Online Prediction Method and System for Oil Content of Camellia oleifera Fruit Based on Near-Infrared Spectroscopic Characteristics

Abstract

This invention discloses an online prediction method and system for the oil content of camellia oleifera fruit based on near-infrared spectral characteristics. Specifically, it relates to the field of near-infrared spectroscopy non-destructive testing technology, and is used to solve the problem that the anisotropic waveguide of the fruit shell causes non-uniform dilution of the oil spectral signal of the kernel and unstable prediction accuracy in the online detection of intact camellia oleifera fruit. The method involves decomposing the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes it. The intensity difference and depth difference between the fruit shell scattering band and the oil absorption band are obtained between the two components. The ratios of the intensity difference ratio and the depth difference ratio are calculated. When the ratios meet the mutual exclusion condition, the average ratio is used as the waveguide contribution rate to determine the dilution amount. The dilution amount is subtracted from the original absorption depth to obtain the corrected absorption depth. The corrected absorption depth is then input into the oil content prediction model to achieve online prediction of the oil content of a single camellia oleifera fruit.

Description

Technical Field

[0001] This invention relates to the field of near-infrared spectroscopy nondestructive testing technology, and more specifically, to a method and system for online prediction of oil content in camellia oleifera fruit based on near-infrared spectral characteristics. Background Technology

[0002] The oil content of camellia oleifera fruit is a core indicator for assessing its economic value and processing suitability. Before large-scale processing, camellia oleifera fruit needs to be sorted online according to its oil content to match different pressing processes and achieve optimal use of the raw material. Since traditional destructive detection methods such as Soxhlet extraction cannot meet the timeliness requirements of online detection, non-destructive analysis based on near-infrared spectroscopy is a major technical approach for online prediction of camellia oleifera fruit oil content. Typically, camellia oleifera fruit is arranged in a single layer on a conveyor belt. As the fruit rapidly passes through the detection station, a near-infrared diffuse reflectance spectroscopy probe is used to collect one or more instantaneous spectra of each fruit. The collected spectral data is then input into a pre-established quantitative prediction model, which outputs the predicted oil content in real time. Utilizing the sensitive response of near-infrared light to the vibrations of hydrogen-containing groups, this method has shown feasibility in the online assessment of camellia oleifera fruit oil content.

[0003] In existing technologies, when using diffuse reflectance spectra from a single acquisition for online prediction of the oil content of intact camellia oleifera fruits, it implicitly assumes that the acquired near-infrared spectra can stably reflect the average oil information of the kernels inside the fruit. However, the shell of the camellia oleifera fruit is not an optically isotropic medium; its dense lignified thick-walled cell layer and vascular bundle network exhibit significant orientation in their structural arrangement. When incident light illuminates the fruit at a fixed angle and diffuse reflection is collected at another fixed angle, the anisotropic structure inside the fruit shell causes selective differences in the light transmission channel. Some photons may form waveguide-like transmission in low-scattering structures such as vascular bundles, bypassing the kernel and directly reaching the detector. This results in the effective absorption signal from the kernel oil being diluted to varying degrees in the spectra collected from different orientations. The overall spectral characteristics not only contain chemical composition information but also structural deviations caused by the geometric orientation inside the fruit shell. These structural deviations are coupled with the spectral response of oil content changes. Conventional spectral preprocessing methods cannot effectively remove these deviations, resulting in the oil content predicted based on a single or limited number of spectral acquisitions not being able to stably represent the overall actual situation of the tested fruit, thus limiting the accuracy of online prediction of oil content in whole camellia oleifera fruits. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides an online prediction method and system for the oil content of Camellia oleifera fruit based on near-infrared spectral characteristics to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The online prediction method for oil content of camellia fruit based on near-infrared spectral characteristics includes the following steps:

[0007] S1: Collect the original near-infrared diffuse reflectance spectrum of a single camellia fruit under online transport conditions;

[0008] S2: Extract the first band response intensity, which is sensitive to the scattering of the lignified structure of the fruit shell, and the second band absorption depth, which characterizes the absorption of oil in the kernel, from the original near-infrared diffuse reflectance spectrum.

[0009] S3: Decompose the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes, obtain the intensity difference between the first band response intensity and the second spectral component, and the depth difference between the second band absorption depth and the first spectral component.

[0010] S4: Calculate the first ratio of the intensity difference to the first band response intensity in the first spectral component, and the second ratio of the depth difference to the second band absorption depth in the second spectral component. When the first ratio and the second ratio satisfy the mutual exclusion condition, the average of the absolute values ​​of the first ratio and the second ratio is used as the waveguide contribution rate. The product of the waveguide contribution rate and the second band absorption depth is determined as the dilution amount.

[0011] S5: Subtract the dilution amount from the second band absorption depth to obtain the corrected absorption depth. Input the corrected absorption depth into the oil content prediction model to obtain the online prediction result of the oil content of a single camellia fruit.

[0012] Furthermore, the raw near-infrared diffuse reflectance spectra of individual camellia fruits under online transport conditions were collected, including:

[0013] A single camellia fruit is irradiated with a near-infrared light source positioned at a first angle to the plane of the conveyor belt. A diffuse reflection fiber optic probe positioned at a second angle to the plane of the conveyor belt receives the near-infrared light emitted from the surface of the single camellia fruit after being transmitted through its interior. The diffuse reflection fiber optic probe transmits the received near-infrared light to a near-infrared spectrometer, which converts the received light signal into the original near-infrared diffuse reflection spectrum.

[0014] Furthermore, the intensity of the first band response sensitive to scattering from the lignified structure of the pericarp and the second band absorption depth characterizing the absorption of kernel oil were extracted from the original near-infrared diffuse reflectance spectrum, including:

[0015] Baseline correction was performed on the original near-infrared diffuse reflectance spectrum. Within the first band range corresponding to the scattering of the lignified structure of the fruit shell, the peak height of the spectrum after baseline correction was taken as the response intensity of the first band. Within the second band range corresponding to the absorption of oil in the kernel, the depression of the local absorption valley of the spectrum after baseline correction relative to the line connecting the two adjacent shoulders was taken as the absorption depth of the second band.

[0016] Further, the original near-infrared diffuse reflectance spectrum is decomposed into a first spectral component that maintains the polarization state and a second spectral component that depolarizes. The intensity difference between the first band response intensity and the second spectral component, and the depth difference between the second band absorption depth and the first spectral component are obtained, including:

[0017] The near-infrared light corresponding to the original near-infrared diffuse reflectance spectrum is passed through a rotatable linear polarizer. The spectrum collected when the transmission axis of the linear polarizer is aligned with a first position parallel to the incident plane is taken as the first spectral component, and the spectrum collected when the transmission axis of the linear polarizer is aligned with a second position perpendicular to the incident plane is taken as the second spectral component. The intensity difference is obtained by subtracting the response intensity of the first band in the second spectral component from the response intensity of the first band in the first spectral component. The depth difference is obtained by subtracting the absorption depth of the second band in the first spectral component from the absorption depth of the second band in the second spectral component.

