A plant growth health state intelligent monitoring method and system based on multi-dimensional spectral feature analysis

Abstract

The present application belongs to the technical field of urban ecology and plant protection, and discloses a plant growth health state intelligent monitoring method and system based on multi-dimensional spectral feature analysis. The method comprises the following steps: constructing a specific multi-stress pathological spectral feature library in advance; acquiring target plant canopy multispectral image data in real time, calling the feature library after pretreatment to complete spectral matching comparison, and outputting canopy physiological state judgment conclusion; simultaneously starting the acoustic flow control aerosol detection device, using the sound field to enrich and focus the biological aerosol particles in the air, performing spectral analysis on the captured particle flow, extracting spectral dimension digital features, and independently judging the aerosol abnormality level; fusing the canopy judgment conclusion and the aerosol abnormality level for two-dimensional cross verification, and outputting the plant health state classification result and the grading early warning instruction. Through the dual-mode parallel monitoring and strong verification fusion decision of canopy spectrum and microenvironment aerosol spectrum, the present application realizes the early accurate classification and automatic early warning of plant stress.

Description

Technical Field

[0001] This invention relates to the field of urban ecology and plant protection technology, and in particular to an intelligent monitoring method and system for plant growth and health status based on multidimensional spectral feature analysis. Background Technology

[0002] In urban greening management, real-time and accurate monitoring of plant growth and health is a key prerequisite for achieving scientific fertilization, water-saving irrigation, and green pest and disease control. In recent years, plant canopy monitoring technology based on hyperspectral imaging has made some progress. By collecting plant canopy reflectance spectra and using machine learning models for disease identification, it can replace traditional manual inspections in some scenarios, initially realizing the automation and non-contact nature of monitoring.

[0003] However, existing technical solutions generally suffer from a fundamental flaw: they rely solely on single-dimensional information from the plant's canopy spectrum, completely ignoring crucial stress precursor signals from the plant microenvironment, such as pathogenic spores and bioaerosols. By the time identifiable abnormalities appear in the canopy spectrum, disease or nutrient imbalances have often already entered the mid-to-late stages of outbreak, and the early warning window has long been missed. Furthermore, single-modal spectral monitoring exhibits overlapping spectral response characteristics when facing different stress types such as nitrogen deficiency, rust infection, and drought stress. Systems lacking cross-validation mechanisms are highly prone to misjudgment, failing to distinguish the root cause of the stress and thus hindering the implementation of precise and targeted measures.

[0004] Therefore, this invention proposes an intelligent monitoring method and system for plant growth and health status based on multidimensional spectral feature analysis. Summary of the Invention

[0005] This invention provides an intelligent monitoring method and system for plant growth and health status based on multidimensional spectral feature analysis. By conducting dual-modal parallel monitoring and strong verification fusion decision-making of canopy multidimensional spectral features and acoustic fluid control aerosol spectra, it can achieve early, accurate, and classified automatic early warning during the stress incubation period. This fundamentally solves the industry pain points of existing single-modal spectral monitoring, such as lag, high misjudgment rate, and inability to distinguish specific stress types.

[0006] This invention provides an intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis, comprising the following steps: A specific multi-stress pathological spectral feature library was pre-constructed. The construction method included: organizing full-band reflectance spectral data that reflected the gradual process of stress, calibrated by real plant physicochemical indicators, into a multi-dimensional feature matrix according to stress type and time sequence. Real-time acquisition of multispectral image data of the target plant canopy; preprocessing of the multispectral image data to obtain reflectance spectral curves; calling a pre-constructed specific multi-stress pathological spectral feature library to complete spectral matching and comparison; and outputting the conclusion of the canopy physiological state determination. While acquiring multispectral images, the acoustic flow control aerosol detection device is activated to continuously draw in air from around the plant. The sound field is used to focus the target particles in the air onto the detection area. The captured particle stream is subjected to spectral analysis to extract the spectral dimension digital features that reflect the chemical composition and physical state of the particles, and the aerosol anomaly level is determined independently. The conclusions of canopy physiological status assessment and aerosol anomaly level are combined for two-dimensional cross-validation. After eliminating interference from a single signal, the plant health status classification results and graded early warning instructions are output.

[0007] Furthermore, the stress features stored in the specific multi-stress pathological spectral feature library include: Nutritional stress time-series spectral fingerprint, covering macro- and micro-element deficiencies; Spectral fingerprints of infectious diseases at each stage of invasion, covering fungal, bacterial, and viral diseases; Spectral fingerprint of environmental stress, covering moisture, salinity, temperature and radiation stress; In addition, the standard fingerprint of bioaerosols released by plants induced by nutritional stress, disease infection or environmental stress.

[0008] Furthermore, the multispectral image data is preprocessed to obtain the reflectance spectral curve, including performing radiometric calibration, atmospheric correction, smoothing and denoising, and normalization. When calling the pre-built specific multi-stress pathological spectral feature library to complete spectral matching and comparison, the vegetation index characterizing the physiological state of vegetation is calculated simultaneously, and the red edge parameter features are extracted as auxiliary discrimination features.

[0009] Furthermore, the acoustic fluid control aerosol detection device utilizes the standing wave field generated by the ultrasonic transducer array to achieve particle focusing. The sound field intensity and distribution are adaptively adjusted within a specified frequency and power range based on the aerodynamic particle size of the target pathogen spores sampled in real time.

