Artificial intelligence-based agricultural pest detection sampling system and method

Abstract

This application provides an artificial intelligence-based agricultural pest and disease detection and sampling system and method. It extracts the temporal variation curves of multispectral index combinations characterizing potato late blight occurrence from temporal hyperspectral image sequences, thereby identifying suspected outbreak time intervals. It also extracts the temporal variation curves of environmental factors from environmental sensor data, thereby determining theoretical high-risk time intervals. A coarse time offset between the suspected outbreak time interval and the theoretical high-risk time interval is determined, followed by a precise time delay between the temporal variation curves of environmental factors and the temporal variation curves characterizing potato late blight occurrence. Based on the precise time delay, the temporal variation curves of environmental factors are shifted along the time axis to predict the spread trend of potato late blight. Based on the spread trend, the priority area for current sampling operations is determined. Using the scheme of this application, the temporal domain prediction results can be accurately converted into spatial domain sampling operation instructions.

Description

Technical Field

[0001] This application relates to the field of agricultural pest and disease detection technology, and in particular to an artificial intelligence-based agricultural pest and disease detection sampling system and method. Background Technology

[0002] Precise control of agricultural pests and diseases is a key link in ensuring food security. Potato late blight, as a typical airborne disease, is characterized by its strong outbreak, rapid spread, and short window of opportunity for prevention and control. Traditional detection of potato late blight mainly relies on manual field inspections and fixed-point sampling. This method is time-consuming and labor-intensive, has limited coverage, and is difficult to respond in time at the early stage of disease outbreak, often leading to delays in prevention and control and excessive application of pesticides.

[0003] However, existing methods for monitoring potato late blight based on hyperspectral imagery and sensor data have significant shortcomings in data fusion and predictive decision-making. On the one hand, there is an inherent time lag between changes in environmental factors and the spectral response of potato late blight. For example, suitable temperature and humidity conditions require a certain incubation period before visible disease symptoms appear on potato leaves. Existing technologies typically perform simple time alignment or resampling of environmental data and remote sensing images, ignoring this delay. This leads to a temporal misalignment between environmental driving information and disease response information, failing to truly reflect the coupling between the environment and the disease during the occurrence of potato late blight. On the one hand, existing technologies rely heavily on complex differential equation models for predicting potato late blight, which are difficult to calibrate and computationally complex. Furthermore, existing methods often only predict the disease's occurrence trend in the time domain, failing to effectively map the time-domain prediction results to the farmland spatial domain. This results in a disconnect between the prediction results and actual sampling operations, failing to resolve the spatiotemporal coupling contradiction between when and where potato late blight occurs during sampling. Therefore, how to accurately convert the time-domain prediction results into spatial domain sampling instructions based on the causal delay relationship between the environment and potato late blight has become a challenge for the industry. Summary of the Invention

[0004] Based on this, this application provides an artificial intelligence-based agricultural pest and disease detection sampling system and method for accurately converting time-domain prediction results into spatial-domain sampling operation instructions.

[0005] Firstly, this application provides a sampling method for detecting potato late blight, comprising the following steps: Simultaneously acquire time-series hyperspectral image sequences of the target farmland and environmental sensor data within the corresponding time window; The temporal variation curves of multispectral index combinations characterizing the occurrence of potato late blight are extracted from the temporal hyperspectral image sequence, thereby identifying the suspected outbreak time intervals where the late blight characteristic values ​​exceed the preset dynamic threshold. The temporal variation curves of environmental factors are extracted from the environmental sensor data, and based on the coupling mechanism model of environment-late epidemic, the theoretical high-risk time intervals corresponding to the environmental factors and the high risk of late epidemic are determined. Determine the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval, and then determine the precise time delay between the time series variation curve of the environmental factor and the time series variation curve characterizing the occurrence of potato late blight based on the approximate time offset; The temporal variation curve of the environmental factor is shifted along the time axis based on the precise time delay, thereby predicting the spread trend of potato late blight and determining the priority area for the current sampling operation based on the spread trend.

[0006] In some embodiments, extracting the time-series variation curves of multispectral index combinations characterizing the occurrence of potato late blight from the time-series hyperspectral image sequence specifically includes: Extract the potato late blight sensitive band dataset from each frame of the time-series hyperspectral image sequence; The multispectral index combination of the corresponding frame image was determined based on the datasets of sensitive bands for each late epidemic disease. The multispectral index combinations of each image frame are weighted and fused to generate late epidemic disease feature values ​​corresponding to each image frame. A time-series variation curve of a multispectral index combination characterizing the occurrence of potato late blight was constructed based on the late blight feature values ​​corresponding to all frames of images.

[0007] In some embodiments, identifying a suspected outbreak time interval where the late epidemic disease characteristic value exceeds a preset dynamic threshold specifically includes: Obtain the preset dynamic threshold; Traverse the time-series variation curves of the multispectral index combination characterizing the occurrence of potato late blight, and compare the late blight characteristic values ​​at each time point with the dynamic threshold. When the late-stage disease characteristic values ​​at multiple consecutive time points exceed the dynamic threshold, the consecutive time period is marked as a suspected outbreak time interval.

[0008] In some embodiments, extracting the time-series variation curves of environmental factors from the environmental sensor data specifically includes: Time-series data of air temperature, relative humidity, soil volumetric water content, and photosynthetically active radiation intensity are extracted from the environmental sensor data. The time-series data is smoothed to generate time-series variation curves of environmental factors.

[0009] In some embodiments, based on the environment-late epidemic coupling mechanism model, determining the theoretical high-risk time intervals corresponding to high late epidemic risk for environmental factors specifically includes: The time-series variation curves of the environmental factors are input into the environmental-late epidemic disease coupling mechanism model to calculate the daily infection risk index sequence; Obtain the pre-set risk threshold; By traversing the daily infection risk index sequence, time periods that continuously exceed the risk threshold are marked as theoretically high-risk time intervals.

[0010] In some embodiments, determining the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval specifically includes: Obtain the midpoint of the suspected outbreak time interval; Obtain the midpoint of the theoretically high-risk time interval; The time difference between the midpoint of the suspected outbreak time interval and the midpoint of the theoretical high-risk time interval is calculated as a rough time offset.

[0011] In some embodiments, determining the precise time delay between the temporal variation curve of the environmental factor and the temporal variation curve characterizing the occurrence of potato late blight, based on the coarse time offset, specifically includes: The time-series variation curves of the environmental factors and the time-series variation curves characterizing the occurrence of potato late blight were normalized. The search window is determined with the approximate time offset as the center. Within the search window, the correlation coefficients between the time-series variation curves of the normalized environmental factors and the time-series variation curves representing the occurrence of potato late blight under different time shifts are calculated. Iterate through all correlation coefficients and select the time shift corresponding to the largest correlation coefficient as the precise time delay.