[0018] Furthermore, when aligning the transmission axis of the linear polarizer with the first position parallel to the incident plane, before acquiring the first spectral component, the method further includes: aligning the transmission axis of the linear polarizer with the first position, acquiring the blank polarization spectrum of the conveyor belt surface before placing a single camellia fruit, using the polarization retention rate in the first position of the blank polarization spectrum as the polarization reference, and normalizing the first and second spectral components respectively using the polarization reference when subsequently acquiring the first and second spectral components.

[0019] Furthermore, the first ratio is calculated by dividing the intensity difference by the response intensity of the first band in the first spectral component, and the second ratio is calculated by dividing the depth difference by the absorption depth of the second band in the second spectral component. The mutual exclusion condition is determined as follows: when the first ratio is greater than zero and the second ratio is greater than zero, and the absolute value of the difference between the first ratio and the second ratio is less than a preset judgment threshold, the first ratio and the second ratio are determined to satisfy the mutual exclusion condition. The waveguide contribution rate is determined by taking the average of the absolute values ​​of the first ratio and the second ratio. The dilution amount is determined by multiplying the waveguide contribution rate by the absorption depth of the second band.

[0020] Furthermore, the preset judgment threshold is determined as follows: a set of isotropic reference camellia fruits that are known to not have anisotropic waveguide characteristics of the fruit shell are selected in advance. S1 to S3 are performed on the isotropic reference camellia fruits respectively to obtain the first ratio and the second ratio corresponding to the isotropic reference camellia fruits. The distribution range of the absolute value of the difference between the first ratio and the second ratio corresponding to the isotropic reference camellia fruits is statistically analyzed, and the upper limit of the distribution range is used as the preset judgment threshold.

[0021] Furthermore, the corrected absorption depth is obtained by subtracting the dilution amount from the second band absorption depth; the oil content prediction model is a quantitative prediction model between the corrected absorption depth and the oil content established in advance based on partial least squares method.

[0022] Furthermore, when inputting the corrected absorption depth into the oil content prediction model, the absorption peak morphology parameters of the second band in the original near-infrared diffuse reflectance spectrum after baseline correction are also input simultaneously. The oil content prediction model is a quantitative prediction model pre-established based on partial least squares method, with the corrected absorption depth and absorption peak morphology parameters as independent variables and the oil content as the dependent variable.

[0023] On the other hand, the present invention provides an online prediction system for the oil content of camellia fruit based on near-infrared spectral characteristics, comprising the following modules:

[0024] The spectral acquisition module is used to acquire the original near-infrared diffuse reflectance spectrum of a single camellia fruit under online transportation conditions;

[0025] The feature extraction module is used to extract the first band response intensity, which is sensitive to the scattering of the lignified structure of the fruit shell, and the second band absorption depth, which characterizes the absorption of oil in the kernel, from the original near-infrared diffuse reflectance spectrum.

[0026] The decomposition and quantization module is used to decompose the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes, to obtain the intensity difference between the first band response intensity and the second spectral component, and the depth difference between the second band absorption depth and the first spectral component.

[0027] The contribution extraction module is used to calculate the first ratio of the intensity difference to the response intensity of the first band in the first spectral component, and the second ratio of the depth difference to the absorption depth of the second band in the second spectral component. When the first ratio and the second ratio satisfy the mutual exclusion condition, the average of the absolute values ​​of the first ratio and the second ratio is used as the waveguide contribution rate, and the product of the waveguide contribution rate and the absorption depth of the second band is determined as the dilution amount.

[0028] The oil content prediction module is used to subtract the dilution amount from the second band absorption depth to obtain the corrected absorption depth. The corrected absorption depth is then input into the oil content prediction model to obtain the online prediction result of the oil content of a single camellia fruit.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. By polarization decomposition, the original near-infrared diffuse reflectance spectrum is separated into a first spectral component that maintains the polarization state and a second spectral component that depolarizes. Utilizing the physical difference between the polarization-maintaining effect of the anisotropic vascular bundle structure of the shell on waveguide photons and the depolarization effect of the kernel tissue on diffuse reflectance photons, the waveguide interference signal and the normal diffuse reflectance signal exhibit different distribution characteristics in the two polarization components. The first ratio and the second ratio are constructed based on the polarization difference of the response intensity in the first band and the polarization difference of the absorption depth in the second band, respectively. Utilizing the mutual exclusion characteristic of waveguide interference producing a positive excess in the structural scattering signal and a reverse deficiency in the oil absorption signal, the existence of interference is determined, and the waveguide contribution rate is quantified by the average of the two ratios. Thus, the spurious attenuation introduced by the anisotropic waveguide of the causal shell is removed from the global second band absorption depth and directly acts on the characteristic parameters of the near-infrared spectrum, without relying on empirical correction factors or batch calibration, so that the spectral absorption depth can truly reflect the oil content of the kernel.

[0031] 2. Under the condition of rapid online detection on the conveyor belt, the degree of waveguide interference of a single camellia fruit can be assessed and compensated by only polarization analysis and ratio analysis of a single acquired spectrum. The corrected second-band absorption depth is then input into the oil content prediction model. Compared with the traditional method of directly using the original absorption depth for prediction, the corrected absorption depth effectively eliminates the instability of spectral characteristics caused by random changes in fruit posture and differences in the internal structure of the fruit shell. This allows the online oil content prediction result to stably represent the actual oil content level of a single camellia fruit kernel, improving the single-fruit accuracy and process adaptability of near-infrared online sorting of whole camellia fruits. Attached Figure Description

[0032] Figure 1 This is a flowchart of the online prediction method for oil content of camellia fruit based on near-infrared spectral characteristics according to the present invention;

[0033] Figure 2 This is a schematic diagram of the online prediction system for oil content of camellia fruit based on near-infrared spectral characteristics according to the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0035] Example 1: Figure 1 The present invention provides an online prediction method for the oil content of camellia fruit based on near-infrared spectral characteristics, which includes the following steps:

[0036] S1: Collect the original near-infrared diffuse reflectance spectrum of a single camellia fruit under online transport conditions;

[0037] S2: Extract the first band response intensity, which is sensitive to the scattering of the lignified structure of the fruit shell, and the second band absorption depth, which characterizes the absorption of oil in the kernel, from the original near-infrared diffuse reflectance spectrum.