[0010] Furthermore, the spectral dimension digital features extracted from the captured particle stream through spectral analysis specifically include: Characteristic peak positions reflect the chemical composition information of the captured bioaerosol particles; Peak intensity ratio characterizes the relative content ratio of different components in particles; Half-width at half-maximum (HWHM) reflects information about particle size distribution and aggregation state.

[0011] Furthermore, the specific multi-stress pathological spectral feature library employs the following update and security mechanisms: The database adopts a combination of local multiple redundant backups and cloud encrypted storage, supports historical version rollback, and is configured with hierarchical permission management, full auditing of operation logs, and anti-tampering mechanisms; It supports remote incremental updates triggered by preset cycles or events, and the update process does not interrupt the execution of monitoring tasks; It supports the automatic loading of localized spectral subsets based on geographical location, climate type, main crop, and dominant pests and diseases.

[0012] Furthermore, the conclusions of the canopy physiological state assessment and the aerosol anomaly level are integrated for two-dimensional cross-validation, including the following strong validation logic: When the canopy spectral characteristics match the probability of a specific nutrient deficiency, and the aerosol detection end does not capture abnormal pathogenic spores, it is determined to be the corresponding nutrient stress. When the canopy spectrum shows early stress variations, and the aerosol detection end determines that the concentration of a specific pathogen spores exceeds the preset risk threshold, it is judged as a high-risk disease infection. When the canopy spectrum indicates a drought response signal, and there is no risk of biological pathogens at the aerosol end, combined with synchronously monitored environmental parameters, it is determined to be drought stress. When no nutritional or pathological fingerprints from the feature library are extracted from the canopy or aerosols, but environmental parameter sensing shows extreme data, it is determined to be physical damage caused by environmental stress.

[0013] Furthermore, the spectral matching comparison employs a fusion of multiple similarity algorithms; Two-dimensional cross-validation is performed through a pre-trained multimodal fusion classification network, which receives canopy spectral features and aerosol features as input, performs feature layer fusion, and outputs the classification result of plant health status.

[0014] Furthermore, the acquisition mode of the canopy multispectral image data was further replaced by active excitation emission using an active pulsed light source with a specific band sequence; By exciting a specific photosystem inside the leaf with a pulse sequence, the chlorophyll fluorescence-induced kinetic curve was captured using a high-speed synchronous detection array. By analyzing the fluorescence quenching parameters and electron transport activity parameters in the kinetic curves, the damage status of photosynthetic organs can be assessed in depth, serving as a substitute and supplementary digital feature for the canopy reflectance spectrum.

[0015] This invention provides an intelligent monitoring system for plant growth and health status based on multidimensional spectral feature analysis, comprising: A multispectral image acquisition module is used to acquire multispectral image data of the target plant canopy in real time. The spectral image preprocessing module is used to perform reflectance inversion and noise reduction normalization on the original multispectral image data to obtain the reflectance spectral curve. The specific multi-stress pathological spectral feature library module stores standard spectral fingerprint data calibrated by gold standard that reflects the gradual process of multiple stresses. It supports redundant backup, remote incremental update, automatic loading of local subsets, and is configured with hierarchical permission management and operation log auditing functions. The canopy feature extraction and comparison module is used to calculate and extract spectral digital features from the reflectance spectrum curve and complete spectral matching and comparison to output the conclusion of the canopy physiological state. The acoustic fluid control aerosol spectral acquisition module, with a built-in acoustic field focusing cavity and spectral detection unit, is used to acquire spectra of enriched bioaerosol particle streams. The aerosol feature extraction module is used to extract the characteristic peak position, peak intensity ratio and full width at half maximum of particles, and to determine the anomaly level of bioaerosols. The dual-factor fusion decision control module receives the conclusion of the canopy physiological state and the aerosol anomaly level, executes a strong verification mechanism of dual-dimensional cross-validation, outputs the plant health status classification result and generates graded early warning instructions. Edge computing units are used to perform local inference in an offline state, with inference latency below a preset threshold; Multi-mode communication unit for remote push of early warning instructions and monitoring reports in multiple formats; The low-power power supply module combines solar energy and energy storage batteries to support long-term continuous operation in unattended scenarios.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: This invention overcomes the fundamental shortcomings of existing technologies, which rely solely on single-dimensional monitoring of plant canopy spectra and are unable to detect abnormalities in a timely manner or accurately distinguish stress types when early stress response signals are weak. By pioneering the parallel deployment of canopy multispectral multidimensional digital feature analysis and acoustic fluid-controlled bioaerosol spectral detection, a dual-evidence chain strong verification fusion decision-making system of "plant body signals" and "microenvironment pathogen precursor signals" is constructed. This allows early weak signals that are easily drowned out by noise or misjudged by single stress features to be confirmed or excluded through independent verification at the aerosol level. Thus, accurate classification and early warning are issued during the disease incubation period or the budding stage of nutrient imbalance, significantly advancing the intervention window, reducing false alarm rates at the source, and providing a reliable decision-making basis for subsequent on-demand fertilization, precision pesticide application, and water-saving irrigation, achieving the comprehensive goals of reducing pesticide and fertilizer use, improving quality and yield, and protecting the ecological environment.