[0012] Secondly, this application provides an artificial intelligence-based agricultural pest and disease detection and sampling system, which includes a potato late blight detection and sampling unit, the potato late blight detection and sampling unit comprising: The acquisition module is used to simultaneously acquire time-series hyperspectral image sequences of the target farmland and environmental sensor data within the corresponding time window; The processing module is used to extract the time-series change curve of the multispectral index combination characterizing the occurrence of potato late blight from the time-series hyperspectral image sequence, and then identify the suspected outbreak time interval where the late blight characteristic value exceeds the preset dynamic threshold. The processing module is also used to extract the time-series variation curves of environmental factors from the environmental sensor data, and based on the coupling mechanism model of environment-late epidemic, determine the theoretical high-risk time intervals of environmental factors corresponding to the high risk of late epidemic. The processing module is also used to determine the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval, and then determine the precise time delay between the time series change curve of the environmental factor and the time series change curve characterizing the occurrence of potato late blight based on the approximate time offset. The execution module is used to shift the time-series change curve of the environmental factors according to the precise time delay, thereby predicting the spread trend of potato late blight, and determining the priority area for the current sampling operation based on the spread trend.

[0013] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described potato late blight detection sampling method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described potato late blight detection sampling method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The artificial intelligence-based agricultural pest and disease detection and sampling system and method provided in this application firstly collects time-series hyperspectral image sequences of the target farmland and environmental sensor data within the corresponding time window. This step enables precise synchronization of the time-series hyperspectral images and environmental sensor data in the time dimension, thereby improving the accuracy of identifying the causal relationship between environmental drivers and disease responses and the reliability of data association in subsequent analysis. Secondly, the system extracts the time-series change curves of multispectral index combinations characterizing potato late blight from the time-series hyperspectral image sequences, thereby identifying the suspected outbreak time intervals where late blight characteristic values ​​exceed preset dynamic thresholds. This step enables the identification of disease outbreak intervals based on dynamic thresholds using multispectral index fusion, thereby improving the sensitivity to early symptoms of potato late blight, avoiding misjudgment by a single index, and accurately locating the time window when the disease begins to appear. Subsequently, the system extracts the time-series change curves of environmental factors from the environmental sensor data, and based on the environment-late blight coupling mechanism model, determines the theoretical high-risk time intervals corresponding to high late blight epidemic risk for environmental factors. This step enables the identification of environmental factors infecting the disease. The method involves quantifying and converting risks to improve the accuracy of identifying time intervals with high late blight epidemic risk. Then, a coarse time offset is determined between the suspected outbreak time interval and the theoretical high-risk time interval. Based on this coarse time offset, a precise time delay is determined between the temporal variation curve of the environmental factor and the temporal variation curve characterizing the occurrence of potato late blight. This step implements a two-level delay determination strategy, from coarse interval estimation to precise curve matching, thereby improving the calculation accuracy of the causal delay relationship between environmental drivers and disease response. Finally, the temporal variation curve of the environmental factor is shifted along the time axis based on the precise time delay to predict the spread trend of potato late blight. The priority area for current sampling operations is determined based on this spread trend. This step achieves a complete mapping from time series alignment to spatial sampling decisions, thereby improving the targeting of sampling operations and prioritizing resource allocation to high-risk areas of disease spread. In summary, the solution of this application can accurately convert time-domain prediction results into spatial-domain sampling operation instructions based on the causal delay relationship between the environment and potato late blight. Attached Figure Description

[0016] Figure 1 This is an exemplary flowchart of a potato late blight detection sampling method according to some embodiments of this application; Figure 2 This is a schematic diagram illustrating an application scenario of a detection and sampling data processing system according to some embodiments of this application; Figure 3 This is a flowchart illustrating the determination of a precise time delay amount according to some embodiments of this application; Figure 4This is a schematic diagram of the structure of a potato late blight detection sampling unit according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a potato late blight detection sampling method according to some embodiments of this application. Detailed Implementation

[0017] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0018] refer to Figure 1 The figure is an exemplary flowchart of a potato late blight detection sampling method according to some embodiments of this application. The potato late blight detection sampling method mainly includes the following steps: In step 101, the time-series hyperspectral image sequence of the target farmland and environmental sensor data within the corresponding time window are acquired simultaneously.

[0019] In practice, the simultaneous acquisition of time-series hyperspectral image sequences of the target farmland and environmental sensor data within the corresponding time window can be achieved in the following way: First, a drone patrol route is planned within the target farmland area, with waypoint spacing set at 20 to 30 meters and flight altitude at 50 to 80 meters. A hyperspectral imager is used to collect data once a day, covering the entire farmland area each time, acquiring a hyperspectral data cube containing wavelengths from 400 nanometers to 1000 nanometers. The spatial resolution of each image frame is controlled within 5 to 10 centimeters, forming a time-series hyperspectral image sequence. Simultaneously, IoT environmental sensor nodes are deployed within the farmland at a density of 2 to 3 per hectare. Each node uploads real-time data on air temperature, relative humidity, and soil volumetric water content via a wireless transmission module. The data collection frequency for the amount of photosynthetically active radiation (RALED) is set to be once every 30 minutes to form environmental sensor data. Then, using the time of each hyperspectral image acquisition as the reference timestamp, the environmental sensor data within a preset time window before and after the reference time is extracted. The arithmetic mean of the sensor data within the time window is taken as the environmental sensor data at the reference time. The length of the preset time window can be set to 2 to 4 hours based on the infection response time of potato late blight. If the proportion of missing environmental sensor data corresponding to a certain reference time exceeds 20%, it is filled by linear interpolation of adjacent time points. Finally, a time-series hyperspectral image sequence with timestamp alignment and a sequence of environmental sensor data within the corresponding time window are formed. Other methods can also be used in other embodiments, which are not limited here.

[0020] It should be noted that the temporal hyperspectral image sequence in this application refers to multiple frames of hyperspectral image data arranged in chronological order, used to provide spectral information of the target farmland at different time points; the environmental sensor data refers to environmental parameter data related to the occurrence of potato late blight collected by Internet of Things sensors deployed in the farmland, used to characterize the driving effect of environmental conditions on the occurrence and development of the disease.

