Soybean crude fat content measuring method and system based on double-spectrum fusion

Abstract

The invention provides a soybean crude fat content measurement method and system based on double-spectrum fusion, and relates to the technical field of soybean crude fat detection, and the method comprises four steps of sample preparation and data acquisition, spectrum pretreatment, data compression and feature fusion, and content prediction output. According to the method, the improvement of the soybean crude fat content detection technology from single feature analysis to multi-dimensional collaborative perception is realized through the collaborative cooperation of a dynamic weighted feature fusion algorithm and a dynamic parameter adjustment mechanism, specifically, the spectrum pretreatment link has adaptive optimization ability by responding to the environmental temperature and humidity change in real time, and the detection accuracy is improved. Meanwhile, multiple scatter correction of the near infrared spectrum effectively overcomes interference of moisture absorption on detection precision through a humidity-driven reference spectrum dynamic selection mechanism, finally, a fusion feature vector is obtained through a dynamic weighted feature fusion algorithm, complementary information of double spectrums is deeply excavated, and therefore structural information of the soybean oil is comprehensively represented.

Description

Technical Field

[0001] This invention relates to the field of soybean crude fat detection technology, and in particular to a method and system for measuring soybean crude fat content based on dual-spectrum fusion. Background Technology

[0002] As an important oilseed crop, the crude fat content of soybeans is a key indicator for assessing quality and breeding value. Traditional crude fat measurement mainly relies on biochemical assays such as Soxhlet extraction. Although this method is listed as a national standard, it suffers from drawbacks such as long extraction time (>10 hours), the need for toxic solvents (such as petroleum ether), and cumbersome operation, making it difficult to meet the timeliness requirements of large-scale germplasm resource screening. To improve detection efficiency, spectroscopic techniques are increasingly being applied to agricultural product quality analysis, with near-infrared spectroscopy (NIR) and Raman spectroscopy becoming research hotspots due to their rapid and non-destructive characteristics.

[0003] However, existing methods for measuring crude fat content in soybeans still have some shortcomings, mainly reflected in the limitations of spectroscopic techniques, the inefficiency of feature extraction, the shallowness of model fusion, the lack of algorithm robustness, and the absence of system integration. Specifically, although near-infrared spectroscopy is widely used in organic component analysis, it relies on the absorption characteristics of polar groups (such as OH and CH), resulting in insufficient sensitivity for detecting non-polar fatty acid components (such as oleic acid and linoleic acid) in soybean oil. Furthermore, it is easily affected by moisture absorption, requiring complex preprocessing (such as multivariate scattering correction) to suppress scattering effects, but it cannot fundamentally solve the problem of accuracy loss in the detection of whole grains.

[0004] Although Raman spectroscopy is sensitive to nonpolar bonds (C=C, CC) and can directly detect intact samples, the intensity of the intrinsic fluorescence background noise in biological samples can be tens of thousands of times that of the effective signal. Traditional denoising methods (such as wavelet transform and Savitzky-Golay filtering) have contradictions of over-smoothing or under-denoising, which leads to the shift or loss of characteristic peaks and reduces the reliability of measurement.

[0005] In terms of feature extraction and model fusion, existing methods mostly employ single feature selection algorithms (principal component analysis or continuous projection algorithms), which fail to efficiently compress redundant variables in high-dimensional spectral data, resulting in weak model generalization ability. Furthermore, the complementary advantages of Raman and near-infrared spectroscopy are not fully utilized. Therefore, this invention proposes a method and system for measuring soybean crude fat content based on dual-spectral fusion to address the problems existing in the prior art. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to propose a method and system for measuring the crude fat content of soybeans based on dual-spectral fusion. This method ensures the maximization of the quality of the original spectral data by introducing a dynamic parameter adjustment mechanism, providing high-quality input for feature fusion. The dynamic weighted feature fusion, through physicochemical weight optimization, deeply mines the complementary information of the dual spectra, thus solving the problems in the prior art.

