissue under its original physical morphology is first converted into equivalent spectral data under the reference physical morphology through a spectral transfer model. Then, the equivalent spectral data is input into the quantitative detection model for prediction. In this way, the model input data of the actual test sample and the modeling data of the quantitative detection model can be matched in spectral morphology without destroying the original morphology of the biological tissue under test. This reduces the spectral inconsistency between the original physical morphology and the reference physical morphology caused by scattering effects and differences in tissue background, reduces the prediction deviation caused by differences in physical morphology, and achieves non-destructive and accurate prediction of fatty acid content in the biological tissue under test, thus meeting the needs of non-invasive detection in clinical or in vivo research. Attached Figure Description

[0021] Figure 1 This is a schematic flowchart of a fatty acid detection method according to an embodiment of this application.

[0022] Figure 2 This is a schematic flowchart illustrating the process of determining a characteristic wavelength window according to an embodiment of this application.

[0023] Figure 3 This is a schematic diagram of the process of training a quantitative detection model according to an embodiment of this application.

[0024] Figure 4 This is a schematic diagram of a fatty acid detection device according to an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0027] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0028] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0029] Specifically, near-infrared spectroscopy is a detection technique that utilizes the absorption information of the harmonics of the stretching vibrations of hydrogen-containing groups (such as CH, OH, NH, SH, etc.) and the combined frequencies of their stretching and bending vibrations to perform qualitative and quantitative analysis of substances. It can analyze physical properties such as density, viscosity, particle size, and hardness, as well as chemical components such as proteins, amino acids, fats, starches, moisture, and other nutrients in a very short time. Due to its advantages such as ease of operation, rapid analysis, relatively accurate results, no sample pretreatment required, and no contamination or loss to the sample, near-infrared spectroscopy has become an ideal alternative to traditional analytical methods and has received widespread attention in the quality evaluation of various foods. In recent years, with the deep application of machine learning algorithms in spectral modeling, the prediction accuracy and robustness of models based on near-infrared spectroscopy have been significantly improved, demonstrating good performance in the classification and identification of single systems (such as grains, oils, and minced meat).

[0030] For example, Figure 1 This is a schematic flowchart of a fatty acid detection method according to an embodiment of this application.

[0031] like Figure 1 As shown, the fatty acid detection method includes: S11: Acquire near-infrared spectral data of the biological tissue under test in its original physical morphology.

[0032] The biological tissues to be tested are biological samples rich in Omega-3 and Omega-6 fatty acids, such as fish tissue, deep-sea shrimp, scallops, squid, and oysters. The original physical morphology usually refers to the natural or near-natural tissue morphology of the biological tissues to be tested during actual testing, such as whole pieces of fresh fish, fresh oysters, or whole pieces of fresh meat.

[0033] S12: Near-infrared spectral data is mapped to matching equivalent spectral data through a spectral transfer model, wherein the equivalent spectral data is the spectral data of the biological tissue to be tested under the baseline physical morphology.

[0034] Specifically, because the biological tissues under test usually have problems such as uneven tissue structure, particle size differences, limited concentration range of target components, and uneven distribution in their original physical form, the water, protein, scattering effect, baseline drift and tissue background in the tissue will seriously interfere with the characteristic absorption signal of fatty acids, thus causing spectral deviation.

[0035] To mitigate the impact of this spectral bias on the quantitative detection of fatty acids, near-infrared spectral data is input into a spectral transfer model that characterizes the spectral correspondence between the original physical form and the reference physical form. Intermorphic spectral correction is then performed on the near-infrared spectral data in the original physical form, mapping it to equivalent spectral data that matches the reference physical form. The reference physical form refers to a physical form suitable for the subsequent quantitative detection model establishment process, such as the extract form or other sample forms with minimal spectral scattering influence and suitable for establishing a quantitative detection model.

[0036] In other words, the equivalent spectral data is not the actual re-acquired baseline physical morphology spectrum, but rather spectral data that matches the baseline physical morphology in spectral characteristics after the near-infrared spectral data under the original physical morphology has been corrected and transformed using a spectral transfer model. By mapping to matching equivalent spectral data, complex pretreatment of biological tissues (hydrolysis, extraction, derivatization, etc.) is eliminated, solving the problems of time-consuming, cumbersome, and sample-destructive processes in related technologies, making them unsuitable for in vivo or rapid on-site detection.

[0037] S13: Input the equivalent spectral data into the trained quantitative detection model to determine the fatty acid content in the biological tissue to be tested. The quantitative detection model is established based on the spectral data under the baseline physical morphology and the corresponding true value of fatty acid content.

[0038] The quantitative detection model is a computer model established based on the sample spectral data in a baseline physical form (such as extract form) and the corresponding true value of fatty acid content, used to determine the fatty acid content of the biological tissue to be tested in the extract form.

[0039] Specifically, in building a quantitative detection model, near-infrared spectral data of multiple samples in a baseline physical state can be obtained, and the true values ​​of the target fatty acid content in each sample can be determined by gas chromatography or other standard chemical analysis methods. Then, the near-infrared spectral data in the baseline physical state can be used as input variables, and the corresponding true values ​​of fatty acid content can be used as response variables to construct a quantitative detection model for predicting fatty acid content.

[0040] The fatty acid content can include the content of at least one target fatty acid among Omega-3 and Omega-6 polyunsaturated fatty acids, such as at least one of eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), alpha-linolenic acid (ALA), gamma-linolenic acid (ALA), and arachidonic acid (RAA). The quantitative detection model can be a partial least squares regression model, an artificial neural network model, or other quantitative prediction models capable of establishing a correspondence between near-infrared spectral data and the true value of fatty acid content.

