A method and apparatus for detecting multiple biochemical components in a leaf

Abstract

This invention discloses a method and instrument for detecting multiple biochemical components in leaves. The method combines pseudo-absorption transformation, Beer's law, and Gauss-Lorentz function for component separation. Pseudo-absorption coefficients are constructed through pseudo-absorption transformation. Based on the coupling of the Gauss-Lorentz function and Beer's law, the problem of overlapping spectral separation and sensitive band selection of multiple biochemical components is solved from the perspective of pseudo-absorption spectroscopy, thereby obtaining the equivalent absorption coefficient of each component in the corresponding band. The leaf under test is irradiated sequentially with multiple discrete band light sources according to a preset time sequence, and the reflected light intensity signal at each band is collected. The reflected light intensity signal is converted into relative reflectance, and pseudo-absorption transformation is performed on the relative reflectance of the sensitive band based on the reference band. The content of different biological components is obtained according to the pseudo-absorption coefficient and the corresponding equivalent absorption coefficient, thus realizing the synchronous, rapid, and non-destructive detection of multiple biochemical components in leaves.

Description

Technical Field

[0001] This invention belongs to the field of plant biochemical parameter detection technology, specifically relating to a method and instrument for detecting multiple biochemical components in leaves. Background Technology

[0002] In satellite remote sensing inversion systems, in-situ acquisition of multi-parameter ground truth values ​​is a crucial verification step. Considering the limitations of traditional chemical analysis, such as its high destructiveness and the inability of existing conventional portable equipment to separate multiple pigments and simultaneously acquire moisture and dry matter content, developing an in-situ multi-component leaf detection instrument has clear engineering application value.

[0003] However, in actual development, there is an inherent contradiction between the complex optical response mechanism of living leaves and the need for simplified hardware structure. Specifically, in the visible light region, the absorption spectra of chlorophyll a and chlorophyll b overlap significantly; in the short-wave infrared region, the absorption signals of dry matter and water are often superimposed with the scattering effects of the leaf's internal structure. While traditional hyperspectral techniques can handle these complex responses, they suffer from data redundancy, high instrument costs, and weak resistance to interference in the field. Therefore, effectively separating overlapping signals and achieving independent extraction of multiple parameters using only a few discrete wavelength bands is the core technical challenge facing the development of such instruments. Summary of the Invention

[0004] This invention proposes a method and instrument for detecting multiple biochemical components in leaves, aiming to solve the problem that it is difficult to independently separate and quantitatively invert the components under the condition of severe overlap of absorption spectra of multiple components in living leaves.

[0005] In a first aspect, the present invention provides a method for detecting multiple biochemical components in leaves, the method comprising:

[0006] A linear inversion model is constructed based on the pseudo-absorption weighting coefficient and the comprehensive constant term; the linear inversion model includes a pigment detection inversion model and a moisture and dry matter detection inversion model;

[0007] The pseudo absorption weighting coefficient is determined by the equivalent absorption coefficients of the target component and the interference component in the corresponding sensitive bands, obtained based on the Gauss-Lorentz mixing function; after the pseudo absorption weighting coefficient is determined, the comprehensive constant term is calibrated based on the sample measured data;

[0008] The blade under test is illuminated sequentially using multiple discrete band light sources, and the reflected light intensity signal is collected and converted into relative reflectance. Based on the reference band, a pseudo absorption transformation is performed on the relative reflectance of the sensitive band to obtain the pseudo absorption coefficient. The pseudo absorption coefficient is input into the corresponding linear inversion model to obtain the content of the first pigment component, the content of the second pigment component, the moisture content, and the dry matter content.

[0009] Preferably, the plurality of discrete band light sources include a first pigment-sensitive band light source, a second pigment-sensitive band light source, and a pigment reference band light source for pigment detection, and a moisture-sensitive band light source, a dry matter-sensitive band light source, and a moisture-dry matter reference band light source for moisture and dry matter detection.

[0010] Preferably, the center wavelength of the first pigment-sensitive band light source is around 650 nm; the center wavelength of the second pigment-sensitive band light source is around 700 nm; the center wavelength of the pigment reference band light source is around 900 nm; the center wavelength of the dry matter-sensitive band light source is around 1204 nm; the center wavelength of the moisture-sensitive band light source is around 1379 nm; and the center wavelength of the moisture-dry matter reference band light source is around 1555 nm.

[0011] Preferably, the method for obtaining the pseudo absorption weight coefficient is as follows: construct an absorption combination relationship based on the equivalent absorption coefficients of the target component and the interfering component in two sensitive bands, and determine the pseudo absorption weight coefficient of the corresponding target component based on the determinant of the absorption combination relationship and the equivalent absorption coefficient of the interfering component in the two sensitive bands.

[0012] Preferably, the method for obtaining the relative reflectivity is as follows:

[0013] The reflected light intensity signal of the tested blade, the dark field background signal, and the reflected light intensity signal of the standard white board were collected for each discrete band. The relative reflectivity of the corresponding discrete band was calculated based on the difference between the reflected light intensity signal of the tested blade and the dark field background signal, as well as the difference between the reflected light intensity signal of the standard white board and the dark field background signal.

[0014] Preferably, when the multiple discrete band light sources irradiate the blade under test in sequence, a dark field interval is set between the irradiation processes of two adjacent discrete band light sources; during the irradiation period of each discrete band light source, multiple frames of reflected light intensity signals are collected and averaged.

[0015] Preferably, each discrete band light source illuminates the blade under test after passing through a narrowband filter; the full width at half maximum (FWHM) of the narrowband filter is 16 nm.

[0016] Preferably, the pseudo-absorption coefficient is the logarithm of the ratio of the relative reflectance of the reference band to the relative reflectance of the sensitive band.

[0017] Preferably, the first pigment component is chlorophyll a; and the second pigment component is chlorophyll b.

[0018] Secondly, the present invention provides a multi-biochemical component detection instrument for leaves, which is used to perform the above-mentioned multi-biochemical component detection method; the multi-biochemical component detection instrument includes a housing and a leaf clamping detection structure, a control processing module, a photoelectric detection module and an active light source module inside the housing;

[0019] The blade clamping detection structure is used to clamp the blade to be tested; the active light source module includes multiple discrete band light sources, which are used to sequentially illuminate the blade to be tested; the multiple discrete band light sources include a first pigment-sensitive band light source, a second pigment-sensitive band light source and a pigment reference band light source for pigment detection, and a moisture-sensitive band light source, a dry matter-sensitive band light source and a moisture and dry matter reference band light source for moisture and dry matter detection.

[0020] The photoelectric detection module is used to collect the reflected light intensity signal generated by the tested leaf under illumination by light sources in various discrete bands; the control processing module is connected to the active light source module and the photoelectric detection module respectively, and is used to convert the reflected light intensity signal into relative reflectivity, perform pseudo-absorption transformation on the relative reflectivity of the sensitive band based on the reference band to obtain the pseudo-absorption coefficient, and input the pseudo-absorption coefficient into the linear inversion model to obtain the content of the first pigment component, the content of the second pigment component, the moisture content, and the dry matter content.