[0007] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a method for measuring the crude fat content of soybeans based on dual-spectral fusion, comprising the following steps:

[0008] Step 1: Sample Preparation and Data Acquisition

[0009] Soybean samples were prepared, and Raman and near-infrared spectral data of the soybean samples were obtained under monitored ambient temperature and humidity. The reference value of crude fat content of soybeans was then determined by Soxhlet extraction.

[0010] Step 2: Spectral Preprocessing

[0011] Hilbert vibrational decomposition denoising was performed on the Raman spectral data, and the spectral peaks were fitted by the Voigt function to remove the fluorescence background and high-frequency noise. Then, multivariate scattering correction and Savitzky-Golay first derivative processing were performed on the near-infrared spectral data. At the same time, a dynamic parameter adjustment mechanism was introduced in both processing methods.

[0012] Step 3: Data Compression and Feature Fusion

[0013] Based on the Hilbert yellow envelope spectroscopy method, the preprocessed Raman and near-infrared spectral data are compressed, significant frequency feature variables are extracted, and a dynamic weighted feature fusion algorithm is used to fuse Raman and near-infrared features to obtain a fused feature vector.

[0014] Step 4: Predicted Content Output

[0015] The fused feature vector is input into a pre-established quantitative relationship model for crude fat content, and the crude fat content value of soybeans is output.

[0016] Further improvements include the following steps:

[0017] Step 5: Development of the graphical user interface system: Develop a graphical user interface system based on the Python platform, integrating spectral preprocessing, feature extraction, and model prediction modules to automate and visualize the measurement process.

[0018] A further improvement is that: in step one, the Raman spectral data is obtained by scanning a whole grain sample, and the near-infrared spectral data is obtained by scanning both the whole grain sample and the powder sample.

[0019] Further improvements are made in the following way: In step two, the specific method of Hilbert vibration decomposition denoising is as follows: the initial parameters of the spectral peak are obtained based on the first component of HVD, the spectral peak is reconstructed using the Voigt function within the moving window, and the denoising effect is optimized by maximizing the fitting function.

[0020] A further improvement is made in the following: In step two, the dynamic parameter adjustment mechanism includes dynamic adjustment of ambient temperature and dynamic adjustment of ambient humidity. The dynamic adjustment of ambient temperature is to dynamically adjust the size of the moving window of the Voigt function fitting spectrum peak according to the ambient temperature. The dynamic adjustment of ambient humidity is to dynamically adjust the window parameters of the reference spectrum selection and derivative processing of the multivariate scattering correction according to the ambient humidity.

[0021] The further improvement lies in the following: In step three, the specific method of the dynamic weighted feature fusion algorithm is as follows: establish the chemical bond correspondence between Raman features and near-infrared features, calculate the bond energy weights through density functional theory, and realize the complementary fusion of dual-spectral features based on the weights. Among them, Raman features cover the vibrational information of C=C and C=C bonds, and near-infrared features cover the information of CH and NH groups.

[0022] A further improvement is made in step four, where the quantitative relationship model is established by performing regression analysis on the reference values ​​measured by the Soxhlet extraction method and the historical fusion feature vector.

[0023] A measurement system based on a dual-spectral fusion method for measuring crude fat content in soybeans includes a data acquisition module for collecting Raman and near-infrared spectral data of soybean samples, and an integrated Soxhlet extraction device for obtaining reference values ​​of crude fat content.

[0024] The spectral preprocessing module is used to process the acquired Raman and near-infrared spectral data;

[0025] The feature extraction and fusion module is used to compress spectral data based on the Hilbert yellow envelope spectroscopy algorithm and execute a dynamic weighted feature fusion algorithm to output a fused feature vector.

[0026] The model fusion module is used to calculate crude fat content based on the fused feature vector and a quantitative relationship model for crude fat content.

[0027] User Interface Module: Used to develop graphical interfaces based on PyQt5 and Tkinter libraries, providing functions for data uploading, preprocessing, feature extraction, feature fusion, and content measurement.