[0041] According to the fatty acid detection method of this application, the near-infrared spectral data of the biological tissue to be tested in its original physical morphology is first converted into equivalent spectral data in the reference physical morphology through a spectral transfer model. Then, the equivalent spectral data is input into the quantitative detection model for prediction. In this way, the model input data of the actual test sample and the modeling data of the quantitative detection model can be matched in spectral morphology without destroying the original morphology of the biological tissue to be tested. This reduces the spectral inconsistency between the original physical morphology and the reference physical morphology caused by scattering effects and differences in tissue background, reduces the prediction deviation caused by differences in physical morphology, and achieves non-destructive and accurate prediction of fatty acid content in the biological tissue to be tested, thus meeting the needs of non-invasive detection in clinical or in vivo studies.

[0042] In some embodiments, acquiring near-infrared spectral data of the biological tissue under test in its original physical morphology includes: measuring the original near-infrared spectral data of the biological tissue under test according to preset acquisition parameters based on near-infrared spectral acquisition technology; and truncating the original near-infrared spectral data according to a preset characteristic wavelength window to obtain target band near-infrared spectral data, which is then used as near-infrared spectral data.

[0043] The preset acquisition parameters may include at least one of the following: spectral acquisition band, resolution, number of scans, gain, and spectral acquisition method. The characteristic wavelength window can be predetermined based on the characteristic absorption response of the target fatty acid (e.g., eicosapentaenoic acid, docosahexaenoic acid, alpha-linolenic acid (CLA), gamma-linolenic acid (CLA), and arachidonic acid) in the near-infrared band, or it can be determined based on the contribution of each wavelength variable in the modeling sample data to the quantitative prediction model of the target fatty acid.

[0044] Since raw near-infrared spectral data typically contains a wide range of wavelength variables, and not all wavelength variables have a significant response to the target fatty acid content, after obtaining the raw near-infrared spectral data, the data can be truncated according to a preset characteristic wavelength window to obtain the target band near-infrared spectral data. This eliminates redundant bands with low correlation to the target fatty acid content, reducing the impact of irrelevant wavelength variables, background absorption, and noise variables on subsequent spectral transfer and quantitative prediction.

[0045] For example, Figure 2 This is a schematic flowchart illustrating the process of determining a characteristic wavelength window according to an embodiment of this application.

[0046] like Figure 2 As shown, determining the characteristic wavelength window includes: S21: Obtain calibration sample data corresponding to the target fatty acid, and divide it into training data and test data. The calibration sample data includes the near-infrared spectral data of the standard solution and the corresponding target fatty acid concentration information.

[0047] Specifically, multiple standard solutions with different target fatty acid concentrations can be pre-prepared, and near-infrared spectroscopy can be performed on each standard solution to obtain corresponding near-infrared spectral data. Simultaneously, based on the prepared concentration, weighed concentration, or concentration determined by gas chromatography of the standard solutions, the target fatty acid concentration information for each standard solution sample can be obtained. By correlating the near-infrared spectral data of multiple standard solutions with the corresponding target fatty acid concentration information, a calibration sample dataset for quantitative analysis of target fatty acids can be formed.

[0048] Thus, based on this corrected sample dataset, the correlation between different wavelength variables and target fatty acid concentrations can be further determined, thereby screening out key characteristic wavelengths that contribute significantly to the quantitative prediction of target fatty acids and determining the characteristic wavelength window.

[0049] Furthermore, in some embodiments, obtaining calibration sample data corresponding to the target fatty acid includes: preparing a standard solution and measuring the original spectral data of the standard solution according to the acquisition parameters based on near-infrared spectroscopy acquisition technology, wherein the standard solution is a pre-prepared solution containing the target fatty acid; performing at least one preprocessing step of denoising, smoothing, and scattering correction on the original spectral data to obtain preprocessed spectral data; and constructing calibration sample data based on the preprocessed spectral data and the target fatty acid concentration information corresponding to each standard solution.

[0050] The standard solution may include a single standard solution of the target fatty acid or a mixed standard solution of the target fatty acids. In some embodiments, preparing the standard solution includes: preparing at least one standard solution of eicosapentaenoic acid, docosahexaenoic acid, alpha-linolenic acid (octadecanoic acid-9,12,5-trienoic acid), gamma-linolenic acid (octadecanoic acid-6,9,12-trienoic acid), and arachidonic acid in different concentration ratios using n-heptane as a solvent. The concentration range of the standard solution covers the concentration range of each target fatty acid in the biological tissue to be tested, and multiple standard solutions with different concentrations are prepared for each target fatty acid.

[0051] For example, using salmon dorsal muscle tissue as the test subject and eicosapentaenoic acid (EPA) as the representative target fatty acid, this demonstrates a method for determining the content of Omega-3 and Omega-6 polyunsaturated fatty acids in biological tissues using Fourier transform near-infrared reflectance spectroscopy. It should be noted that the determination methods for other fatty acids (such as DHA, ALA (Alpha-Linolenic Acid), GLA (Gamma-Linolenic Acid), and AA (Arachidonic Acid)) are the same as in this example.

[0052] Using n-heptane (chromatographic grade, purity ≥99.9%) as a solvent, 53 EPA (eicosapentaenoic acid) single-standard solutions, 15 DHA single-standard solutions, and 24 EPA / DHA mixed standard solutions were first prepared for near-infrared reflectance spectroscopy. The concentration range of the solvents was 50–6000 ppm. This concentration gradient covers and exceeds the actual fatty acid content range that may be present in salmon tissue. The concentration gradient of each solvent was different and randomly determined. The number of standard solutions is not limited and can be 60, 92, 140, or other values, which can be adjusted according to the actual situation. By preparing 60 gradient standard solutions in the range of 50–6000 ppm, the problem of limited concentration range of target components in real biological tissues is effectively solved, giving the model a wider quantitative range and better generalization ability.

