Hyperspectral crop disease and pest early dynamic detection method and system

Abstract

The application relates to the field of testing and analysis technology, and discloses a hyperspectral crop disease and pest early dynamic detection method and system, which comprises the following steps: constructing a circadian rhythm response vector baseline by collecting baseline hyperspectral images of crops in a key period; calculating the deviation degree of the circadian rhythm response vector of the day from the baseline in a monitoring period; and generating a warning signal when the deviation degree continuously abnormally changes. The application realizes the super-early identification of the initial stage of disease and pest infection by dynamically monitoring the physiological rhythm change of crops instead of static spectral characteristics. Meanwhile, the directionality characteristics of the asynchronous phase vector are used to analyze the stress type, and the diagnosis specificity is improved through micro-disturbance response verification, so that the traditional disease identification is changed from post-judgment to pre-warning, and reliable technical support is provided for precision agriculture.

Description

Technical Field

[0001] This invention relates to a method and system for early dynamic detection of crop diseases and pests using hyperspectral imaging, belonging to the field of testing and analysis technology. Background Technology

[0002] In the field of agricultural image recognition, using optical methods to analyze the health status of crops is one of the mainstream technologies in this field. Existing solutions mainly rely on static comparison of spectral features of a single time phase or on judging abnormalities by setting a preset vegetation index threshold, such as marking diseases when the NDVI value is below 0.3 or judging the status by the matching degree between daily spectral data and health templates.

[0003] While such methods can identify visible symptoms, they have two fundamental limitations: 1. Pest and disease infection is essentially a gradual physiological disorder process, such as chlorophyll synthesis lagging behind water transport disorder. Static analysis only captures the absolute value difference at a certain moment and cannot perceive the temporal coordinated fluctuations across wavebands; 2. In order to suppress interference such as gradual changes in light intensity and differences in soil background, existing technologies generally introduce multi-sensor fusion or complex compensation algorithms, which leads to a surge in system complexity and makes it difficult for edge devices to handle.

[0004] In recent years, with the upgrading of precision agriculture's demand for early warning, the industry has attempted to build time-series models through continuous multi-day sampling, such as LSTM networks to predict spectral trends. However, these improvements have exposed new contradictions: model training requires massive amounts of labeled samples, but early cases in the field are rare; the dramatic increase in parameters leads to inference latency exceeding the milliwatt-level power consumption constraints of agricultural IoT nodes, drastically reducing the feasibility of actual deployment. Deep-seated technical contradictions are thus highlighted: existing technologies fail to reveal the dynamic correlation between the intrinsic physical properties of crops and progressive physiological disorder; their detection mechanisms either remain at the comparison of static physical quantities or introduce complex external models, failing to effectively capture the inherent rhythmic changes of materials from the perspective of measurement principles; and environmental noise suppression relies on additional hardware or algorithms, while the stringent cost constraints of agricultural edge scenarios force the system to be extremely simple. Therefore, how to achieve lightweight perception of dynamic anomalies in the intrinsic physiological rhythms of crops without relying on complex models and multi-source hardware has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a method and system for early dynamic detection of crop diseases and pests using hyperspectral imaging. Its main purpose is to solve the problem that existing technologies cannot simultaneously achieve dynamic rhythm perception and lightweight edge implementation.

[0006] To achieve the above objectives, the present invention provides a hyperspectral method for early dynamic detection of crop diseases and pests, comprising the following steps:

[0007] Step a: During a baseline learning cycle, acquire baseline hyperspectral images of the target crop daily at a first specific time period and a second specific time period; the first specific time period is defined as after sunrise, the initial stage of photosynthesis, and the second specific time period is defined as noon, the peak period of photosynthesis.

[0008] Step b: Based on the intensity values ​​of the chlorophyll absorption peak band sensitive to photosynthesis and the intensity values ​​of the water absorption peak band sensitive to water status in the baseline hyperspectral image, calculate the normalized intensity value change rate vector corresponding to each pixel from the first specific time period to the second specific time period.

[0009] Step c: Perform time series smoothing on the normalized intensity value change rate vector obtained for each day within the baseline learning period to dynamically establish a daily rhythmic response vector baseline representing the health status of the target crop.

[0010] Step d: Within a monitoring cycle, hyperspectral images are collected daily during the first and second specific time periods. Based on the same chlorophyll absorption peak band intensity value and water absorption peak band intensity value in the hyperspectral images, the daily circadian rhythm response vector is calculated.

[0011] Step e: Based on the daily circadian rhythm response vector and the daily circadian rhythm response vector baseline, calculate the asynchronous phase vector, which represents the vector difference between the daily circadian rhythm response vector and the daily circadian rhythm response vector baseline.

[0012] Step f involves comparing the magnitude of the asynchronous phase vector with a dynamic statistical threshold determined based on the inherent fluctuation characteristics of the daily rhythm response vector baseline.

[0013] Step g: When the magnitude of the asynchronous phase vector is greater than the dynamic statistical threshold for two or more consecutive days, it is determined that the target crop has an abnormal physiological rhythm and an early warning signal for pests and diseases is generated.

[0014] Preferably, the normalized intensity value change rate vector is constructed by calculating the normalized intensity change of the chlorophyll absorption peak band intensity value from the first specific time period to the second specific time period, and the normalized intensity change of the water absorption peak band intensity value from the first specific time period to the second specific time period; or, it is constructed by calculating the relative change rate of the intensity values ​​of the chlorophyll absorption peak band and the water absorption peak band from the first specific time period to the second specific time period, and using the relative change rate as a component of the normalized intensity value change rate vector.

[0015] Preferably, asynchronous phase vector The calculation method is as follows: ,in, This represents the daily rhythm response vector for that day. Represents the baseline of the daily rhythm response vector; asynchronous phase vector Length of the module Reflects the overall intensity of rhythm deviation, asynchronous phase vector The direction reflects an abnormality in the synchronicity between different physiological activities.

[0016] Preferably, before acquiring the monitoring hyperspectral image in step d, a step for verifying the effectiveness of illumination conditions is included: continuously sampling the intensity value of the chlorophyll absorption peak band at high frequency to obtain an instantaneous intensity sequence; calculating the instantaneous rate of change variance of the instantaneous intensity sequence; when the instantaneous rate of change variance continuously exceeds a preset illumination mutation threshold, the current illumination condition is determined to be an invalid illumination mutation condition; when determined to be an invalid illumination mutation condition, compensation processing is performed on the currently acquired monitoring hyperspectral image, which includes abandoning the current data acquisition and using the effective rhythm state of the previous day, or re-sampling after the illumination intensity returns to stability.