[0028] The beneficial effects of this invention are as follows: By combining a dynamic weighted feature fusion algorithm and a dynamic parameter adjustment mechanism, this invention improves the detection technology of soybean crude fat content from single feature analysis to multi-dimensional collaborative perception. Specifically, by responding to changes in environmental temperature and humidity in real time, the spectral preprocessing stage has adaptive optimization capabilities. At the same time, the multivariate scattering correction of near-infrared spectroscopy effectively overcomes the interference of moisture absorption on detection accuracy through a humidity-driven dynamic selection mechanism of the reference spectrum, maintaining stable detection performance even in complex environments. Finally, the fused feature vector is obtained through the dynamic weighted feature fusion algorithm, deeply mining the complementary information of the two spectra, thereby comprehensively characterizing the structural information of soybean oil and providing an efficient and stable solution for breeding. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the steps of the present invention.

[0030] Figure 2 This is a schematic diagram of obtaining Raman spectral data by non-destructively scanning an intact grain sample according to the present invention.

[0031] Figure 3 This is a schematic diagram of the near-infrared spectrum obtained by destructively scanning an intact grain sample according to the present invention.

[0032] Figure 4 This is a schematic diagram of the near-infrared spectrum obtained by destructively scanning an intact grain sample according to the present invention.

[0033] Figure 5 This is a schematic diagram of the denoising algorithm based on HVD decomposition of the present invention.

[0034] Figure 6 This is a schematic diagram of the data compression process based on Hilbert's yellow envelope spectrum according to the present invention. Detailed Implementation

[0035] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0036] according to Figure 1 As shown, this embodiment proposes a method for measuring the crude fat content of soybeans based on dual-spectral fusion, including the following steps:

[0037] Step 1: Sample Preparation and Data Acquisition

[0038] Soybean samples were prepared, and Raman and near-infrared spectral data were acquired. Raman spectral data were obtained through non-destructive scanning of intact seed samples, while near-infrared spectral data were obtained from both intact seed and powder samples. Simultaneously, the Soxhlet extraction method was used to determine the reference value for soybean crude fat content. The Raman spectral data was acquired using a 785nm laser source with a scanning range of 200cm². -1 ~3200cm -1 The integration time is 10 seconds, and the average spectrum of the whole grain sample is processed.

[0039] The near-infrared spectral data was acquired by scanning over a range of 4000 cm⁻¹. -1 ~11520cm -1 8cm resolution -1 And perform multiple scans and average the results.

[0040] Step 2: Spectral Preprocessing

[0041] Hilbert vibrational decomposition (HVD) denoising was performed on the Raman spectral data. The peaks were fitted using the Voigt function to remove fluorescence background and high-frequency noise. Next, multivariate scattering correction and Savitzky-Golay first-order derivative processing were applied to the near-infrared spectral data, while moisture-sensitive bands were removed. The specific method of HVD denoising involves obtaining initial peak parameters based on the first component of HVD, reconstructing the peaks using the Voigt function within a moving window, and optimizing the denoising effect by maximizing the fit function. A dynamic parameter adjustment mechanism was introduced in both processing methods.

[0042] The dynamic parameter adjustment mechanism includes dynamic adjustment of ambient temperature and ambient humidity. The dynamic adjustment of ambient temperature involves adjusting the size of the moving window of the Voigt function fitting spectrum peak according to the ambient temperature. Specifically, when the ambient temperature rises above 24°C, the Voigt function fitting window for HVD denoising is adjusted to 3cm. -1 Up to 4cm -1 To suppress thermal noise, adjust the window to 5cm when the ambient temperature is below 20℃. -1 Up to 6cm -1 To preserve signal integrity at low temperatures, a standard window parameter of 4.5 cm is used for the temperature range of 20°C to 24°C. -1 Maintain processing stability;

[0043] The dynamic adjustment of ambient humidity involves dynamically adjusting the reference spectrum selection and derivative processing window parameters for multivariate scattering correction based on ambient humidity. Specifically, when the ambient humidity is higher than 55% RH, the MSC correction of near-infrared spectroscopy uses the spectrum of a dry sample as a reference, and the Savitzky-Golay derivative window is adjusted to 9 to 10 points to enhance moisture interference suppression. When the ambient humidity is lower than 45% RH, the derivative window is adjusted to 11 to 13 points to optimize signal smoothness. For the humidity range between 45% RH and 55% RH, a standard derivative window of 11 points is used and matched with a medium humidity reference spectrum to ensure processing efficiency under normal conditions.