[0021] The beneficial effects of this invention are:

[0022] 1. Based on the pseudo-absorption transformation of leaf reflectance, this invention introduces the Gauss-Lorentz function to physically model the absorption spectra of each component, transforming the absorption coefficient from an unknown parameter into a computable function with clear spectroscopic meaning, thereby distinguishing the characteristic absorption contribution of each component from the overlapping total absorption envelope. At the same time, this invention combines the spectral morphology description capability of the Gauss-Lorentz function with the quantitative framework of Beer's law, making up for the inherent shortcomings of single methods in the separation of overlapping spectra, and realizing the separable expression and stable high-precision inversion of multiple components under finite discrete band conditions.

[0023] 2. This invention, by setting pigment-sensitive bands and pigment reference bands, and performing linear separation based on the pseudo-absorption coefficients normalized to the reference bands, can distinguish the overlapping spectral information of chlorophyll a and chlorophyll b under limited discrete band conditions, avoiding the problem of insufficient pigment discrimination ability caused by only detecting total chlorophyll or using empirical spectral indices. At the same time, by setting moisture-sensitive bands, dry matter-sensitive bands, and moisture-dry matter reference bands, this invention can separate the overlapping absorption information of moisture and dry matter in the short-wave infrared region, reducing the masking effect of strong moisture absorption on dry matter signals.

[0024] 3. This invention uses a limited number of discrete band light sources to replace the continuous beam splitting structure of traditional hyperspectral instruments, which can reduce equipment cost, size and data processing complexity, making it suitable for portable in-situ field detection. At the same time, the detection instrument of this invention can be combined with a light-shielding clamping structure, constant current drive, dual-channel photoelectric detection and diffuse reflection measurement structure with tilted incident / normal reception to improve the stability and anti-interference ability of in-situ detection. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of Embodiment 1 of the present invention.

[0026] Figure 2 This is a two-dimensional R² distribution diagram of the pseudo-absorption dual-band combination and Ca and Cb in Embodiment 1 of the present invention.

[0027] Figure 3 This is a scatter plot of the predicted and measured values ​​of Ca and Cb from the inversion model in Embodiment 1 of the present invention.

[0028] Figure 4 This is a two-dimensional R² distribution diagram of the pseudo-absorption dual-band combination and Cw and Cm in Embodiment 1 of the present invention.

[0029] Figure 5 This is a scatter plot of the predicted and measured values ​​of Cw and Cm in inversion in Embodiment 1 of the present invention.

[0030] Figure 6 This is an R² two-dimensional matrix diagram of different spectral indices and Ca and Cb in Embodiment 1 of the present invention.

[0031] Figure 7 The figure shows the fitting and verification results of the RDI exponential inversion model for Ca and Cb in Embodiment 1 of the present invention.

[0032] Figure 8 This is a two-dimensional R² matrix diagram of different spectral indices and Cw and Cm contents in Example 1 of the present invention.

[0033] Figure 9 This is a scatter plot showing the fitting and verification of the optimal spectral index inversion model in Embodiment 1 of the present invention.

[0034] Figure 10 This is a comparison diagram of different ranges of resampling of the visible light feature center band in Embodiment 1 of the present invention.

[0035] Figure 11 This is a comparison diagram of different ranges of resampling in the near-infrared feature center band in Embodiment 1 of the present invention.

[0036] Figure 12 This is a scatter plot of biochemical components obtained by 16nm bandwidth resampling and characteristic peak inversion in Example 1 of the present invention.

[0037] Figure 13 This is a schematic diagram of the multi-biochemical component detection instrument in Embodiment 2 of the present invention.

[0038] Figure 14 This is a schematic diagram of the internal layout of the multi-biochemical component detection instrument in Embodiment 2 of the present invention.

[0039] Figure 15 This is a spectral response curve of the sensor in the visible and near-infrared bands in Embodiment 2 of the present invention.

[0040] Figure 16 This is a scatter plot verifying the detection accuracy of the multi-biochemical component detection instrument in Embodiment 2 of the present invention. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings.

[0042] Example 1

[0043] like Figure 1 As shown, a method for detecting multiple biochemical components in leaves is used for in-situ, non-destructive detection of chlorophyll a, chlorophyll b, water, and dry matter in leaves. The method includes the following steps:

[0044] Step 1: Construct a spectral separation model

[0045] Step 1-1. Constructing the pseudo absorption coefficient

[0046] Leaf reflectance spectra obtained under natural conditions not only contain absorption information of biochemical components but are also affected by various factors such as leaf surface reflection, internal multiple scattering, and measurement system errors. These factors often lead to a complex nonlinear relationship between reflectance and biochemical component content, thereby increasing the uncertainty of the inversion model. To suppress these nonlinear effects and highlight the absorption contribution in the spectrum, a logarithmic transformation is performed on the reflectance spectrum to construct a pseudo-absorption coefficient. Its expression is:

[0047] in, The reference band reflectivity is typically located in the near-infrared region; The reflectivity is the target band.

[0048] According to Beer–Lambert's law, the intensity of light absorption by a substance is linearly related to its concentration. Ideally, when multiple absorbing substances coexist, the total absorption can be expressed as a linear superposition of the absorptions of each component. Therefore, for leaf spectra, the pseudo-absorption coefficient… Further expressed as:

[0049]

[0050] in, For the m-th biochemical component at wavelength The equivalent absorption coefficient per unit concentration at a given wavelength reflects its absorption at a given wavelength. Absorption capacity at the site; The content of the m-th biochemical component; The background term mainly originates from scattering from the internal structure of the blades and measurement system errors.

[0051] Step 1-2. Construct the equivalent absorption coefficient

[0052] In actual blades, the absorption bands of each component are not ideal narrow peaks, but rather exhibit broadening, and there is often significant overlap between components. This phenomenon mainly stems from two aspects: non-uniform broadening caused by differences in the microenvironment of molecules, and uniform broadening caused by molecular energy level lifetimes and interactions. In spectroscopy, these two types of broadening are usually described by Gaussian and Lorentz functions, respectively; therefore, the absorption lines of a single component can be considered as a mixture of both. The Gaussian-Lorentz mixing function is used to evaluate the equivalent absorption coefficient. Parametric representation:

[0053] in, These are the weighting coefficients. ; For Gaussian components, ; The wavelength of the absorption peak; For non-uniformly widened width parameters; For the Lorentz component, ; The width parameter is for uniform expansion.

[0054] The equivalent absorption coefficients of the target component and the interfering component are obtained based on formula (3). This realizes the transformation from "spectral morphology description" to "quantitative expression of absorption coefficient", laying the foundation for subsequent concentration inversion.