[0021] In some embodiments, reference Figure 2 As shown in the figure, this figure is a schematic diagram of the application scenario of the detection and sampling data processing system shown in some embodiments of this application. The figure includes three main components: acquisition device, server and data storage device. The acquisition device is responsible for collecting the time-series hyperspectral image sequence of the target farmland and the environmental sensor data within the corresponding time window, and sending the acquired time-series hyperspectral image sequence and the corresponding environmental sensor data to the server through the communication network. The detection and sampling data processing system runs in the server, and the server stores the processing results in the data storage device and visualizes them.

[0022] In step 102, the temporal variation curve of the multispectral index combination characterizing the occurrence of potato late blight is extracted from the temporal hyperspectral image sequence, thereby identifying the suspected outbreak time interval where the late blight characteristic value exceeds the preset dynamic threshold.

[0023] In some embodiments, extracting the temporal variation curve of the multispectral index combination characterizing the occurrence of potato late blight from the temporal hyperspectral image sequence can be achieved by the following steps: Extract the potato late blight sensitive band dataset from each frame of the time-series hyperspectral image sequence; The multispectral index combination of the corresponding frame image was determined based on the datasets of sensitive bands for each late epidemic disease. The multispectral index combinations of each image frame are weighted and fused to generate late epidemic disease feature values ​​corresponding to each image frame. A time-series variation curve of a multispectral index combination characterizing the occurrence of potato late blight was constructed based on the late blight feature values ​​corresponding to all frames of images.

[0024] It should be noted that, in this application, potato late blight refers to the main potato disease caused by Phytophthora infestans, which is the target detection object of the sampling method in this application. Its occurrence and development are jointly influenced by environmental conditions and crop growth status. The multispectral index combination refers to the set of multiple spectral indices extracted from hyperspectral images, used to comprehensively characterize the occurrence degree of potato late blight. The late blight feature value refers to the quantitative value obtained by weighted fusion of the multispectral index combination of each frame of image, used to quantitatively describe the overall occurrence degree of potato late blight in the farmland at the time of image acquisition.

[0025] In specific implementation, the extraction of potato late blight sensitive band dataset from each frame of the time-series hyperspectral image sequence can be achieved in the following manner: For each frame of the time-series hyperspectral image sequence, firstly, based on the spectral response characteristics of potato late blight on leaf tissue, the late blight sensitive band range is determined to be the green light reflectance peak band of 550 nm to 560 nm, the red light absorption trough band of 670 nm to 680 nm, the near-infrared high reflectance plateau band of 750 nm to 800 nm, and the red edge band of 710 nm to 730 nm; then, by reading the pixel values ​​of the corresponding bands in the hyperspectral data cube, the reflectance data of the above four band ranges is extracted pixel by pixel to form the potato late blight sensitive band dataset for each frame of the image. This late blight sensitive band dataset contains a two-dimensional reflectance matrix of four bands consistent with the spatial resolution of the image. Other methods can also be used in other embodiments, which are not limited here.

[0026] In specific implementation, the multispectral index combination of the corresponding frame image can be determined based on the late blight sensitive band datasets as follows: For each frame image, the multispectral index combination is calculated pixel by pixel based on the late blight sensitive band dataset corresponding to the image. The multispectral index combination includes the normalized vegetation index, the red-edge normalized index, and the disease stress index. The normalized vegetation index is obtained by calculating the difference between the near-infrared band reflectance and the red band reflectance and dividing by the sum of the two. The red-edge normalized index is obtained by calculating the difference between the near-infrared band reflectance and the red-edge band reflectance and dividing by the sum of the two. The disease stress index is obtained by calculating the ratio of the green band reflectance to the red band reflectance. The calculation results of the above three indices are stored according to the pixel position to form a three-channel index image with the same spatial resolution as the image. Each channel corresponds to a spectral index, which serves as the multispectral index combination of the frame image, thereby obtaining the multispectral index combination of each frame image. Other methods can also be used in other embodiments, which are not limited here.

[0027] In practice, the multispectral index combination of each frame of image is weighted and fused separately to generate the late blight feature value corresponding to each frame of image. This can be achieved in the following way: For the multispectral index combination of each frame of image, a weighted fusion calculation is performed pixel by pixel. First, the weight coefficients are pre-calibrated based on the sensitivity differences of potato late blight to each spectral index at different growth stages. The weight coefficient of the normalized vegetation index is set to 0.20 to 0.25, the weight coefficient of the red edge normalized index is set to 0.30 to 0.35, and the weight coefficient of the disease stress index is set to 0.45 to 0.50, and the sum of the weight coefficients of the three is 1. The specific values ​​can be dynamically adjusted within their respective ranges according to the current growth stage. Then, each The late blight feature value of a pixel is obtained by multiplying its normalized vegetation index, red-edge normalized index, and disease stress index by their respective weighting coefficients and summing the results. All pixels in the image frame are traversed to generate a late blight feature value matrix with the same spatial resolution as the image frame, thus obtaining the late blight feature value matrix for each image frame. For each image frame's late blight feature value matrix, the arithmetic mean of the late blight feature values ​​of all pixels in the matrix is ​​calculated as the late blight feature value corresponding to that image frame. In other embodiments, the mean or median of a specific region in the feature value matrix, such as a diseased area or a high-risk area, can also be selected as the late blight feature value corresponding to that image frame; this is not limited here.

[0028] In specific implementation, the time-series variation curve of the multispectral index combination representing the occurrence of potato late blight can be constructed based on the late blight feature values ​​corresponding to all frames of images in the following way: the late blight feature values ​​at each time point are arranged sequentially according to the acquisition time of each frame of images, forming a two-dimensional sequence with the acquisition time as the horizontal axis and the late blight feature values ​​as the vertical axis. This two-dimensional sequence is the time-series variation curve of the multispectral index combination representing the occurrence of potato late blight. Other methods can also be used in other embodiments, which are not limited here.

[0029] In some embodiments, identifying a suspected outbreak time interval where late-stage disease characteristic values ​​exceed a preset dynamic threshold can be achieved using the following steps: Obtain the preset dynamic threshold; Traverse the time-series variation curves of the multispectral index combination characterizing the occurrence of potato late blight, and compare the late blight characteristic values ​​at each time point with the dynamic threshold. When the late-stage disease characteristic values ​​at multiple consecutive time points exceed the dynamic threshold, the consecutive time period is marked as a suspected outbreak time interval.