[0044] Step 3: Data Compression and Feature Fusion

[0045] The Hilbert yellow envelope spectroscopy method is used to compress preprocessed Raman and near-infrared spectral data and extract significant frequency feature variables. Specifically, the Hilbert yellow envelope spectroscopy data compression method is as follows: empirical mode decomposition is performed on the spectral signal to obtain intrinsic mode functions, then Hilbert transform is performed on the intrinsic mode functions to generate the envelope spectrum, significant frequency variables are screened by one-way ANOVA, and Raman and near-infrared features are fused using a dynamic weighted feature fusion algorithm to obtain a fused feature vector.

[0046] Specifically, the dynamic weighted feature fusion algorithm works as follows: establish the chemical bond correspondence between Raman features and near-infrared features, calculate the bond energy weights using density functional theory, and then achieve complementary fusion of the dual-spectral features based on the weights. The Raman features cover the vibrational information of C and C=C bonds, while the near-infrared features cover the information of CH and NH groups.

[0047] Step 4: Predicted Content Output

[0048] The fused feature vector is input into a pre-established quantitative relationship model for crude fat content, and the crude fat content value of soybean is output. The quantitative relationship model is established by using a regression analysis to calibrate the reference value measured by Soxhlet extraction and the historical fused feature vector.

[0049] Step 5: Development of the Graphical User Interface System

[0050] A graphical user interface system was developed based on the Python platform, integrating modules for spectral preprocessing, feature extraction, and model prediction to automate and visualize the measurement process.

[0051] A measurement system based on a dual-spectral fusion method for measuring crude fat content in soybeans includes a data acquisition module for collecting Raman and near-infrared spectral data of soybean samples, and an integrated Soxhlet extraction device for obtaining reference values ​​of crude fat content.

[0052] The spectral preprocessing module is used to process the acquired Raman and near-infrared spectral data;

[0053] The feature extraction and fusion module is used to compress spectral data based on the Hilbert yellow envelope spectroscopy algorithm and execute a dynamic weighted feature fusion algorithm to output a fused feature vector.

[0054] The model fusion module is used to calculate crude fat content based on the fused feature vector and a quantitative relationship model for crude fat content.

[0055] User Interface Module: Used to develop graphical interfaces based on PyQt5 and Tkinter libraries, providing functions for data uploading, preprocessing, feature extraction, feature fusion, and content measurement.

[0056] Example 2

[0057] according to Figures 2-6 As shown, this embodiment proposes a method for measuring the crude fat content of soybeans based on dual-spectral fusion, including the following steps: Step 1: Sample Preparation and Data Acquisition This embodiment selects 12 soybean varieties with genetic differences and clear origins, as shown in Table 1 below: specifically including Kennong 30, Kennong 14, Heihe 43, Henong 71, Henong 95, Henong 48, Kendou 43, Dongnong 42, Heike 60, Henong 531, Suinong 52 and Hefeng 45. All samples are from qualified seed suppliers.

[0060] Table 1. Statistical Table of Soybean Samples

[0061]

[0062]

[0063] Furthermore, this embodiment uses three types of equipment. The first type is a Raman spectrometer, such as the Haiguang Optoelectronics PR785-TEC3200 series Raman spectrometer, to acquire Raman spectral data. The configuration is as follows: exposure time of 10 seconds and scanning range of 200 cm. -1 ~3200cm -1 The laser radiation was 785nm and the power was 100mW. The whole grain sample was subjected to average spectral processing.

[0064] The second type of equipment is the TANGO NIR spectrometer from Bruker, which acquires near-infrared spectral data. It is configured with a scanning range of 4000 cm⁻¹ to 11520 cm⁻¹ and a resolution of 8 cm⁻¹. It performs multiple scans and averages the data. Specifically, 32 spectral scans constitute one scan cycle.

[0065] The third type of equipment is the Soxhlet extractor, which is used to obtain chemical reference values ​​for the crude fat content of soybeans.