[0038] S3: Decompose the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes, obtain the intensity difference between the first band response intensity and the second spectral component, and the depth difference between the second band absorption depth and the first spectral component.

[0039] S4: Calculate the first ratio of the intensity difference to the first band response intensity in the first spectral component, and the second ratio of the depth difference to the second band absorption depth in the second spectral component. When the first ratio and the second ratio satisfy the mutual exclusion condition, the average of the absolute values ​​of the first ratio and the second ratio is used as the waveguide contribution rate. The product of the waveguide contribution rate and the second band absorption depth is determined as the dilution amount.

[0040] S5: Subtract the dilution amount from the second band absorption depth to obtain the corrected absorption depth. Input the corrected absorption depth into the oil content prediction model to obtain the online prediction result of the oil content of a single camellia fruit.

[0041] An example of collecting the original near-infrared diffuse reflectance spectrum of a single camellia fruit under online transport conditions is as follows.

[0042] In this embodiment, the camellia oleifera fruits are arranged in a single layer on a conveyor belt moving at a constant speed. The conveyor belt transports the camellia oleifera fruits through the inspection station at a speed of, for example, 0.3 meters per second to 1.5 meters per second. Near-infrared light sources and diffuse reflection fiber optic probes are respectively installed above and to the side of the inspection station.

[0043] The near-infrared light source is configured at a first angle ranging from 30 to 60 degrees with the plane of the conveyor belt, for example, 45 degrees. The diameter of the emitted light spot of the near-infrared light source covers the cross-sectional size of a single camellia fruit, and the emitted wavelength range covers near-infrared light in the 1100 nm to 2500 nm band. The near-infrared light source illuminates the single camellia fruit passing through the detection station. The reason for choosing this first angle range is that when the incident light illuminates the camellia fruit in a direction deviating from the normal, the proportion of incident light that enters the fruit interior after refraction on the fruit shell surface is increased, reducing the probability of surface specular reflection directly entering the diffuse reflection fiber optic probe, thereby facilitating the acquisition of diffuse reflection light signals carrying information about the internal structure of the fruit.

[0044] The diffuse reflection fiber optic probe is configured at a second angle to the conveyor belt plane, for example, between 60 and 90 degrees, such as 75 degrees. The normal direction of the receiving end face of the diffuse reflection fiber optic probe points towards the light emission area of ​​the surface of the irradiated camellia fruit. The second angle is different from the first angle, and the sum of the second angle and the first angle is not 180 degrees, to avoid the diffuse reflection fiber optic probe directly receiving specular reflection light generated by the near-infrared light source on the surface of the camellia fruit. The diffuse reflection fiber optic probe receives the near-infrared light emitted from the surface after being transmitted through the interior of a single camellia fruit. Near-infrared light emitted from the surface of a single camellia fruit after transmission within the fruit refers to diffuse reflection light emitted from a near-infrared light source. This light is emitted from the fruit surface after a portion of photons penetrate the fruit shell surface and enter the fruit interior. After undergoing multiple scattering or waveguide transmissions in the complex medium composed of the thick-walled lignified cell layer, vascular bundle network, and kernel tissue of the fruit shell, it is emitted from the fruit surface at a position different from the direction of specular reflection. The spectrum of this diffuse reflection light carries information about the absorption of near-infrared light of specific wavelengths by the kernel oil and the scattering information of near-infrared light by the lignified structure of the fruit shell.

[0045] The diffuse reflection fiber optic probe transmits the received near-infrared light to the near-infrared spectrometer via a multimode silica fiber. The core diameter of the multimode silica fiber is, for example, 200 micrometers or 400 micrometers, to ensure that the attenuation of the near-infrared light signal during transmission is within an acceptable range. The near-infrared spectrometer converts the received light signal into the original near-infrared diffuse reflectance spectrum. The near-infrared spectrometer is equipped with a grating beam splitter and an indium gallium arsenide (IGaAs) detector array. The grating beam splitter separates the near-infrared light entering the spectrometer according to wavelength dispersion. Each pixel of the IGaAs detector array simultaneously receives light intensity signals at different wavelengths and accumulates charge within a set integration time, such as 5 to 50 milliseconds. The integration time is set based on the conveyor belt speed and the time window of the light spot irradiation of a single camellia fruit through the detection station. Specifically, the integration time is not greater than the time it takes for a single camellia fruit to go from entering the light spot irradiation area to leaving the light spot irradiation area. This time is obtained by dividing the length of the light spot in the direction of the conveyor belt movement by the conveyor belt speed. For example, when the light spot length is 20 millimeter and the conveyor belt speed is 1 meter per second, the time is 20 milliseconds, and the integration time can be set to 15 milliseconds to ensure that each camellia fruit completes at least one complete spectral acquisition during its stay in the light spot irradiation area. After analog-to-digital conversion, the electrical signal output from the indium gallium arsenide (IGaAs) detector array is plotted on the x-axis as the detector pixel number corresponding to the wavelength and on the y-axis as the relative light intensity value calculated from the accumulated charge of each pixel, forming a continuous raw near-infrared diffuse reflectance spectrum curve. Each data point on the raw near-infrared diffuse reflectance spectrum curve corresponds to the relative intensity of diffuse reflected light at a wavelength position. This relative intensity is expressed as a dimensionless count or voltage ratio, and the magnitude of the relative intensity is proportional to the number of photons emitted from the surface of the camellia fruit at that wavelength and collected by the diffuse reflectance fiber optic probe.

[0046] To eliminate interference from spectrometer dark current noise and ambient stray light on the raw near-infrared diffuse reflectance spectrum, dark spectra and reference spectra were acquired before or after each acquisition of the raw near-infrared diffuse reflectance spectrum of camellia oleifera fruit. The dark spectrum was acquired by using the same integration time as the raw near-infrared diffuse reflectance spectrum acquisition, with the near-infrared light source off, to obtain the dark current response values ​​of each detector pixel of the near-infrared spectrometer. The dark spectrum was used to subsequently subtract the detector's own background signal from the raw near-infrared diffuse reflectance spectrum. The reference spectrum is acquired by placing a standard diffuse reflectance reference white plate at the testing station. The standard diffuse reflectance reference white plate is made of a material with a diffuse reflectance close to a constant in the 1100 nm to 2500 nm wavelength range, such as a polytetrafluoroethylene sintered white plate. The near-infrared spectrum of the standard diffuse reflectance reference white plate is obtained with the same geometric configuration conditions between the near-infrared light source and the diffuse reflectance fiber optic probe, the same near-infrared light source intensity, and the same integration time as the original near-infrared diffuse reflectance spectrum. The reference spectrum is used for subsequent calculation of the diffuse reflectance of the camellia fruit to eliminate the influence of near-infrared light source energy fluctuations and spectrometer response function on the original near-infrared diffuse reflectance spectrum.