[0030] It should be noted that the preset dynamic threshold in this application refers to the baseline for judging whether the late blight characteristic value is abnormally high, and its value changes dynamically with the crop growth period or farmland condition; the suspected outbreak time interval refers to the continuous time period identified based on the late blight characteristic value exceeding the dynamic threshold, which is used to characterize the time window when the disease symptoms observed by remote sensing begin to appear or significantly worsen.

[0031] In specific implementation, obtaining the preset dynamic threshold can be achieved in the following way: A healthy plant reference area is pre-defined in the target farmland. This healthy plant reference area can be confirmed through manual surveying or historical images of healthy plants from the same period, ensuring that no late blight occurs within the reference area; from the time-series variation curve of the multispectral index combination characterizing potato late blight occurrence, the late blight characteristic value sequence corresponding to the healthy plant reference area within a preset time window is extracted. This preset time window can be set to 10 to 15 days based on the crop growth period; the arithmetic mean and standard deviation of the late blight characteristic value sequence are calculated, and the dynamic threshold is set as the arithmetic mean plus 2.5 times the standard deviation; if the healthy plant... If the reference area for the plant cannot be obtained due to objective reasons, healthy period data from the same historical period in the same plot are used as a substitute, and the calculation method is the same. It should be noted that the dynamic threshold is updated every 7 days as the crop growth period progresses: if a healthy plant reference area is used, the late blight characteristic value sequence of the most recent 7 days is extracted from the reference area to calculate a new threshold each time it is updated; if historical data from the same period is used, a rolling time series dataset is constructed using the same growth period window of the past three years, i.e., healthy period data of 3 days before and after, and the mean and standard deviation within the time window are calculated every 7 days to achieve dynamic updating of the threshold. Other methods can also be used in other embodiments, which are not limited here.

[0032] In specific implementation, the time-series change curve of the multispectral index combination representing the occurrence of potato late blight is traversed, and the late blight characteristic value at each time point is compared with the dynamic threshold. This can be achieved in the following way: in the order of collection time, the late blight characteristic value of each time point in the time-series change curve of the multispectral index combination representing the occurrence of potato late blight is read sequentially; the late blight characteristic value at the current time point is compared with the dynamic threshold to determine whether the late blight characteristic value at the current time point is greater than the dynamic threshold; the comparison result of each time point is recorded, and a Boolean label sequence corresponding to the time axis is generated, where time points exceeding the dynamic threshold are marked as 1, and time points not exceeding the dynamic threshold are marked as 0. Other methods can also be used in other embodiments, which are not limited here.

[0033] In specific implementation, when the late blight characteristic values ​​at multiple consecutive time points exceed the dynamic threshold, marking the consecutive time period as a suspected outbreak time interval can be achieved in the following way: traverse the Boolean marker sequence and identify time periods marked with consecutive 1s. The time period marked with consecutive 1s is defined as the time interval between two adjacent time points not exceeding the image acquisition cycle and without any time points marked with 0 in between. For each time period marked with consecutive 1s, count the number of consecutive time points within that time period. When the number of consecutive time points reaches a preset minimum consecutive point threshold, mark the time interval defined by the start and end time points corresponding to that consecutive time period as a suspected outbreak time interval. The minimum consecutive point threshold can be set to 3 consecutive time points based on the incubation period of potato late blight from infection to symptom onset. If multiple discontinuous suspected outbreak time intervals are identified, they are marked as different suspected outbreak time intervals. Other methods can also be used in other embodiments, which are not limited here.

[0034] In step 103, the time-series variation curves of environmental factors are extracted from the environmental sensor data, and based on the coupling mechanism model of environment-late epidemic, the theoretical high-risk time intervals corresponding to the environmental factors and the high risk of late epidemic are determined.

[0035] In some embodiments, extracting the time-series variation curves of environmental factors from the environmental sensor data can be achieved by the following steps: Time-series data of air temperature, relative humidity, soil volumetric water content, and photosynthetically active radiation intensity are extracted from the environmental sensor data. The time-series data is smoothed to generate time-series variation curves of environmental factors.

[0036] It should be noted that the photosynthetically active radiation intensity in this application refers to the radiative energy in solar radiation with wavelengths in the range of 400 nanometers to 700 nanometers that can be utilized by plant photosynthesis, as one of the environmental factors affecting the disease resistance of potato plants and the reproduction of late blight pathogen.

[0037] In specific implementation, the extraction of time-series data of air temperature, relative humidity, soil volumetric water content, and photosynthetically active radiation intensity from the environmental sensor data can be achieved in the following way: from the time-stamped environmental sensor data, extract the air temperature value, relative humidity value, soil volumetric water content value, and photosynthetically active radiation intensity value at each sampling time in chronological order. Combine the four values ​​at the same sampling time into a four-dimensional vector. Arrange the four-dimensional vectors of all sampling times in chronological order to form the time-series data of air temperature, relative humidity, soil volumetric water content, and photosynthetically active radiation intensity. Other methods can also be used in other embodiments, which are not limited here.

[0038] In specific implementation, the time-series data is smoothed to generate time-series variation curves of environmental factors. This can be achieved by performing a three-point moving average smoothing process on the time-series data of air temperature, air relative humidity, soil volumetric water content, and photosynthetically active radiation intensity. Specifically, for each time-series data, the arithmetic mean of the values ​​at three consecutive time points is calculated. This arithmetic mean is used as the smoothed value at the middle time point. The first and last time points are averaged at two points or the original values ​​are retained. After smoothing, the time-series variation curves of air temperature, air relative humidity, soil volumetric water content, and photosynthetically active radiation intensity are formed. These four curves are collectively referred to as the time-series variation curves of environmental factors. Other methods can also be used in other embodiments, which are not limited here.

[0039] In some embodiments, determining the theoretical high-risk time intervals corresponding to high late epidemic risk for environmental factors based on the environment-late epidemic coupling mechanism model can be achieved through the following steps: The time-series variation curves of the environmental factors are input into the environmental-late epidemic disease coupling mechanism model to calculate the daily infection risk index sequence; Obtain the pre-set risk threshold; By traversing the daily infection risk index sequence, time periods that continuously exceed the risk threshold are marked as theoretically high-risk time intervals.

[0040] It should be noted that the environment-late blight coupling mechanism model in this application refers to a computational model constructed based on the infection biological laws of *Pseudomonas aeruginosa*, used to transform environmental factor inputs into infection risk outputs, serving as a computational bridge connecting environmental conditions and disease occurrence risk. The daily infection risk index sequence refers to the risk value sequence formed after inputting environmental factors into the coupling mechanism model, calculating on a daily basis, and normalizing the data. It is used to quantitatively describe the suitability of daily environmental conditions for late blight occurrence. The risk threshold refers to the baseline for judging whether the daily infection risk index has reached a high-risk level, serving as the basis for identifying theoretically high-risk time intervals from the risk index sequence. The theoretically high-risk time interval refers to the period of time when environmental conditions are most suitable for disease occurrence, calculated according to the environment-late blight coupling mechanism model, used to characterize the time window with a theoretically high disease infection risk.