[0055] Steps 1-3. Construction and solution of the dual-band linear model

[0056] Considering the significant overlap in the absorption spectra of various biochemical components in plant leaves, a single wavelength band is insufficient to effectively distinguish the contributions of the target component, interfering components, and background scattering effects. Therefore, two characteristic wavelength bands with different sensitivities to the target and interfering components were selected. and characteristic bands By eliminating the influence of interfering components through linear combination, a system of simultaneous equations for absorbance is constructed:

[0057]

[0058] in, and Characteristic bands and characteristic bands The pseudo-absorption coefficient below; and The equivalent absorption coefficient of the target component; The target component concentration; Concentration of interfering components; and The equivalent absorption coefficient of the interfering component; and This is used to measure noise or non-specific background constants in the system.

[0059] By performing an algebraic derivation of formula (4), the concentration of the target component is obtained. The expression:

[0060] From equation (5), it can be seen that the concentration of the target component is... This can be expressed as a linear combination of the pseudo-absorption coefficients of two characteristic bands. To facilitate subsequent parameter optimization using the particle swarm optimization algorithm, let the determinant of the equation system be... Therefore, the coefficient terms in equation (5) consisting of the physical absorption coefficient are simplified to model weights. And the constant terms are combined into the intercept. Thus, a general dual-band linear inversion model is established:

[0061] (6)

[0062] in, and These are the pseudo-absorption weighting coefficients for the corresponding bands (all positive numbers), and their mathematical form consists of the absorption coefficients of the interfering components and the determinant. Decision (i.e.) , ); This is a comprehensive constant term that includes system noise and background correction information.

[0063] Steps 1-4. Parameter optimization

[0064] In terms of model parameter optimization, a particle swarm optimization algorithm is introduced for global optimization, based on spectral band position ( , Using the root mean square error (RMSE) between the model predictions and the measured values ​​as the fitness function, the optimal combination of bands and the corresponding physical model parameters that minimizes the inversion error are searched across the entire band range.

[0065] (7)

[0066] Where n is the number of test samples; and represents the predicted and measured values ​​of the biochemical component content of the i-th leaf sample, respectively.

[0067] Based on the above, band selection and model verification were carried out using measured data for biochemical parameters such as chlorophyll, water, and dry matter.

[0068] Step 2: Feature Band Screening

[0069] 2-1. Screening for characteristic bands of chlorophyll a and chlorophyll b

[0070] The measured dataset was divided into training and validation sets at a 3:1 ratio using stratified random sampling. To effectively avoid the influence of strong absorption peaks of interfering components such as anthocyanins near 550 nm on chlorophyll inversion, the data analysis selected 600 nm–1000 nm as the research interval, which covers both the strong absorption centers of chlorophyll a and chlorophyll b in the visible light region and the near-infrared plateau reflecting the internal structural characteristics of leaves. Based on this, the dual-band combinations across the entire spectrum were traversed and pseudo-absorption coefficients were calculated. The aim was to utilize the differences in the sensitivity of different bands to absorption and scattering characteristics to maximize the extraction of component separation signals while effectively offsetting background interference from leaf surface reflections and non-target components. The coefficient of determination (R²) for different band combinations was calculated. 2 Distribution as follows Figure 2 As shown.

[0071] from Figure 2 As shown in (a) and (b), the highly correlated regions are not uniformly distributed across the entire spectrum, but rather exhibit a clear concentration trend, primarily located in the transition region from red light to the red edge. Specifically, when the combined wavelengths approach 700 nm, continuous highly correlated bands are often formed, indicating a strong correlation and stable fitting ability between the spectral information near this band and the contents of chlorophyll a and chlorophyll b. Meanwhile, locally high correlation regions can also be observed in the near-infrared band, but their distribution is relatively discrete. Combined with the leaf spectral response mechanism, it is known that chlorophyll has significant absorption characteristics in the 670 nm-680 nm range, while near 700 nm, chlorophyll absorption weakens, and spectral reflectance becomes more sensitive to changes in pigment concentration, while avoiding the saturation effect that may occur in the strong absorption region. Therefore, 700 nm, avoiding the strong absorption saturation region and being relatively sensitive to pigment changes, provides a clear basis as a sensitive wavelength for chlorophyll a.

[0072] When determining the sensitive wavelength band of chlorophyll b, it is necessary to establish a certain spectral interval with the 700 nm band while ensuring that it is responsive to chlorophyll, in order to reduce the mutual interference between the bands. Since the main absorption bands of chlorophyll a and chlorophyll b in the red light region intersect around 650 nm, this band can reflect the differences in the absorption characteristics of the two pigments. Combined with the correlation distribution results, it can be found that the region corresponding to the combination of 650 nm and 700 nm also shows a high correlation level. Therefore, 650 nm is selected as the second characteristic band, combined with 700 nm, to distinguish chlorophyll a from chlorophyll b.

[0073] Based on the above analysis, 650 nm and 700 nm were ultimately determined as the core band combination for distinguishing chlorophyll a and chlorophyll b. Simultaneously, to reduce the influence of leaf surface reflection and internal tissue scattering on the spectral signal, a 900 nm band, located in the near-infrared plateau region and largely unaffected by pigment absorption modulation, was introduced as a reference band to correct for the structural scattering term. Based on this, independent quantitative inversion models for chlorophyll a and chlorophyll b were constructed, as shown in equations (8) and (9):

[0074] (8)

[0075] (9)

[0076] in, and These represent the contents of chlorophyll a and chlorophyll b, respectively. , , This represents the reflectivity of the corresponding waveband.

[0077] To verify the stability and applicability of the constructed inversion model, a sample set was used to validate the inversion results of chlorophyll a and chlorophyll b. The predicted values ​​calculated by the model were compared with the measured values, and a scatter plot was plotted, as shown below. Figure 3 As shown in the figure, the model demonstrates good predictive ability for both chlorophyll a and chlorophyll b. The scatter plots of the predicted and measured values ​​generally fit the 1:1 reference line, indicating strong model stability. Most samples of chlorophyll a are concentrated in the mid-to-high value range (RMSE = 3.845 μg·cm⁻²), with only slight dispersion at extreme values; the scatter plot distribution of chlorophyll b is relatively more concentrated (RMSE = 1.963 μg·cm⁻²). Based on the combined validation results of the two sets, it can be concluded that the inversion model constructed based on the 650 nm, 700 nm, and 900 nm bands can accurately characterize the variation characteristics of chlorophyll a and chlorophyll b. The performance of the theoretical model on independent samples is consistent with the results of the modeling stage, showing no significant systematic bias and demonstrating good accuracy.