[0053] Next, after the standard solution is prepared, the raw spectral data of the standard solution is measured according to the acquisition parameters based on near-infrared spectroscopy acquisition technology. The specific acquisition parameters are as follows: Fourier transform near-infrared reflectance spectroscopy is used to acquire the standard solutions of different concentrations. Each concentration is repeatedly acquired 3 times and the average spectrum is taken. The spectral range is 4000-12800cm-1, the number of scans is 64, the resolution is 8cm-1, and the gain is 2.

[0054] Next, two methods were used for spectral preprocessing. One method was to use Savitzky-Golay filtering (a polynomial least squares convolution smoothing algorithm used to denoise and smooth the original spectral curve, eliminating random noise caused by instrument and environmental disturbances during spectral acquisition, and retaining effective information of spectral characteristic absorption peaks) for smoothing and first derivative calculation. The window width was 11 wavelength points and the polynomial order was 2. Subsequently, SNV scattering correction (Standard Normal Variate, used to correct spectral baseline shifts and scale differences caused by particle size, thickness, and light scattering effects in biological tissue samples) and second-order extended multiplicative scattering correction (EMSC, which can simultaneously correct spectral shifts caused by matrix background such as moisture and temperature while eliminating sample scattering interference) were performed to identify and remove outliers.

[0055] Another approach is to use Norris derivative filtering (a moving window average smoothing algorithm, also used for spectral denoising and curve smoothing), with a window width of 5 wavelength points. First-order derivative filtering is performed, followed by SNV scattering correction and second-order EMSC correction to identify and remove outliers.

[0056] After the first method of Savitzky-Golay filtering, the spectral baseline fluctuations are stable and the characteristic peaks of the target fatty acid are clearly presented. After the second method of Norris processing, the baseline is completely flattened, but high-frequency noise is amplified at the same time, and abnormal fluctuations occur at the end of the spectrum. Therefore, Savitzky-Golay filtering is the optimal preprocessing method.

[0057] In other words, Savitzky-Golay filtering or Norris is used for spectral denoising and smoothing, SNV scattering correction is applied, and extended multivariate scattering correction (EMSC) is used to subtract absorption interference from the n-heptane solvent, preserving the overall variation trend of characteristic peaks to obtain preprocessed spectral data. A multi-layered interference subtraction framework is designed, employing a hierarchical interference elimination process to sequentially subtract the effects of particle scattering, baseline drift, tissue moisture, and protein, maximizing the preservation of the intrinsic absorption characteristics of fatty acids and significantly improving the signal-to-noise ratio and model robustness.

[0058] Finally, based on the preprocessed spectral data and the target fatty acid concentration information corresponding to each standard solution, calibration sample data is constructed. Specifically, taking the EPA single standard solution as an example, the known preparation concentration of EPA in each standard solution is used as the response variable, i.e., the dependent variable Y; the Fourier near-infrared full spectrum data of each standard solution collected simultaneously and preprocessed as the prediction variable, i.e., the independent variable matrix X, is used to construct the calibration sample data. Then, the Kennard-Stone algorithm is used to divide the training data and test data in an 8:2 ratio according to the spectral distribution characteristics. Dividing the calibration sample data into training data and test data makes the determination of the characteristic wavelength window and the establishment of the quantitative prediction model more reliable.

[0059] In this way, by preparing standard solutions of known concentrations and measuring their near-infrared spectra, the accurate correspondence between the spectra and the true concentration values ​​is ensured. Furthermore, the original spectra are preprocessed to remove non-target interference, so that the spectra retain only the effective information related to fatty acid concentration. This provides standardized training data for the quantitative detection model, thereby ensuring the prediction accuracy of the model.

[0060] S22: Establish a quantitative prediction model for the target fatty acid based on the training data, and determine the contribution of each wavelength variable to the quantitative prediction model. The contribution of each wavelength variable to the quantitative prediction model is characterized by the variable importance projection value.

[0061] The training data is used to establish a quantitative prediction model for the target fatty acid and to determine the contribution of each wavelength variable to the model. The test data is used to evaluate the predictive performance of the model. Specifically, near-infrared spectral data from the test data can be input into the quantitative prediction model based on the training data to obtain the predicted concentration of the target fatty acid. This predicted concentration is then compared with the corresponding target fatty acid concentration in the test data to evaluate the accuracy and stability of the model.

[0062] For example, within the training dataset, PLSR (Partial Least Squares Regression) is first used to establish a quantitative prediction model for EPA based on the full-spectrum variables in the training data. Then, based on this quantitative prediction model, the VIP value (Variable Importance in Projection) corresponding to each wavelength variable is calculated, which can identify key wavelength variables that are highly correlated with the EPA concentration response and remove redundant spectral variables that contribute less to the EPA concentration response, in order to determine the degree of contribution of each wavelength variable to the quantitative prediction model. The degree of contribution of each wavelength variable to the quantitative prediction model is represented by the variable importance projection value.

[0063] S23: Spectral variables are screened according to the contribution of each wavelength variable to obtain the key characteristic wavelengths corresponding to the target fatty acid, and the characteristic wavelength window is determined based on the key characteristic wavelengths.

[0064] Referring to the example above, the full-spectrum variables in the training set are then screened using a VIP value ≥ 0.9 as the screening criterion. Redundant wavelength variables that contribute little to the EPA concentration response are removed, and several key feature wavelengths that contribute significantly to the prediction of EPA concentration are obtained. These feature wavelengths will serve as the input variable window for the subsequent quantitative detection model of the extract.

[0065] Based on the above screening, the key characteristic wavelengths of EPA are mainly concentrated in the range of 5800–6050 cm⁻¹. -1 and 4250~4350 cm -1 Two spectral windows, the locations of these characteristic wavelengths, will serve as input variable windows for subsequent modeling.