[0047] An example of extracting the first band response intensity sensitive to scattering of the lignified structure of the fruit shell and the second band absorption depth characterizing the absorption of oil in the kernel from the original near-infrared diffuse reflectance spectrum is as follows.

[0048] Baseline correction is performed on the original near-infrared diffuse reflectance spectrum. The purpose of baseline correction is to eliminate spectral baseline drift caused by differences in scattering from the fruit shell surface, fruit size variations, and minor fluctuations in detection distance, ensuring that the corrected spectral curve primarily reflects the combined effects of absorption and scattering by the internal medium of the fruit. Baseline correction is achieved using an iterative polynomial fitting method. Specifically, the difference between each point in the original near-infrared diffuse reflectance spectrum data and the initial estimated baseline is compared for convergence. In each iteration, data points with values ​​higher than the current estimated baseline are identified, and their spectral values ​​are replaced with the corresponding values ​​of the current estimated baseline. This process gradually reduces and approximates the actual spectral base profile. The iteration terminates when the maximum deviation between the estimated baselines obtained from two consecutive iterations is less than a preset convergence tolerance. The preset convergence tolerance is set as follows: the standard deviation of the ordinate values ​​of all data points in the original near-infrared diffuse reflectance spectrum is calculated, and the preset convergence tolerance is set to one-thousandth of the standard deviation. The estimated baseline obtained at the end of the iteration is subtracted point by point from the original near-infrared diffuse reflectance spectrum to obtain the baseline-corrected spectrum.

[0049] After baseline correction, the peak height of the spectrum within the first band corresponding to the scattering from the lignified structure of the fruit shell is taken as the first band response intensity. The selection of the first band corresponding to the scattering from the lignified structure of the fruit shell is based on the following criteria: the thick-walled lignified cell layer of the camellia fruit shell exhibits significant scattering of near-infrared light, with the scattering intensity decreasing with increasing wavelength. The overtone absorption of CH stretching vibrations near 1670 nm by lignin and the absorption of OH deformation vibrations near 1490 nm by cellulose in the fruit shell modulate the scattering characteristics. Therefore, a band range far from the strong absorption of lignin and cellulose, such as the 1100 nm to 1300 nm band, is selected as the first band. Within this first band, the fluctuations in the spectral signal are mainly dominated by changes in the scattering intensity of the lignified structure of the fruit shell, with less interference from the absorption of chemical components in the fruit shell. Within this first band, the baseline-corrected spectrum exhibits one or more scattering peaks; the peak height corresponding to the highest peak intensity is taken as the first band response intensity. The peak height is obtained as follows: Within the first band, all local maxima of the baseline-corrected spectral curve are identified. A local maximum is determined when the ordinate of the data point is greater than the ordinates of several adjacent data points (e.g., five data points to the left and right). The local maximum is selected as the scattering peak. The difference between the ordinate of the scattering peak and the baseline at that wavelength is the first band response intensity. The first band response intensity is a dimensionless value, reflecting the relative luminous flux collected by the diffuse reflection fiber optic probe after multiple scattering of near-infrared light in the lignified thick-walled cell layer of the fruit shell.

[0050] Within the second wavelength range corresponding to the absorption of oil in camellia seeds, the depth of absorption in the second wavelength range is determined by the degree of depression of the local absorption valley relative to the line connecting the adjacent shoulders of the baseline-corrected spectrum. The selection method for the second wavelength range corresponding to the absorption of oil in camellia seeds is as follows: The main oil component in camellia seeds is unsaturated fatty acid triglycerides. Their absorption in the near-infrared band is generated by the overtones and combination frequencies of the stretching vibrations of carbon-hydrogen bonds in linoleic and oleic acids. Absorption combination frequencies exist in the 1700 nm to 1800 nm wavelength range and near 2300 nm. One wavelength range with significant absorption characteristics is selected as the second wavelength range, for example, the 1700 nm to 1760 nm wavelength range. Within this second wavelength range, the absorption of oil in the seeds is relatively concentrated, and there is sufficient spacing between this range and the absorption peak of moisture in the pericarp near 1940 nm to reduce interference. Within the second wavelength range, the baseline-corrected spectrum exhibits local absorption valleys, which are shaped as a concave spectral curve segment located between two locally higher spectral points. The absorption depth of the second band is obtained as follows: Within the second band, the point with the smallest ordinate value on the baseline-corrected spectral curve is identified as the absorption valley point. To the left and right of the absorption valley point, the local maxima closest to it are identified as the left shoulder point and right shoulder point, respectively. The criteria for determining the local maxima points are the same as those described in the first band. A straight line segment connecting the left shoulder point and the right shoulder point is formed as the local baseline. The difference between the ordinate value of the absorption valley point and the interpolated ordinate value of the local baseline at the corresponding wavelength of the absorption valley point is calculated. This difference is the absorption depth of the second band. The absorption depth of the second band is a dimensionless value. The magnitude of the absorption depth reflects the intensity of absorption by the kernel oil during the transmission of incident near-infrared light within the camellia fruit. Under ideal conditions without waveguide interference, the absorption depth of the second band has a stable positive correlation with the kernel oil content.

[0051] When obtaining the response intensity of the first band and the absorption depth of the second band, the baseline-corrected spectrum is used as the processing object. The baseline-corrected spectrum is obtained by baseline correction of the previously acquired raw near-infrared diffuse reflectance spectrum. The raw near-infrared diffuse reflectance spectrum is obtained by converting the near-infrared light received by the diffuse reflectance fiber optic probe by the near-infrared spectrometer.

[0052] An example of decomposing the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes, and obtaining the intensity difference between the first and second spectral components for the first band response intensity, and the depth difference between the first and second spectral components for the second band absorption depth, is as follows.

[0053] A rotatable linear polarizer is inserted into the optical path between the near-infrared spectrometer and the diffuse reflection fiber optic probe. The linear polarizer is made of a metal wire grid polarizer with a high extinction ratio in the 1100 nm to 2500 nm wavelength range and is mounted on an electrically driven rotating frame with precisely controllable rotation angle. The near-infrared light corresponding to the original near-infrared diffuse reflection spectrum passes through the rotatable linear polarizer before entering the near-infrared spectrometer. The near-infrared light corresponding to the original near-infrared diffuse reflection spectrum refers to the near-infrared light carrying information about the internal structure of a single camellia fruit, received by the diffuse reflection fiber optic probe and transmitted to the near-infrared spectrometer via multimode quartz fiber. This near-infrared light is the same light signal that enters the near-infrared spectrometer when the original near-infrared diffuse reflection spectrum is acquired in S1.

