Rice growth state monitoring method based on hyperspectral imaging
By combining hyperspectral imaging with continuous wavelet transform and bio-fingerprint constrained ICA, the problems of feature aliasing and background noise overwhelmance in rice growth status monitoring have been solved, enabling accurate and robust monitoring of rice growth status and providing a scientific quantitative assessment basis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- RICE RES INST GUANGDONG ACADEMY OF AGRI SCI
- Filing Date
- 2026-04-03
- Publication Date
- 2026-06-19
AI Technical Summary
Existing technologies cannot accurately distinguish between early diseases and nutrient deficiency symptoms during rice growth, and weak disease signals are easily drowned out by background noise, leading to misdiagnosis and missed detection. Traditional monitoring methods have low accuracy in complex field environments.
A hyperspectral imaging-based approach was adopted, combined with continuous wavelet transform and independent component analysis constrained by bio-fingerprint. The optimal feature scale was automatically selected by the scale-specific response intensity index, and effective stress signal components were screened by the bio-fingerprint consistency factor. The comprehensive growth inhibition index was calculated for monitoring.
It achieves precise unmixing of weak and overlapping spectral features of rice canopy, improves the detection accuracy of early growth stress, provides scientific quantitative monitoring basis, and enhances the robustness and accuracy of the monitoring system in complex environments.
Smart Images
Figure CN121963103B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of image processing technology, specifically relating to a method for monitoring the growth status of rice based on hyperspectral imaging. Background Technology
[0002] Rice is a vital food crop globally, and monitoring its health during its growth process is crucial for ensuring yield. Currently, monitoring the growth status of rice mainly faces two categories of problems: one is biotic stress, such as diseases like rice blast and sheath blight; the other is abiotic stress, such as nutrient or environmental problems like nitrogen deficiency and water shortage.
[0003] Current technologies typically utilize hyperspectral imaging to acquire rice canopy data and calculate vegetation indices to retrieve growth parameters. However, in the complex real-world field environment, existing techniques have significant limitations, primarily manifested in feature aliasing. For example, the chlorosis caused by early-stage rice blast lesions and the yellowing caused by nitrogen deficiency exhibit remarkably similar responses across a wide spectral band. Traditional vegetation indices cannot distinguish between these two distinct stress sources, easily leading to misdiagnosis. For instance, nitrogen deficiency may be mistaken for disease, resulting in pesticide application, which not only wastes resources but may also delay disease treatment.
[0004] Furthermore, early disease signals are often hidden within specific narrow-band high-frequency variations, easily masked by field background noise, leading to missed detections. Although Independent Component Analysis (ICA) is commonly used for unmixing mixed signals, traditional ICA algorithms only pursue statistical independence, and the separated components often lack clear biological significance, failing to directly correspond to specific diseases or nutritional indicators. This results in monitoring results that only indicate abnormalities without knowing the cause. Factors such as changes in field light, soil reflection, and weed interference further exacerbate the confusion of spectral characteristics, significantly reducing the accuracy of traditional monitoring methods and making it difficult to meet the actual needs of large-scale, precision paddy field management, posing potential risks to rice yield assurance. Summary of the Invention
[0005] This invention provides a method for monitoring the growth status of rice based on hyperspectral imaging, in order to solve the technical problems in the prior art, such as the inability to accurately distinguish different sources of stress due to feature aliasing, the easy submersion of weak disease signals by background noise leading to missed detection, and the lack of biological interpretation of blind source separation results.
[0006] This invention provides a method for monitoring the growth status of rice based on hyperspectral imaging, comprising the following steps:
[0007] The hyperspectral reflectance curve of the rice canopy was obtained. The hyperspectral reflectance curve reflects the spectral response characteristics of rice in different wavelength bands.
[0008] The hyperspectral reflectance curve is mapped to the time-frequency space using continuous wavelet transform to obtain the wavelet coefficient matrix. Based on the scale-specific response intensity at different scales, the optimal feature scale is determined from the wavelet coefficient matrix and the feature vector is extracted.