[0066] Furthermore, the quantitative prediction model is reconstructed based on the characteristic wavelength window. Specifically, based on the key characteristic wavelength positions obtained by screening, spectral variables at the corresponding wavelengths are extracted from the near-infrared spectral data of the training set, and the characteristic variable matrix for EPA quantitative prediction is reconstructed. Based on this characteristic variable matrix, the quantitative prediction model of EPA is re-established, thereby reducing the impact of interfering components such as DHA and redundant spectral variables on the model while retaining the effective response information of EPA, and improving the stability and prediction accuracy of the EPA quantitative prediction model.

[0067] In addition, the performance of the quantitative prediction model can be evaluated, specifically by using the coefficient of determination R. 2 The performance of a model is evaluated using root mean square error (RMSEC, RMSECV, and RMSEP) and relative analytical error (RPD). The evaluation criteria are as follows: RPD > 2.0 indicates that the model is excellent and can be used for precise quantitative analysis; 1.5 ≤ RPD ≤ 2.0 indicates that the model is usable and suitable for coarse screening; and RPD < 1.5 indicates that the model is unusable.

[0068] The performance of the model after Savitzky-Golay filtering is: calibration set R 2_c=0.7566, RMSEC=301.5334ppm, RMSECV=365.1000 ppm; prediction set R 2 _p=0.7178, RMSEP=282.8557 ppm; RPD=1.9403.

[0069] The performance of the model after Norris processing is: calibration set R 2 _c=0.9382, RMSEC=151.9687ppm, RMSECV=349.7040 ppm; prediction set R 2 _p=0.3388, RMSEP=432.9396 ppm; RPD=1.2677.

[0070] Comparison shows that the Savitzky-Golay filter results in an RPD of 1.94, making the model suitable for coarse screening and approximate quantitative analysis of EPA concentrations. The Norris filter, however, results in an RPD of 1.27, which, while flattening the baseline, amplifies high-frequency noise, leading to overfitting. Therefore, this step determines the Savitzky-Golay filter as the optimal preprocessing method, and the characteristic wavelength positions selected using this method are used for subsequent modeling.

[0071] In some embodiments, near-infrared spectral data is mapped to matching equivalent spectral data through a spectral transfer model, including: determining a spectral transfer model, and based on the spectral transfer model, determining a reference physical morphology corresponding to the original physical morphology, wherein the reference physical morphology is adapted to the quantitative detection model; and performing intermorphological spectral correction on the near-infrared spectral data according to the spectral transfer model to obtain equivalent spectral data under the reference physical morphology.

[0072] Specifically, the spectral transfer model is used to characterize the correspondence between near-infrared spectral data under different physical forms. Since the biological tissues to be tested usually have strong tissue scattering effects, baseline drift and background interference in their original physical forms, while quantitative detection models are usually established based on reference physical forms with less scattering influence or higher spectral stability, it is necessary to use the spectral transfer model to correct the spectral differences under different physical forms.

[0073] Furthermore, in some embodiments, determining the spectral transfer model includes: acquiring near-infrared spectral data of sample biological tissue under various physical morphologies; establishing spectral transfer functions between adjacent physical morphologies according to the spectral scattering differences corresponding to various physical morphologies; and constructing a spectral transfer model based on each spectral transfer function.

[0074] For example, fresh biological tissue samples are first extracted, and three different physical forms of samples are prepared in sequence. The near-infrared reflectance spectra of these samples are collected. The specific acquisition parameters are as follows: Fourier transform near-infrared reflectance spectroscopy is used to collect the above standards at different concentrations. Each concentration is collected three times and the average spectrum is taken. The measurement spectral range is 4000-12800 cm-1, the number of scans is 64, the resolution is 8 cm-1, and the gain is 2.

[0075] The process of collecting near-infrared spectral data in three different physical forms is as follows: Block tissue: A cylindrical meat block with a diameter of 3cm and a thickness of 1.75cm was cut using a stainless steel punch to ensure that the surface was flat and undamaged. The sample rotating cup was used to collect the average spectrum of the meat block on both sides three times for each side to fully reflect the original tissue state.

[0076] Coarsely chopped meat: The above-mentioned blocky tissue is chopped with a stainless steel knife to obtain meat mince with a particle size of about 2-5 mm. The integrity of the tissue is moderately damaged to change the light scattering characteristics. The meat mince is spread evenly in a rotating sample cup (thickness not less than 1 cm), the surface is gently flattened, and the average spectrum is taken from each side 3 times as the spectrum of coarsely chopped meat mince.

[0077] Finely chopped minced meat: The coarsely chopped minced meat was further finely chopped using a tissue homogenizer (5000 rpm, 2 min) until it reached a homogeneous paste state. The homogenized minced meat was placed into a sample cup, and after smoothing the surface, three spectral acquisitions were performed. The average spectrum was taken as the spectrum of the finely chopped minced meat. The spectral data of the above three forms were used for subsequent scattering effect analysis and cross-morphological spectral transfer modeling. Three parallel samples were taken for each salmon sample, and the spectral acquisition and true value determination of each parallel sample for the three forms were performed independently.

[0078] The spectral transfer function (PSF) is used to characterize the spectral correspondence between two adjacent physical forms. This PSF can be established using a piecewise direct standardization (PDS) method. Specifically, within a local window of the target domain spectrum, a correspondence with the source domain spectrum is established through multiple linear regression. This yields a transfer function used to correct or convert spectral data from one physical form to equivalent spectral data from another. This transfer function corrects for scattering effects caused by differences in physical forms, allowing the quantitative detection model used for the baseline physical form to be applied to fish tissue. This solves the problem of inconsistent spectral scattering effects between real fish tissue and extracts due to differences in physical forms, enabling direct and non-destructive prediction of omega-3 and omega-6 fatty acid content in biological tissues.