[0017] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained through the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis in an embodiment of the present invention. Figure 2 This is a logic diagram for the two-dimensional cross-validation strong verification decision in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] refer to Figure 1 and Figure 2 This invention provides an embodiment of an intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis, comprising the following steps: A specific multi-stress pathological spectral feature library was pre-constructed. The construction method included: organizing full-band reflectance spectral data that reflected the gradual process of stress, calibrated by real plant physicochemical indicators, into a multi-dimensional feature matrix according to stress type and time sequence. Real-time acquisition of multispectral image data of the target plant canopy; preprocessing of the multispectral image data to obtain reflectance spectral curves; calling a pre-constructed specific multi-stress pathological spectral feature library to complete spectral matching and comparison; and outputting the conclusion of the canopy physiological state determination. While acquiring multispectral images, the acoustic flow control aerosol detection device is activated to continuously draw in air from around the plant. The sound field is used to focus the target particles in the air onto the detection area. The captured particle stream is subjected to spectral analysis to extract the spectral dimension digital features that reflect the chemical composition and physical state of the particles, and the aerosol anomaly level is determined independently. The conclusions of canopy physiological status assessment and aerosol anomaly level are combined for two-dimensional cross-validation. After eliminating interference from a single signal, the plant health status classification results and graded early warning instructions are output.

[0022] In this embodiment, the specific multi-stress pathological spectral feature library is a multi-dimensional feature matrix composed of a large amount of full-band reflectance spectral data calibrated by real plant physicochemical indicators. During library construction, monitoring points were set up in controlled experimental environments and natural stress areas in the field. Nutrient solutions with deficiencies such as nitrogen, phosphorus, and potassium were artificially applied to healthy plants in a gradient manner. Simultaneously, specific pathogens such as powdery mildew, rust, and rice blast were inoculated in a minimally invasive quantitative manner to induce a complete disease process from the incubation period to the late stage of disease. Equipment was also deployed in natural drought, salinity, high and low temperature stress areas in the field. Throughout the stress process, canopy hyperspectral images and laser-induced spectra of acoustically enriched aerosols were continuously and synchronously collected at high frequency. Each set of spectral data was labeled with the gold standard true value by combining leaf total nutrient measurement, chlorophyll a / b measurement, PCR nucleic acid detection, or ELISA antigen analysis. After outlier removal and standardization, the data were organized and entered into the library according to stress type and time sequence, forming a multi-dimensional feature matrix. The stress type dimension distinguishes different types of stress, while the time series dimension records the gradual process of the same stress from healthy to severe stress. The database establishes multi-level indexes according to crop type, stress type, disease stage, and collection time, supporting millisecond-level fast retrieval.

[0023] In this embodiment, the multispectral image data is preprocessed to obtain the reflectance spectral curve. Specifically, the process includes: first, performing absolute radiometric calibration on the original DN value to convert it into physical reflectance; then, performing atmospheric correction using the 6S radiative transfer model to eliminate interference from illumination changes and atmospheric scattering; subsequently, using Savitzky-Golay smoothing filtering combined with wavelet transform threshold denoising to remove high-frequency noise and outliers; and finally, normalizing the data distribution scale using standard normal variable transformation.

[0024] In this embodiment, spectral matching and comparison are performed by calling a pre-constructed specific multi-stress pathological spectral feature library. This involves extracting multi-dimensional spectral digital features from the pre-processed reflectance spectral curve and then calculating the similarity with the standard spectral fingerprints in the feature library. The extracted multi-dimensional features include: reflectance at key wavelength positions in the visible light 400-780 nm band and near-infrared 780-1700 nm band; red edge position, red edge slope, and red edge area calculated using first-derivative spectroscopy; normalized vegetation index, enhanced vegetation index, photochemical reflectance index, structurally insensitive pigment index, improved chlorophyll absorption reflectance index, optimized soil-regulated vegetation index, and chlorophyll index. During matching, similarity is calculated using spectral angle mapping algorithm, cosine similarity algorithm, Euclidean distance algorithm, and spectral information divergence fusion. Simultaneously, support vector machine, convolutional neural network, long short-term memory network, lightweight gradient booster, and stacked fusion model are used for classification reasoning, outputting the probability distribution and confidence level of health, nutrient deficiency, disease, or environmental stress as the conclusion for determining the physiological state of the canopy.

[0025] In this embodiment, the working process of the acoustic fluid control aerosol detection device is as follows: a flat optical glass microchannel is selected as the acoustic fluid control cavity; a micro brushless air pump is used to draw air from the surrounding area of ​​the plant at a stable flow rate; after the airflow enters the cavity, the PZT piezoelectric transducer generates ultrasonic standing waves in the 1 to 5 MHz frequency band, focusing pathogenic spores and bioaerosol particles with a particle size range of 0.3 to 100 micrometers in the air onto the center of the channel, with a focusing accuracy of no more than 30 micrometers; when a single particle passes through the detection window, it is irradiated by a laser of a specific wavelength, and the scattering spectrum or autofluorescence spectrum generated by the particle is collected. The extracted aerosol digital features include: characteristic peak position, reflecting the chemical composition of the particle; peak intensity ratio, characterizing the relative content ratio of different components; and half-width at half-maximum (WHM), reflecting the particle size distribution. The extracted features are compared with the standard fingerprints of pathogenic spores and stress-induced volatile organic compound-derived aerosols in the feature library. When specific pathogenic spore characteristics appear and the intensity is abnormal, the aerosol abnormality level is independently determined to be increased.