[0078] 2-2. Screening for characteristic wavelengths of moisture and dry matter

[0079] Using the aforementioned spectral separation method based on Beer-Lambert's law and pseudo-absorption transform, pseudo-absorption weighting coefficients are set for the target component and interfering components respectively, thereby separating the contributions of different components in the mixed spectrum and finally obtaining the characteristic absorption bands of dry matter and water. Figure 4 As shown in the two-dimensional coefficient of determination distribution diagram, after band separation, water and dry matter exhibit high correlation within specific band combinations. Through three-dimensional parameter space optimization, a common reference baseline of 1555 nm was determined within the 600 nm–1700 nm range, and 1379 nm and 1204 nm were extracted as the optimal characteristic absorption bands. Based on the above band combinations, the following linear inversion equation is established:

[0080]

[0081]

[0082] in, and These represent the content of moisture and dry matter, respectively.

[0083] As can be seen from equations (10) and (11), 1204 nm and 1379 nm are each other's sensitive band and background band in the model: when inverting Cw, 1379 nm is introduced as a background correction term to eliminate the strong absorption interference of water; while when inverting Cm, 1204 nm is used to subtract the spectral contribution of dry matter. This dual-band strategy based on a unified baseline effectively removes the overlap effect in the mixed spectrum, thus providing a feasible solution for simplifying the structure of the portable instrument and controlling the cost.

[0084] Combination Figure 5 Scatter validation results show that the RMSE for Cw is 0.001421 cm⁻¹, and the RMSE for Cm is 0.000967 g·cm⁻². These results demonstrate that even within the limited spectral range of 400 nm to 1700 nm, this separation method effectively removes the masking effect of moisture background on dry matter signals, significantly improving inversion accuracy. Furthermore, the model relies on only three bands (1204 nm, 1379 nm, and 1555 nm), reducing the data processing complexity and hardware requirements.

[0085] To verify the performance of the spectral separation method based on Beer-Lambert's law and pseudo-absorption transform in terms of inversion accuracy and stability, commonly used spectral indices (Ratio Index (RI), Difference Index (DI), Normalized Difference Index (NDI), and Reciprocal Difference Index (RDI)) were selected for comparative analysis. Since the spectral index method utilizes a few fixed bands and aligns with the design principles of portable instrument development, it was used as a reference method for comparison with the spectral separation method based on Beer-Lambert's law and pseudo-absorption transform. The specific calculation formulas for each index are shown in Table 1. To ensure fairness in the comparison, all indices adopted the same training and validation set partitioning method as the pseudo-absorption dual-band separation method, and the accuracy was evaluated using the same dataset.

[0086] Table 1. Spectral indices and their calculation formulas

[0087]

[0088] Table Notes: , Let i be the reflectance at bands i and j in the hyperspectral spectrum.

[0089] Step 3: Verification of the Inversion Model

[0090] 3-1. Validation of the inversion model for chlorophyll a and chlorophyll b

[0091] Based on the spectral reflectance (400nm-1000 nm) of the same dataset, four classical spectral indices—the ratio, difference index, normalized difference index, and inverse difference index—were calculated for any two-band combinations. Then, these indices were compared with Ca and Cb using Ri... 2 The traversal calculations constructed a two-dimensional correlation distribution matrix, as shown below. Figure 6 As shown. From Figure 6 The two-dimensional heatmap clearly shows that the responses of the four spectral indices to chlorophyll a and chlorophyll b contents exhibit similar distributions in two-dimensional space. Highly correlated regions all show obvious cross-shaped clustering characteristics. Specifically, the sensitive bands are highly concentrated in the transition region from visible to near-infrared (700-750 nm), which closely matches the spectral characteristics of plant leaves exhibiting typical drastic changes in the red edge position in this region. Furthermore, in the near-infrared plateau region of 750-900 nm, each index also maintains a relatively broad correlation response.

[0092] Among the four indices, RDI exhibits the most concentrated area of ​​high correlation (e.g., Figure 6 As shown in (d) and (h) in the figure, background noise was most effectively suppressed. Therefore, RDI was finally determined as the optimal feature extraction index. The modeling results (as shown in the figure) Figure 7 As shown in (a) and (b) in the figure, for Ca, the band combination of RDI in the red edge region (731-762 nm) shows the highest correlation (R 2 = 0.809); for Cb, the optimal band combination also lies in the red-edge region (735-766 nm), with a determination coefficient of 0.7603. The inversion model established based on the extracted optimal red-edge band combination shows excellent statistical fitting potential on the modeling set and can be used for accuracy evaluation on the subsequent independent validation set.

[0093] To verify the reliability of the Ca and Cb inversion model built based on the RDI index, modeling validation was performed using blade samples from the validation set. The estimation accuracy of the model was comprehensively evaluated by calculating the RMSE and plotting scatter plots of predicted and measured values. The validation results are as follows: Figure 7 As shown in (c) and (d), the scatter plots for Ca and Cb are generally distributed around the 1:1 relationship line, with RMSEs of 16.20 μg·cm⁻² and 5.43 μg·cm⁻², respectively. Further analysis reveals that the prediction model for Cb is relatively stable; while the model for Ca, although some predicted values ​​are significantly lower than the measured values ​​when the measured content is high, indicating a certain degree of underestimation, the model overall meets the requirements for predicting chlorophyll content.

[0094] 3-2. Validation of the inversion model for moisture and dry matter

[0095] For the spectral inversion of moisture and dry matter content, using spectral reflectance data (400-1700 nm), four classical spectral indices were calculated, and their coefficients of determination were calculated by comparing them with the measured values ​​of moisture and dry matter. The resulting two-dimensional correlation distribution matrix (as shown in the figure) was then obtained. Figure 8 ).

[0096] Depend on Figure 8 It can be seen that the overall correlation of Cw inversion is higher than that of Cm. The optimal index for moisture inversion is NDI (1567, 1683), with R0. 2 The correlation coefficient reached 0.9322. (From the correlation distribution plot...) Figure 8As shown in (a)-(d)), the band region with a high correlation to Cw is widely distributed in the short-wave infrared region of 1400nm~1700nm, which coincides with the strong absorption band of moisture near 1450nm. This optimal combination avoids the absorption saturation region and uses the secondary absorption characteristics of 1550–1650 nm to better characterize the moisture gradient change. In addition, the correlation distribution characteristics of the four indices in Cw inversion are basically consistent. In comparison, the optimal index for dry matter estimation is RI (1390, 1564), with R0... 2 The value is only 0.7280. This is because the strong absorption of water in living leaves masks the spectral response of dry matter, resulting in a high correlation region for the retrieved Cm. Figure 8 The (e)-(h) bands are significantly narrowed. The RI combination mainly uses dual-band computation to reduce the interference of moisture background, thereby indirectly extracting dry matter information.

[0097] like Figure 9 Figures (a) and (b) show the fitting results and distribution of Cw and Cm, respectively. For the inversion results of Cw ( Figure 9 In (c) of the model, the scatter plots of the measured and predicted values ​​are generally distributed around the 1:1 reference line, with an inversion RMSE of 0.001955 cm, indicating that the model has good estimation accuracy in the low to medium moisture range. However, when the measured Cw is high, the scatter plots tend to shift downwards towards the 1:1 line, indicating that the model underestimates the moisture content in the high moisture region. This may be related to the tendency of spectral absorption to saturate under high moisture conditions. For the scatter plots of the measured and predicted values ​​of Cm inversion (… Figure 9 In (d), the inversion RMSE was 0.001295 g·cm⁻², and most samples were also distributed near the diagonal, indicating a good overall fit.