[0079] For example, using spectral data from three physical morphologies, a stepwise transfer mapping relationship from "homogeneous state" to "original tissue state" is constructed. The transfer path is: extract spectrum (homogeneous, no scattering) → finely cut homogeneous surimi spectrum (homogeneous, weak scattering) → coarsely cut surimi spectrum (heterogeneous, moderate scattering) → block tissue spectrum (original state, strong scattering). Using the spectral data of two adjacent morphologies as input and output, a piecewise direct correction (PDS) or slope / intercept correction (S / B) algorithm is used to establish spectral transfer functions between morphologies. For example, using the finely cut homogeneous surimi spectrum as a benchmark, a transfer model from the coarsely cut surimi spectrum to the finely cut homogeneous surimi spectrum is established, and so on, thus forming a complete spectral transfer link. After establishing spectral transfer functions between multiple adjacent physical morphologies, the spectral transfer functions can be cascaded or combined according to the order of scattering differences between physical morphologies to construct a spectral transfer model.

[0080] In this way, the near-infrared spectral data of the biological tissue under test in its original physical morphology can be corrected by the spectral transfer model to obtain equivalent spectral data that matches the baseline physical morphology. This enables the quantitative detection model based on the extract to be successfully applied to blocky biological tissues, decomposing complex migration into multiple small-scale progressive transformations. This significantly improves the conversion accuracy from quantitative detection model adapted to extract to tissue application, achieving non-destructive and direct prediction of intact tissues.

[0081] In some embodiments, before establishing the spectral transfer function between adjacent physical morphologies, the method further includes: performing spectral preprocessing on the near-infrared spectral data of the sample biological tissue under various physical morphologies within a preset characteristic wavelength window to obtain preprocessed near-infrared spectral data, wherein the spectral preprocessing includes at least one of smoothing and denoising, baseline correction and scattering correction, and the spectral correction between morphologies is based on the preprocessed near-infrared spectral data.

[0082] Specifically, the following preprocessing steps are performed sequentially within the characteristic wavelength window: Norris derivative filtering is used to smooth and denoise the spectral window near the characteristic peak to eliminate high-frequency random noise; the airPLS algorithm (Adaptive Iteratively Reweighted Penalized Least Squares) is used for baseline correction to subtract baseline drift caused by tissue inhomogeneity; and standard normal variable transformation (SNV) is used to correct the scattering effect caused by particle size differences, preserving the absorption characteristics of the target fatty acids to the greatest extent while effectively removing tissue background interference.

[0083] In some embodiments, the fatty acid detection method further includes: acquiring near-infrared spectral data of validation set samples in their original physical state and determining the true content of target fatty acids in the validation set samples; inputting the near-infrared spectral data into a quantitative detection model to obtain direct prediction results; and converting the near-infrared spectral data into equivalent spectral data through a spectral transfer model and inputting the equivalent spectral data into a quantitative detection model to obtain migration prediction results; and evaluating the migration effect of the spectral transfer model based on the differences between the direct prediction results and the migration prediction results and the true content values.

[0084] In other words, to verify the migration effect of the spectral transfer model, independent validation set samples (samples that did not participate in any previous modeling steps) were used. Simultaneously, the near-infrared tissue spectra of the block-shaped independent validation set samples were measured, and the true content of the target fatty acids in the extract samples obtained by standard processing of the block tissue was determined (this true content was obtained by gas chromatography). The near-infrared spectral data were input into the quantitative detection model to obtain the direct prediction results, and the equivalent spectral data were input into the quantitative detection model to obtain the migration prediction results. The accuracy difference between the direct prediction results and the migration prediction results was compared, using root mean square error (RMSEP) and relative analysis error (RPD) as evaluation indicators to verify the migration effect of the spectral transfer model.

[0085] For example, Figure 3 This is a schematic diagram of the process of training a quantitative detection model according to an embodiment of this application.

[0086] like Figure 3 As shown, the training process of the quantitative detection model specifically includes: S31: Determine the sample data and perform at least one of the following preprocessing steps on the near-infrared spectral data in the sample data: smoothing and denoising, derivative transformation, baseline correction, and background interference correction. The sample data includes the near-infrared spectral data of each sample and the true values ​​of the target fatty acid content.

[0087] Among these methods, smoothing and denoising are used to reduce random noise during spectral acquisition; derivative transformation is used to enhance spectral detail and reduce baseline drift; baseline correction is used to subtract baseline shifts caused by changes in instrument status, sample background, or optical path; and background interference correction is used to reduce the influence of solvents, matrices, or other coexisting components on the spectral response of the target fatty acid. The true content of the target fatty acid is determined by gas chromatography. This preprocessed near-infrared spectral data can be used as an input variable for the quantitative detection model, and the true content of the target fatty acid determined by gas chromatography can be used as an output reference value or response variable for the quantitative detection model.

[0088] By performing the above preprocessing on the near-infrared spectral data, the consistency and comparability of spectral data between different samples can be improved, the effective expression of the target fatty acid-related spectral information can be enhanced, and the impact of noise, baseline drift and background interference on the establishment of subsequent quantitative detection models can be reduced.

[0089] Furthermore, in some embodiments, determining sample data includes: pre-treating symmetrically taken sample biological tissues to obtain fatty acid extract; preparing fatty acid extract samples of different concentrations based on the fatty acid extract; and performing near-infrared spectroscopy acquisition and gas chromatography analysis on the fatty acid extract samples to obtain near-infrared spectral data and the true value of the target fatty acid content, respectively, as sample data.