[0066] Furthermore, Raman spectral data were obtained through non-destructive scanning of intact soybean samples. Specifically, an Oceanhood PR785-TEC3200 series high-sensitivity Raman spectrometer was used to perform non-destructive scanning of soybean samples in a constant temperature laboratory environment (22℃±3℃). One hundred and ten batches of morphologically intact soybeans (moisture content 11.4%–18.7%) without obvious damage were selected. Each soybean was placed below a fiber optic probe (core diameter 200μm, NA=0.22), and spectra were acquired using a contact vertical scanning mode. The integration time was set to 10 seconds per spectrum, and the average of three scans was taken to suppress random noise. The average spectrum of 25 soybeans was then calculated as the Raman spectrum of this batch for subsequent analysis. All 110 Raman spectra are shown below. Figure 2 As shown. The dataset was randomly divided into a training set (70 spectra) and a test set (40 spectra) at a 3:2 ratio to ensure that the fat content distribution of the two classes (training set and test set) covered the same range. Due to the fluorescence effect of biological samples excited by lasers and the instability of optical paths, circuits and mechanical systems, such as Figure 2 As shown, the original Raman spectrum exhibits high-intensity fluorescence background and noise, almost completely obscuring the effective signal. Therefore, effective removal of Raman spectral background and noise is crucial for further feature extraction and effective data analysis.

[0067] Near-infrared spectroscopy includes two methods: whole grain and powder samples, namely non-destructive acquisition and destructive acquisition. The non-destructive acquisition method is as follows: Each batch of samples consisted of 25 soybean seeds of the same variety (moisture content 11.4%–18.7%), placed in a quartz sample cup (46 mm diameter light-transmitting window, 20 mm optical path). The cup was wiped with anhydrous ethanol and purged with nitrogen to prevent fingerprint and dust contamination. A stainless steel light-shielding plate (45 mm × 5 mm aperture array) was placed over the sample surface to eliminate stray light interference, and a stainless steel cover was used to press the sample firmly, ensuring a tight arrangement of the seeds to reduce differences in light scattering. The scanning range was set to 4000 cm⁻¹. -1 ~11520cm -1 (Covering the fundamental and overtone regions of functional group vibrations such as CH, NH, and OH), with 32 consecutive scans per scan cycle (integration time 50 ms / scan, scan speed 10 kHz), and peak position shift > ±2 cm. -1 Using a threshold value, the system automatically removes abnormally fluctuating spectra and calculates the arithmetic mean of 32 scans to generate the near-infrared spectral curve for a single batch of samples. The near-infrared spectra of all 110 batches of intact grain samples are shown below. Figure 3As shown, the dataset is randomly divided into a training set (70 spectra) and a test set (40 spectra) in a 3:2 ratio to ensure that the fat content distribution of the two classes (training set and test set) covers the same range.

[0069] The destructive acquisition method is:

[0070] A high-speed rotary grinder (speed ≥12,000 rpm, blade material tungsten carbide) was used to process 25 seed samples per batch. After each batch, the samples were cleaned to avoid cross-contamination. After 2-3 grinding cycles, the powder was passed through a 60-mesh standard sieve (pore size 250 μm), retaining particles with a diameter of 2.5 ± 0.3 mm. The sieving efficiency was ≥80% (vibration frequency ≥50 Hz, sieving time 3 minutes). Three random samples of the sieved powder were then used to detect the particle size distribution using a laser particle size analyzer. When D90 ≤ 2.8 mm, the powder particles were considered to have uniformity (RSD < 5%).