[0098] Overall, both models have good inversion capabilities within the normal sample range, but there is still a certain degree of saturation effect in the high value region.

[0099] 3-3. Comparison of Inversion Accuracy between Overlapping Spectral Separation Method and Spectral Index Method

[0100] To evaluate the applicability of the two inversion methods in portable spectroscopic detection, a comparative analysis was conducted on the Beer-Lambert law-based overlapping spectral separation method and the traditional spectral index method. From a theoretical perspective, the overlapping spectral separation method, based on the Beer-Lambert law, converts reflectance into pseudo-absorbance through pseudo-absorption spectral transformation, making the contributions of different biochemical components in the spectrum approximately linearly superimposed. Simultaneously, the introduction of reference bands (such as 900 nm and 1555 nm) insensitive to the absorption of the target component can reduce the influence of leaf structure scattering and light variations, and the separation of spectral information of different components is achieved through dual-band algebraic elimination. In contrast, the traditional spectral index method only enhances the target signal through calculations such as difference, ratio, or normalized difference, and its essence belongs to an empirical statistical model. When the absorption characteristics of different biochemical components overlap, the index method struggles to completely distinguish the spectral contributions of each component.

[0101] The accuracy results of the inversion verification of the two methods based on the same measured dataset are as follows (e.g.) Figure 9 As shown in the figure, the overlapping spectral separation method based on Beer-Lambert's law has higher prediction accuracy in the inversion of various biochemical components. For pigment parameters, the RMSE of this method for inverting Ca and Cb are 3.845 μg·cm⁻² and 1.963 μg·cm⁻², respectively, which are lower than the best-performing RDI model, whose RMSE are 16.20 μg·cm⁻² and 5.43 μg·cm⁻², respectively.

[0102] In shortwave infrared parameter inversion, the separation method achieved RMSEs of 0.001421 cm⁻¹ and 0.000967 g·cm⁻² for Cw and Cm, respectively, which are also superior to the spectral index model. The RMSEs for NDI and RI were 0.001955 cm⁻¹ and 0.001295 g·cm⁻², respectively. From the scatter plot distribution characteristics, the predicted values ​​obtained by the separation method are distributed along the 1:1 reference line, showing no significant systematic bias on independent samples. In contrast, the spectral index method mainly relies on empirical statistical relationships, and its discriminative ability is limited when the absorption characteristics of different components overlap.

[0103] Overall, the Beer-Lambert law-based overlapping spectral separation method can effectively distinguish the spectral information of different biochemical components under limited wavelength conditions and has advantages in inversion accuracy. The modeling and verification results of this method can provide a theoretical basis for the band selection and embedded biochemical component inversion model of subsequent portable spectroscopic instruments.

[0104] 3-4. The impact of hardware spectral bandwidth selection on inversion accuracy

[0105] To achieve low-cost, portable detection, a hardware alternative combining narrowband filters and photoelectric sensors is proposed. Within this framework, the impact of bandwidth variations on the retention of characteristic information of biochemical parameters needs to be analyzed. The limited transmission bandwidth of narrowband filters inevitably reduces spectral resolution, potentially weakening the characterization of key biochemical information. Therefore, this paper systematically evaluates the impact of bandwidth compression on feature extraction based on spectral response mechanisms. Since the absorption characteristics of biochemical components such as chlorophyll and water in plant leaves mainly exhibit broad-spectrum absorption over a continuous wavelength range, rather than sharp features at a single wavelength, and strong correlations exist between adjacent wavelengths within specific sensitive regions, this indicates that appropriately reducing spectral resolution may still preserve the main absorption characteristic information, providing a theoretical feasibility for using limited-bandwidth filters.

[0106] Therefore, drawing on existing spectral resampling methods, this paper constructs a consistency analysis of hyperspectral and spectral information under different bandwidth conditions, and uses the accuracy of the inversion model of hyperspectral and spectral information with different bandwidths as the evaluation standard to quantitatively compare the inversion performance of each bandwidth parameter. In this way, the filter configuration scheme that achieves the optimal trade-off between information retention capability, hardware implementation feasibility and system cost can be determined.

[0107] To conduct a bandwidth comparison experiment, representative leaf samples were selected using random sampling. For the sensitive characteristic wavelengths of leaf biochemical components (650nm, 700nm, 900nm, 1204nm, 1379nm, 1555nm), spectral resampling was performed with bandwidths of 10nm, 16nm, and 30nm, respectively. The results are as follows: Figure 10 and Figure 11 As shown in the figure. Experimental results show that the original hyperspectral characteristic peaks of different leaf samples and the resampled broadband data are highly consistent in both morphology and amplitude. Specifically, the waveforms of the resampled spectra with bandwidths of 10 nm and 16 nm almost completely overlap with the original spectra, while the 30 nm bandwidth shows a slight spectral smoothing effect at some absorption peaks, but still maintains the main peak-valley structure. The correlation coefficient R between the resampled spectral reflectance and the original characteristic peak values ​​was selected as the criterion. 2 Using the root mean square error (RMSE) as the quantification standard, the results are shown in Table 2.

[0108] Table 2. Consistency analysis results at different resolutions

[0109]

[0110] Table 2 shows the determination coefficients R0 of the spectral inversion models under the three bandwidth configurations. 2The RMSE values ​​were all relatively high, and the overall RMSE remained at a low level, indicating that all three bandwidths could preserve feature information to some extent. However, the RMSE of the 30 nm bandwidth at the 700 nm characteristic peak was 0.01466, significantly higher than that of 10 nm (0.00210) and 16 nm (0.00422); at 650 nm, the RMSE of the 30 nm bandwidth was 0.00108, also significantly higher than that of 10 nm (0.00025) and 16 nm (0.00041). This is mainly because the 650 nm and 700 nm areas are the main absorption regions and red edge regions of chlorophyll, with large spectral gradients. Overly wide bandwidths would smooth out the steep features of these regions, thus introducing larger reconstruction errors. In contrast, the differences in RMSE among the three bandwidths are relatively small in the near-infrared and short-wave infrared bands such as 900 nm, 1204 nm, 1379 nm, and 1555 nm (for example, the RMSEs at 10 nm, 16 nm, and 30 nm at 900 nm are 0.00007, 0.00008, and 0.00016, respectively), indicating that the absorption characteristics of these bands are more moderate and less sensitive to bandwidth changes.