[0090] For example, the pretreatment of symmetrically taken biological tissue samples to obtain fatty acid extract is as follows: Take fresh salmon back muscle, remove the skin and visible fat, chop and mix evenly. Accurately weigh 1.65 g of the homogenized salmon sample into a 250 mL round-bottom flask, add 10 mL of 8.3 mol / L hydrochloric acid solution, mix well, and hydrolyze in an 80℃ water bath for 40 min, shaking once every 10 min during the process. After hydrolysis, remove the sample and cool to room temperature. Add 10 mL of 95% ethanol, mix well, and transfer the hydrolysate to a separatory funnel. Rinse the flask and stopper with 50 mL of a mixture of diethyl ether and petroleum ether (V / V, 1:1). Combine the rinsing solution in the separatory funnel, shake for 5 min, let stand for 10 min, and collect the ether extract. Repeat the extraction of the hydrolysate three times, and rinse the separatory funnel again with a mixture of diethyl ether and petroleum ether. Combine all extracts in a 250 mL flask and concentrate to dryness by rotary evaporation to obtain the fat extract. Add 8 mL of 2% sodium hydroxide methanol solution and reflux in a water bath at 80℃±1℃ until the oil droplets disappear. Then add 7 mL of 14% boron trifluoride methanol solution from the top of the condenser and continue reflux for 2 min. After cooling, accurately add 15 mL of n-heptane and shake for 2 min. Then add saturated sodium chloride aqueous solution, let stand for separation, and take about 5 mL of the upper n-heptane extract. Add 4 g of anhydrous sodium sulfate and dry. Shake for 1 minute. After standing for 5 minutes, the supernatant is collected to obtain the unsaturated fatty acid extract.

[0091] Next, the unsaturated fatty acid extract was concentrated by rotary evaporation in a 40°C water bath, redissolved in n-heptane, and brought to a final volume of 20 mL to obtain the stock solution. This stock solution was then serially diluted with n-heptane to prepare 60 calibration samples with different concentration gradients, ensuring that the EPA concentration range covered 50–6000 ppm. Each gradient sample was divided into two portions: one for near-infrared spectroscopy acquisition and the other for gas chromatography true value determination. Near-infrared spectral data and the true values ​​of the target fatty acid content were obtained as sample data, and a gradient concentration database was established.

[0092] The near-infrared spectral acquisition parameters are as follows: Fourier transform near-infrared reflectance spectroscopy is used to acquire the above standards at different concentrations. Each concentration is repeatedly acquired 3 times and the average spectrum is taken. The measurement spectral range is 4000-12800cm-1, the number of scans is 64, the resolution is 8cm-1, and the gain is 2x.

[0093] The content of EPA, DHA, ALA, GLA, and AA in unsaturated fatty acid extracts was analyzed by gas chromatography. Specifically, the external standard method was used to prepare standard solutions, which were then injected and detected under the same gas chromatographic instrument conditions. A linear standard curve (calibration line) was obtained by fitting the standard concentration as the x-axis and the corresponding peak area as the y-axis. Then, the salmon extract was analyzed under the same instrument parameters, and the peak areas of each fatty acid in the sample were read and substituted into the standard curve equation to calculate the corresponding contents of EPA, DHA, ALA, GLA, and AA in the extract. The gas chromatographic column used was SP-2560 (specifications: 100m x 0.25mm x 0.2um), with a flame ionization detector (FID). The injection port temperature was 270℃, the detector temperature was 280℃, and the split ratio was 100:1. Temperature program: Initial temperature 100℃, hold for 13 min; increase to 180℃ at 10℃ / min, hold for 6 min; increase to 200℃ at 1℃ / min, hold for 20 min; increase to 230℃ at 4℃ / min, hold for 10.5 min. Injection volume 1 μL, carrier gas high-purity nitrogen (purity ≥99.999%), flow rate 1.0 mL / min. Qualitative (retention time control) and quantitative (external standard method, linear regression of peak area and concentration) analyses were performed using EPA standard solutions of different concentrations, with units expressed in ppm.

[0094] S32: Extract the spectral data within the characteristic wavelength window corresponding to the target fatty acid from the preprocessed near-infrared spectral data, and use it as the input variable for the quantitative detection model.

[0095] For sample data consisting of near-infrared spectral data and the true values ​​of target fatty acid content (obtained by gas chromatography analysis), Savitzky-Golay smoothing, first derivative processing, airPLS baseline correction, and EMSC solvent interference subtraction were performed sequentially. Based on this, only the spectral windows near the key characteristic wavelengths of each target fatty acid were extracted as input variables for the quantitative detection model, eliminating the need for full spectrum analysis. This reduces the number of variables, lowers the risk of overfitting, and improves the model's computational efficiency.

[0096] For example, the total number of samples is 30, calculated as 10 gradients × 3 repetitions. These are randomly divided into a training set (22 samples) and a prediction set (8 samples) at a 3:1 ratio. The spectral data in this set is preprocessed as follows: Savitzky-Golay smoothing (window width 15, polynomial order 2), first derivative (window width 15, polynomial order 2), and airPLS baseline correction (λ=10). 5 (Maximum number of iterations: 10) and EMSC solvent interference subtraction. Based on the above preprocessing, only the EPA characteristic wavelength window (5800–6050 cm⁻¹) is extracted. -1 and 4250~4350 cm -1 Spectral variables within the range of 4000–12800 cm⁻¹ are used as input, eliminating the need for inputting the full spectrum (4000–12800 cm⁻¹). -1 ) to conduct analysis.

[0097] S33: Divide the sample data into a training set and a prediction set, and train a quantitative detection model based on the input variables and the corresponding true values ​​of the target fatty acid content in the training set. Evaluate the predictive performance of the quantitative detection model based on the prediction set.

[0098] In other words, the quantitative detection model is trained by using the spectral data of each sample in the training set as the input variable and the true value of the target fatty acid content of each sample (such as the standard value determined by gas chromatography) as the output target. The quantitative detection model is trained by regression modeling algorithm (such as partial least squares regression or neural network) so that the quantitative detection model learns and establishes the mapping relationship between spectral features and fatty acid content.