[0071] Using a calibrated (0.001 g) balance, weigh 20.00 ± 0.02 g of sieved powder and evenly spread it in a quartz sample cup (46 mm diameter, 10 mm path length), taking care to avoid accumulation that could cause density differences. Cover with a stainless steel baffle (45 mm × 5 mm aperture array) and compact the powder to ensure a dense and uniform distribution, reducing differences in light scattering. Set the scanning range to 4000 cm⁻¹. -1 ~11520cm -1 A scan cycle consists of 32 consecutive scans (integration time 50ms / scan, scan speed 10kHz), with a peak position shift > ±2cm. -1 Using a threshold value, the system automatically removes abnormally fluctuating spectra and calculates the arithmetic mean of 32 scans to generate the near-infrared spectral curve for a single batch of samples. The near-infrared spectra of all 110 batches of powder samples are shown below. Figure 4 As shown, the dataset is randomly divided into a training set (70 spectra) and a test set (40 spectra) in a 3:2 ratio to ensure that the fat content distribution of the two classes (training set and test set) covers the same range.

[0072] Then, the Soxhlet extraction method was used to determine the reference values ​​for the crude fat content of soybeans, obtaining the standard reference values ​​for 110 batches of 12 varieties. The statistical results are shown in Table 2.

[0073] Table 2. Statistics on Fat Content of Soybean Samples

[0074]

[0075] Max:Maximum.Min:Minimum.SD:Standard deviation.CV:Coefficient ofvariation.

[0076] Step 2: Spectral Preprocessing

[0077] Hilbert vibrational decomposition (HVD) denoising was performed on the Raman spectral data. The spectral peaks were fitted by the Voigt function to remove fluorescence background and high-frequency noise. Then, multivariate scattering correction (MSC) and Savitzky-Golay first derivative processing were performed on the near-infrared spectral data, while moisture-sensitive bands were removed.

[0078] like Figure 5 As shown, the specific method of Hilbert vibration decomposition (HVD) denoising is as follows: the initial parameters of the spectral peak are obtained based on the first component of HVD, the spectral peak is reconstructed using the Voigt function within the moving window, and the denoising effect is optimized by maximizing the fitting function.

[0079] Step 3: Data Compression and Feature Fusion

[0080] The preprocessed Raman and near-infrared spectra were compressed using the Hilbert-Huang envelope spectroscopy method, and significant frequency characteristic variables were extracted. The compression ratio of the Raman spectrum reached 800:1, and the compression ratio of the near-infrared spectrum reached 283:1.

[0081] Specific examples Figure 6 As shown, the Hilbert envelope spectrum data compression method is as follows: the Raman spectrum is decomposed using the EMD (Empirical Mode Decomposition) method, transforming it into a finite number of intrinsic mode functions (IMFs). These IMF components contain local feature information of the original signal at different time scales, which helps to extract subtle differences in the local spectral regions. Subsequently, Hilbert envelope spectral transformation is performed on each IMF component, and point-by-point one-way ANOVA is conducted on the normalized envelope spectrum data. By establishing the correspondence between frequency feature variables and chemical bond vibration parameters, the compressed frequency features are converted into feature parameters reflecting the chemical structure, providing a physically meaningful input basis for subsequent feature fusion.

[0082] Based on the completed feature transformation, a dynamic weighted feature fusion algorithm is used to fuse Raman and near-infrared features to obtain a fused feature vector with clear physicochemical significance. Specifically, the dynamic weighted feature fusion algorithm works as follows: by establishing the chemical bond correspondence between the C and C=C bond vibration features of the Raman spectrum and the CH and NH group features of the near-infrared spectrum, the bond energy weights are calculated based on density functional theory to form a feature weight mapping table. Then, based on the bond energy weight coefficients in the feature weight mapping table, the significant frequency features of the preprocessed Raman spectrum and the near-infrared spectral features are weighted and calculated. Subsequently, the feature reconstruction stage is entered, where the weighted dual-spectral feature vectors are spliced ​​and normalized to obtain the fused feature vector.

[0083] Step 4: Predicted Content Output

[0084] The fused feature vector is input into a pre-established quantitative relationship model for crude fat content, and the crude fat content value of soybean is output. The quantitative relationship model is established by using a regression analysis to calibrate the reference value measured by Soxhlet extraction and the historical fused feature vector.