[0054] The spectrum collected when the transmission axis of the linear polarizer is aligned with the incident plane in a first orientation is taken as the first spectral component. The incident plane is a plane jointly determined by the direction of the incident light from the near-infrared light source onto a single camellia fruit and the direction of the received light from the diffuse reflection fiber optic probe onto the single camellia fruit. The first orientation is when the transmission axis of the linear polarizer is parallel to the incident plane. In the first orientation, the component of near-infrared light with a polarization direction parallel to the incident plane is transmitted through the linear polarizer to the maximum extent and reaches the near-infrared spectrometer, while the component of near-infrared light with a polarization direction perpendicular to the incident plane is blocked by the linear polarizer. The spectrum collected when the transmission axis of the linear polarizer is aligned with the incident plane in a second orientation is taken as the second spectral component. The second orientation is when the transmission axis of the linear polarizer is perpendicular to the incident plane. In the second orientation, the component of near-infrared light with a polarization direction perpendicular to the incident plane is transmitted through the linear polarizer to the maximum extent and reaches the near-infrared spectrometer, while the component of near-infrared light with a polarization direction parallel to the incident plane is blocked by the linear polarizer. The first spectral component mainly collects photons that have undergone fewer scattering events inside the camellia fruit and maintained their original polarization state. These photons include those that maintain their polarization state through waveguide transmission via the anisotropic vascular bundle structure of the fruit shell. The second spectral component mainly collects photons that have undergone sufficient multiple scattering and depolarized inside the camellia fruit. These photons have undergone a more uniform scattering process, and the spectral characteristics of the second spectral component are more representative of the average optical properties of the kernel and shell tissue inside the fruit.

[0055] The intensity difference is obtained by subtracting the first band response intensity of the second spectral component from the first band response intensity of the first spectral component. The method for obtaining the first band response intensity of the first spectral component is the same as that used in S2 to extract the first band response intensity from the baseline-corrected spectrum; that is, baseline correction is performed on the first spectral component, and the peak height of the baseline-corrected spectrum within the first band is taken as the first band response intensity of the first spectral component. Similarly, the method for obtaining the first band response intensity of the second spectral component is the same: baseline correction is performed on the second spectral component, and the peak height of the baseline-corrected spectrum within the first band is taken as the first band response intensity of the second spectral component. The first band range is the same as the first band range corresponding to the scattering from the lignified structure of the fruit shell in S2, for example, the 1100 nm to 1300 nm band range. A positive intensity difference indicates that the scattering signal from the lignified structure of the fruit shell in the first spectral component is stronger than that in the second spectral component, and the magnitude of the intensity difference reflects the degree of polarization selectivity enhancement of the scattered signal by waveguide transmission.

[0056] The depth difference is obtained by subtracting the absorption depth of the second band in the first spectral component from the absorption depth of the second band in the second spectral component. The method for obtaining the absorption depth of the second band in the first spectral component is the same as that used in S2 to extract the absorption depth from the baseline-corrected spectrum; that is, baseline correction is performed on the first spectral component, and the depression of the local absorption valley of the baseline-corrected spectrum relative to the line connecting the two adjacent shoulders within the second band is taken as the absorption depth of the second band in the first spectral component. Similarly, the method for obtaining the absorption depth of the second band in the second spectral component is the same as that used in S2, for example, the 1700 nm to 1760 nm band. A positive depth difference indicates that the absorption signal of the kernel oil in the second spectral component is stronger than that in the first spectral component, and the magnitude of the depth difference reflects the degree of dilution of the oil absorption signal by waveguide transmission.

[0057] Before acquiring the first and second spectral components, a step of normalizing the first and second spectral components using a polarization reference is included. The purpose of normalization is to eliminate the differences in transmittance of the linear polarizer to light of different polarization states and the differences in the response of the near-infrared spectrometer to light of different polarization states. The polarization reference is obtained as follows: With the transmission axis of the linear polarizer aligned to the first orientation, a blank polarization spectrum of the conveyor belt surface is acquired before placing a single camellia fruit. The blank polarization spectrum is the near-infrared spectrum of the conveyor belt surface with the linear polarizer in the first orientation, acquired under the same near-infrared light source illumination, the same diffuse reflection fiber optic probe receiving position, and the same near-infrared spectrometer acquisition parameters as in S1, without any camellia fruit on the conveyor belt surface. The polarization retention rate in the first orientation of the blank polarization spectrum is used as the polarization reference. The polarization retention rate is a parameter reflecting the inherent proportional relationship between the polarized and depolarized components in the light scattered from the conveyor belt surface under conditions without camellia fruit. The polarization retention rate is calculated as follows: after baseline correction of the blank polarized spectrum, the ratio of the integrated intensity of the blank polarized spectrum in the first band to the integrated intensity in the second band is taken. This ratio physically represents the difference in polarization response of the light scattered from the conveyor belt surface in the two characteristic bands. When subsequently acquiring the first and second spectral components, they are normalized using a polarization reference. The normalization process involves dividing the intensity value of each wavelength of the first spectral component by the corresponding intensity value in the polarization reference, and multiplying the intensity value of each wavelength of the second spectral component by the corresponding intensity value in the polarization reference, thus equalizing the polarization responses of the first and second spectral components under the conditions of light scattered from the conveyor belt surface to the same reference level.

[0058] An example of calculating the intensity difference as a first ratio of the response intensity of the first band in the first spectral component, calculating the depth difference as a second ratio of the absorption depth of the second band in the second spectral component, and determining the waveguide contribution rate and dilution amount when the first ratio and the second ratio satisfy a mutually exclusive condition is as follows.

[0059] The first ratio is calculated by dividing the intensity difference by the response intensity of the first band in the first spectral component. The intensity difference, obtained from S3, is the difference between the response intensity of the first band in the first spectral component and the response intensity of the first band in the second spectral component. The response intensity of the first band in the first spectral component is the peak height extracted within the first band range after baseline correction of the first spectral component in S3. The first band range is the same as the first band range corresponding to the scattering from the lignified structure of the fruit shell in S2, for example, the 1100 nm to 1300 nm band range. The formula for calculating the first ratio is: R1 = ΔS1 / S1 para Where R1 represents the first ratio, ΔS1 represents the strength difference, and S1 paraThis represents the response intensity of the first band in the first spectral component. The first ratio is a dimensionless value, and its physical meaning is the proportion of the excess of the structure-scattered signal in the polarization-preserving component caused by the anisotropic waveguide with a causal shell to the response intensity of the first band in the first spectral component. When waveguide interference is present, the response intensity of the first band in the first spectral component is higher due to the contribution of waveguide photons, resulting in a positive intensity difference and a first ratio greater than zero. When there is no waveguide interference, the response intensity of the first band in the first spectral component is close to that in the second spectral component, the intensity difference approaches zero, and the first ratio approaches zero.