[0009] Based on the bio-fingerprint constraint, the feature vector is decomposed into multiple independent components by independent component analysis. Then, by calculating the bio-fingerprint consistency factor of each independent component, the effective stress signal components are screened out.
[0010] Based on the mixing weights of each effective stress signal component in the rice canopy and the corresponding hazard coefficients, a comprehensive growth inhibition index is calculated to complete the monitoring of rice growth status.
[0011] Its effect is that by deeply integrating the time-frequency analysis capability of continuous wavelet transform with ICA under the constraint of biological fingerprint, it overcomes the problem in existing technologies that it is difficult to distinguish between early diseases and nutrient deficiency symptoms caused by the phenomenon of foreign objects having the same spectrum. It achieves accurate unmixing of weak and overlapping spectral features of rice canopy, thereby significantly improving the detection accuracy of early growth stress in complex field backgrounds and providing a scientific basis for the leap from empirical diagnosis to quantitative monitoring of rice.
[0012] Furthermore, the hyperspectral reflectance curve of the rice canopy was obtained, including:
[0013] Collect raw hyperspectral images of rice paddies;
[0014] Radiometric correction was performed on the original hyperspectral image using standard whiteboard data and dark current data to obtain a reflectance image;
[0015] Based on the normalized vegetation index, a mask threshold is set to remove the background area in the reflectance image and extract the region of interest for rice.
[0016] The average reflectance across the entire wavelength range of the rice region of interest is calculated to obtain the hyperspectral reflectance curve.
[0017] Its effect is that, through multiple preprocessing methods such as radiation correction and normalized vegetation index mask threshold, the interference of soil background, water reflection and ambient light fluctuations on the original data is effectively eliminated, ensuring that the subsequently extracted hyperspectral reflectance curve has extremely high purity and physical consistency, thereby reducing the misleading effect of data noise on health assessment results from the source.
[0018] Furthermore, scale-specific response intensity is used to measure the significance of biosignal signals relative to background noise at a specific scale;
[0019] The logic for determining the scale-specific response intensity is as follows: calculate the difference between the wavelet coefficient modulus and the background noise reference value, calculate the ratio of this difference to the sum of the local volatility and the numerical stability constant, and then weight the square of this ratio using a logarithmic term that includes the scale factor to obtain the scale-specific response intensity.
[0020] Among them, the scale-specific response intensity is positively correlated with the difference and negatively correlated with the local volatility, and the logarithmic term is used to suppress the high-frequency noise weights at small scales.
[0021] Its effects are as follows: a scale-specific response intensity index is constructed, which can quantify the significance of biological feature signals relative to background noise, solve the problem that traditional methods easily drown out early weak disease features across the entire spectrum, realize automatic focusing and enhancement of specific diagnostic scales, and greatly improve the sensitivity of the monitoring system to narrow-band high-frequency changes induced by diseases.
[0022] Furthermore, the background noise baseline value is obtained by calculating the mean value of wavelet coefficients in non-plant areas of the image at the same scale; the local fluctuation is obtained by calculating the standard deviation of wavelet coefficients within a sliding window of a preset width centered on the current wavelength.
[0023] Its effect is that by dynamically calculating the noise benchmark and the local standard deviation of the spectrum in non-plant areas, a real-time adaptive reference benchmark is provided for the monitoring model, which enables the method to maintain stable feature extraction capabilities under different light conditions and different field backgrounds, and enhances the robustness of the algorithm in actual production environments.
[0024] Furthermore, based on the biometric fingerprint constraint, the feature vector is unmixed using independent component analysis, including:
[0025] The eigenvectors are unmixed using the independent component analysis algorithm to obtain multiple independent component vectors;
[0026] Introduce prior weights for standard stress types and standard stress spectral fingerprints;
[0027] The similarity between each independent component vector and the standard stress spectral fingerprint is calculated. The bio-fingerprint consistency factor is calculated by combining the prior weights and the statistical properties of the independent component vectors.