[0041] In specific implementation, the environment-late blight coupling mechanism model can be constructed as follows: Based on the biological laws of potato late blight infection, an infection risk determination rule set is established with air temperature, relative humidity, soil volumetric water content, and photosynthetically active radiation intensity as input variables. This infection risk determination rule set includes: when the relative humidity exceeds 90% for more than 10 consecutive hours and the air temperature is between 10°C and 25°C during the same period, it is marked as a valid infection event; when the soil volumetric water content exceeds 80%, the risk intensity of the current valid infection event is weighted and amplified. The weighting factor is set to 1.2 to 1.5 times. When the photosynthetically active radiation intensity is less than 200 micromoles per square meter per second, the risk intensity of the current effective infection event is amplified by a second weighting factor, which is set to 1.1 to 1.3 times. The above infection risk judgment rule set is encapsulated into an executable calculation module. This calculation module takes the time series data of each environmental factor every hour as input and outputs the effective infection event marker and the corresponding weighted infection intensity value every hour, which serves as the coupling mechanism model of the environment-late epidemic disease. Other methods can also be used in other embodiments, which are not limited here.

[0042] In specific implementation, the time-series variation curves of the environmental factors are input into the environment-late epidemic disease coupling mechanism model. The calculation of the daily infection risk index sequence can be achieved in the following way: the time-series variation curves of the environmental factors are resampled at hourly resolution. The resampling method is arithmetic mean, that is, the average value of sensor data collected between every two adjacent 30-minute intervals is calculated as the value for that hour, thereby obtaining the hourly air temperature value, air relative humidity value, soil volumetric water content value, and photosynthetically active radiation intensity value; the resampled hourly air temperature value, air relative humidity value, soil volumetric water content value, and photosynthetically active radiation intensity value are then used to calculate the hourly air temperature value, air relative humidity value, soil volumetric water content value, and photosynthetically active radiation intensity value. The values ​​are input into the environment-late epidemic coupling mechanism model. The coupling mechanism model determines whether each hour is a valid infection event based on the infection risk judgment rule set and outputs the infection intensity value for that hour. The infection intensity values ​​of all hours within each day are summed to obtain the cumulative infection intensity value for that day. The cumulative infection intensity value for that day is divided by the maximum possible infection intensity value for that day and normalized to obtain a value in the range of 0 to 1, which serves as the daily infection risk index for that day. The daily infection risk indices for all days are arranged in chronological order to form a daily infection risk index sequence. Other methods can also be used in other embodiments, which are not limited here.

[0043] In specific implementation, the risk threshold can be preset in the following way: based on the daily infection risk index distribution corresponding to the actual outbreak time of potato late blight in historical years, the minimum infection risk index value for the three consecutive days before the outbreak is statistically analyzed, and the lower quartile of the statistical values ​​of multiple historical samples is taken as the initial risk threshold. Alternatively, based on the experience of field plant protection experts, the initial risk threshold can be set to 0.65. During system operation, the initial risk threshold can be dynamically fine-tuned in combination with the actual disease incidence in that year, with a fine-tuning step size of 0.05. Finally, the adjusted initial risk threshold, i.e., the risk threshold, is used for the subsequent determination of theoretical high-risk time intervals. Other methods can also be used in other embodiments, which are not limited here.

[0044] In specific implementation, traversing the daily infection risk index sequence and marking time periods that continuously exceed the risk threshold as theoretical high-risk time intervals can be achieved in the following way: traversing the daily infection risk index sequence in chronological order, comparing the daily infection risk index of each day with the risk threshold, recording the start and end days of continuous exceedance of the risk threshold, and marking the continuous time period as a theoretical high-risk time interval when the number of consecutive days exceeding the risk threshold reaches a preset minimum consecutive days threshold. The minimum consecutive days threshold can be set to 2 days based on the cumulative effect of potato late blight infection. If there are multiple non-continuous time periods that continuously exceed the risk threshold, they are marked as different theoretical high-risk time intervals. Other methods can also be used in other embodiments, which are not limited here.

[0045] In step 104, a rough time offset is determined between the suspected outbreak time interval and the theoretical high-risk time interval, and then the precise time delay between the time-series change curve of the environmental factor and the time-series change curve characterizing the occurrence of potato late blight is determined based on the rough time offset.

[0046] In some embodiments, determining the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval can be achieved by the following steps: Obtain the midpoint of the suspected outbreak time interval; Obtain the midpoint of the theoretically high-risk time interval; The time difference between the midpoint of the suspected outbreak time interval and the midpoint of the theoretical high-risk time interval is calculated as a rough time offset.

[0047] It should be noted that the rough time offset in this application refers to the estimated time difference obtained by using the time difference between the midpoint of the suspected outbreak time interval and the theoretical high-risk time interval, which is used to provide a search center reference for the calculation of the precise time delay.

[0048] In specific implementation, the midpoint time of the suspected outbreak time interval can be obtained in the following ways: when there is only one suspected outbreak time interval, the arithmetic mean of the start time and end time of the suspected outbreak time interval is directly used as the midpoint time of the suspected outbreak time interval; when there are multiple suspected outbreak time intervals, the latest suspected outbreak time interval is selected according to the time order, and the arithmetic mean of its start time and end time is used as the midpoint time of the interval. Other methods can also be used in other embodiments, which are not limited here.

[0049] In specific implementation, the midpoint of the theoretical high-risk time interval can be obtained in the following ways: when there is only one theoretical high-risk time interval, the arithmetic mean of the start and end times of the theoretical high-risk time interval is directly used as the midpoint of the theoretical high-risk time interval; when there are multiple theoretical high-risk time intervals, the earliest theoretical high-risk time interval is selected in chronological order, and the arithmetic mean of its start and end times is used as the midpoint of the interval. Other methods can also be used in other embodiments, which are not limited here.