[0017] Preferably, after generating the early pest and disease warning signal in step g, the following stress type attribution step is further included: acquiring and storing multiple consecutive asynchronous phase vectors within a preset time window to form an asynchronous phase time-series trajectory; extracting the geometric morphological features of the asynchronous phase time-series trajectory, including the linear fit goodness of the trajectory, the average curvature of the trajectory, the vector rotation direction of the trajectory, and the rate of change of the magnitude of the asynchronous phase vector over time; performing feature distance matching between the geometric morphological features and a preset stress prototype trajectory library; and determining the stress attribution type of the early pest and disease warning signal based on the matching results.

[0018] Preferably, the stress prototype trajectory library is a lookup table containing at least the following prototype trajectory features defined based on plant pathology knowledge: acute disease prototypes, characterized by high linearity, high speed and a trajectory pointing to a specific physiological quadrant; nutrient imbalance prototypes, characterized by low linearity, high curvature or spiral and low speed trajectories; and transient environmental stress recovery prototypes, characterized by a trajectory that shows an arc shape returning to the origin.

[0019] Preferably, after generating the early pest and disease warning signal in step g, the method further includes a root zone stress analysis step: periodically applying a standardized, short-term humidity micro-perturbation to the root zone of the target crop, the humidity micro-perturbation being achieved by the intelligent irrigation system stopping water supply in advance in the early morning or precisely spraying a standard amount of clean water at noon; during a preset recovery period after the application of the humidity micro-perturbation, continuously monitoring the intensity value of the water absorption peak band and the intensity value of a band on the near-infrared high reflectivity platform as a reference baseline at a high frequency; calculating the response recovery rate of the water absorption peak band intensity value relative to the reference baseline band intensity value; and determining whether the early pest and disease warning signal is related to root zone stress by comparing the response recovery rate with a first health recovery rate threshold.

[0020] Preferably, when the response recovery rate is less than the first health recovery rate threshold, it is determined that the early pest and disease warning signal is highly correlated with root zone stress.

[0021] Preferably, the method further includes a step of sensing soil moisture status: within the baseline learning period, acquiring the diurnal rhythm residual sequence of the diurnal rhythm response vectors corresponding to the chlorophyll absorption peak band intensity value and the water absorption peak band intensity value relative to the diurnal rhythm response vector baseline; calculating the correlation coefficient between the chlorophyll absorption peak band residual and the water absorption peak band residual in the diurnal rhythm residual sequence; determining the soil moisture status by comparing the correlation coefficient with a preset health correlation coefficient range, the preset health correlation coefficient range being -0.2 to 0.2; when the correlation coefficient continuously exceeds the preset health correlation coefficient range, determining that the soil moisture status is abnormal, and outputting a soil moisture warning signal in parallel; when the correlation coefficient is less than -0.2, determining that the soil has begun to dry; when the correlation coefficient is greater than 0.2, determining that the soil is approaching saturation.

[0022] A hyperspectral dynamic detection system for early-stage crop diseases and pests, characterized by comprising: an image acquisition module configured to acquire hyperspectral images of the target crop daily during a first specific time period and a second specific time period within a baseline learning period and a monitoring period; the first specific time period is defined as after sunrise, the initial stage of photosynthesis, and the second specific time period is defined as noon, the peak period of photosynthesis; and a data processing module connected to the image acquisition module, configured to calculate, based on the intensity values ​​of chlorophyll absorption peak bands sensitive to photosynthesis and water absorption peak bands sensitive to water status in the acquired hyperspectral images, the normalized intensity value change rate vector corresponding to each pixel from the first specific time period to the second specific time period; and to process the data acquired daily within the baseline learning period. The normalized intensity change rate vector is subjected to time series smoothing to dynamically establish a daily rhythm response vector baseline representing the health status of the target crop. The anomaly detection module, connected to the data processing module, is configured to calculate an asynchronous phase vector based on the daily rhythm response vector and the daily rhythm response vector baseline. The asynchronous phase vector represents the vector difference between the daily rhythm response vector and the daily rhythm response vector baseline. It is also configured to compare the magnitude of the asynchronous phase vector with a dynamic statistical threshold determined based on the fluctuation characteristics of the daily rhythm response vector baseline. The early warning generation module, connected to the anomaly detection module, is configured to generate an early pest and disease warning signal when the magnitude of the asynchronous phase vector is continuously greater than the dynamic statistical threshold for two or more consecutive days.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] 1. By capturing the spectral response changes of crops during key periods such as the photosynthetic initiation and peak periods, a diurnal rhythm response vector is constructed. This transforms the traditional static test based on a single physical quantity into a comprehensive analysis method based on the dynamic correlation characteristics of multiple physical quantities. When the asynchronous phase vector continuously deviates from the healthy baseline, the system can keenly identify physiological disorder states that are not visible to the naked eye, thereby achieving early warning before tissue lesions are caused in the early stage of pathogen infection. This mechanism transmits health information through changes in the crop's own rhythmic patterns, avoiding the environmental interference problem that relies on absolute value comparison.

[0025] 2. The directional characteristics of asynchronous phase vectors and geometric trajectory analysis form a synergistic logic: the directional deviation of the vector reveals the priority anomalies of the photosynthetic or water system; the curvature and linearity differences of the vector trajectory over multiple consecutive days, such as high linearity trajectory pointing to acute diseases and spiral trajectory pointing to chronic malnutrition; the combination of the two enables the system not only to determine the existence of anomalies, but also to analyze the type of stress, providing a deeper basis for precision agricultural decision-making.

[0026] 3. By applying standardized humidity micro-perturbations to the reused irrigation system and analyzing the dynamic characteristics of the 970nm water absorption band during the recovery period, such as the delayed recovery rate, the latent state of root function can be inverted. After triggering an initial abnormal alarm, this mechanism actively verifies whether the source of stress is related to the root zone, upgrading passive monitoring to a hypothesis-verification scientific diagnostic closed loop, thereby improving the credibility of the alarm. Based on the changes in the correlation coefficient of the diurnal rhythm response residuals, such as the sudden increase in the negative correlation between the 680nm and 970nm residuals, the system simultaneously captures soil moisture anomalies. This mechanism transforms the noise byproducts of the main scheme calculation process into an indicator signal of soil moisture transport status, expanding the parallel monitoring dimension of the crop growth environment without adding new hardware and data acquisition. Attached Figure Description

[0027] Figure 1 This is a flowchart of the hyperspectral method for early dynamic detection of crop diseases and pests according to the present invention.

[0028] Figure 2 This is a dynamic response curve diagram for pest and disease early warning in this invention;

[0029] Figure 3 This is a flowchart of the hyperspectral crop pest and disease detection process of the present invention.