[0028] Its effect is that it strongly introduces prior weights and standard spectral fingerprints as navigation in the blind source separation process, transforming the purely statistical signal separation into directional extraction guided by the laws of agricultural biology, ensuring that each separated signal component has a clear agronomic interpretation meaning, and completely solving the technical bottleneck of blind and uncontrollable separation results of the traditional ICA algorithm.
[0029] Furthermore, the bio-fingerprint consistency factor is used to measure the biological reliability of isolated independent components;
[0030] The logic for determining the bio-fingerprint consistency factor is as follows: calculate the weighted sum of the fingerprint similarity between independent components and each known stress type, and use the square of the weighted sum as the numerator; calculate the product of the absolute value of the kurtosis of the independent components and the adjustment coefficient, and use the sum of the product and the preset constant as the denominator; calculate the square root of the ratio of the numerator to the denominator to obtain the bio-fingerprint consistency factor.
[0031] Among them, the biometric fingerprint consistency factor is positively correlated with fingerprint similarity and is subject to the inverse constraint of the absolute value of kurtosis, in order to penalize components that only have statistical independence but lack biometric features.
[0032] Furthermore, fingerprint similarity is calculated using cosine similarity; kurtosis absolute value is used to measure the non-Gaussianity of the signal, and by introducing kurtosis absolute value, a soft constraint is applied to noise components that excessively pursue the maximization of non-Gaussianity but deviate from biological reality.
[0033] Furthermore, the comprehensive growth inhibition index is used to quantitatively assess the overall negative impact on rice;
[0034] The logic for determining the comprehensive growth inhibition index is as follows: For each screened effective stress signal component, calculate the product of its bio-fingerprint consistency factor and the mixing matrix coefficient, normalize the product using a normalization factor, and then multiply it by the hazard coefficient corresponding to the effective stress signal component to obtain the single stress contribution value of the effective stress signal component; sum up the single stress contribution values of all effective stress signal components and add the environmental baseline pressure value to obtain the comprehensive growth inhibition index.
[0035] Furthermore, the normalization factor is taken as the maximum value of the full-band reflectance of healthy rice samples to eliminate the influence of light intensity; the mixing matrix coefficients are determined by the weight of the corresponding pixel in the mixing matrix obtained by the inverse transformation of independent component analysis on the current independent component, representing the relative concentration of the effective stress signal component.
[0036] Furthermore, after calculating the comprehensive growth inhibition index, it also includes:
[0037] The current health level of rice is assessed based on the value of the comprehensive growth inhibition index.
[0038] Based on the biofinite consistency factors and mixing weights of each effective stress signal component, the main stress sources leading to the increase in the comprehensive growth inhibition index are traced, and targeted field management recommendations are given.
[0039] The beneficial effects are as follows: This invention proposes a monitoring method combining continuous wavelet transform and bio-fingerprint constrained ICA. By constructing a scale-specific response intensity index, it achieves automatic selection of the optimal feature scale and effectively extracts weak biological stress features from background noise. By introducing a bio-fingerprint consistency factor, it improves the traditional blind source separation algorithm, enabling the successful decomposition of mixed spectra into independent signal sources when rice is simultaneously subjected to multiple stresses. Finally, it achieves quantitative assessment of growth status through a comprehensive growth inhibition index, providing precise scientific basis for agricultural production. Attached Figure Description
[0040] Figure 1 This is a flowchart of a method for monitoring the growth status of rice based on hyperspectral imaging.
[0041] Figure 2 This is a schematic diagram of the multi-scale feature screening analysis based on continuous wavelet transform in this invention.
[0042] Figure 3 This is a schematic diagram of the blind source separation and decoupling distribution of multiple stress signals in this invention. Detailed Implementation
[0043] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0044] An embodiment of the rice growth status monitoring method based on hyperspectral imaging provided by the present invention:
[0045] like Figure 1 As shown, the method for monitoring rice growth status based on hyperspectral imaging includes the following steps:
[0046] S1, Data Acquisition and Region of Interest Extraction.