[0026] In this embodiment, the conclusions of the canopy physiological state assessment and the aerosol anomaly level are combined for two-dimensional cross-validation, executing the following strong verification judgment logic: when the canopy spectrum matches a high probability of nitrogen deficiency and no pathogenic spores are captured in the aerosol, it is judged as nitrogen deficiency stress; when the canopy spectrum shows small variations in early stress and the concentration of rust spores in the aerosol exceeds a preset risk threshold, it is judged as high-risk rust infection; when the canopy spectrum characterizes a drought response signal and there is no biological pathogenic risk in the aerosol, it is judged as drought stress in combination with soil volumetric moisture content; when no nutritional or pathological fingerprints are extracted from the canopy or aerosol, but soil moisture content, conductivity, pH, air temperature and humidity, and light parameters show extreme values, it is judged as physical damage caused by environmental stress. The two-factor cross-validation is not a simple superposition of two signals, but a strong verification mechanism of mutual verification. When only one dimension is abnormal while the other dimension is normal, the system does not trigger an early warning and continues to accumulate evidence, thereby effectively eliminating false alarms caused by interference from a single signal.

[0027] Furthermore, the stress features stored in the specific multi-stress pathological spectral feature library include: Nutritional stress time-series spectral fingerprint, covering macro- and micro-element deficiencies; Spectral fingerprints of infectious diseases at each stage of invasion, covering fungal, bacterial, and viral diseases; Spectral fingerprint of environmental stress, covering moisture, salinity, temperature and radiation stress; In addition, the standard fingerprint of bioaerosols released by plants induced by nutritional stress, disease infection or environmental stress.

[0028] In this embodiment, the stress features stored in the specific multi-stress pathological spectral feature library specifically include four categories: The first category is the spectral fingerprint of nutritional stress, covering macro-element deficiencies such as nitrogen, phosphorus, and potassium, as well as micro-element deficiencies such as calcium, magnesium, sulfur, iron, manganese, zinc, copper, boron, and molybdenum. Each type of deficiency records a gradual spectrum from mild to severe deficiency; the second category is the spectral fingerprint of each stage of infectious disease invasion, covering true diseases such as powdery mildew, rust, rice blast, and sheath blight. The four categories of stress characteristics are: 1) complete spectral changes from the incubation period and early stage of bacterial diseases such as bacterial wilt and viral diseases such as mosaic virus, to full outbreak; 2) environmental stress spectral fingerprints, covering the spectral responses to drought, waterlogging, salinity, alkali damage, high temperature, low temperature, and UV-B radiation stress; and 3) standard fingerprints of bioaerosols released by plants induced by the above-mentioned nutrient stress, disease infection, and environmental stress, including the characteristic spectra of pathogen spores and the characteristic spectra of stress-specific volatile organic compound-derived aerosols. All four types of stress characteristics are stored in a multi-dimensional feature matrix in the form of standard spectral curves, indexed by crop type, stress type, disease stage, and collection time, providing a complete reference for subsequent canopy spectral matching and aerosol anomaly determination.

[0029] Furthermore, the multispectral image data is preprocessed to obtain the reflectance spectral curve, including performing radiometric calibration, atmospheric correction, smoothing and denoising, and normalization. When calling the pre-built specific multi-stress pathological spectral feature library to complete spectral matching and comparison, the vegetation index characterizing the physiological state of vegetation is calculated simultaneously, and the red edge parameter features are extracted as auxiliary discrimination features.

[0030] In this embodiment, the multispectral image data is preprocessed to obtain the reflectance spectral curve. Specifically, this includes: first, performing absolute radiometric calibration on the original DN values ​​of the multispectral images to convert the original signals into physical reflectance; then, performing atmospheric correction using the 6S radiative transfer model to eliminate environmental interference such as changes in sunlight, atmospheric scattering, dust, and fog; subsequently, using a combination algorithm of Savitzky-Golay smoothing filtering and wavelet transform threshold denoising to remove high-frequency noise and abnormal abrupt changes; and finally, normalizing the distribution scale of data collected at different times and under different lighting conditions through standard normal variable transformation to make the full-band reflectance spectral curves comparable.

[0031] In this embodiment, when calling a pre-constructed specific multi-stress pathological spectral feature library to complete spectral matching and comparison, vegetation indices characterizing the physiological state of vegetation are calculated simultaneously, and red-edge parameter features are extracted as auxiliary discrimination features. Vegetation indices include the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EDI), Photochemical Reflectance Index (PRRI), Structure-Insensitive Pigment Index (SRI), Improved Chlorophyll Absorption Reflectance Index (ERRI), Optimized Soil-Regulating Vegetation Index (SRI), and Chlorophyll Index. Red-edge parameter features include the red-edge position, slope, and area calculated using first-derivative spectroscopy. These indices and parameters reflect the physiological state of plants from different perspectives, such as photosynthetic capacity, pigment ratio, cell structure integrity, and light energy utilization efficiency. Together with the reflectance spectral curve itself, they constitute a multi-dimensional discrimination criterion, effectively improving the sensitivity to distinguishing early stresses after being input into the spectral matching and comparison process.

[0032] Furthermore, the acoustic fluid control aerosol detection device utilizes the standing wave field generated by the ultrasonic transducer array to achieve particle focusing. The sound field intensity and distribution are adaptively adjusted within a specified frequency and power range based on the aerodynamic particle size of the target pathogen spores sampled in real time.