[0030] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0031] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0032] This application provides a hyperspectral method for early dynamic detection of crop diseases and pests, enabling proactive and high-precision health monitoring of crop growth status. This avoids the fundamental limitations of existing technologies in early disease and pest identification, such as the static nature of dynamic processes and sensitivity to environmental disturbances. In a specific application scenario, such as the refined management of high-value-added economic crops like wine grapes, the operation of this hyperspectral method and system for early dynamic detection of crop diseases and pests can be described as follows: The core of this system and method lies in abandoning the static comparison of absolute spectral values ​​at a single time point. Instead, it deeply explores and quantifies the inherent physiological rhythm dynamics of the plant in response to changes in sunlight, thereby capturing pathogens in the very early stages of infection or environmental stress, before macroscopic symptoms appear. To detect rhythmic anomalies caused by physiological disorder, this method begins with a baseline learning cycle. This cycle is designed to establish a physiological rhythm reference standard representing the absolute health state of a target crop in a specific plot or greenhouse. The cycle duration needs to cover a complete and stable growth phase. For example, for grapes, 7 to 14 consecutive sunny days without extreme weather can be selected in the early fruiting stage. During this cycle, image acquisition modules deployed in the field, such as tracked robots or fixed gantry cranes equipped with hyperspectral imagers, will acquire hyperspectral images of the target crop canopy at two key first and second specific time periods each day to obtain a series of baseline hyperspectral images. The first specific time period is precisely defined as after sunrise, at the initial stage of photosynthesis. At this moment, after a night's rest, chloroplasts begin to respond to light, and their physiological activities change most dramatically. This is a key window for observing their activation efficiency. In engineering practice, this window can be set to 30 to 60 minutes after sunrise. The second specific time period is defined as noon, which is the peak period of photosynthesis, usually within one hour before or after the local solar altitude angle is at its maximum. At this time, the crop's photosynthesis and transpiration reach their peak, reflecting the ultimate carrying capacity of its physiological functions. The technical purpose of selecting these two time periods is to maximize the capture of the range of intraday physiological activity changes, thereby amplifying the rhythmic differences between healthy and potential stress states.

[0033] The acquired hyperspectral images are data cubes containing information from hundreds of consecutive bands. The core of the method lies in not utilizing all spectral information, but focusing on characteristic bands directly related to the two core life activities of crops: photosynthesis and water metabolism. Specifically, the data processing module extracts the intensity value of the chlorophyll absorption peak band, which is most sensitive to the state of photosynthesis, from the hyperspectral data of each pixel. This band is usually located in the red light region around 680 nm. At the same time, it extracts the intensity value of the water absorption peak band, which is extremely sensitive to the water state within the plant. This band is usually selected in the near-infrared region around 970 nm. Then, it calculates the normalized intensity change rate vector for each pixel. The construction of this vector aims to transform the spectral intensity readings at two key moments in a day into a two-dimensional vector that can characterize the dynamics of intraday physiological rhythms. In a specific implementation, this vector consists of two components: one is the normalized intensity change of the chlorophyll absorption peak band intensity value from the first specific time period to the second specific time period; the other is the normalized intensity change of the water absorption peak band intensity value within the same time span. Normalization is crucial for measuring light intensity, as it avoids systematic errors introduced by variations in daily total light intensity or minute changes in the distance between the sensor and the crop. For example, it can be calculated by subtracting the peak intensity from the initiation intensity and then adding the peak intensity to the initiation intensity for each band, making the rate of change a dimensionless relative value that focuses on the trend rather than absolute readings. Thus, each pixel obtains a two-dimensional vector for the day, which represents the diurnal rhythm response of that point in the physiological space of photosynthetic change-water state change. Next, time-series smoothing is performed on all normalized intensity value rate of change vectors obtained daily throughout the baseline learning period. This process aims to filter out random errors caused by minor meteorological fluctuations or sensor noise during daily acquisition, extracting a stable and reliable diurnal rhythm response vector baseline that represents the healthiest physiological state of the target crop at the current growth stage, denoted as vector B. This processing can use algorithms such as exponentially weighted moving average or Kalman filtering. The final baseline B is a two-dimensional vector whose direction and magnitude together define the standard diurnal rhythm pattern of a healthy crop.

[0034] After baseline learning is completed, the system enters a continuous monitoring cycle. The image acquisition module acquires hyperspectral images daily using the exact same time period and method as the baseline learning. The data processing module then uses the same bands and algorithms to calculate the daily circadian rhythm response vector, denoted as... To ensure the validity of the monitoring data, a light condition validity verification step can be included. Short-term, drastic changes in light, such as rapid cloud cover, can cause dramatic changes in spectral intensity. These changes are not due to crop physiology, and without verification, misjudgment can occur. This step is implemented as follows: Before formally acquiring and monitoring hyperspectral images, a short-term, high-frequency continuous sampling of the chlorophyll absorption peak intensity value is performed, for example, sampling once per second within one minute, to obtain an instantaneous intensity sequence containing 60 data points. Then, the instantaneous rate of change of this sequence is calculated, i.e., the sequence of intensity differences between adjacent sampling points, and the variance of this rate of change sequence is further calculated. When this variance continuously exceeds a preset light change threshold, the current light condition is determined to be zero. The threshold for determining invalid light change conditions is set by balancing the sensitivity to identifying dramatic changes in actual light intensity with the tolerance for minor fluctuations in normal light intensity. A threshold that is too low will lead to excessively frequent data acquisition interruptions, while a threshold that is too high may overlook light interference that significantly affects data quality. Therefore, this threshold is usually set based on historical meteorological data and sensor noise characteristics, as a reasonable engineering value that can distinguish between random noise and significant cloud cover events. When an invalid light change condition is determined, the system will perform compensation processing. Compensation processing may include abandoning the current data acquisition and using the valid circadian rhythm state from the previous day, or resampling after the light intensity returns to stability, to ensure data consistency and reliability and obtain the valid daily circadian rhythm response vector for the current day. Then, the asynchronous phase vector is calculated, i.e. The calculation method is as follows , here This refers to the previously established daily rhythm response vector baseline. The physical meaning of this vector subtraction operation is to translate the daily rhythm response vector into the physiological state space with the endpoint of the healthy baseline vector as the origin; therefore, the asynchronous phase vector... It visually represents the direction and extent to which the daily rhythm response deviates from the healthy baseline, and its magnitude... It reflects the overall intensity of the rhythm deviation, that is, the degree of abnormality in the physiological state; while its direction reflects the abnormality in the synchronicity between the two major physiological activities of photosynthesis and water metabolism, containing deeper diagnostic information, for example... If the direction is predominantly negative on the water change axis, it may suggest that stress initially affects the plant's water transport system, and subsequently, the system will adjust the magnitude of the asynchronous phase vector. The data is compared to a dynamic statistical threshold, which is not a fixed empirical value but is dynamically determined based on the fluctuation characteristics of the daily rhythm response vector baseline itself within the learning period. Specifically, the system calculates the daily vector and the final smoothed baseline within the baseline learning period. The deviations between the mean and the standard deviation of the modulus are calculated, and a reasonable upper limit for tolerance is set based on the statistical distribution of these deviations, such as the mean plus three times the standard deviation. This adaptive threshold strategy enables the detection method to adapt to the inherent rhythmic fluctuations of crops of different varieties and at different growth stages, greatly improving its universality and accuracy.

