A method for early identification of crop diseases and variable application based on multi-modal phenotype time sequence fusion

Abstract

The application discloses a crop disease early identification and variable pesticide application method based on multi-modal phenotype time sequence fusion, and relates to the technical field of intelligent agriculture. The growth period of target crops, the insect population density of target insects and multi-modal phenotype data at multiple collection time points are acquired; the single-factor phenotype change interval corresponding to each mode is determined according to the growth period and the insect population density; the phenotype change amount and the out-of-bound amount of each mode between adjacent collection time points are determined; the out-of-bound amount of different modes is time sequence fused according to the collection time to form a multi-modal phenotype out-of-bound time sequence feature; the disease risk and the variable pesticide application parameter of each operation area are determined according to the feature, and variable pesticide application is performed. The application can reduce the interference of abnormal target insect stress phenotypes on disease risk judgment, and make the variable pesticide application parameter correspond to the disease risk.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and in particular to a method for early identification of crop diseases and variable-rate application of pesticides based on multimodal phenotypic temporal fusion. Background Technology

[0002] In the early stages of crop disease, physiological traits such as leaf pigment content, tissue structure, and water status can change before obvious lesions appear. These changes are reflected in leaf spectral reflectance characteristics, canopy temperature, and visible light phenotypes. Hyperspectral imaging technology can continuously record narrow-band reflectance information and has been used to detect the latent stage of crop diseases. The paper "Detection Method for the Late Stage of Tomato Leaf Mold Based on Hyperspectral Imaging," published in the *Transactions of the Chinese Society for Agricultural Machinery* (Vol. 56, No. 8, 2025), uses characteristic wavelengths to screen hyperspectral curves of leaves with latent tomato leaf mold and healthy leaves, and identifies latent diseases based on the selected spectral features. Chinese invention patent application CN119992327A further discloses acquiring visible light images, temperature images, and spectral data of crop leaves at preset time intervals, performing temporal alignment and multimodal feature fusion on continuously acquired data, and determining the disease type and severity based on the fused features. Therefore, continuous observation of spectral, temperature, and visible light phenotype changes during crop disease occurrence can be used for early disease identification.

[0003] When crops are eaten by insects, changes occur in leaf tissue damage, chlorophyll status, and spectral reflectance characteristics. These phenotypic changes are already used in pest monitoring to identify the degree of insect stress. The paper "Research on Intelligent Identification Method of Tea Tree Pests Based on Hyperspectral Imaging with Multimodal Feature Fusion," published in the *Journal of Shenyang Agricultural University*, Vol. 56, No. 6, 2025, collects spectral reflectance, texture parameters, and chlorophyll fluorescence characteristics of tea trees and identifies pests based on multimodal phenotypic characteristics under different pest conditions. Chinese invention patent application CN121686520A discloses the collection of visible light information and canopy spectral information of crops, the identification of diseases and pests through multimodal feature fusion, and the determination of corresponding pesticide application information based on the identification results. It is evident that crop disease phenotypic identification, pest phenotypic identification, and variable-rate pesticide application based on identification results can already be combined and applied in field plant protection operations.

[0004] However, in field conditions where the target crop is simultaneously subjected to target insect stress, existing early disease identification methods usually determine disease risk directly based on the collected multimodal phenotypic abnormalities. It is still difficult to determine whether the multimodal phenotypic abnormalities are within the range of phenotypic changes that the target insect single-factor stress can produce under the corresponding insect population density. This can easily lead to the inclusion of some insect phenotypic abnormalities in the disease risk and affect the determination of subsequent variable pesticide application rates. Summary of the Invention

[0005] The purpose of this invention is to provide a method for early identification of crop diseases and variable application of pesticides based on multimodal phenotypic time-series fusion. This method can solve the problem that phenotypic changes caused by target insect stress participate in disease risk assessment and affect the determination of variable application parameters, reduce the interference of abnormal target insect stress phenotypic patterns on disease risk assessment, and make variable application parameters correspond to disease risk.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for early identification of crop diseases and variable-rate pesticide application based on multimodal phenotypic time-series fusion includes: Acquire the growth period of the target crop, the population density of the target insect, and multimodal phenotypic data at multiple collection times within each operational area of ​​the target field; Based on the reproductive period and insect population density, the range of single-factor phenotypic variation corresponding to each modality of multimodal phenotypic data is determined from the pre-established single-factor phenotypic reference data that are only subjected to target insect stress. The phenotypic changes of each modality between adjacent acquisition times are determined based on multimodal phenotypic data; The out-of-bounds amount for each mode is determined based on the magnitude of the phenotypic change relative to the boundary of the corresponding single-factor phenotypic change interval. The out-of-bounds amounts of different modalities are fused temporally according to the collection time to form multimodal phenotypic out-of-bounds temporal features; Determine the disease risk of each work area based on the time sequence characteristics of multimodal phenotypic boundary crossing; Determine the variable application parameters for each work area based on disease risk; Variable-rate pesticide application is performed on the corresponding work area according to variable-rate pesticide application parameters.