[0033] In this embodiment, the acoustic fluid control aerosol detection device utilizes a standing wave field generated by an ultrasonic transducer array to achieve particle focusing. The sound field intensity and distribution are adaptively adjusted based on the aerodynamic particle size of the target pathogen spores sampled in real time. Specifically, the device uses a front-mounted optical scattering particle size sensor to measure the aerodynamic particle size distribution of aerosol particles entering the microchannel in real time. Based on the measurement results, it automatically selects the optimal resonant frequency within the 1 to 5 MHz frequency band and dynamically adjusts the driving voltage amplitude within the corresponding power range to match the acoustic radiation force of the standing wave field with the size and mass of the currently dominant particle size. When the field microenvironment is dominated by small-sized bacterial pathogens or viral vectors, the system automatically increases the resonant frequency and enhances the sound field intensity to maintain focusing accuracy; when the field microenvironment is dominated by large-sized fungal spores such as rust urediniospores, the system automatically decreases the resonant frequency and appropriately reduces the power to reduce shear damage to particles within the flow channel. This adaptive adjustment mechanism enables bioaerosol particles of different sizes to be stably focused at the center of the channel, with a focusing accuracy that never exceeds 30 micrometers. This ensures the accuracy and stability of subsequent spectral analysis under complex and variable field microenvironment conditions.

[0034] Furthermore, the spectral dimension digital features extracted from the captured particle stream through spectral analysis specifically include: Characteristic peak positions reflect the chemical composition information of the captured bioaerosol particles; Peak intensity ratio characterizes the relative content ratio of different components in particles; Half-width at half-maximum (HWHM) reflects information about particle size distribution and aggregation state.

[0035] In this embodiment, the spectral dimensions extracted from the captured particle stream through spectral analysis specifically include three items: characteristic peak position, i.e., the wavelength position corresponding to the absorption or emission peak on the spectral curve. Different pathogenic spores will form characteristic absorption or fluorescence emission peaks at specific wavelengths due to differences in chemical components such as cell wall polysaccharides, proteins, and lipids, thereby distinguishing pathogen types; peak intensity ratio, i.e., the intensity ratio between different characteristic peaks. Since the relative content ratio of each component within the same pathogenic spore is relatively stable, abnormal changes in the peak intensity ratio can indicate spore activity or infection stage; and half-width at half-maximum (HWHM), i.e., the width of the characteristic peak at half-peak height. Monodisperse, uniformly sized spore groups correspond to a narrower HWHM, while the HWHM increases significantly when the particle size distribution is broad or particles aggregate, thereby determining the size distribution and aggregation state of aerosol particles. The combined interpretation of these three features can comprehensively characterize the captured particles from three dimensions: chemical composition, content ratio, and physical state, achieving accurate identification of pathogenic spores.

[0036] Furthermore, the specific multi-stress pathological spectral feature library employs the following update and security mechanisms: The database adopts a combination of local multiple redundant backups and cloud encrypted storage, supports historical version rollback, and is configured with hierarchical permission management, full auditing of operation logs, and anti-tampering mechanisms; It supports remote incremental updates triggered by preset cycles or events, and the update process does not interrupt the execution of monitoring tasks; It supports the automatic loading of localized spectral subsets based on geographical location, climate type, main crop, and dominant pests and diseases.

[0037] In this embodiment, the specific multi-stress pathological spectral feature library adopts the following update and security mechanisms: Regarding data backup and security protection, the database employs a dual-redundancy backup system with a local embedded multimedia storage card and an external storage card, along with a three-tiered linkage scheme of edge gateway local caching and cloud-encrypted storage. Data loss at any node can be fully recovered from other nodes, supporting historical version rollback and retrospective. Simultaneously, hierarchical permission management is configured, allowing different roles to access corresponding levels of spectral data according to their authorized scope. All operations are audited and recorded by the system throughout the process, and core spectral files are subject to tamper-proof digital signatures to prevent malicious modification or illegal copying. Regarding updates, the database supports remote incremental updates. Update strategies include monthly data repair updates based on preset cycles, quarterly additions of regionally specific diseases and new varieties' spectra, and annual model iterations and library version upgrades. All update packages undergo encrypted transmission and legality verification, and the update process does not interrupt the execution of front-end monitoring tasks. In terms of localization adaptation, the system automatically filters and loads the corresponding localized spectral subsets from the cloud database based on the geographical location, climate type, soil conditions, main crop varieties, and historical pest and disease occurrence patterns of the deployment site. This ensures that the edge matching and comparison always operates within a feature space that is highly relevant to the local stress scenario, thereby improving the recognition accuracy and reducing computational overhead.

[0038] Furthermore, the conclusions of the canopy physiological state assessment and the aerosol anomaly level are integrated for two-dimensional cross-validation, including the following strong validation logic: When the canopy spectral characteristics match the probability of a specific nutrient deficiency, and the aerosol detection end does not capture abnormal pathogenic spores, it is determined to be the corresponding nutrient stress. When the canopy spectrum shows early stress variations, and the aerosol detection end determines that the concentration of a specific pathogen spores exceeds the preset risk threshold, it is judged as a high-risk disease infection. When the canopy spectrum indicates a drought response signal, and there is no risk of biological pathogens at the aerosol end, combined with synchronously monitored environmental parameters, it is determined to be drought stress. When no nutritional or pathological fingerprints from the feature library are extracted from the canopy or aerosols, but environmental parameter sensing shows extreme data, it is determined to be physical damage caused by environmental stress.