[0035] Finally, when the magnitude of the asynchronous phase vector is greater than the dynamic statistical threshold for two or more consecutive days, the system determines that the target crop has experienced a physiological rhythm abnormality and immediately generates an early warning signal for pests and diseases. Setting the judgment condition of two or more consecutive days is a filtering mechanism in the time dimension. Its purpose is to further eliminate the transient rhythm disturbances caused by special environmental factors on a single day, such as slight light abnormalities or short-term drastic temperature differences that cannot be completely filtered out. This ensures that the warning signal points to a continuous abnormal state driven by internal physiological disorder, thereby significantly improving the reliability of the warning. After the system generates an early warning signal for pests and diseases, in order to provide managers with more instructive decision-making basis, a stress type attribution step can be further executed. This step first acquires and stores multiple consecutive asynchronous phase vectors within a preset time window, such as the past 5-7 days, to form an asynchronous phase time series trajectory. Subsequently, the system extracts the geometric features of the trajectory, including: the linear fit goodness of the trajectory, used to measure the stability and directionality of the deviation process; the average curvature of the trajectory, used to measure the volatility of the deviation process; the vector rotation direction of the trajectory; and the rate of change of the magnitude of the asynchronous phase vector over time. These extracted geometric features are matched with a pre-defined stress prototype trajectory library by feature distance matching. This library is a lookup table built based on plant pathology knowledge, which stores typical trajectory features corresponding to different stress types. The library includes at least: acute disease prototypes, characterized by high linearity, high speed, and trajectories pointing to specific physiological quadrants, which are consistent with the rapid and destructive characteristics of diseases such as downy mildew; and nutrient imbalance prototypes, characterized by low linearity, high curvature, or spiral and low speed trajectories, which reflect the slow and nonlinear physiological compensation process of plants due to nutrient imbalance.The system also identifies a prototype for recovery from brief environmental stress, characterized by an arc-shaped trajectory returning to the origin. This corresponds to the plant's recovery process after experiencing a brief stress (such as irrigation after drought). By calculating the characteristic distance between the current trajectory and the prototype trajectories in the database, the system can determine the most likely stress attribution type for this warning signal with a certain degree of confidence. Furthermore, if the warning signal may be caused by root problems, the system can initiate a root zone stress identification step. This is an active, verified, closed-loop diagnostic process. Through a networked intelligent irrigation system, the system periodically applies a standardized, brief humidity perturbation to the root zone of the target crop. This perturbation is designed to elicit an observable response rather than create new stress on the crop. For example, during irrigation... The system stops water supply in the morning or precisely sprays a standard amount of clean water at noon. During a preset recovery period after the disturbance, the system continuously monitors the intensity of the water absorption peak band at a high frequency, and simultaneously monitors the intensity of a band on a near-infrared high-reflectivity platform as a reference baseline. By calculating the response recovery rate of the former relative to the latter, the root system's water absorption function can be quantified. This rate is compared to a first health recovery rate threshold, which is calibrated by conducting the same micro-disturbance experiment on confirmed healthy crops. When the measured response recovery rate is less than the first health recovery rate threshold, it can be determined that this early pest and disease warning signal is highly correlated with root zone stress, such as root rot or soil compaction.

[0036] This invention can also achieve simultaneous sensing of soil moisture status without increasing any additional hardware or data acquisition costs. This function utilizes a byproduct of the main process calculation. Specifically, the system records the daily rhythm residual sequence corresponding to the intensity values ​​of the chlorophyll absorption peak band and the water absorption peak band, i.e., the daily... Components and baseline Based on the difference between the corresponding components, the system calculates the correlation coefficient between the two residual sequences within the most recent sliding time window. The underlying physiological principle is that under healthy soil moisture conditions, the daily fluctuations in photosynthetic efficiency and water transport efficiency are relatively decoupled. The correlation coefficient between their residuals will be within a preset healthy correlation coefficient range close to zero, i.e., between -0.2 and 0.2. When the correlation coefficient continuously exceeds the preset healthy correlation coefficient range, the soil moisture state is judged to be abnormal. Specifically, when the soil begins to dry out, water becomes the main limiting factor, which will simultaneously inhibit water transport and photosynthesis, resulting in a strong negative correlation between the two residuals. That is, when the correlation coefficient is less than -0.2, it is judged that the soil has begun to dry out. Conversely, when the soil approaches saturation, root hypoxia may simultaneously affect both systems, resulting in a positive correlation between the residuals. That is, when the correlation coefficient is greater than 0.2, it is judged that the soil is approaching saturation. Therefore, the system can output soil moisture early warning signals in parallel. Corresponding to the above method, a high The spectral crop pest and disease early dynamic detection system comprises four core modules: an image acquisition module, configured to acquire hyperspectral images at specified times; a data processing module, connected to the image acquisition module, which is the system's computational core, configured to extract key band intensities from hyperspectral data, calculate the normalized intensity value change rate vector, and dynamically establish the daily rhythm response vector baseline through time series smoothing; an anomaly detection module, connected to the data processing module, configured to calculate the asynchronous phase vector based on the daily rhythm response vector and the daily rhythm response vector baseline, and compare its magnitude with a dynamic statistical threshold; and finally, an early warning generation module, connected to the anomaly detection module, configured to generate an early pest and disease warning signal when the comparison result of the anomaly detection module shows that the magnitude of the asynchronous phase vector is continuously greater than the dynamic statistical threshold for two or more consecutive days. These four modules work together to form a complete technical closed loop from data acquisition, baseline learning, dynamic monitoring to intelligent early warning.

[0037] Example 1: This example is set in an intensive, high-yield facility agriculture production environment. Its system characteristics include extremely high requirements for monitoring crop health status, coupled with complex environmental and physiological fluctuations caused by instantaneous cloud cover, changes in physiological background at different growth stages, and fine-tuning of irrigation strategies. The superposition of these technical conditions leads to an inherent contradiction between high false alarm rates and low sensitivity in traditional detection methods that rely on fixed spectral thresholds or static health templates. The root of this contradiction lies in the fact that traditional methods attempt to separate a weak absolute spectral signal caused by early diseases from a background full of environmental noise. This process is computationally extremely complex and unreliable. This invention does not attempt to suppress or compensate for dynamic changes in the external environment, but rather establishes the crop's response pattern to the most important environmental driver, namely the day cycle, as... The core monitoring signal, by constructing a daily rhythm response vector baseline, establishes a dynamic and self-consistent health reference system for crops. Therefore, the detection task is no longer to find signals from noise, but to reconstruct the identification of deviations from a stable dynamic pattern. This transformation allows common-mode interference introduced by environmental factors such as changes in light intensity to be naturally canceled out in vector subtraction operations, thereby achieving the purification of the monitoring signal without increasing any additional hardware complexity. Under this architecture, the system further resolves the technical conflict between sensitivity and specificity that is common in early warning. The system's anomaly detection module first achieves extremely high monitoring sensitivity by continuously comparing the magnitude of the asynchronous phase vector with a dynamic statistical threshold established based on the volatility of the health baseline itself, ensuring that any slight signs of deviation from the normal physiological rhythm can be captured in a timely manner.