[0060] The second ratio is calculated by dividing the depth difference by the absorption depth of the second band in the second spectral component. The depth difference, obtained from S3, is the difference between the absorption depth of the second band in the second spectral component and the absorption depth of the second band in the first spectral component. The absorption depth of the second band in the second spectral component is the depression of the local absorption valley extracted within the second band range after baseline correction of the second spectral component in S3, relative to the line connecting the two adjacent shoulders. The second band range is the same as the second band range corresponding to the absorption of seed oil in S2, for example, the 1700 nm to 1760 nm band range. The formula for calculating the second ratio is: R2 = ΔD2 / D2 para Where R2 represents the second ratio, ΔD2 represents the depth difference, and D2 para This represents the absorption depth of the second band in the second spectral component. The second ratio is a dimensionless value. Physically, the second ratio represents the proportion of the loss of absorption signal in the depolarized component caused by the anisotropic waveguide of the causal shell relative to the absorption signal lost by waveguide photons bypassing the seed kernel, relative to the absorption depth of the second band in the second spectral component. When waveguide interference exists, the absorption depth of the second band in the first spectral component is lower due to waveguide photons bypassing the seed kernel, resulting in a positive depth difference and a second ratio greater than zero. When there is no waveguide interference, the absorption depths of the second band in the first and second spectral components are close, the depth difference approaches zero, and the second ratio approaches zero.

[0061] The mutual exclusion condition is determined as follows: when both the first ratio and the second ratio are greater than zero, and the absolute value of the difference between the first and second ratios is less than a preset threshold, the first and second ratios are deemed to satisfy the mutual exclusion condition. The physical basis for satisfying the mutual exclusion condition is that the signal redistribution induced by the anisotropic waveguide of the kernel shell exhibits homogeneous symmetry. The waveguide photon population simultaneously causes an excess of structural scattering signal in the first spectral component and a deficiency of oil absorption signal in the second spectral component. Furthermore, the excess and deficiency are relatively close in proportion because they originate from the same batch of waveguide photons bypassing the kernel. Therefore, the first and second ratios are not only positive but also have similar values, and the absolute value of the difference between them falls within a small range. If the absolute value of the difference between the first and second ratios exceeds the preset threshold, it indicates that the response difference between the first and second bands is not dominated by the same waveguide effect but may be caused by other interference factors. In this case, the mutual exclusion condition is not satisfied, and subsequent correction is not triggered.

[0062] The preset judgment threshold is determined as follows: A group of isotropic reference camellia fruits known to lack anisotropic waveguide characteristics in their shells are pre-selected. Steps S1 to S3 are performed on each isotropic reference camellia fruit to obtain the first ratio and second ratio corresponding to the isotropic reference camellia fruit. The distribution range of the absolute value of the difference between the first ratio and the second ratio corresponding to the isotropic reference camellia fruit is calculated, and the upper limit of this distribution range is used as the preset judgment threshold. The selection method for isotropic reference camellia fruits is as follows: A certain number of fruits are extracted from the same batch of camellia fruits. The internal structure of the shell of each fruit is analyzed, and the orientation and distribution of the vascular bundles in the shell are analyzed. Fruits with uniform vascular bundle orientation in all spatial directions and no significant directional arrangement are selected as isotropic reference camellia fruits. For each isotropic reference camellia fruit, steps S1 to S3 are repeated multiple times under different conveyor belt passing postures. For example, each fruit passes the detection station 5 times in different orientations. The first spectral component and the second spectral component are collected, and the first ratio and the second ratio are calculated for each passing time in the manner described above. This yields a set of absolute values ​​of the difference between the first ratio and the second ratio. The distribution range of these absolute values ​​is statistically analyzed, and the upper limit of the distribution range is taken, for example, the maximum value of all absolute values ​​of the differences is taken as the preset judgment threshold. In the subsequent online prediction process, for the camellia fruit to be tested, if the absolute value of the difference between the calculated first ratio and the second ratio is less than the preset judgment threshold, then the mutual exclusion condition is determined to be met.

[0063] The waveguide contribution rate is determined by averaging the absolute values ​​of the first ratio and the second ratio. The waveguide contribution rate is calculated as: Cwg = (|R1| + |R2|) / 2, where Cwg represents the waveguide contribution rate. The waveguide contribution rate is a dimensionless value, reflecting the average proportion of signal redistribution caused by waveguide photons in the total signal.

[0064] The dilution amount is determined by multiplying the waveguide contribution rate by the absorption depth in the second band. The formula for calculating the dilution amount is: ΔD dilute =Cwg×D2 total , where ΔD dilute Indicates the dilution amount, D2 total The second-band absorption depth is the global second-band absorption depth extracted from the original near-infrared diffuse reflectance spectrum in S2 without polarization decomposition. The physical meaning of dilution is the spurious attenuation that should be subtracted from the second-band absorption depth caused by the anisotropic waveguide effect of the causal shell. Subtracting the dilution from the global second-band absorption depth allows for the restoration of the corrected absorption depth that only reflects the absorption of seed oil. This method of determining mutual exclusion and quantifying contribution based on waveguide physics can adaptively compensate for the actual waveguide interference level of each camellia fruit, correcting the impact of waveguide interference on oil content prediction without damaging the camellia fruit or introducing an additional imaging system.

[0065] An example of obtaining the corrected absorption depth by subtracting the dilution amount from the second-band absorption depth, and inputting the corrected absorption depth into the oil content prediction model to obtain the online prediction result of the oil content of a single camellia fruit is as follows.

[0066] The corrected absorption depth is obtained by subtracting the dilution amount from the second-band absorption depth. The second-band absorption depth is the absorption depth characterizing the absorption of kernel oil extracted from the original near-infrared diffuse reflectance spectrum in S2. The dilution amount is the product of the waveguide contribution rate determined in S4 and the second-band absorption depth. The formula for calculating the corrected absorption depth is: D2 corr =D2 total -ΔD dilute D2 corr D2 represents the corrected absorption depth. total Indicates the absorption depth of the second band, ΔD dilute This indicates the dilution amount. The corrected absorption depth is a dimensionless value that reflects the true absorption depth produced solely by the absorption of the kernel oil components after removing the spurious attenuation of the absorption signal caused by the anisotropic waveguide effect of the causal shell from the original near-infrared diffuse reflectance spectrum.