[0111] Considering both instrument development cost and accuracy requirements, while a 10 nm bandwidth offers slightly higher accuracy, the corresponding narrowband filter manufacturing cost is significantly increased, and the light throughput is lower, which is detrimental to improving the signal-to-noise ratio in portable detection. Conversely, while a 30 nm bandwidth offers the lowest cost, the accuracy loss in the chlorophyll-sensitive bands (650 nm and 700 nm) is unacceptable. Therefore, 16 nm was selected as the central bandwidth for instrument development. This configuration achieves the best balance between spectral information fidelity, hardware implementation difficulty, and system cost. To further verify the accuracy of the inversion model under the 16 nm bandwidth configuration, a subset of samples was randomly selected. Based on their hyperspectral data and resampled spectral data, the same model was used for inversion, and the accuracy was analyzed. The results are as follows: Figure 12 As shown, the RMSE values ​​for the retrieved Ca, Cb, Cm, and Cw were 0.285 μg·cm⁻², 0.074 μg·cm⁻², 0.000019 g·cm⁻², and 0.0001233 cm⁻², respectively. The retrieval accuracy of each parameter did not significantly decrease compared to the results based on the original hyperspectral data, with the RMSE values ​​for moisture and dry matter remaining at a low level. These results indicate that when retrieving Ca, Cb, Cm, and Cw using lower-resolution spectral data, the retrieval model exhibits a reasonable error level, meeting the requirements of practical applications. This further verifies the engineering feasibility of using a 16nm bandwidth configuration for in-situ detection of biochemical components in tea gardens.

[0112] Step 4: Acquire the reflected light intensity signal

[0113] Multiple discrete wavelength light sources are driven sequentially in a preset order to illuminate the blade under test. The discrete wavelength light sources with center wavelengths of 650nm, 700nm, 900nm, 1204nm, 1379nm, and 1555nm are driven sequentially. During the illumination period of each discrete wavelength light source, a preset illumination duration is maintained to allow the photoelectric detection signal to reach a stable state.

[0114] A dark field interval is set between the illumination processes of two adjacent discrete band light sources. During the dark field interval, each discrete band light source is turned off, and the dark field background signal is acquired. By setting the dark field interval, the influence of residual radiation from the previous band light source, detector response hysteresis, and light source crosstalk on the detection signal of the subsequent band can be reduced.

[0115] During each discrete wavelength band illumination period, the reflected light intensity signal generated by the tested blade under the corresponding discrete wavelength band is acquired. To improve detection stability, multiple frames of reflected light intensity signals can be acquired during each discrete wavelength band illumination period, and the average of the multiple frames of reflected light intensity signals can be processed to obtain the original reflected light intensity signal of the corresponding discrete wavelength band.

[0116] Step 5: Obtain the relative reflectance

[0117] For any discrete wavelength band λ, the relative reflectivity of the corresponding discrete wavelength band is calculated based on the original reflected light intensity signal of the tested blade, the dark field background signal, and the reflected light intensity signal of the standard white board. It is represented as:

[0118]

[0119] in, The original reflected light intensity signal of the blade under test in band λ; This represents the dark field background signal corresponding to band λ. This represents the light intensity signal reflected by a standard diffuse whiteboard in band λ.

[0120] By using dark field background subtraction and standard white board calibration, the impact of detector dark current, environmental background, light source intensity differences, and optical path transmission loss on the detection results can be reduced.

[0121] Step 6: Perform pseudo-absorption transformation on the relative reflectivity of the sensitive band based on the reference band.

[0122] For pigment detection, the wavelength around 900nm is used as the reference wavelength for pigments, and the pseudo-absorption coefficient is calculated based on the relative reflectance of the wavelengths around 650nm and 700nm, respectively. Let R be the relative reflectance of the wavelengths around 650nm, 700nm, and 900nm. 650 R 700 and R900 Then we can calculate:

[0123]

[0124]

[0125] Among them, A 650 A represents the pseudo-absorption coefficient in the band around 650 nm relative to a reference band around 900 nm. 700 This is the pseudo-absorption coefficient in the 700nm band relative to the reference band in the 900nm band.

[0126] For moisture and dry matter detection, the wavelength around 1555 nm is used as the reference wavelength for moisture and dry matter. The pseudo-absorption coefficient is calculated based on the relative reflectance of the wavelengths around 1204 nm and 1379 nm, respectively. Let R1 be the relative reflectance of the wavelengths around 1204 nm, 1379 nm, and 1555 nm. 1204 R 1379 and R 1555 Then we can calculate:

[0127]

[0128]

[0129] Among them, A 1204 A is the pseudo-absorption coefficient in the band around 1204 nm relative to the reference band around 1555 nm. 1379 It represents the pseudo-absorption coefficient of the band near 1379 nm relative to the reference band near 1555 nm.

[0130] The above pseudo-absorption transform normalizes the sensitive band with the reference band, converting the reflectance signal into a characteristic quantity that can characterize the absorption contribution, which is beneficial for separating the spectral contributions of different biochemical components in overlapping bands.

[0131] Step 7: Calculate the chlorophyll a content, chlorophyll b content, water content, and dry matter content based on the pseudo-absorption coefficient. The chlorophyll a content (Ca) and chlorophyll b content (Cb) are calculated using the following linear model:

[0132]

[0133]

[0134] Wherein, a1, a2, a3, b1, b2, and b3 are model coefficients obtained through calibration using sample measured data. The above model, through the combination of two pigment-sensitive bands and one pigment reference band, enables the separation of the overlapping absorption contributions of chlorophyll a and chlorophyll b in a linear combination manner, thereby achieving independent calculation of the two pigment parameters.

[0135] Moisture content Cw and dry matter content Cm are calculated according to the following linear model:

[0136]

[0137]

[0138] Wherein, c1, c2, c3, d1, d2, and d3 are model coefficients obtained through calibration using sample measured data. The above model separates the overlapping absorption contributions of water and dry matter under finite shortwave infrared conditions by combining the water-sensitive band, the dry matter-sensitive band, and the water-dry matter reference band.

[0139] Step 8: Output the detection results

[0140] After completing the inversion calculation, the system outputs the chlorophyll a content, chlorophyll b content, moisture content, and dry matter content. The test results can be displayed in real time on a screen, written to local storage, or transmitted to an external terminal.

[0141] The above detection method enables rapid in-situ detection of multiple biochemical components in leaves using a limited number of discrete wavelengths without picking or damaging the leaves.

[0142] Example 2

[0143] like Figure 13 and Figure 14 As shown, a multi-biochemical component detection instrument for leaves is used to perform the detection method described in Example 1. The multi-biochemical component detection instrument includes a housing 1 and a leaf clamping and detection structure 2 installed inside the housing 1, a result output module 3, a control and processing module 4, a photoelectric detection module 5, a storage module 6, a power supply module 7, and an active light source module 8.

[0144] The housing 1 is made of light-shielding material, and a detection window is provided at its front end. The blade clamping and detection structure 2 is located at the detection window, and includes an upper clamping part and a lower clamping part. A clamping gap is formed between the upper and lower clamping parts to accommodate the blade being tested. The clamping gap and the detection window together constitute a light-shielding detection space. The blade clamping and detection structure adopts an elastic clamping form to accommodate blades of different thicknesses and reduce the possibility of damage to the blade during the clamping process.