[0039] In this embodiment, the conclusions of the canopy physiological state assessment and the aerosol anomaly level are combined for two-dimensional cross-validation, and the following four strong validation judgment logics are executed: The first rule states that when the canopy spectral characteristics match a specific nutrient deficiency probability and the aerosol detection device does not capture any abnormal pathogenic spores, the system determines that the plant is under corresponding nutrient stress. For example, if the canopy reflectance spectrum shows a blue shift at the red edge and the normalized vegetation index decreases, exhibiting typical nitrogen deficiency spectral characteristics, and the acoustic flow-controlled aerosol detection device does not detect any abnormal pathogenic spore characteristic peaks during the same period, after these two independent pieces of evidence are mutually verified, the system confirms that the plant is currently under nitrogen deficiency nutrient stress rather than disease infection, thus guiding the precise application of nitrogen fertilizer rather than blindly spraying fungicides.

[0040] Article 2: When the canopy spectrum shows early stress variations and the aerosol detection end determines that the concentration of a specific pathogenic spore exceeds a preset risk threshold, it is judged as a high-risk disease infection. For example, the canopy spectrum shows slight reflectance anomalies in the 550 nm green peak band and the 750 nm near-infrared plateau band, and no visible lesions have yet formed. However, the aerosol detection end simultaneously captures the characteristic peak position of rust urediniospores, and the peak intensity ratio continues to rise, with the concentration exceeding the historical statistical threshold. After the two pieces of evidence corroborate each other, the system issues a high-risk rust infection warning during the disease incubation period.

[0041] Article 3: When the canopy spectrum indicates a drought response signal and there is no risk of biological pathogens at the aerosol end, drought stress is determined by combining the synchronously monitored soil volumetric moisture content. For example, a decrease in reflectance and a reduction in red edge slope in the near-infrared band of the canopy spectrum indicates reduced leaf water content and damaged cell structure, while there are no abnormal characteristics of pathogenic spores at the aerosol end. At the same time, the volumetric moisture content returned by the soil moisture sensor buried in the root zone is lower than the preset threshold for the crop's growth stage. After cross-confirmation of the three pieces of information, drought stress is determined and precision irrigation recommendations are triggered.

[0042] Article 4. When no nutrient deficiency or pathological fingerprints from the feature library are extracted from the canopy or aerosols, but the synchronous monitoring data of soil moisture content, electrical conductivity, pH, air temperature and humidity, and light parameters show extreme values, it is determined to be physical damage caused by environmental stress. For example, if the overall canopy spectrum is abnormal but there is no specific nutrient deficiency or disease fingerprint matching, and aerosol detection shows no signs of pathogenic spores, while environmental sensors show that the temperature continuously exceeds 40 degrees Celsius or the soil electrical conductivity is far beyond the normal range, the system comprehensively judges it to be physical damage caused by high temperature heat damage or saline-alkali stress.

[0043] The above four judgment logics together constitute the core of the two-factor strong verification mechanism. The two independent evidences, canopy spectroscopy and aerosol detection, are not simply superimposed but mutually verified: when only one signal is abnormal while the other is normal, the system will not trigger an early warning and will continue to accumulate evidence, thereby effectively eliminating false alarms caused by interference such as changes in illumination, dust obstruction, and instantaneous environmental fluctuations in single-modal monitoring, and outputting highly reliable stress classification results.

[0044] Furthermore, the spectral matching comparison employs a fusion of multiple similarity algorithms; Two-dimensional cross-validation is performed through a pre-trained multimodal fusion classification network, which receives canopy spectral features and aerosol features as input, performs feature layer fusion, and outputs the classification result of plant health status.

[0045] Furthermore, the acquisition mode of the canopy multispectral image data was further replaced by active excitation emission using an active pulsed light source with a specific band sequence; By exciting a specific photosystem inside the leaf with a pulse sequence, the chlorophyll fluorescence-induced kinetic curve was captured using a high-speed synchronous detection array. By analyzing the fluorescence quenching parameters and electron transport activity parameters in the kinetic curves, the damage status of photosynthetic organs can be assessed in depth, serving as a substitute and supplementary digital feature for the canopy reflectance spectrum.

[0046] In this embodiment, spectral matching and comparison employs a fusion of multiple similarity algorithms, specifically including spectral angle mapping, cosine similarity, Euclidean distance, and spectral information divergence. The spectral angle mapping algorithm uses the angle formed by the measured spectral vector and the standard spectral vector in the feature library in high-dimensional space as a similarity measure; the smaller the angle, the closer the shapes of the two spectral curves are. The cosine similarity algorithm assesses the directional consistency of the two vectors in the feature space. The Euclidean distance algorithm calculates the absolute deviation of the two spectral curves in reflectance values ​​at each band. Spectral information divergence measures the degree of information deviation between the two spectra from the perspective of probability distribution differences. After each of the four algorithms independently calculates its similarity, a weighted voting mechanism is used to output a comprehensive matching probability, avoiding the problem of insufficient sensitivity of a single algorithm under specific spectral shapes.