[0067] The oil content prediction model is a quantitative prediction model based on partial least squares (PLS) established beforehand, relating the corrected absorption depth to the oil content. The establishment process of this quantitative prediction model is as follows: A sample set of camellia oleifera fruits is prepared for modeling. The sample set contains no fewer than 50 camellia oleifera fruits, covering a distribution range from low to high oil content. For each camellia oleifera fruit in the sample set, the original near-infrared diffuse reflectance spectrum under online transport conditions is collected according to method S1. The second band absorption depth is extracted according to methods S2 to S5, and the corrected absorption depth is calculated to obtain the corrected absorption depth corresponding to each camellia oleifera fruit. The kernel oil content of each camellia oleifera fruit is determined using Soxhlet extraction as a reference oil content. A partial least squares regression model is constructed using the corrected absorption depth of all camellia oleifera fruits in the sample set as the independent variable and the corresponding reference oil content as the dependent variable. The number of latent variables in the partial least squares regression model is determined using leave-one-out cross-validation, selecting the number of latent variables that minimizes the root mean square error of the cross-validation. After modeling, the result is of the form: Y = b × D² corr The quantitative prediction model for +b0, where Y represents the predicted oil content as a percentage by mass, and D2 corr The value represents the corrected absorption depth, and b and b0 are the regression coefficients obtained from partial least squares regression. This quantitative prediction model is stored in a computer connected to the near-infrared spectrometer for use during online prediction.

[0068] When inputting the corrected absorption depth into the oil content prediction model, the absorption peak morphology parameters of the second band of the original near-infrared diffuse reflectance spectrum after baseline correction are also input simultaneously. The absorption peak morphology parameters are quantitative indicators describing the shape characteristics of the absorption valleys in the baseline-corrected spectrum within the second band. These parameters include at least one of the absorption valley asymmetry and the half-width at half-maximum (HWHM) of the absorption valley. The absorption valley asymmetry is obtained as follows: within the second band, the absorption valley is divided into a left half-valley and a right half-valley, with the valley floor as the boundary. The widths of the left and right half-valleys are measured, and the asymmetry is calculated as the ratio of the width of the right half-valley to the width of the left half-valley. The HWHM of the absorption valley is obtained as follows: within the second band, two wavelength positions are found where the ordinate value recovers to half the absorption valley depth from the valley floor. The wavelength interval between these two positions is taken as the HWHM of the absorption valley. The reason for introducing the absorption peak morphology parameter is that the slow changes in the fatty acid composition of the kernels of Camellia oleifera fruit over time after harvest will cause changes in the morphology of the oil absorption peak. This change is not entirely independent of the change in oil content, and using the corrected absorption depth alone may produce systematic prediction biases for fruits with different post-harvest times. By using the corrected absorption depth and the absorption peak morphology parameter together as independent variables, the oil content prediction model can comprehensively capture the intensity and morphological information of oil absorption, thereby improving the adaptability of the oil content prediction model to the post-harvest physiological changes of Camellia oleifera fruit.

[0069] The oil content prediction model incorporating absorption peak morphology parameters is a pre-established quantitative prediction model based on partial least squares (PLS) with corrected absorption depth and absorption peak morphology parameters as independent variables and oil content as the dependent variable. The establishment method of this quantitative prediction model is similar to that of the quantitative prediction model using only corrected absorption depth. Using the same sample set of camellia oleifera fruits for modeling, while obtaining the corrected absorption depth for each fruit, absorption peak morphology parameters are also extracted according to the aforementioned acquisition method. The corrected absorption depth and absorption peak morphology parameters together constitute the independent variable matrix, with reference oil content as the dependent variable. PLS regression is then performed to obtain a multiple linear quantitative prediction model, for example: Y = b1 × D2 corr The formula is: +b2×P+b0, where P represents the absorption peak morphology parameter, and b1, b2, and b0 are regression coefficients obtained from partial least squares regression. During online prediction, the corrected absorption depth and absorption peak morphology parameters obtained from S1 to S5 of the individual camellia fruit to be tested are substituted into this quantitative prediction model to calculate the online prediction result of the oil content of a single camellia fruit. The unit of the oil content prediction result is mass percentage.

[0070] Example 2: Figure 2 A schematic diagram of the online prediction system for oil content of camellia fruit based on near-infrared spectral characteristics is provided. The online prediction system for oil content of camellia fruit based on near-infrared spectral characteristics includes the following modules:

[0071] The spectral acquisition module is used to acquire the original near-infrared diffuse reflectance spectrum of a single camellia fruit under online transportation conditions;

[0072] The feature extraction module is used to extract the first band response intensity, which is sensitive to the scattering of the lignified structure of the fruit shell, and the second band absorption depth, which characterizes the absorption of oil in the kernel, from the original near-infrared diffuse reflectance spectrum.

[0073] The decomposition and quantization module is used to decompose the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes, to obtain the intensity difference between the first band response intensity and the second spectral component, and the depth difference between the second band absorption depth and the first spectral component.

[0074] The contribution extraction module is used to calculate the first ratio of the intensity difference to the response intensity of the first band in the first spectral component, and the second ratio of the depth difference to the absorption depth of the second band in the second spectral component. When the first ratio and the second ratio satisfy the mutual exclusion condition, the average of the absolute values ​​of the first ratio and the second ratio is used as the waveguide contribution rate, and the product of the waveguide contribution rate and the absorption depth of the second band is determined as the dilution amount.

[0075] The oil content prediction module is used to subtract the dilution amount from the second band absorption depth to obtain the corrected absorption depth. The corrected absorption depth is then input into the oil content prediction model to obtain the online prediction result of the oil content of a single camellia fruit.

[0076] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0077] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0078] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics, characterized in that, Includes the following steps: S1: Collect the original near-infrared diffuse reflectance spectrum of a single camellia fruit under online transport conditions; S2: Extract the first band response intensity, which is sensitive to the scattering of the lignified structure of the fruit shell, and the second band absorption depth, which characterizes the absorption of oil in the kernel, from the original near-infrared diffuse reflectance spectrum. S3: Decompose the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes, obtain the intensity difference between the first band response intensity and the second spectral component, and the depth difference between the second band absorption depth and the first spectral component. S4: Calculate the first ratio of the intensity difference to the first band response intensity in the first spectral component, and the second ratio of the depth difference to the second band absorption depth in the second spectral component. When the first ratio and the second ratio satisfy the mutual exclusion condition, the average of the absolute values ​​of the first ratio and the second ratio is used as the waveguide contribution rate. The product of the waveguide contribution rate and the second band absorption depth is determined as the dilution amount. S5: Subtract the dilution amount from the second band absorption depth to obtain the corrected absorption depth. Input the corrected absorption depth into the oil content prediction model to obtain the online prediction result of the oil content of a single camellia fruit.

2. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 1, characterized in that, The raw near-infrared diffuse reflectance spectra of individual camellia fruits were collected during online transport, including: A single camellia fruit is irradiated with a near-infrared light source positioned at a first angle to the plane of the conveyor belt. A diffuse reflection fiber optic probe positioned at a second angle to the plane of the conveyor belt receives the near-infrared light emitted from the surface of the single camellia fruit after being transmitted through its interior. The diffuse reflection fiber optic probe transmits the received near-infrared light to a near-infrared spectrometer, which converts the received light signal into the original near-infrared diffuse reflection spectrum.

3. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 1, characterized in that, The first band response intensity, sensitive to scattering from the lignified structure of the fruit shell, and the second band absorption depth, characterizing the absorption of oil in the kernel, were extracted from the original near-infrared diffuse reflectance spectrum, including: Baseline correction was performed on the original near-infrared diffuse reflectance spectrum. Within the first band range corresponding to the scattering of the lignified structure of the fruit shell, the peak height of the spectrum after baseline correction was taken as the response intensity of the first band. Within the second band range corresponding to the absorption of oil in the kernel, the depression of the local absorption valley of the spectrum after baseline correction relative to the line connecting the two adjacent shoulders was taken as the absorption depth of the second band.

4. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 1, characterized in that, The original near-infrared diffuse reflectance spectrum is decomposed into a first spectral component that maintains the polarization state and a second spectral component that depolarizes. The intensity difference between the first and second spectral components is obtained for the first band response intensity, and the depth difference between the first and second spectral components for the second band absorption depth, including: The near-infrared light corresponding to the original near-infrared diffuse reflectance spectrum is passed through a rotatable linear polarizer. The spectrum collected when the transmission axis of the linear polarizer is aligned with a first position parallel to the incident plane is taken as the first spectral component, and the spectrum collected when the transmission axis of the linear polarizer is aligned with a second position perpendicular to the incident plane is taken as the second spectral component. The intensity difference is obtained by subtracting the response intensity of the first band in the second spectral component from the response intensity of the first band in the first spectral component. The depth difference is obtained by subtracting the absorption depth of the second band in the first spectral component from the absorption depth of the second band in the second spectral component.

5. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 4, characterized in that, When aligning the transmission axis of the linear polarizer with the first position parallel to the incident plane, before acquiring the first spectral component, the process includes: aligning the transmission axis of the linear polarizer with the first position, acquiring the blank polarization spectrum of the conveyor belt surface before placing a single camellia fruit, using the polarization retention rate in the first position of the blank polarization spectrum as the polarization reference, and normalizing the first and second spectral components respectively using the polarization reference when acquiring the first and second spectral components.

6. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 1, characterized in that, The first ratio is calculated by dividing the intensity difference by the response intensity of the first band in the first spectral component, and the second ratio is calculated by dividing the depth difference by the absorption depth of the second band in the second spectral component. The mutual exclusion condition is determined as follows: when the first ratio is greater than zero and the second ratio is greater than zero, and the absolute value of the difference between the first ratio and the second ratio is less than a preset threshold, the first ratio and the second ratio are determined to satisfy the mutual exclusion condition. The waveguide contribution rate is determined by taking the average of the absolute values ​​of the first ratio and the second ratio. The dilution amount is determined by multiplying the waveguide contribution rate by the absorption depth of the second band.

7. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 6, characterized in that, The preset judgment threshold is determined as follows: a set of isotropic reference camellia fruits that are known to not have anisotropic waveguide characteristics of the fruit shell are selected in advance. S1 to S3 are executed on the isotropic reference camellia fruits respectively to obtain the first ratio and the second ratio corresponding to the isotropic reference camellia fruits. The distribution range of the absolute value of the difference between the first ratio and the second ratio corresponding to the isotropic reference camellia fruits is calculated, and the upper limit of the distribution range is used as the preset judgment threshold.

8. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 1, characterized in that, The corrected absorption depth is obtained by subtracting the dilution amount from the second band absorption depth; the oil content prediction model is a quantitative prediction model between the corrected absorption depth and the oil content established in advance based on partial least squares method.

9. The method for online prediction of oil content in camellia fruit based on near-infrared spectral characteristics according to claim 8, characterized in that, When inputting the corrected absorption depth into the oil content prediction model, the absorption peak morphology parameters of the second band in the original near-infrared diffuse reflectance spectrum after baseline correction are also input. The oil content prediction model is a quantitative prediction model established in advance based on partial least squares method, with the corrected absorption depth and absorption peak morphology parameters as independent variables and the oil content as the dependent variable.

10. An online prediction system for the oil content of Camellia oleifera fruit based on near-infrared spectral characteristics, used to implement the online prediction method for the oil content of Camellia oleifera fruit based on near-infrared spectral characteristics as described in any one of claims 1-9, characterized in that, Includes the following modules: The spectral acquisition module is used to acquire the original near-infrared diffuse reflectance spectrum of a single camellia fruit under online transportation conditions; The feature extraction module is used to extract the first band response intensity, which is sensitive to the scattering of the lignified structure of the fruit shell, and the second band absorption depth, which characterizes the absorption of oil in the kernel, from the original near-infrared diffuse reflectance spectrum. The decomposition and quantization module is used to decompose the original near-infrared diffuse reflectance spectrum into a first spectral component that maintains the polarization state and a second spectral component that depolarizes, to obtain the intensity difference between the first band response intensity and the second spectral component, and the depth difference between the second band absorption depth and the first spectral component. The contribution extraction module is used to calculate the first ratio of the intensity difference to the response intensity of the first band in the first spectral component, and the second ratio of the depth difference to the absorption depth of the second band in the second spectral component. When the first ratio and the second ratio satisfy the mutual exclusion condition, the average of the absolute values ​​of the first ratio and the second ratio is used as the waveguide contribution rate, and the product of the waveguide contribution rate and the absorption depth of the second band is determined as the dilution amount. The oil content prediction module is used to subtract the dilution amount from the second band absorption depth to obtain the corrected absorption depth. The corrected absorption depth is then input into the oil content prediction model to obtain the online prediction result of the oil content of a single camellia fruit.

Images