[0047] This step aims to obtain pure rice canopy spectral data, eliminating environmental interference.
[0048] First, data acquisition: using pushbroom hyperspectral imagers mounted on drones or on the ground, raw hyperspectral image cubes of rice paddies are acquired.
[0049] Secondly, radiation correction: using dark current data collected with a standard white board and the lens cap closed, the raw digital values are converted into reflectance data. This step is crucial to ensuring the clear physical meaning of the data. The formula is as follows: ,in This is a dimensionless reflectance value.
[0050] Finally, ROI extraction and dimensionality reduction: A threshold was set using the normalized vegetation index (NVI) to generate a mask to remove water and soil background. For each retained rice pixel, the average reflectance across all wavelengths of the retained rice pixels was calculated, resulting in a hyperspectral reflectance curve representing the rice canopy, denoted as [missing information]. ,in This indicates the wavelength, typically ranging from 400nm to 1000nm.
[0051] Through rigorous radiometric correction and background removal, spectral data with clear physical meaning and free from environmental noise were obtained, thus solving the problem of impure data sources.
[0052] S2, wavelet feature enhancement based on multi-scale energy difference.
[0053] To address the issue of weak and inconsistent scale characteristics in early-stage disease, this step utilizes continuous wavelet transform to transform the one-dimensional spectrum. Mapping to a two-dimensional time-frequency space yields the wavelet coefficient matrix. ,in The scale factor is dimensionless. To automatically find the optimal scale that best reflects the characteristics of biological stress, this invention constructs a scale-specific response intensity index. .
[0054] The specific calculation process follows the following formula:
[0055]
[0056] in, This represents the total number of bands in the spectrum. In scale and wavelength Wavelet coefficient modulus at; The baseline value for background noise is obtained by calculating the mean value of wavelet coefficients in non-plant areas of the image at the same scale. For local fluctuations, in wavelength Centered on, take a width of Calculate the standard deviation of the wavelet coefficients within a sliding window; This is a numerical stability constant to prevent the denominator from being zero; It is a natural constant.
[0057] The core of this formula revolves around highlighting biometric signals and suppressing background noise. It achieves accurate identification of effective feature scales through multi-dimensional calculations. The formula is based on... The summation result across all bands is normalized to eliminate the influence of differences in the number of bands and ensure the comparability of different spectral data. The numerator is quantified using the difference between the wavelet coefficient modulus and the background noise baseline value to determine the effective amplitude of the biosignal signal. Local volatility and a numerical stability constant are added to the denominator to mitigate the interference of local signal fluctuations and avoid calculation problems where the denominator is zero. The square of the ratio of these two factors further amplifies the difference between the effective signal and noise, enhancing feature identification. Simultaneously, [the following is introduced]... The logarithmic weighting term utilizes the properties of the logarithmic function to suppress the high-frequency noise weights at small scales, highlighting the effective biological signals at medium and high scales. This allows the calculation results to accurately measure the relative significance of biological features to noise at each scale, enabling automatic selection of the optimal feature scale and laying the foundation for subsequent extraction of weak stress features.
[0058] Calculation example:
[0059] Assuming at scale Next, we focus on wavelength. Data at that location; the wavelet coefficient modulus at that location was measured. The background noise baseline value is calculated from the background shadow area. Local fluctuations near this band Take the stability constant .
[0060] The signal-to-noise ratio term for this band is: ; Scale weighting term is Assuming a total number of bands If we take the average contribution across the entire band as an example, then .
[0061] The physical meaning of this index is clear: it amplifies significant features through the square of the signal-to-noise ratio and uses the logarithmic term to suppress small-scale high-frequency noise, selecting... The scale corresponding to the peak Wavelet coefficients at this scale are extracted as feature vectors. .