[0038] However, to avoid unnecessary alarms caused by transient disturbances or non-critical stresses, the system incorporates a deterministic, hierarchically triggered diagnostic logic. Only after the initial, high-sensitivity warning conditions are met—that is, after the early pest and disease warning signal is formally generated—will the system selectively initiate higher-level analysis processes aimed at improving diagnostic specificity. For example, it might initiate a stress type attribution step, distinguishing stress types by analyzing the geometric morphology of asynchronous phase time-series trajectories, or initiate a root zone stress identification step, verifying whether the anomaly originates from the roots by actively applying humidity micro-perturbations and observing their response recovery rate. This mechanism, combining universal high-sensitivity monitoring with trigger-based high-specificity diagnosis, achieves optimal allocation of system resources within a single technical architecture, ensuring the complete capture of potential risks and guaranteeing the high reliability of the final diagnostic conclusion. Furthermore, the design of this scheme demonstrates… A profound system synergy effect is achieved by transforming computational byproducts in the main process into independent and valuable monitoring dimensions. During the calculation of asynchronous phase vectors, the system naturally generates diurnal rhythm residual sequences of chlorophyll absorption peak bands and water absorption peak bands. In traditional data processing workflows, these residuals are often regarded as random noise that needs to be filtered out. However, the accompanying soil moisture sensing step in this scheme uses these two residual sequences as its core inputs. By calculating the correlation coefficient between the two, the system can gain insight into the changes in the coupling relationship between photosynthesis and water metabolism caused by changes in soil moisture. For example, when soil drought causes water to become a limiting factor, the two residual sequences will show a significant negative correlation. This design allows the system to perform its core pest and disease early warning task while outputting high-value information about soil moisture status in parallel without any additional sensors or data collection.

[0039] Example 2: This example objectively verifies the superiority of the present invention compared to traditional static spectral analysis methods under real stress conditions, and its effectiveness in attribution analysis of different stress types. A comparative verification experiment was designed and executed. The experiment focused on two points: first, quantifying the time advantage of this scheme in issuing early warnings before visible symptoms appear in crops; second, verifying the reliability of stress type matching based on the geometric characteristics of asynchronous phase time-series trajectories. The experimental platform was built in a plant growth chamber where environmental parameters could be precisely controlled. Tomato seedlings with clear physiological responses were selected as the experimental subjects, divided into three groups of 20 seedlings each, grown in their respective independent hydroponic systems. Experiment A... Group A was designated as the acute stress group, Group B as the chronic stress group, and Group C as the control group, which maintained optimal growth conditions throughout the experimental period. All plants first underwent a 10-day baseline learning period. The duration of this period was fundamentally designed to achieve an optimal balance between the stability of the baseline model and the speed of experimental deployment. For fast-growing herbaceous plants like tomatoes, 10 days is sufficient to cover the establishment of their physiological rhythms from the seedling stage to the stable vegetative growth stage, ensuring that the established daily rhythm response vector baseline accurately reflects their dynamic patterns under healthy conditions. An excessively long period might introduce unnecessary physiological background changes as the plant enters the next developmental stage. During this period, hyperspectral images were acquired daily using the method of this invention at the initial and peak stages of photosynthesis, establishing healthy daily rhythmic response vector baselines for all plants. After baseline learning, the experiment officially began on a specific day. Group A's nutrient solution was inoculated with a quantitative suspension of downy mildew spores to simulate the rapid infection process of an acute disease; Group B's nutrient solution was replaced with a formula containing half the nitrogen concentration to simulate the gradual effects of chronic nutrient imbalance. Subsequently, two data monitoring methods were performed daily on all three groups of plants: first, the asynchronous phase vector (APV) modulus of each plant was calculated using the method of this invention; second, the normalized vegetation index (NVI), commonly used in remote sensing, was calculated and recorded. This data served as a transmission mechanism. Using the static detection method as a reference, the experiment lasted for 12 days. At T0+8, the leaves of Group A plants first showed very slight, visible chlorotic spots on their edges. In Group B plants, at T0+10, the lower, older leaves began to show a identifiable, uniform pale yellow color. The key data trend throughout the process was that the APV modulus of Groups A and B began to be significantly and consistently higher than the natural fluctuation range of Group C from the second day after stress application. However, the NDVI values ​​of both groups were not statistically different from Group C at this point. It wasn't until T0+8 that the mean NDVI value of Group A showed its first statistically significant decrease. To more clearly demonstrate this core difference, the monitoring data at some key points are summarized below:

[0040] Table 1: Changes in APV modulus and NDVI value under different stresses.

[0041]

[0042] The underlying mechanism of the aforementioned data phenomena lies in the fact that the asynchronous phase vector monitored in this invention represents the asynchrony of the diurnal rhythm response between the two core physiological systems of plant photosynthesis and water use. Pathogen infection or nutrient imbalance initially triggers this functional disorder, which only accumulates over sufficient time and evolves into macroscopic physical symptoms such as decreased chlorophyll content or cell structure damage. Therefore, abnormal APV modulus length can capture stress signals at least 6 days earlier than the NDVI index, which depends on chlorophyll content. Furthermore, during the period from T0 to T0+8 days, the asynchronous phase time trajectory of experimental group A... The physiological trajectory exhibits an approximately straight line with a stable slope, which highly matches the characteristics of the acute disease prototype defined in the stress prototype trajectory library. In contrast, the trajectory of experimental group B shows a slow speed, uncertain direction, and high curvature, which is consistent with the characteristics of the nutrient imbalance prototype. This experiment, by monitoring the deviation of the crop's internal diurnal rhythm response vector, can provide an effective early warning window for pests, diseases, or physiological stress several days earlier than the traditional static spectral index method. At the same time, the experimental results also verify the effectiveness of stress attribution based on the geometric characteristics of asynchronous phase time-series trajectories, and can clearly distinguish between acute and chronic stress.