[0050] In specific implementation, the time difference between the midpoint of the suspected outbreak time interval and the midpoint of the theoretical high-risk time interval is calculated as a coarse time offset. This can be achieved by subtracting the midpoint of the theoretical high-risk time interval from the midpoint of the suspected outbreak time interval to obtain the time difference value, which is the coarse time offset. If the coarse time offset is positive, it indicates that the time-series change curve of the environmental factor leads the time-series change curve representing the occurrence of potato late blight. If it is negative, it indicates that the time-series change curve of the environmental factor lags behind the time-series change curve representing the occurrence of potato late blight. Other methods can also be used in other embodiments, which are not limited here.

[0051] In some embodiments, reference Figure 3 As shown in the figure, this is a flowchart illustrating the determination of the precise time delay in some embodiments of this application. In this embodiment, the determination of the precise time delay between the temporal variation curve of the environmental factor and the temporal variation curve characterizing the occurrence of potato late blight based on the coarse time offset can be achieved by the following steps: In step 1041, the time-series variation curves of the environmental factors and the time-series variation curves characterizing the occurrence of potato late blight are normalized. In step 1042, a search window is determined with the coarse time offset as the center; In step 1043, within the search window, the correlation coefficient between the time-series variation curve of the normalized environmental factors and the time-series variation curve representing the occurrence of potato late blight under different time shifts is calculated. In step 1044, all correlation coefficients are iterated, and the time shift corresponding to the largest correlation coefficient is selected as the precise time delay.

[0052] It should be noted that the time shift in this application refers to the different shifts attempted when shifting the environmental factor curve along the time axis during the calculation of the precise time delay, used to traverse each candidate time offset value within the search window; the precise time delay refers to the time shift determined by correlation analysis within the search window that optimizes the alignment between the environmental factor curve and the disease characteristic curve, used to characterize the causal delay relationship between environmental driving forces and disease response.

[0053] In specific implementation, the normalization of the time-series variation curves of the environmental factors and the time-series variation curves characterizing the occurrence of potato late blight can be achieved in the following way: Obtain the minimum and maximum values ​​of the environmental factors at all time points in the time-series variation curves of the environmental factors, and the minimum and maximum values ​​of the late blight characteristic values ​​at all time points in the time-series variation curves characterizing the occurrence of potato late blight; for the time-series variation curves of the environmental factors, subtract the minimum value from the environmental factor value at each time point and divide by the difference between the maximum and minimum values ​​to obtain a range from 0 to 1. The normalized environmental factor values ​​are calculated as follows: For the time-series curve characterizing the occurrence of potato late blight, the late blight characteristic value at each time point is subtracted from its minimum value and then divided by the difference between its maximum and minimum values ​​to obtain the normalized late blight characteristic value in the range of 0 to 1; The normalized environmental factor values ​​are arranged in chronological order to form the normalized environmental factor time-series curve, and the normalized late blight characteristic values ​​are arranged in chronological order to form the normalized time-series curve characterizing the occurrence of potato late blight. Other methods can also be used in other embodiments, which are not limited here.

[0054] In specific implementation, the search window can be determined using the coarse time offset as the center in the following manner: The half-width of the search window is set to 3 to 5 days. The coarse time offset is used as the center point of the search window. The half-width is extended to the left from this center point to obtain the starting boundary of the search window, and the half-width is extended to the right from this center point to obtain the ending boundary of the search window. The time range defined by the starting and ending boundaries of the search window is the search window itself. If the starting boundary of the search window exceeds the time range of the environmental factor time-series change curve, the starting boundary is truncated to the earliest time point of the environmental factor time-series change curve. If the ending boundary of the search window exceeds the time range of the environmental factor time-series change curve, the ending boundary is truncated to the latest time point of the environmental factor time-series change curve. Other methods can also be used in other embodiments, which are not limited here.

[0055] In specific implementation, within the search window, calculating the correlation coefficient between the normalized environmental factor time-series variation curve and the normalized time-series variation curve representing the occurrence of potato late blight at different time shifts can be achieved in the following way: all possible time shifts are traversed within the search window using a preset time shift step size, where the time shift step size is set to 1 day; for each time shift, the normalized environmental factor time-series variation curve is shifted along the time axis by that time shift, so that the shifted environmental factor curve correlates with the normalized time-series variation curve representing the occurrence of potato late blight. The sequence change curves are aligned in the time dimension; within the time interval covered by both curves, the Pearson correlation coefficient between the normalized environmental factor values ​​and the normalized late epidemic disease characteristic values ​​is calculated point by point in time, and this correlation coefficient is used as the correlation coefficient corresponding to the time shift; each time shift and its corresponding correlation coefficient are recorded to form a set of correspondences between time shifts and correlation coefficients; if the number of time points covered by both curves under a certain time shift is less than 3, the correlation coefficient of the time shift is marked as invalid. Other methods can also be used in other embodiments, which are not limited here.

[0056] In step 105, the time-series change curve of the environmental factor is shifted according to the precise time delay to predict the spread trend of potato late blight, and the priority area for the current sampling operation is determined based on the spread trend.

[0057] In some embodiments, the prediction of the spread trend of potato late blight can be achieved by shifting the time-series variation curve of the environmental factor based on the precise time delay amount, thereby using the following steps: The time-series variation curves of the environmental factors are shifted along the time axis based on the precise time delay, so that the shifted time-series variation curves of the environmental factors are aligned with the time-series variation curves characterizing the occurrence of potato late blight in the time dimension, thus obtaining a synchronous time-series feature set. Based on the aforementioned synchronous temporal feature set, the spread trend of potato late blight is predicted.

[0058] It should be noted that the synchronous time series feature set in this application refers to the two-dimensional feature vector sequence formed by translating the environmental factor curve according to the precise time delay and pairing it with the disease feature curve at the same time point, which is used as input data for subsequent diffusion trend prediction.

[0059] In specific implementation, the time-series change curve of the environmental factor is shifted along the time axis based on the precise time delay, so that the shifted time-series change curve of the environmental factor is aligned with the time-series change curve representing the occurrence of potato late blight in the time dimension, thus obtaining a synchronized time-series feature set. This can be achieved in the following way: using the precise time delay as the shift amount, the entire time-series change curve of the environmental factor is shifted along the time axis. If the precise time delay is positive, it indicates that the time-series change curve of the environmental factor leads the time-series change curve representing the occurrence of potato late blight. In this case, the time-series change curve of the environmental factor is shifted along the time axis. The precise time delay is shifted in the negative direction. If the precise time delay is negative, it indicates that the temporal change curve of the environmental factor lags behind the temporal change curve characterizing the occurrence of potato late blight. In this case, the temporal change curve of the environmental factor is shifted in the positive direction of the time axis by the absolute value of the precise time delay. After the shift, for each time point, the shifted environmental factor value corresponding to that time point is paired with the late blight characteristic value at the same time point to form a two-dimensional feature vector. The two-dimensional feature vectors of all time points are arranged in chronological order to form a synchronous temporal feature set. Other methods can also be used in other embodiments, which are not limited here.