[0145] The active light source module 8 generates detection light in multiple discrete wavelength bands and includes an LED array, a narrowband filter, and a constant current drive circuit. The LED array comprises six groups of LED light sources, with center wavelengths located near 650nm, 700nm, 900nm, 1204nm, 1379nm, and 1555nm, respectively. The wavelengths near 650nm and 700nm are used to extract differential absorption information between chlorophyll a and chlorophyll b; the wavelength near 900nm serves as a reference wavelength for pigments; the wavelength near 1204nm is used to extract dry matter absorption information; the wavelength near 1379nm is used to extract moisture absorption information; and the wavelength near 1555nm serves as a reference wavelength for moisture and dry matter. Narrowband filters are positioned at the front end of each LED's emission path, with a full width at half maximum (FWHM) of 16nm. By using narrowband filters to constrain the bandwidth of the LED emitted light, the possibility of broadband stray light entering the detection optical path can be reduced, making the incident light more concentrated in the target wavelength band, thereby improving the stability of pseudo-absorption feature calculation. A constant current drive circuit is used to drive each LED light source and maintain a stable LED drive current when the supply voltage drops or external power supply conditions change, thus reducing the impact of LED luminous intensity variations and center wavelength drift on the detection results. The constant current drive circuit can also be equipped with overvoltage and overcurrent protection circuits to improve the safety and stability of the instrument during field use.

[0146] The photoelectric detection module is arranged correspondingly to the detection window to collect the reflected light intensity signals generated by the tested leaf under illumination from various discrete wavelength light sources. The photoelectric detection module includes a visible light detection channel and a near-infrared detection channel. The visible light detection channel collects reflected light intensity signals in the vicinity of 650nm, 700nm, and 900nm. The near-infrared detection channel collects reflected light intensity signals in the vicinity of 1204nm, 1379nm, and 1555nm. The combination of the visible light and near-infrared detection channels covers the wavelength range required for pigment detection, as well as moisture and dry matter detection, avoiding the problem of uneven response of a single detector across a wide wavelength range. The spectral response curves of the visible light and near-infrared detection channels are shown below. Figure 15 As shown.

[0147] The active light source module and the photoelectric detection module adopt a diffuse reflection measurement structure with oblique incidence and normal reception. Specifically, each LED light source in the active light source module illuminates the blade at an oblique angle relative to the surface of the blade being measured, and the photoelectric detection module receives the reflected light intensity signal along the normal direction of the surface of the blade being measured.

[0148] In this embodiment, the active light source module illuminates the leaf at an incident angle of approximately 45°, and the photoelectric detection module receives the reflected light intensity signal along the 0° direction. This structure reduces the influence of specular reflection from the cuticle layer on the detection signal, allowing the photoelectric detection module to primarily receive diffusely reflected light carrying biochemical information from within the leaf.

[0149] The control and processing module includes a microcontroller. The microcontroller is connected to the active light source module, photoelectric detection module, storage module, and result output module. The microcontroller controls multiple LED light sources to illuminate sequentially according to a preset timing sequence, and reads the light intensity signal output by the photoelectric detection module during the illumination of each group of LED light sources. The microcontroller is also used to perform dark field subtraction, white board calibration, relative reflectance calculation, pseudo-absorption coefficient calculation, and biochemical component inversion calculation.

[0150] During a single detection cycle, the control processing module sequentially controls LED light sources emitting light at wavelengths near 650nm, 700nm, 900nm, 1204nm, 1379nm, and 1555nm. During each LED light source emission period, the control processing module acquires multiple frames of light intensity data and performs mean filtering on these frames to obtain a stable light intensity signal for that wavelength band. Between the emission processes of two adjacent LED light sources, the control processing module controls the active light source module to enter a dark field interval state and acquires the dark field background signal.

[0151] The storage module stores standard whiteboard calibration data, dark-field background data, model coefficients, and test results. Before use, the testing instrument collects the reflected light intensity signals of the whiteboard at each wavelength using a standard diffuse reflection whiteboard and stores these signals as standard whiteboard calibration data in the storage module. During actual testing, the control and processing module calls upon the standard whiteboard calibration data to convert the original reflected light intensity signal of the tested blade into relative reflectivity.

[0152] After obtaining the relative reflectance of each band, the control processing module uses the band near 900nm as the pigment reference band and calculates the pseudo absorption coefficients corresponding to the bands near 650nm and 700nm; it uses the band near 1555nm as the water and dry matter reference band and calculates the pseudo absorption coefficients corresponding to the bands near 1204nm and 1379nm. Subsequently, the control processing module calls the model coefficients in the storage module to calculate the chlorophyll a content, chlorophyll b content, water content, and dry matter content, respectively.

[0153] The results output module is used to output the test results. The results output module can use a display screen, such as an OLED display. The results output module can display chlorophyll a content, chlorophyll b content, moisture content, and dry matter content. The power supply module can use a rechargeable battery to power the active light source module, photoelectric detection module, control processing module, storage module, and results output module.

[0154] The specific process for detecting the content of multiple biochemical components in the tested leaves using a multi-biochemical component detection instrument is as follows:

[0155] Because non-ideal physical factors such as manufacturing tolerances of electronic components, temperature drift, dark current of sensors, and slight fluctuations in light source intensity can cause certain systematic deviations in the theoretical calculation model during actual testing, after the instrument was developed, the theoretical inversion model was further reconstructed and optimized using measured sample data.

[0156] First, a multi-biochemical component detection instrument was used to collect the raw reflected light intensity signals of live leaves at various characteristic wavelengths, and the relative reflectance of each wavelength was calculated by combining it with standard white board reference data. Simultaneously, the chlorophyll a content, chlorophyll b content, moisture content, and dry matter content measured by standard chemical analysis methods on the same sample of leaves were used as reference values. The spectral response characteristics acquired by the instrument were fitted with the chemical measurement results using multiple linear regression, thereby calibrating the inversion model parameters. The calibrated model coefficients were written into the storage unit of the control processing module and directly called by the microcontroller during actual detection to complete the real-time calculation of the multi-biochemical component content of the leaves.

[0157] After reconstruction and optimization, the leaf to be tested is clamped at the detection window, placing it within the light-shielding detection space. Subsequently, the control and processing module sequentially drives light sources at wavelengths of 650 nm, 700 nm, 900 nm, 1204 nm, 1379 nm, and 1555 nm according to a preset timing sequence, and collects the reflected light intensity signals of the leaf at each wavelength. Then, the control and processing module sequentially performs dark field subtraction, white board calibration, relative reflectance calculation, pseudo-absorption coefficient calculation, and multi-biochemical component inversion calculation to obtain predicted values ​​for chlorophyll a content, chlorophyll b content, moisture content, and dry matter content. Finally, the root mean square error relative accuracy is used as the instrument's performance evaluation index, calculated as follows: Where C* represents the predicted value from the multi-biochemical component detection instrument; C represents the chemically determined value, i.e., the true value. denoted as the average chemical measurement value; n represents the number of validation samples. The biochemical component content of the same batch of leaf samples was determined using standard chemical analysis methods, and the chemical measurement results were used as reference values ​​for error analysis of the instrument prediction results.