[0085] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the present invention without departing from its framework and scope of application, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for measuring crude fat content in soybeans based on dual-spectral fusion, characterized in that: Includes the following steps: Step 1: Sample Preparation and Data Acquisition Soybean samples were prepared, and Raman and near-infrared spectral data of the soybean samples were obtained under monitored ambient temperature and humidity. The reference value of crude fat content of soybeans was then determined by Soxhlet extraction. Step 2: Spectral Preprocessing Hilbert vibrational decomposition denoising was performed on the Raman spectral data, and the spectral peaks were fitted by the Voigt function to remove the fluorescence background and high-frequency noise. Then, multivariate scattering correction and Savitzky-Golay first derivative processing were performed on the near-infrared spectral data. At the same time, a dynamic parameter adjustment mechanism was introduced in both processing methods. Step 3: Data Compression and Feature Fusion Based on the Hilbert yellow envelope spectroscopy method, the preprocessed Raman and near-infrared spectral data are compressed, significant frequency feature variables are extracted, and a dynamic weighted feature fusion algorithm is used to fuse Raman and near-infrared features to obtain a fused feature vector. Step 4: Predicted Content Output The fused feature vector is input into a pre-established quantitative relationship model for crude fat content, and the crude fat content value of soybeans is output.

2. The method for measuring crude fat content in soybeans based on dual-spectral fusion according to claim 1, characterized in that: It also includes the following steps: Step 5: Development of the graphical user interface system: Develop a graphical user interface system based on the Python platform, integrating spectral preprocessing, feature extraction, and model prediction modules to automate and visualize the measurement process.

3. The method for measuring crude fat content in soybeans based on dual-spectrum fusion according to claim 1, characterized in that: In step one, the Raman spectral data was obtained by scanning the whole grain sample, and the near-infrared spectral data was obtained by scanning the whole grain sample and the powder sample.

4. The method for measuring crude fat content in soybeans based on dual-spectral fusion according to claim 1, characterized in that: In step two, the specific method of Hilbert vibration decomposition denoising is as follows: the initial parameters of the spectral peak are obtained based on the first component of HVD, the spectral peak is reconstructed using the Voigt function within the moving window, and the denoising effect is optimized by maximizing the fitting degree function.

5. The method for measuring crude fat content in soybeans based on dual-spectral fusion according to claim 1, characterized in that: In step two, the dynamic parameter adjustment mechanism includes dynamic adjustment of ambient temperature and dynamic adjustment of ambient humidity. The dynamic adjustment of ambient temperature is to dynamically adjust the size of the moving window of the Voigt function fitting spectrum peak according to the ambient temperature. The dynamic adjustment of ambient humidity is to dynamically adjust the window parameters of the reference spectrum selection and derivative processing of the multivariate scattering correction according to the ambient humidity.

6. The method for measuring crude fat content in soybeans based on dual-spectral fusion according to claim 1, characterized in that: In step three, the specific method of the dynamic weighted feature fusion algorithm is as follows: establish the chemical bond correspondence between Raman features and near-infrared features, calculate the bond energy weights through density functional theory, and realize the complementary fusion of dual-spectral features based on the weights. Among them, Raman features cover the vibrational information of C=C and C=C bonds, and near-infrared features cover the information of CH and NH groups.

7. The method for measuring crude fat content in soybeans based on dual-spectral fusion according to claim 1, characterized in that: In step four, the quantitative relationship model is established by performing regression analysis on the reference values ​​measured by Soxhlet extraction and the historical fusion feature vector.

8. A measurement system applied to the soybean crude fat content measurement method based on dual-spectral fusion as described in claim 1, characterized in that: It includes a data acquisition module for collecting Raman and near-infrared spectral data of soybean samples, and integrates a Soxhlet extraction device to obtain reference values ​​for crude fat content; The spectral preprocessing module is used to process the acquired Raman and near-infrared spectral data. The feature extraction and fusion module is used to compress spectral data based on the Hilbert yellow envelope spectroscopy algorithm and execute a dynamic weighted feature fusion algorithm to output a fused feature vector. The model fusion module is used to calculate crude fat content based on the fused feature vector and a quantitative relationship model for crude fat content. User Interface Module: Used to develop graphical interfaces based on PyQt5 and Tkinter libraries, providing functions for data uploading, preprocessing, feature extraction, feature fusion, and content measurement.

Images