[0062] Through the above calculation logic, the effectiveness of signals at different scales can be quantitatively evaluated, and the feature scale containing the most information can be automatically locked.
[0063] S3, blind source separation based on biometric fingerprint constraints.
[0064] To address the issue that components separated by traditional ICA lack physical meaning, this step utilizes the FastICA algorithm to analyze the feature vectors. To demix the fingerprints and guide the algorithm to separate the true biological signals, prior knowledge is introduced, and the biometric fingerprint consistency factor is calculated. To screen for effective ingredients.
[0065] The specific calculation process follows the following formula:
[0066]
[0067] in, For the separated first 1 independent component vector; For the first The prior weights for each known stress type are set by experts based on the seasonal disease prevalence. for With the Cosine similarity between standard stress spectral fingerprints, ranging from 0 to 1; For the first The absolute value of the kurtosis of each independent component; This is the adjustment coefficient.
[0068] The core of this formula revolves around integrating prior agronomic knowledge with statistical characteristics to achieve quantitative screening of the biological reliability of independent components, addressing the problem that traditional ICA only pursues statistical independence, resulting in separation results lacking agronomic significance. The numerator of the formula is the weighted sum and square of the prior weights of each known stress type and the cosine similarity of the component fingerprint. The prior weights are set by agronomic experts based on disease prevalence, and the cosine similarity measures the matching degree between the component and the standard stress spectral fingerprint. Weighting and amplifying the component signals with real biological characteristics, and squaring further enhance the identification of effective features. The denominator introduces the product of the absolute value of kurtosis and the adjustment coefficient, plus a constant. The absolute value of kurtosis measures the non-Gaussianity of the signal, penalizing noise components that are only statistically independent but deviate from biological reality through inverse constraints. The constraint strength can be flexibly adjusted by adjusting the coefficients. The overall normalization mapping of factors is achieved through square root extraction, allowing the calculation results to accurately quantify the biological attribution credibility of independent components, providing a scientific basis for screening effective stress signal components.
[0069] Calculation example:
[0070] Assuming the first component is separated Calculate its absolute kurtosis value Set adjustment coefficient .
[0071] There are two main known stresses: rice blast ( ) and nitrogen deficiency ( ). Calculated Cosine similarity to standard rice blast fingerprint Cosine similarity to standard nitrogen-deficient fingerprints The numerator represents the weighted similarity: The square of the numerator is The denominator is the non-Gaussian penalty. Final score: If the threshold is set to 0.3, this component is considered a valid dominant signal of rice blast.
[0072] This formula not only utilizes the statistical independence of ICA, but also forcibly introduces biological fingerprints to ensure that the isolated components have a clear biological classification.
[0073] S4, Growth Status Assessment and Comprehensive Analysis.
[0074] The comprehensive growth inhibition index is calculated based on the selected effective components and their mixing weights in the current sample. This transforms complex unmixed signals into a single health indicator that farmers can understand.
[0075] The specific calculation process follows the following formula:
[0076]
[0077] in, A collection of effective independent components; For the first The hazard coefficient corresponding to each component is a constant, such as 1.0 for diseases and 0.6 for nutrient deficiency; For consistency score; These are the coefficients of the mixing matrix; Normalization factor; This represents the environmental baseline pressure value.
[0078] The core of this formula revolves around quantifying the combined effects of multiple stresses and achieving a quantitative assessment of growth status. It transforms unmixed independent stress signals into intuitive agronomic assessment indicators, taking into account signal validity, stress severity, and environmental baseline impacts. The logical design aligns with actual field monitoring needs. The formula integrates the individual contribution values of all effective stress components in a cumulative manner, then superimposes the environmental baseline stress value. This study aims to reconstruct the true growth state of rice in a natural field, influenced by both specific stresses and the basic environmental factors. The calculation of the contribution value of a single stress is progressively layered, first considering the biometric fingerprint consistency factor... With mixing matrix coefficients Multiplication ensures that only biologically reliable and effective signals are included, while also reflecting the relative concentration of the components of that effective stress signal; then, the maximum reflectance of healthy rice across the entire wavelength band is used. Normalization eliminates interference from environmental factors such as lighting on the spectral signal, ensuring the comparability of data under different monitoring scenarios. Finally, it is multiplied by a hazard coefficient. By assigning different weights based on the type of stress, the index highlights the dominant influence of high-risk stresses on rice growth, allowing it to accurately reflect the actual degree of harm caused by stress. The overall formula transforms abstract spectral unmixed signals into a quantifiable, comparable, and agronomically significant single index, providing direct and scientific numerical basis for rice health assessment and field management decisions.