[0099] Furthermore, in some embodiments, the quantitative detection model is established based on the partial least squares regression algorithm, and the performance of the quantitative detection model is evaluated based on at least one of the coefficient of determination, root mean square error, and relative analysis error.

[0100] After the quantitative detection model is trained, the spectral data of each sample in the prediction set are input into the trained model to obtain the predicted fatty acid content of each sample. By comparing the predicted values ​​with the corresponding true values, the prediction error index is calculated to evaluate the model's prediction accuracy, stability, and generalization ability. If the evaluation results meet the preset accuracy requirements, the quantitative detection model can be used for actual fatty acid content prediction; if not, the model parameters need to be adjusted or the training strategy optimized until the expected performance is achieved.

[0101] Specifically, model performance is comprehensively evaluated using the following metrics: coefficient of determination R0 2 (including the calibration set R) 2 _cal, cross-validation R 2 _cv, prediction set R 2The model is evaluated using the following parameters: root mean square error (RMSE), root mean square error (RMSEC) of the calibration set, root mean square error of the cross-validation set (RMSECV), and root mean square error of the prediction set (RMSEP), and relative analysis error (RPD). An RPD > 2.0 indicates an excellent model suitable for precise quantitative analysis; 1.5 ≤ RPD ≤ 2.0 indicates a usable model suitable for preliminary screening; and RPD < 1.5 indicates an unusable model. The optimal number of latent variable factors (7 in this example) is determined through five-fold cross-validation.

[0102] For example, the performance of the EPA extract quantitative detection model is as follows: R 2 _cal=0.978, R 2 _cv=0.965, R 2 Pred = 0.959; RMSEC = 0.42 mg / mL, RMSECV = 0.55 mg / mL, RMSEP = 0.61 mg / mL; RPD = 4.3. As can be seen, RPD > 2.0, indicating that the model is excellent and can be used for accurate quantitative analysis of EPA content in salmon extract.

[0103] It should be noted that quantitative detection models for DHA, ALA, GLA, and AA can be established using the same method, and the performance of all models meets the standard of RPD>2.0.

[0104] In summary, this application systematically solves the problem that "high-precision models based on extracts cannot be directly applied to intact biological tissues." Through cascaded PDS cross-morphological spectral transfer and multi-layer interference subtraction framework, it achieves rapid, non-destructive, and accurate quantification of EPA, DHA, ALA, GLA, and AA in intact block tissues. It has made significant progress in terms of the morphology of the detected object (block → intact tissue), model transferability, systematic interference handling, concentration coverage, and prediction accuracy.

[0105] For example, Figure 4 This is a schematic diagram of a fatty acid detection device according to an embodiment of this application.

[0106] like Figure 4 As shown, the fatty acid detection device 40 includes: an acquisition module 401 for acquiring near-infrared spectral data of the biological tissue to be tested in its original physical form; a mapping module 402 for mapping the near-infrared spectral data into matching equivalent spectral data through a spectral transfer model, wherein the equivalent spectral data is the spectral data of the biological tissue to be tested in a reference physical form; and a determination module 403 for inputting the equivalent spectral data into a trained quantitative detection model to determine the fatty acid content in the biological tissue to be tested, wherein the quantitative detection model is established based on the spectral data in the reference physical form and the corresponding true value of fatty acid content.

[0107] According to the fatty acid detection device of this application, the near-infrared spectral data of the biological tissue to be tested in its original physical morphology is first converted into equivalent spectral data in the reference physical morphology through a spectral transfer model. Then, the equivalent spectral data is input into the quantitative detection model for prediction. In this way, the model input data of the actual test sample and the modeling data of the quantitative detection model can be matched in spectral morphology without destroying the original morphology of the biological tissue to be tested. This reduces the spectral inconsistency between the original physical morphology and the reference physical morphology caused by scattering effects and differences in tissue background, reduces the prediction deviation caused by differences in physical morphology, and achieves non-destructive and accurate prediction of fatty acid content in the biological tissue to be tested, thus meeting the needs of non-invasive detection in clinical or in vivo research.

[0108] In addition, those skilled in the art will understand that the structure of the computing device shown in the above figures does not constitute a limitation on the computing device. The computing device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0109] This application also provides a computer storage medium, characterized in that the storage medium stores at least one instruction, which is executed by a processor to implement the fatty acid detection method of any of the above embodiments.

[0110] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this application can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transfer of a computer program from one place to another. Storage media can be any available medium accessible to a general-purpose or special-purpose computer.

[0111] It should be noted that the technical solutions described in this application can be combined arbitrarily without conflict. The above description is only a specific embodiment of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A method for detecting fatty acids, characterized in that, include: Acquire near-infrared spectral data of the biological tissue under test in its original physical morphology; The near-infrared spectral data is mapped to matching equivalent spectral data through a spectral transfer model, wherein the equivalent spectral data is the spectral data of the biological tissue under test in a baseline physical morphology. The equivalent spectral data is input into a trained quantitative detection model to determine the fatty acid content in the biological tissue to be tested. The quantitative detection model is established based on the spectral data under the baseline physical morphology and the corresponding true values ​​of fatty acid content.

2. The fatty acid detection method according to claim 1, characterized in that, The acquisition of near-infrared spectral data of the biological tissue under test in its original physical morphology includes: Based on near-infrared spectroscopy acquisition technology, the raw near-infrared spectral data of the biological tissue to be tested are measured according to preset acquisition parameters; According to the preset characteristic wavelength window, the original near-infrared spectral data is truncated to obtain the target band near-infrared spectral data, which is then used as the near-infrared spectral data.