[0158] The scatter distribution of measured and predicted values ​​of various biochemical parameters for multi-biochemical component detection instruments is as follows: Figure 16 As shown. By Figure 16It can be seen that the predicted results of each biochemical parameter have good consistency with the chemical measurement results, and most sample points are distributed near the 1:1 reference line, indicating that the multi-biochemical component detection instrument can achieve good in-situ prediction of multiple biochemical components in leaves. Specifically, the root mean square error (RMSE) for chlorophyll a detection is 1.6528 μg·cm⁻², with a relative accuracy of 95.71%; the RMS error for chlorophyll b detection is 1.3058 μg·cm⁻², with a relative accuracy of 90.26%; the RMS error for moisture content detection is 0.001039 cm⁻², with a relative accuracy of 93.07%; and the RMS error for dry matter content detection is 0.001089 g·cm⁻², with a relative accuracy of 86.94%.

[0159] In summary, the multi-biochemical component detection instrument completes a comprehensive detection loop, from multi-band light source timing control, reflected light intensity signal acquisition, relative reflectivity calculation, pseudo-absorption feature extraction to embedded inversion output of multi-biochemical components. Overall testing results show that the instrument's relative accuracy in detecting chlorophyll a, chlorophyll b, moisture, and dry matter in leaves remains stable within the range of 86% to 96%, meeting the application requirements for rapid, in-situ, and non-destructive detection of multiple biochemical parameters in tea leaves from mountainous tea gardens.

Claims

1. A method for detecting multiple biochemical components in leaves, characterized in that: The method includes: A linear inversion model is constructed based on the pseudo-absorption weighting coefficient and the comprehensive constant term; the linear inversion model includes a pigment detection inversion model and a moisture and dry matter detection inversion model; The pseudo absorption weighting coefficient is determined by the equivalent absorption coefficients of the target component and the interference component in the corresponding sensitive bands, obtained based on the Gauss-Lorentz mixing function; after the pseudo absorption weighting coefficient is determined, the comprehensive constant term is calibrated based on the sample measured data; The blade under test is illuminated sequentially using multiple discrete band light sources, and the reflected light intensity signal is collected and converted into relative reflectance. Based on the reference band, a pseudo absorption transformation is performed on the relative reflectance of the sensitive band to obtain the pseudo absorption coefficient. The pseudo absorption coefficient is input into the corresponding linear inversion model to obtain the content of the first pigment component, the content of the second pigment component, the moisture content, and the dry matter content.

2. The method for detecting multiple biochemical components in leaves according to claim 1, characterized in that: The plurality of discrete band light sources include a first pigment-sensitive band light source, a second pigment-sensitive band light source, and a pigment reference band light source for pigment detection, as well as a moisture-sensitive band light source, a dry matter-sensitive band light source, and a moisture-dry matter reference band light source for moisture and dry matter detection.

3. The method for detecting multiple biochemical components in leaves according to claim 2, characterized in that: The center wavelength of the first pigment-sensitive band light source is around 650 nm; the center wavelength of the second pigment-sensitive band light source is around 700 nm; the center wavelength of the pigment reference band light source is around 900 nm; the center wavelength of the dry matter-sensitive band light source is around 1204 nm; the center wavelength of the moisture-sensitive band light source is around 1379 nm; and the center wavelength of the moisture-dry matter reference band light source is around 1555 nm.

4. The method for detecting multiple biochemical components in leaves according to claim 1, characterized in that: The method for obtaining the pseudo absorption weight coefficient is as follows: construct an absorption combination relationship based on the equivalent absorption coefficients of the target component and the interfering component in two sensitive bands, and determine the pseudo absorption weight coefficient of the corresponding target component based on the determinant of the absorption combination relationship and the equivalent absorption coefficient of the interfering component in the two sensitive bands.

5. The method for detecting multiple biochemical components in leaves according to claim 1, characterized in that: The method for obtaining the relative reflectivity is as follows: The reflected light intensity signal of the tested blade, the dark field background signal, and the reflected light intensity signal of the standard white board were collected for each discrete band. The relative reflectivity of the corresponding discrete band was calculated based on the difference between the reflected light intensity signal of the tested blade and the dark field background signal, as well as the difference between the reflected light intensity signal of the standard white board and the dark field background signal.

6. The method for detecting multiple biochemical components in leaves according to claim 1, characterized in that: When the multiple discrete band light sources sequentially illuminate the blade under test, a dark field interval is set between the illumination processes of two adjacent discrete band light sources; during the illumination period of each discrete band light source, multiple frames of reflected light intensity signals are collected and averaged.

7. The method for detecting multiple biochemical components in leaves according to claim 1, characterized in that; Each discrete band light source illuminates the tested blade after passing through a narrowband filter; the full width at half maximum (FWHM) of the narrowband filter is 16 nm.

8. The method for detecting multiple biochemical components in leaves according to claim 1, characterized in that: The pseudo-absorption coefficient is the logarithm of the ratio of the relative reflectance of the reference band to the relative reflectance of the sensitive band.

9. The method for detecting multiple biochemical components in leaves according to claim 1, characterized in that: The first pigment component is chlorophyll a; the second pigment component is chlorophyll b.

10. A multi-biochemical component detection instrument for leaves, characterized in that: The instrument is used to perform the multi-biochemical component detection method for leaves as described in claim 1. The multi-biochemical component detection instrument includes a housing (1) and a leaf clamping and detection structure (2) inside the housing (1), a control and processing module (4), a photoelectric detection module (5) and an active light source module (8). The blade clamping detection structure (2) is used to clamp the blade to be tested; the active light source module (8) includes multiple discrete band light sources, which are used to sequentially irradiate the blade to be tested; the multiple discrete band light sources include a first pigment-sensitive band light source, a second pigment-sensitive band light source and a pigment reference band light source for pigment detection, as well as a moisture-sensitive band light source, a dry matter-sensitive band light source and a moisture and dry matter reference band light source for moisture and dry matter detection; The photoelectric detection module (5) is used to collect the reflected light intensity signal generated by the tested leaf under illumination by light sources in each discrete band; the control processing module (4) is connected to the active light source module (8) and the photoelectric detection module (5) respectively, and is used to convert the reflected light intensity signal into relative reflectivity, perform pseudo absorption transformation on the relative reflectivity of the sensitive band based on the reference band to obtain the pseudo absorption coefficient, and input the pseudo absorption coefficient into the linear inversion model to obtain the content of the first pigment component, the content of the second pigment component, the moisture content and the dry matter content.

Images