[0079] Calculation example:
[0080] Assume that after S3 screening, there are two effective components.
[0081] Component A, characteristics of rice blast: damage coefficient Consistency score Mixed matrix coefficients Component B, nitrogen deficiency characteristics: hazard factor Consistency score Mixed matrix coefficients .
[0082] Normalization factor Environmental baseline stress value Component A contribution value: Component B contribution value: Overall Index: The index clearly indicates that rice is currently under moderate compound stress, with both diseases and nutrient deficiencies contributing to the problem.
[0083] Through the above calculations, the abstract spectral signal is transformed into a specific quantitative value, providing farmers with a direct basis for decision-making on whether or not to fertilize or spray pesticides.
[0084] To visually demonstrate the technical effects of this invention, Figure 2 The results of multi-scale feature screening analysis based on continuous wavelet transform are presented. The horizontal axis of the figure represents the wavelet transform scaling factor, and the vertical axis represents the scale-specific response index. A significant single peak appears on the curve at a wavelet scale of approximately 28, which is marked as the optimal feature scale. This visually verifies the effectiveness of the scale-specific response intensity index formula, indicating that the present invention can automatically identify specific frequency scales containing the richest biological features and accordingly set thresholds to filter out low signal-to-noise ratio background noise. Figure 3 The blind source separation and decoupling distribution of multiple stress signals is shown. The sample points in the figure are clearly divided into three regions: the healthy growth characteristic region, the nitrogen deficiency characteristic region, and the disease stress characteristic region. This intuitively verifies the decoupling ability of the biofingerprint consistency factor relationship and step S3. After the constrained ICA processing of this invention, the nitrogen deficiency signal and the disease signal are mapped onto orthogonal feature space axes, achieving perfect linear separability.
[0085] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A method for monitoring the growth state of rice based on hyperspectral imaging, characterized in that, Includes the following steps: The hyperspectral reflectance curve of the rice canopy was obtained. The hyperspectral reflectance curve reflects the spectral response characteristics of rice in different wavelength bands. The hyperspectral reflectance curve is mapped to the time-frequency space using continuous wavelet transform to obtain the wavelet coefficient matrix. Based on the scale-specific response intensity at different scales, the optimal feature scale is determined from the wavelet coefficient matrix and the feature vector is extracted. Scale-specific response intensity is used to measure the significance of a biological feature signal relative to background noise at a specific scale; The logic for determining the scale-specific response intensity is as follows: calculate the difference between the wavelet coefficient modulus and the background noise reference value, calculate the ratio of this difference to the sum of the local volatility and the numerical stability constant, and then weight the square of this ratio using a logarithmic term that includes the scale factor to obtain the scale-specific response intensity. Among them, the scale-specific response intensity is positively correlated with the difference and negatively correlated with the local volatility, and the logarithmic term is used to suppress the high-frequency noise weights at small scales. Based on the constraints of bio-fingerprint, the feature vector is decomposed into multiple independent components by independent component analysis. The effective stress signal components are screened out by calculating the bio-fingerprint consistency factor of each independent component. The bio-fingerprint consistency factor is used to measure the credibility of the isolated independent components in a biological sense. The logic for determining the bio-fingerprint consistency factor is as follows: calculate the weighted sum of the fingerprint similarity between independent components and each known stress type, and use the square of the weighted sum as the numerator; calculate the product of the absolute value of the kurtosis of the independent components and the adjustment coefficient, and use the sum of the product and the preset constant as the denominator; calculate the square root of the ratio of the numerator to the denominator to obtain the bio-fingerprint consistency factor. Among them, the biometric fingerprint consistency factor is positively correlated with fingerprint similarity and is inversely constrained by the absolute value of kurtosis to penalize components that only have statistical independence but lack biometric features. Based on the mixing weights of each effective stress signal component in the rice canopy and the corresponding hazard coefficients, a comprehensive growth inhibition index is calculated to complete the monitoring of rice growth status.