[0060] In specific implementation, the prediction of the spread trend of potato late blight based on the synchronous temporal feature set can be achieved in the following way: A historical spread trend feature library is pre-constructed, containing multiple historical samples. Each historical sample includes a historical synchronous temporal feature set, a corresponding historical spread rate, and a historical spread direction. The historical spread rate is expressed in meters per day, and the historical spread direction is expressed as an angle relative to the prevailing wind direction over the farmland. The current synchronous temporal feature set is matched with each historical synchronous temporal feature set in the historical spread trend feature library to calculate the dynamic time warping distance between the current synchronous temporal feature set and each historical synchronous temporal feature set. The three historical samples with the smallest dynamic time warping distance are selected as similar reference samples. The arithmetic mean of the historical spread rates of these three similar reference samples is used as the predicted spread rate, and the mode of the historical spread directions of these three similar reference samples is used as the predicted spread direction. The predicted spread rate and the predicted spread direction together constitute the spread trend of potato late blight. Other methods can also be used in other embodiments, which are not limited here.

[0061] In some embodiments, determining the priority area for the current sampling operation based on the diffusion trend can be achieved by the following steps: The diffusion trend is projected spatially to generate a sampling priority distribution map; Based on the sampling priority distribution map, the priority area for the current sampling operation is determined.

[0062] In specific implementation, the sampling priority distribution map generated by projecting the diffusion trend in space can be achieved in the following way: Obtain the late blight feature value matrix of the latest frame of hyperspectral image corresponding to the temporal change curve representing the occurrence of potato late blight. The spatial resolution of this late blight feature value matrix is ​​consistent with the spatial resolution of the hyperspectral image. The late blight feature value of each pixel represents the current severity of the disease at that location. Using the currently confirmed lesion center as the origin, draw a ray along the predicted diffusion direction. Multiply the predicted diffusion rate in the diffusion trend by a preset number of prediction days on the ray to obtain the predicted diffusion distance. Mark the area between the origin and the endpoint of the predicted diffusion distance as the diffusion influence zone. For each pixel in the late blight feature value matrix, calculate the distance weight between the pixel and the diffusion influence zone and the distance weight between the pixel and the lesion center. Multiply the late blight feature value by the distance weights to obtain the sampling priority value of the pixel. After traversing all pixels, generate a sampling priority distribution map corresponding to the farmland spatial grid. Other methods can also be used in other embodiments, which are not limited here.

[0063] In specific implementation, the priority region for the current sampling operation can be determined based on the sampling priority distribution map in the following way: A priority threshold is set in the sampling priority distribution map, and the spatial grid corresponding to pixels with sampling priority values ​​higher than the priority threshold is marked as a priority region; connected component analysis is performed on the priority region, and isolated regions with areas smaller than a preset minimum sampling area threshold are removed, while connected regions with areas reaching the preset minimum sampling area threshold are retained as the final priority region; within the final priority region, sampling points are generated according to a preset sampling density; it should be noted that the preset sampling density can be dynamically adjusted based on the average sampling priority value within the priority region, with a higher average priority value resulting in a higher sampling density; the coordinates of the generated sampling points are output to the sampling operation equipment; other methods can also be used in other embodiments, which are not limited here.

[0064] In specific implementation, the priority threshold can be preset in the following ways: Based on actual data from potato late blight sampling operations in historical years, the correspondence between the disease detection rate at sampling points within the priority area and the sampling priority value can be statistically analyzed. The sampling priority value corresponding to a disease detection rate of 80% can be used as the initial priority threshold. Alternatively, based on the experience of field plant protection experts in classifying disease risk levels, risk levels can be divided into high, medium, and low levels. The lowest sampling priority value corresponding to high risk can be used as the initial priority threshold. During system operation, the initial priority threshold can be dynamically adjusted in step size of 0.05, taking into account the actual disease occurrence area and sampling cost constraints of the year. When the disease occurrence area exceeds the preset area limit, the threshold is lowered to expand the sampling coverage. When the sampling cost exceeds the budget, the initial priority threshold is raised to focus on high-risk areas. Finally, the adjusted initial priority threshold is used as the priority threshold for priority area determination. The minimum sampling area threshold can be preset in the following way: based on the single-operation coverage capacity of the sampling equipment. The minimum effective sampling area for field sampling points is determined, wherein the sampling equipment includes drones or ground robots. The single-operation coverage capacity refers to the maximum grid area that the equipment can effectively sample in a single task. The minimum effective sampling area refers to the smallest continuous plot area that can represent the disease occurrence status of the area. The ratio of the minimum effective sampling area to the single-operation coverage capacity is multiplied by a preset proportional coefficient as the initial minimum sampling area threshold. The preset proportional coefficient can be set to 0.1 to 0.3 according to the economic requirements of the sampling operation, or it can be set to 2 to 3 times the area of ​​the grid unit divided by the farmland space grid as the initial minimum sampling area threshold. During system operation, the initial minimum sampling area threshold can be dynamically corrected in combination with the measured efficiency of the sampling operation. When no disease is detected in the priority area after multiple consecutive samplings, the initial minimum sampling area threshold is appropriately increased to exclude noise areas. Finally, the adjusted initial minimum sampling area threshold is used as the minimum sampling area threshold. Other methods can also be used in other embodiments, which are not limited here.

[0065] Furthermore, in another aspect of this application, in some embodiments, this application provides an artificial intelligence-based agricultural pest and disease detection and sampling system, which includes a potato late blight detection and sampling unit, referencing... Figure 4 The figure is a schematic diagram of the structure of a potato late blight detection sampling unit according to some embodiments of this application. The potato late blight detection sampling unit includes: a collection module 401, a processing module 402, and an execution module 403, which are described below: Acquisition module 401, in this application, is mainly used to synchronously acquire the time-series hyperspectral image sequence of the target farmland and environmental sensor data within the corresponding time window; Processing module 402, in this application, is mainly used to extract the time-series change curve of the multispectral index combination characterizing the occurrence of potato late blight from the time-series hyperspectral image sequence, and then identify the suspected outbreak time interval where the late blight characteristic value exceeds the preset dynamic threshold. The processing module 402 described in this application is also used to extract the time-series variation curves of environmental factors from the environmental sensor data, and based on the coupling mechanism model of environment-late epidemic, determine the theoretical high-risk time intervals of environmental factors corresponding to high late epidemic risk. The processing module 402 described in this application is also used to determine the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval, and then determine the precise time delay between the time series change curve of the environmental factor and the time series change curve characterizing the occurrence of potato late blight based on the approximate time offset. The execution module 403 in this application is mainly used to shift the time-series change curve of the environmental factors according to the precise time delay, thereby predicting the spread trend of potato late blight, and determining the priority area of ​​the current sampling operation based on the spread trend.