[0047] In this embodiment, dual-dimensional cross-validation is performed through a pre-trained multimodal fusion classification network. The network is constructed as follows: a one-dimensional convolutional neural network branch receives and processes the canopy reflectance spectral curve, automatically extracting hierarchical waveform features from the spectral curve; a fully connected neural network branch receives structured parameters such as vegetation indices, red-edge parameters, and aerosol digital features; the feature vectors extracted by the two branches are fused at a splicing layer to form a joint feature vector containing both global spectral waveform information and multi-dimensional parameter quantification information, which is then fed into a classifier consisting of a fully connected layer and a Softmax classification layer, outputting the probability distribution of plant health status as healthy, nutrient deficient, diseased, or under environmental stress. During network training, the input consists of paired samples labeled with stress types. Each sample contains a canopy reflectance spectral curve and a corresponding set of aerosol three-dimensional features. The output is a classification result consistent with the labeled type. The network parameters are iteratively optimized through backpropagation until convergence. In application, the canopy spectral features and aerosol features are used as dual-channel inputs to the network. After feature layer fusion, the classification layer outputs the final plant health status classification result.

[0048] This invention provides an embodiment of an intelligent monitoring system for plant growth and health status based on multidimensional spectral feature analysis, comprising: A multispectral image acquisition module is used to acquire multispectral image data of the target plant canopy in real time. The spectral image preprocessing module is used to perform reflectance inversion and noise reduction normalization on the original multispectral image data to obtain the reflectance spectral curve. The specific multi-stress pathological spectral feature library module stores standard spectral fingerprint data calibrated by gold standard that reflects the gradual process of multiple stresses. It supports redundant backup, remote incremental update, automatic loading of local subsets, and is configured with hierarchical permission management and operation log auditing functions. The canopy feature extraction and comparison module is used to calculate and extract spectral digital features from the reflectance spectrum curve and complete spectral matching and comparison to output the conclusion of the canopy physiological state. The acoustic fluid control aerosol spectral acquisition module, with a built-in acoustic field focusing cavity and spectral detection unit, is used to acquire spectra of enriched bioaerosol particle streams. The aerosol feature extraction module is used to extract the characteristic peak position, peak intensity ratio and full width at half maximum of particles, and to determine the anomaly level of bioaerosols. The dual-factor fusion decision control module receives the conclusion of the canopy physiological state and the aerosol anomaly level, executes a strong verification mechanism of dual-dimensional cross-validation, outputs the plant health status classification result and generates graded early warning instructions. Edge computing units are used to perform local inference in an offline state, with inference latency below a preset threshold; Multi-mode communication unit for remote push of early warning instructions and monitoring reports in multiple formats; The low-power power supply module combines solar energy and energy storage batteries to support long-term continuous operation in unattended scenarios.

[0049] In this embodiment, the overall workflow of the system's various modules working together is as follows: the multispectral image acquisition module acquires canopy multispectral image data in real time; the spectral image preprocessing module performs radiometric calibration, atmospheric correction, smoothing, denoising, and normalization on the raw data to obtain the reflectance spectral curve; the canopy feature extraction and comparison module extracts spectral digital features from the reflectance spectral curve and calls the standard spectral fingerprint data stored in the specific multi-stress pathological spectral feature library module to complete the matching and comparison, and outputs the canopy physiological state judgment conclusion; at the same time, the acoustic fluid control aerosol spectral acquisition module uses the acoustic field focusing cavity to focus airborne particles and then the spectral detection unit collects the particle spectrum; the aerosol feature extraction module extracts the characteristic peak position, peak intensity ratio, and half-width at half-maximum and determines the aerosol anomaly level; the dual-factor fusion decision control module receives the above two independent conclusions and executes a strong verification mechanism, outputs the plant health status classification result through the multimodal fusion classification network, and generates graded early warning instructions; the edge computing unit completes all local inference in offline mode, the multimodal communication unit remotely pushes the early warning instructions and monitoring reports to the user terminal, and the low-power power supply module ensures the long-term continuous operation of the system in unattended scenarios.

[0050] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent monitoring of plant growth and health status based on multidimensional spectral feature analysis, characterized in that, Includes the following steps: A specific multi-stress pathological spectral feature library was pre-constructed. The construction method included: organizing full-band reflectance spectral data that reflected the gradual process of stress, calibrated by real plant physicochemical indicators, into a multi-dimensional feature matrix according to stress type and time sequence. Real-time acquisition of multispectral image data of the target plant canopy; preprocessing of the multispectral image data to obtain reflectance spectral curves; calling a pre-constructed specific multi-stress pathological spectral feature library to complete spectral matching and comparison; and outputting the conclusion of the canopy physiological state determination. While acquiring multispectral images, the acoustic flow control aerosol detection device is activated to continuously draw in air from around the plant. The sound field is used to focus the target particles in the air onto the detection area. The captured particle stream is subjected to spectral analysis to extract the spectral dimension digital features that reflect the chemical composition and physical state of the particles, and the aerosol anomaly level is determined independently. The conclusions of canopy physiological status assessment and aerosol anomaly level are combined for two-dimensional cross-validation. After eliminating interference from a single signal, the plant health status classification results and graded early warning instructions are output.

2. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1, characterized in that, The stress features stored in the specific multi-stress pathological spectral feature library include: Nutritional stress time-series spectral fingerprint, covering macro- and micro-element deficiencies; Spectral fingerprints of infectious diseases at each stage of invasion, covering fungal, bacterial, and viral diseases; Spectral fingerprint of environmental stress, covering moisture, salinity, temperature and radiation stress; In addition, the standard fingerprint of bioaerosols released by plants induced by nutritional stress, disease infection or environmental stress.

3. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1, characterized in that, The multispectral image data is preprocessed to obtain the reflectance spectral curve, including radiometric calibration, atmospheric correction, smoothing and denoising, and normalization. When calling the pre-built specific multi-stress pathological spectral feature library to complete spectral matching and comparison, the vegetation index characterizing the physiological state of vegetation is calculated simultaneously, and the red edge parameter features are extracted as auxiliary discrimination features.

4. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1, characterized in that, The acoustic fluid control aerosol detection device uses the standing wave field generated by the ultrasonic transducer array to achieve particle focusing. The sound field intensity and distribution are adaptively adjusted within a specified frequency and power range based on the aerodynamic particle size of the target pathogen spores sampled in real time.

5. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1, characterized in that, The spectral dimension digital features extracted from the captured particle stream through spectral analysis specifically include: Characteristic peak positions reflect the chemical composition information of the captured bioaerosol particles; Peak intensity ratio characterizes the relative content ratio of different components in particles; Half-width at half-maximum (HWHM) reflects information about particle size distribution and aggregation state.

6. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1, characterized in that, The specific multi-stress pathological spectral feature library employs the following update and security mechanisms: The database adopts a combination of local multiple redundant backups and cloud encrypted storage, supports historical version rollback, and is configured with hierarchical permission management, full auditing of operation logs, and anti-tampering mechanisms; It supports remote incremental updates triggered by preset cycles or events, and the update process does not interrupt the execution of monitoring tasks; It supports the automatic loading of localized spectral subsets based on geographical location, climate type, main crop, and dominant pests and diseases.

7. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1, characterized in that, The conclusions of the canopy physiological status assessment and the aerosol anomaly level are combined for two-dimensional cross-validation, including the following strong validation logic: When the canopy spectral characteristics match the probability of a specific nutrient deficiency, and the aerosol detection end does not capture abnormal pathogenic spores, it is determined to be the corresponding nutrient stress. When the canopy spectrum shows early stress variations, and the aerosol detection end determines that the concentration of a specific pathogen spores exceeds the preset risk threshold, it is judged as a high-risk disease infection. When the canopy spectrum indicates a drought response signal, and there is no risk of biological pathogens at the aerosol end, combined with synchronously monitored environmental parameters, it is determined to be drought stress. When no nutritional or pathological fingerprints from the feature library are extracted from the canopy or aerosols, but environmental parameter sensing shows extreme data, it is determined to be physical damage caused by environmental stress.

8. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1 or 7, characterized in that, Spectral matching and comparison employs a fusion of multiple similarity algorithms; Two-dimensional cross-validation is performed through a pre-trained multimodal fusion classification network, which receives canopy spectral features and aerosol features as input, performs feature layer fusion, and outputs the classification result of plant health status.

9. The intelligent monitoring method for plant growth and health status based on multidimensional spectral feature analysis according to claim 1, characterized in that, The acquisition mode of canopy multispectral image data has been further changed to active excitation emission using an active pulsed light source with a specific band sequence; By exciting a specific photosystem inside the leaf with a pulse sequence, the chlorophyll fluorescence-induced kinetic curve was captured using a high-speed synchronous detection array. By analyzing the fluorescence quenching parameters and electron transport activity parameters in the kinetic curves, the damage status of photosynthetic organs can be assessed in depth, serving as a substitute and supplementary digital feature for the canopy reflectance spectrum.

10. A smart monitoring system for plant growth and health status based on multidimensional spectral feature analysis, characterized in that, include: A multispectral image acquisition module is used to acquire multispectral image data of the target plant canopy in real time. The spectral image preprocessing module is used to perform reflectance inversion and noise reduction normalization on the original multispectral image data to obtain the reflectance spectral curve. The specific multi-stress pathological spectral feature library module stores standard spectral fingerprint data calibrated by gold standard that reflects the gradual process of multiple stresses. It supports redundant backup, remote incremental update, automatic loading of local subsets, and is configured with hierarchical permission management and operation log auditing functions. The canopy feature extraction and comparison module is used to calculate and extract spectral digital features from the reflectance spectrum curve and complete spectral matching and comparison to output the conclusion of the canopy physiological state. The acoustic fluid control aerosol spectral acquisition module, with a built-in acoustic field focusing cavity and spectral detection unit, is used to acquire spectra of enriched bioaerosol particle streams. The aerosol feature extraction module is used to extract the characteristic peak position, peak intensity ratio and full width at half maximum of particles, and to determine the anomaly level of bioaerosols. The dual-factor fusion decision control module receives the conclusion of the canopy physiological state and the aerosol anomaly level, executes a strong verification mechanism of dual-dimensional cross-validation, outputs the plant health status classification result and generates graded early warning instructions. Edge computing units are used to perform local inference in an offline state, with inference latency below a preset threshold; Multi-mode communication unit for remote push of early warning instructions and monitoring reports in multiple formats; The low-power power supply module combines solar energy and energy storage batteries to support long-term continuous operation in unattended scenarios.

Images