[0043] Example 3: This example combines Figures 1 to 3 This paper describes the implementation of a hyperspectral method and system for early dynamic detection of crop diseases and pests. For example... Figure 1 As shown, during the baseline learning period (7–14 days), the administrator initiates the data acquisition task. The image acquisition module acquires images twice daily during the first specific time period (30–60 minutes after sunrise) and the second specific time period (midday peak). Continuous hyperspectral images are transmitted to the data processing module, which extracts the intensity values ​​of the 680nm chlorophyll absorption peak band and the 970nm water absorption peak band, calculates the normalized intensity change rate vector, and performs time-series smoothing on the multi-day data. Finally, a baseline B for the diurnal rhythm response vector of healthy crops is established. After entering the monitoring period, the image acquisition module acquires monitoring images daily during the same time period and transmits the images to the data processing module, extracting the corresponding characteristic band intensities and calculating the diurnal rhythm response vector for that day. Then The data is transmitted to the anomaly detection module to calculate the asynchronous phase vector. And further calculate its modulus. Then compare it with the dynamic statistical threshold set by the system: if If the value exceeds the threshold for two consecutive days or more, an early warning condition is triggered, and the early warning generation module generates an early pest and disease warning signal accordingly; if If the crop does not exceed the threshold, it is determined that the crop is in normal health and the system continues to monitor it daily until the administrator terminates the task.

[0044] like Figure 2 As shown in the figure, the horizontal axis represents the time (in days) after induction, marking various time points starting from time T0, indicating the changes in pest and disease response at different times after induction; the vertical axis represents... Modulus length (relative value) is used to characterize the degree of abnormality in the physiological rhythm of crops at different time points. The figure shows three different sets of experimental data: Group A (acute stress), Group B (chronic stress), and Group C (control group), represented by different symbols: Group A is represented by a solid black line, Group B by a box symbol with a dashed line, and Group C by a triangle symbol with a dotted line. Based on the curves in the figure, it can be seen that Group A (acute stress)... The modulus length rapidly exceeded the dynamic statistical threshold on day T0+2 and continued to increase thereafter, forming a clear early warning trigger point; this indicates that acute diseases or acute stress can rapidly alter the physiological rhythms of crops, thereby triggering early warning signals of pests and diseases, while group B (chronic stress)... The modulus length changed relatively gradually, only beginning to deviate from the dynamic statistical threshold on T0+8 day, indicating that the effect of chronic stress on crop physiological rhythms is relatively slow and gradual. Group C (control group) The modulus length basically did not exceed the dynamic statistical threshold, which shows the normal physiological rhythm changes of healthy crops; the dynamic statistical threshold (dashed line) in the figure is the reference standard set by the system, which is used to detect whether there are significant abnormalities in the physiological rhythm of crops.

[0045] like Figure 3 As shown, the process includes four main stages: baseline learning, daily monitoring, anomaly detection and diagnosis, and diagnostic report output. In the baseline learning stage, the system first establishes a baseline for the daily rhythm response vector of healthy crops by acquiring hyperspectral images. In the daily monitoring stage, the system acquires hyperspectral images of the target crop daily and calculates the daily rhythm response vector. During this stage, the system determines whether to proceed to the anomaly detection and diagnosis stage by judging whether the asynchronous phase vector (i.e., the deviation of the daily rhythm response) is abnormal. If the system detects an anomaly in the crop's rhythm, such as abnormal changes for more than two consecutive days, it generates an early warning signal. In the anomaly detection and diagnosis stage, the system further analyzes the stress type attribution based on the anomaly, including analyzing the dynamic characteristics of stress and combining root zone humidity analysis for diagnosis. If the stress type attribution is clear, the system outputs a corresponding diagnostic report and agricultural recommendations to help agricultural managers take appropriate countermeasures.

[0046] Example 4: During the deployment phase of this example, several key parameters and thresholds need to be systematically calibrated. Regarding the dynamic statistical threshold, the determination procedure aims to establish an objective standard that can define the range of normal physiological fluctuations. This procedure is executed after the baseline learning cycle ends. First, all historical asynchronous phase vectors between the daily rhythm response vectors within this cycle and the finally established daily rhythm response vector baseline B are summarized, and the magnitudes of these vectors are calculated. This forms an empirical distribution dataset of magnitudes representing the rhythm deviation in a healthy state. The specific value of the dynamic statistical threshold is determined as a high quantile statistical value of this dataset. For example, the 95th percentile ensures that the threshold is adaptively generated based on the inherent physiological fluctuation characteristics of a specific crop at a specific growth stage, and serves as a fixed judgment benchmark in subsequent monitoring cycles. Similarly, the preset light mutation threshold on which the effectiveness of light conditions is determined is established as follows: During the baseline learning period, under multiple clear weather conditions without cloud cover, high-frequency continuous sampling of the chlorophyll absorption peak intensity is performed multiple times to obtain multiple instantaneous intensity sequences. For each sequence, the variance of its instantaneous rate of change is calculated. These variance values ​​together constitute a representation of the system's stability under normal light conditions. The background noise distribution includes both noisy and slightly fluctuating environmental noise. A preset threshold for sudden light changes is set as a statistically significant upper limit for this background noise distribution, such as the mean of the distribution plus four standard deviations. This effectively separates drastic light changes caused by events like rapid cloud movement from the background noise. Regarding the optional stress type attribution step, the feature distance matching logic of the stress prototype trajectory library is crucial. Its core lies in transforming multi-dimensional geometric features into a single, comparable distance metric. This matching process first compares the geometric feature values ​​extracted from the trajectory to be analyzed, such as trajectory linearity and average curvature. Normalization is performed to map the vectors to a uniform numerical range to eliminate dimensional differences. Then, the comprehensive difference between the normalized feature vector of the trajectory to be analyzed and the feature vector of each prototype trajectory in the library is calculated. This comprehensive difference is determined by calculating the absolute value of the difference between the corresponding components of the two vectors one by one, and then summing the absolute values ​​of the differences of all components by weight. The weight coefficients of each component are set based on prior knowledge of plant pathology, which aims to amplify the influence of the features that are most distinguishable to stress types. Finally, the prototype trajectory with the smallest comprehensive difference from the trajectory to be analyzed is determined to be the most likely stress attribution type.

[0047] In the root zone stress identification step, the calculation of the response recovery rate is refined. After applying a micro-perturbation of humidity, the high-frequency sequence of water absorption peak intensity collected by the system is first compared point-by-point with the intensity value of a near-infrared high-reflectivity plateau band that has the same trend of light influence, generating a ratio sequence that has eliminated the influence of light fluctuations. This ratio sequence depicts the net physiological response curve of water state recovering from the perturbation point to steady state. The response recovery rate is then quantified as the linear fitting slope of this net physiological response curve in the initial recovery stage. This slope value objectively reflects the response efficiency of root water transport function to changes in demand. Finally, regarding the steps accompanying the sensing of soil moisture status, the preset range of health correlation coefficients can be refined through calibration in specific applications. This calibration process includes, within the baseline learning period, subjecting soil moisture to a controlled process of change from optimal to slight drought or near saturation, and simultaneously recording the correlation coefficients of the daily rhythm residual sequences. By analyzing the actual distribution range of these correlation coefficients under different moisture conditions, a more accurate range of health correlation coefficients representing the suitable soil moisture state can be established for a specific combination of target crop and soil, thereby improving the reliability of early warning of abnormal soil moisture.