3. The fatty acid detection method according to claim 2, characterized in that, The determination of the characteristic wavelength window includes: Obtain calibration sample data corresponding to the target fatty acid, and divide it into training data and test data. The calibration sample data includes near-infrared spectral data of the standard solution and the corresponding target fatty acid concentration information. A quantitative prediction model for the target fatty acid is established based on the training data, and the contribution of each wavelength variable to the quantitative prediction model is determined. The contribution of each wavelength variable to the quantitative prediction model is characterized by the variable importance projection value. The spectral variables are screened according to the contribution of each wavelength variable to obtain the key characteristic wavelengths corresponding to the target fatty acid, and the characteristic wavelength window is determined based on the key characteristic wavelengths.

4. The fatty acid detection method according to claim 3, characterized in that, The acquisition of the corrected sample data corresponding to the target fatty acid includes: Prepare a standard solution and, based on the near-infrared spectroscopy acquisition technology, determine the raw spectral data of the standard solution according to the acquisition parameters, wherein the standard solution is a pre-prepared solution containing the target fatty acid; The original spectral data is subjected to at least one preprocessing step, including denoising, smoothing, and scattering correction, to obtain preprocessed spectral data. Based on the preprocessed spectral data and the target fatty acid concentration information corresponding to each standard solution, the calibration sample data is constructed.

5. The fatty acid detection method according to claim 4, characterized in that, The standard solution to be prepared includes: Using n-heptane as a solvent, at least one standard solution of eicosapentaenoic acid, docosahexaenoic acid, alpha-linolenic acid, gamma-linolenic acid, and arachidonic acid is prepared in different concentration ratios. The concentration range of the standard solution covers the concentration range of each target fatty acid in the biological tissue to be tested, and multiple standard solutions with different concentrations are prepared for each target fatty acid.

6. The fatty acid detection method according to claim 1, characterized in that, The step of mapping the near-infrared spectral data into matching equivalent spectral data through a spectral transfer model includes: The spectral transfer model is determined, and based on the spectral transfer model, the reference physical form corresponding to the original physical form is determined, wherein the reference physical form is adapted to the quantitative detection model. Based on the spectral transfer model, the near-infrared spectral data is subjected to intermorphic spectral correction to obtain the equivalent spectral data under the reference physical morphology.

7. The fatty acid detection method according to claim 6, characterized in that, The determination of the spectral transfer model includes: Near-infrared spectral data of biological tissue samples under various physical morphologies were collected. Based on the spectral scattering differences corresponding to the various physical forms, a spectral transfer function is established between adjacent physical forms. Based on the spectral transfer functions, the spectral transfer model is constructed.

8. The fatty acid detection method according to claim 7, characterized in that, Before establishing the spectral transfer function between adjacent physical forms, the method further includes: Within a preset characteristic wavelength window, the near-infrared spectral data of the sample biological tissue under various physical morphologies are subjected to spectral preprocessing to obtain preprocessed near-infrared spectral data. The spectral preprocessing includes at least one of smoothing and denoising, baseline correction and scattering correction. The morphological spectral correction is performed based on the preprocessed near-infrared spectral data.

9. The fatty acid detection method according to claim 7, characterized in that, The method further includes: Near-infrared spectral data of the validation set samples in their original physical state were collected, and the true values ​​of the target fatty acid content in the validation set samples were determined. The near-infrared spectral data is input into the quantitative detection model to obtain direct prediction results. The near-infrared spectral data is converted into equivalent spectral data through the spectral transfer model, and the equivalent spectral data is input into the quantitative detection model to obtain migration prediction results. The migration effect of the spectral transfer model is evaluated based on the differences between the direct prediction results and the migration prediction results and the true content values.

10. The fatty acid detection method according to claim 1, characterized in that, The training process of the quantitative detection model specifically includes: The sample data is determined, and at least one of the following preprocessing methods is applied to the near-infrared spectral data in the sample data: smoothing and denoising, derivative transformation, baseline correction, and background interference correction. The sample data includes the near-infrared spectral data of each sample and the true values ​​of the target fatty acid content. From the preprocessed near-infrared spectral data, the spectral data within the characteristic wavelength window corresponding to the target fatty acid are extracted and used as the input variable of the quantitative detection model; The sample data is divided into a training set and a prediction set. The quantitative detection model is trained based on the input variables and the corresponding true values ​​of the target fatty acid content in the training set. The predictive performance of the quantitative detection model is evaluated based on the prediction set.

11. The fatty acid detection method according to claim 10, characterized in that, The quantitative detection model is established based on the partial least squares regression algorithm, and its performance is evaluated based on at least one of the coefficient of determination, root mean square error, and relative analysis error.

12. The fatty acid detection method according to claim 10, characterized in that, The determined sample data includes: Symmetrically taken biological tissue samples were pretreated to obtain fatty acid extract; Fatty acid extract samples of different concentrations were prepared based on the fatty acid extract. The fatty acid extract sample was subjected to near-infrared spectroscopy and gas chromatography analysis to obtain the near-infrared spectral data and the true content of the target fatty acid, which were used as the sample data.

13. A fatty acid detection device, characterized in that, include: The acquisition module is used to acquire near-infrared spectral data of the biological tissue under test in its original physical morphology; The mapping module is used to map the near-infrared spectral data into matching equivalent spectral data through a spectral transfer model, wherein the equivalent spectral data is the spectral data of the biological tissue under the reference physical morphology. The determination module is used to input the equivalent spectral data into a trained quantitative detection model to determine the fatty acid content in the biological tissue to be tested, wherein the quantitative detection model is established based on the spectral data under the reference physical morphology and the corresponding true value of fatty acid content.

14. A computer storage medium, characterized in that, The storage medium stores at least one instruction, which is executed by a processor to implement the fatty acid detection method as claimed in any one of claims 1 to 12.

Images