2. The method for monitoring the growth status of rice based on hyperspectral imaging according to claim 1, characterized in that, Obtain the hyperspectral reflectance curve of the rice canopy, including: Collect raw hyperspectral images of rice paddies; Radiometric correction was performed on the original hyperspectral image using standard whiteboard data and dark current data to obtain a reflectance image; Based on the normalized vegetation index, a mask threshold is set to remove the background area in the reflectance image and extract the region of interest for rice. The average reflectance across the entire wavelength range of the rice region of interest is calculated to obtain the hyperspectral reflectance curve. 3.The method for monitoring the growth state of rice based on hyperspectral imaging according to claim 1, characterized in that, The background noise baseline value is obtained by calculating the mean value of wavelet coefficients in non-plant areas of the image at the same scale; the local fluctuation is obtained by calculating the standard deviation of wavelet coefficients within a sliding window of a preset width centered on the current wavelength. 4.The method for monitoring the growth state of rice based on hyperspectral imaging according to claim 1, characterized in that, Independent component analysis (ICA) unmixing of feature vectors based on biometric fingerprint constraints includes: The eigenvectors are unmixed using the independent component analysis algorithm to obtain multiple independent component vectors; Introduce prior weights for standard stress types and standard stress spectral fingerprints; The similarity between each independent component vector and the standard stress spectral fingerprint is calculated. The bio-fingerprint consistency factor is calculated by combining the prior weights and the statistical properties of the independent component vectors.
5. The method for monitoring the growth status of rice based on hyperspectral imaging according to claim 4, characterized in that, Fingerprint similarity is calculated using cosine similarity; kurtosis absolute value is used to measure the non-Gaussianity of the signal. By introducing kurtosis absolute value, a soft constraint is applied to the noise components that excessively pursue the maximization of non-Gaussianity but deviate from biological reality. 6.The method for monitoring the growth state of rice based on hyperspectral imaging according to claim 1, characterized in that, The comprehensive growth inhibition index is used to quantitatively assess the overall negative impact on rice. The logic for determining the comprehensive growth inhibition index is as follows: For each screened effective stress signal component, calculate the product of its bio-fingerprint consistency factor and the mixing matrix coefficient, normalize the product using a normalization factor, and then multiply it by the hazard coefficient corresponding to the effective stress signal component to obtain the single stress contribution value of the effective stress signal component; sum up the single stress contribution values of all effective stress signal components and add the environmental baseline pressure value to obtain the comprehensive growth inhibition index.
7. The method for monitoring the growth status of rice based on hyperspectral imaging according to claim 6, characterized in that, The normalization factor is taken as the maximum value of the full-band reflectance of healthy rice samples to eliminate the influence of light intensity; the mixing matrix coefficients are determined by the weight of the corresponding pixel in the mixing matrix obtained by the inverse transformation of independent component analysis on the current independent component, representing the relative concentration of the effective stress signal component.
8. The method for monitoring rice growth status based on hyperspectral imaging according to claim 1, characterized in that, After calculating the comprehensive growth inhibition index, the following is also included: The current health level of rice is assessed based on the value of the comprehensive growth inhibition index. Based on the biofinite consistency factors and mixing weights of each effective stress signal component, the main stress sources leading to the increase in the comprehensive growth inhibition index are traced, and targeted field management recommendations are given.