[0066] Each module in the aforementioned potato late blight detection and sampling unit can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0067] In another embodiment, this application provides a computer device, which may be a server, and its internal structure diagram may be as follows. Figure 5 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores potato late blight detection sampling data. The network interface communicates with external terminals via a network connection. When the processor executes the computer program, it implements a potato late blight detection sampling method.

[0068] Those skilled in the art will understand that Figure 5The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0069] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above embodiments of the potato late blight detection sampling method.

[0070] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above embodiments of the potato late blight detection sampling method.

[0071] In one embodiment, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps described in the above embodiment of the potato late blight detection sampling method.

[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0073] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0074] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A sampling method for detecting potato late blight, applied to an artificial intelligence-based agricultural pest and disease detection and sampling system, characterized in that, Includes the following steps: Simultaneously acquire time-series hyperspectral image sequences of the target farmland and environmental sensor data within the corresponding time window; The temporal variation curves of multispectral index combinations characterizing the occurrence of potato late blight are extracted from the temporal hyperspectral image sequence, thereby identifying the suspected outbreak time intervals where the late blight characteristic values ​​exceed the preset dynamic threshold. The temporal variation curves of environmental factors are extracted from the environmental sensor data, and based on the coupling mechanism model of environment-late epidemic, the theoretical high-risk time intervals corresponding to the environmental factors and the high risk of late epidemic are determined. Determine the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval, and then determine the precise time delay between the time series variation curve of the environmental factor and the time series variation curve characterizing the occurrence of potato late blight based on the approximate time offset; The temporal variation curve of the environmental factor is shifted along the time axis based on the precise time delay, thereby predicting the spread trend of potato late blight and determining the priority area for the current sampling operation based on the spread trend.

2. The method as described in claim 1, characterized in that, The extraction of time-series variation curves of multispectral index combinations characterizing the occurrence of potato late blight from the aforementioned time-series hyperspectral image sequence specifically includes: Extract the potato late blight sensitive band dataset from each frame of the time-series hyperspectral image sequence; The multispectral index combination of the corresponding frame image was determined based on the datasets of sensitive bands for each late epidemic disease. The multispectral index combinations of each image frame are weighted and fused to generate late epidemic disease feature values ​​corresponding to each image frame. A time-series variation curve of a multispectral index combination characterizing the occurrence of potato late blight was constructed based on the late blight feature values ​​corresponding to all frames of images.

3. The method as described in claim 1, characterized in that, The specific time intervals for identifying suspected outbreaks where late-stage disease characteristic values ​​exceed preset dynamic thresholds include: Obtain the preset dynamic threshold; Traverse the time-series variation curves of the multispectral index combination characterizing the occurrence of potato late blight, and compare the late blight characteristic values ​​at each time point with the dynamic threshold. When the late-stage disease characteristic values ​​at multiple consecutive time points exceed the dynamic threshold, the consecutive time period is marked as a suspected outbreak time interval.

4. The method as described in claim 1, characterized in that, The extraction of time-series variation curves of environmental factors from the environmental sensor data specifically includes: Time-series data of air temperature, relative humidity, soil volumetric water content, and photosynthetically active radiation intensity are extracted from the environmental sensor data. The time-series data is smoothed to generate time-series variation curves of environmental factors.

5. The method as described in claim 1, characterized in that, Based on the coupling mechanism model of environment and late epidemic disease, the theoretical high-risk time intervals corresponding to high late epidemic disease risk for environmental factors are determined to include: The time-series variation curves of the environmental factors are input into the environmental-late epidemic disease coupling mechanism model to calculate the daily infection risk index sequence; Obtain the pre-set risk threshold; By traversing the daily infection risk index sequence, time periods that continuously exceed the risk threshold are marked as theoretically high-risk time intervals.

6. The method as described in claim 1, characterized in that, Determining the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval specifically includes: Obtain the midpoint of the suspected outbreak time interval; Obtain the midpoint of the theoretically high-risk time interval; The time difference between the midpoint of the suspected outbreak time interval and the midpoint of the theoretical high-risk time interval is calculated as a rough time offset.

7. The method as described in claim 1, characterized in that, The precise time delay between the temporal variation curve of the environmental factor and the temporal variation curve characterizing the occurrence of potato late blight, determined based on the coarse time offset, specifically includes: The time-series variation curves of the environmental factors and the time-series variation curves characterizing the occurrence of potato late blight were normalized. The search window is determined with the approximate time offset as the center. Within the search window, the correlation coefficients between the time-series variation curves of the normalized environmental factors and the time-series variation curves representing the occurrence of potato late blight under different time shifts are calculated. Iterate through all correlation coefficients and select the time shift corresponding to the largest correlation coefficient as the precise time delay.

8. An artificial intelligence-based agricultural pest and disease detection and sampling system, comprising a potato late blight detection and sampling unit, characterized in that, The potato late blight detection sampling unit includes: The acquisition module is used to simultaneously acquire time-series hyperspectral image sequences of the target farmland and environmental sensor data within the corresponding time window; The processing module is used to extract the time-series change curve of the multispectral index combination characterizing the occurrence of potato late blight from the time-series hyperspectral image sequence, and then identify the suspected outbreak time interval where the late blight characteristic value exceeds the preset dynamic threshold. The processing module is also used to extract the time-series variation curves of environmental factors from the environmental sensor data, and based on the coupling mechanism model of environment-late epidemic, determine the theoretical high-risk time intervals of environmental factors corresponding to the high risk of late epidemic. The processing module is also used to determine the approximate time offset between the suspected outbreak time interval and the theoretical high-risk time interval, and then determine the precise time delay between the time series change curve of the environmental factor and the time series change curve characterizing the occurrence of potato late blight based on the approximate time offset. The execution module is used to shift the time-series change curve of the environmental factors according to the precise time delay, thereby predicting the spread trend of potato late blight, and determining the priority area for the current sampling operation based on the spread trend.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the potato late blight detection sampling method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the potato late blight detection sampling method as described in any one of claims 1 to 7.

Images