[0048] Example 5: During the deployment phase of this example, several key parameters and thresholds need to be systematically calibrated. The procedure for determining the dynamic statistical threshold aims to establish an objective standard that can define the range of normal physiological fluctuations. This procedure is executed after the baseline learning cycle ends. First, the daily rhythm response vectors within that cycle are summarized and compared with the finally established daily rhythm response vector baseline. The method calculates the magnitude of each of the historical asynchronous phase vectors between the two phase vectors, thereby forming an empirical distribution dataset of magnitudes representing the rhythm deviation under healthy conditions. The specific value of the dynamic statistical threshold is determined as the 95th percentile of this dataset. This method ensures that the threshold is adaptively generated based on the inherent physiological fluctuation characteristics of a specific crop at a specific growth stage, and it serves as a fixed judgment benchmark in subsequent monitoring cycles. Similarly, the preset light change threshold on which the effectiveness of light conditions is determined is established as follows: During the baseline learning cycle, under multiple clear weather conditions without cloud cover, high-frequency continuous sampling of the chlorophyll absorption peak intensity is performed multiple times to obtain multiple instantaneous intensity sequences. The variance of the instantaneous change rate of each sequence is calculated. These variance values ​​together constitute a background noise distribution representing the inherent noise of the system and weak environmental fluctuations under normal light conditions. The preset light change threshold is set as the mean of this background noise distribution plus four times the standard deviation, thereby effectively separating the drastic light changes caused by rapid cloud movement from the background noise.

[0049] Regarding the stress type attribution step, the core of the stress prototype trajectory library construction and feature distance matching logic lies in transforming multi-dimensional geometric features into a single comparable distance metric. The construction of this library begins with a series of independent, controlled greenhouse experiments. For different known stress types on the target crop, such as specific pathogen infection or nutrient deficiency of specific elements, artificial induction is performed, and data is continuously collected during this process to generate typical asynchronous phase-series trajectories under each stress. For each trajectory, the system quantifies and calculates its geometric feature vector, where the linear fit goodness of the trajectory is determined by calculating the coefficient of determination of the trajectory point coordinate sequence. The average curvature of the trajectory is obtained by calculating the arithmetic mean of the angles between adjacent asynchronous phase vectors on the trajectory; the rate of change is the derivative of the asynchronous phase vector magnitude with respect to time. All eigenvalues ​​are processed by min-max normalization and mapped to... The intervals are used to form dimensionless feature vectors, which are then stored in the database. When stress attribution is required, the system first calculates the normalized feature vectors of the trajectory to be analyzed in the same way. Subsequently, this vector is compared with the feature vector of each prototype trajectory in the library. Weighted Manhattan distance between To determine the degree of matching, In this formula, These are weighting coefficients determined during the model calibration phase. Their values ​​are adjusted through optimization algorithms to maximize the accuracy of distinguishing training samples of known stress types. Their design aims to amplify the influence of features most discriminative of stress types. In applications distinguishing between acute diseases and chronic nutritional imbalances, the weights of trajectory linearity and magnitude change rate are... Therefore, it is assigned a higher value. Ultimately, it has the smallest overall difference. The prototype trajectory is the one that is judged to be the most likely type of stress attribution.

[0050] In the root zone stress identification step, the calculation of the response recovery rate is refined. After applying a micro-perturbation of humidity, the high-frequency sequence of water absorption peak intensity collected by the system is first compared point-by-point with the intensity value of a near-infrared high-reflectivity plateau band, which has the same trend as light influence, to generate a ratio sequence that has eliminated the influence of light fluctuations. This ratio sequence depicts the net physiological response curve of water state recovering from the perturbation point to steady state. The response recovery rate is quantified as the slope of the linear fit obtained by the least squares method of this net physiological response curve in the initial recovery stage. This slope value objectively reflects the response efficiency of root water transport function to changes in demand. Finally, for the step of sensing soil moisture status, the preset health correlation coefficient range is... In specific applications, this can be refined through calibration. The calibration process involves subjecting soil moisture to controlled variations from optimal to slightly dry or near-saturation conditions within a baseline learning period, while simultaneously recording the correlation coefficients of the diurnal rhythm residual sequences. By analyzing the actual distribution range of these correlation coefficients under different moisture conditions, a more precise range of health correlation coefficients representing suitable soil moisture levels can be established for specific combinations of target crops and soils, thereby improving the reliability of early warnings for abnormal soil moisture conditions.

[0051] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A hyperspectral method for early dynamic detection of crop diseases and pests, characterized in that, The method includes the following steps: Step a: During a baseline learning cycle, acquire baseline hyperspectral images of the target crop daily at a first specific time period and a second specific time period; the first specific time period is defined as after sunrise, the initial stage of photosynthesis, and the second specific time period is defined as noon, the peak period of photosynthesis. Step b: Based on the intensity values ​​of the chlorophyll absorption peak band sensitive to photosynthesis and the intensity values ​​of the water absorption peak band sensitive to water status in the baseline hyperspectral image, calculate the normalized intensity value change rate vector corresponding to each pixel from the first specific time period to the second specific time period. Step c: Perform time series smoothing on the normalized intensity value change rate vector obtained for each day within the baseline learning period to dynamically establish a daily rhythmic response vector baseline representing the health status of the target crop. Step d: Within a monitoring cycle, hyperspectral images are collected daily during the first and second specific time periods. Based on the same chlorophyll absorption peak band intensity value and water absorption peak band intensity value in the hyperspectral images, the daily circadian rhythm response vector is calculated. Step e: Based on the daily circadian rhythm response vector and the daily circadian rhythm response vector baseline, calculate the asynchronous phase vector, which represents the vector difference between the daily circadian rhythm response vector and the daily circadian rhythm response vector baseline. Step f involves comparing the magnitude of the asynchronous phase vector with a dynamic statistical threshold determined based on the inherent fluctuation characteristics of the daily rhythm response vector baseline. Step g: When the magnitude of the asynchronous phase vector is greater than the dynamic statistical threshold for more than two consecutive days, it is determined that the target crop has an abnormal physiological rhythm and an early warning signal for pests and diseases is generated. Furthermore, the normalized intensity value change rate vector is constructed by calculating the normalized intensity change of the chlorophyll absorption peak band intensity value from the first specific time period to the second specific time period, and the normalized intensity change of the water absorption peak band intensity value from the first specific time period to the second specific time period; or by calculating the relative change rate of the intensity values ​​of the chlorophyll absorption peak band and the water absorption peak band from the first specific time period to the second specific time period, and using the relative change rate as a component of the normalized intensity value change rate vector.

2. The method for early dynamic detection of crop diseases and pests using hyperspectral imaging according to claim 1, characterized in that, Asynchronous phase vector The calculation method is as follows: ,in, This represents the daily rhythm response vector for that day. Represents the baseline of the daily rhythm response vector; asynchronous phase vector Length of the module Reflects the overall intensity of rhythm deviation, asynchronous phase vector The direction reflects an abnormality in the synchronicity between different physiological activities.

3. The method for early dynamic detection of crop diseases and pests using hyperspectral imaging according to claim 1, characterized in that, Before acquiring the monitoring hyperspectral image in step d, a step for verifying the effectiveness of illumination conditions is also included: high-frequency continuous sampling of the intensity value of the chlorophyll absorption peak band to obtain an instantaneous intensity sequence; calculation of the instantaneous rate of change variance of the instantaneous intensity sequence; when the instantaneous rate of change variance continuously exceeds a preset illumination mutation threshold, the current illumination condition is determined to be an invalid illumination mutation condition; when determined to be an invalid illumination mutation condition, compensation processing is performed on the currently acquired monitoring hyperspectral image, which includes abandoning the current data acquisition and using the effective rhythm state of the previous day, or re-sampling after the illumination intensity returns to stability.

4. The method for early dynamic detection of crop diseases and pests using hyperspectral imaging according to claim 1, characterized in that, After generating the early pest and disease warning signal in step g, the following stress type attribution steps are also included: acquiring and storing multiple consecutive asynchronous phase vectors within a preset time window to form an asynchronous phase time-series trajectory; extracting the geometric morphological features of the asynchronous phase time-series trajectory, including the linear fit goodness of the trajectory, the average curvature of the trajectory, the vector rotation direction of the trajectory, and the rate of change of the magnitude of the asynchronous phase vector over time; performing feature distance matching between the geometric morphological features and a preset stress prototype trajectory library; and determining the stress attribution type of the early pest and disease warning signal based on the matching results.

5. The method for early dynamic detection of crop diseases and pests using hyperspectral imaging according to claim 4, characterized in that, The stress prototype trajectory database is a lookup table containing at least the following prototype trajectory features defined based on plant pathology knowledge: acute disease prototypes, Its characteristics include high linearity, high speed, and a trajectory pointing to a specific physiological quadrant; The prototype of nutritional imbalance Its characteristics include low linearity, high curvature or spiral and low speed trajectory; it recovers its original shape under short-term environmental stress, and its trajectory is characterized by an arc shape that returns to the origin.

6. The method for early dynamic detection of crop diseases and pests using hyperspectral imaging according to claim 1, characterized in that, After generating the early pest and disease warning signal in step g, the process also includes a root zone stress analysis step: periodically applying a standardized, short-term humidity micro-perturbation to the root zone of the target crop. The humidity micro-perturbation is achieved by the intelligent irrigation system stopping water supply in advance in the early morning or precisely spraying a standard amount of clean water at noon. During a preset recovery period after the humidity micro-perturbation is applied, the intensity value of the water absorption peak band and the intensity value of a band on the near-infrared high reflectivity platform used as a reference baseline are continuously monitored at high frequency. The response recovery rate of the water absorption peak band intensity value relative to the reference baseline band intensity value is calculated. By comparing the response recovery rate with a first health recovery rate threshold, it is determined whether the early pest and disease warning signal is related to root zone stress.

7. The method for early dynamic detection of crop diseases and pests using hyperspectral imaging according to claim 6, characterized in that, When the response recovery rate is less than the first health recovery rate threshold, it is determined that the early pest and disease warning signal is highly correlated with root zone stress.

8. The method for early dynamic detection of crop diseases and pests using hyperspectral imaging according to claim 1, characterized in that, It also includes a step of sensing soil moisture status: within the baseline learning period, the diurnal rhythm residual sequence of the diurnal rhythm response vector corresponding to the chlorophyll absorption peak band intensity value and the water absorption peak band intensity value relative to the diurnal rhythm response vector baseline is obtained; the correlation coefficient between the chlorophyll absorption peak band residual and the water absorption peak band residual in the diurnal rhythm residual sequence is calculated; the soil moisture status is determined by comparing the correlation coefficient with a preset health correlation coefficient range, the preset health correlation coefficient range is -0.2 to 0.2; when the correlation coefficient continuously exceeds the preset health correlation coefficient range, the soil moisture status is determined to be abnormal, and a soil moisture warning signal is output in parallel; When the correlation coefficient is less than -0.2, the soil is considered to be starting to dry out; when the correlation coefficient is greater than 0.2, the soil is considered to be approaching saturation.

9. A hyperspectral dynamic detection system for early-stage crop diseases and pests, characterized in that, include: The image acquisition module is configured to acquire hyperspectral images of the target crop daily during a first specific time period and a second specific time period within the baseline learning period and the monitoring period. The first specific time period is defined as after sunrise, the initial stage of photosynthesis, and the second specific time period is defined as noon, the peak period of photosynthesis. The data processing module, connected to the image acquisition module, is configured to calculate the normalized intensity change rate vector for each pixel from a first specific time period to a second specific time period based on the intensity values ​​of the chlorophyll absorption peak band (sensitive to photosynthesis) and the water absorption peak band (sensitive to water status) in the acquired hyperspectral image. The normalized intensity change rate vector is constructed by calculating the normalized intensity change of the chlorophyll absorption peak band intensity value from the first specific time period to the second specific time period, and the normalized intensity change of the water absorption peak band intensity value from the first specific time period to the second specific time period; or by calculating the relative change rate of the intensity values ​​of the chlorophyll absorption peak band and the water absorption peak band from the first specific time period to the second specific time period, and using this relative change rate as a component of the normalized intensity change rate vector; and by performing time-series smoothing processing on the normalized intensity change rate vector obtained for each day within the baseline learning period to dynamically establish a daily rhythm response vector baseline representing the health status of the target crop. The anomaly detection module, connected to the data processing module, is configured to calculate the asynchronous phase vector based on the daily circadian rhythm response vector and the daily circadian rhythm response vector baseline. The asynchronous phase vector represents the vector difference between the daily circadian rhythm response vector and the daily circadian rhythm response vector baseline. It is configured to compare the magnitude of the asynchronous phase vector with a dynamic statistical threshold determined based on the fluctuation characteristics of the daily rhythm response vector baseline itself; The early warning generation module is connected to the anomaly detection module and is configured to generate an early pest and disease warning signal when the magnitude of the asynchronous phase vector is continuously greater than the dynamic statistical threshold for more than two consecutive days.

Images