[0007] Preferably, the multimodal phenotypic data includes visible light phenotypic data, spectral phenotypic data, and canopy temperature phenotypic data.

[0008] Preferably, the pre-established single-factor phenotypic reference data subjected only to the target insect stress are established according to the following steps: Select reference plants that are the same variety as the target crop in the target field and have negative disease test results; Reference plants were set according to multiple preset growth periods; Under each preset growth stage, the target insects were subjected to single-factor stress on the reference plants according to multiple preset insect population density ranges; During single-factor stress, multimodal phenotypic data of reference plants were acquired at multiple collection points; Multimodal phenotypic data of reference plants were collected according to preset growth period and preset insect population density range to form single-factor phenotypic reference data.

[0009] Preferably, the range of single-factor phenotypic variation is determined according to the following steps: From the single-factor phenotypic reference data corresponding to the same preset growth period and the same preset insect population density range, the reference phenotypic feature values ​​of each modality at multiple collection times are determined; The reference phenotypic change of each mode is determined based on the reference phenotypic feature values ​​at adjacent acquisition times; The minimum value among multiple reference phenotypic changes corresponding to the same mode is determined as the lower boundary of the interval, and the maximum value is determined as the upper boundary of the interval; The range between the lower and upper boundaries of the interval is defined as the single-factor phenotypic variation range under the corresponding preset reproductive period and the corresponding preset insect population density range.

[0010] Preferably, based on the growth period and insect population density, the range of single-factor phenotypic variation corresponding to each modality of the multimodal phenotypic data is determined from pre-established single-factor phenotypic reference data subjected only to the target insect stress, including: Determine the corresponding pre-defined reproductive period from multiple pre-defined reproductive periods; Determine the preset insect population density range that includes the insect population density from multiple preset insect population density ranges; From the multiple single-factor phenotypic variation intervals that correspond to both the preset reproductive period and the preset insect population density interval, the single-factor phenotypic variation intervals corresponding to each modality are determined.

[0011] Preferably, determining the phenotypic change of each modality between adjacent acquisition times based on multimodal phenotypic data includes: Determine the phenotypic feature values ​​corresponding to each modality from the multimodal phenotypic data at each acquisition time; Phenotypic feature values ​​corresponding to the same modality are arranged according to the collection time; Calculate the difference between the phenotypic feature value of the later acquisition time and the phenotypic feature value of the earlier acquisition time in adjacent acquisition time, and determine the difference as the phenotypic change of the corresponding mode at the later acquisition time.

[0012] Preferably, the out-of-bounds amount for each mode is determined based on the magnitude of the phenotypic change relative to the boundary of the corresponding single-factor phenotypic change interval, including: Determine the lower and upper boundaries of the single-factor phenotypic variation interval for the corresponding modality; When the phenotypic change is between the lower boundary and the upper boundary of the interval, the out-of-bounds amount of the corresponding mode is set to zero; When the phenotypic change is less than the lower boundary of the interval, the difference between the lower boundary of the interval and the phenotypic change is determined as the out-of-bounds amount of the corresponding mode. When the phenotypic change is greater than the upper boundary of the interval, the difference between the phenotypic change and the upper boundary of the interval is determined as the out-of-bounds amount of the corresponding mode.

[0013] Preferably, the out-of-bounds amounts of different modalities are time-series fused according to the acquisition time to form multimodal phenotypic out-of-bounds temporal features, including: Determine the interval width of the single-factor phenotypic variation interval corresponding to each modality; The ratio between the out-of-bounds amount of each mode and the corresponding interval width is determined as the standardized out-of-bounds amount; Arrange the standardized out-of-bounds quantities corresponding to the same acquisition time according to the preset modal arrangement order to form the multimodal out-of-bounds features corresponding to the acquisition time. The multimodal out-of-bounds features corresponding to multiple acquisition times are arranged according to the acquisition time to form multimodal phenotypic out-of-bounds temporal features.

[0014] Preferably, the disease risk of each work area is determined based on the time sequence characteristics of multimodal phenotypic boundary crossing, including: The multimodal phenotypic out-of-bounds temporal features corresponding to each work area are input into the disease identification model. The disease identification model is trained from the multimodal phenotypic out-of-bounds temporal features corresponding to the training samples with disease occurrence status labels and disease severity labels. The probability of disease occurrence and the severity of disease in each work area are determined by a disease identification model. Based on the preset disease risk assessment rules, as well as the probability and severity of disease occurrence, the disease risk of the corresponding work area is determined.

[0015] Preferably, the variable application parameters for each work area are determined based on the disease risk, including: Match the disease risk of each work area with multiple preset disease risk intervals to determine the disease risk level of the corresponding work area; Based on the pre-established correspondence between disease risk levels and variable application parameters, determine the variable application parameters corresponding to the disease risk levels.

[0016] The present invention discloses the following beneficial effects: (1) Based on the growth period of the target crop and the population density of the target insect, this invention determines the range of single-factor phenotypic changes corresponding to each modality from the reference data of single-factor phenotypic changes that are only under the stress of the target insect, so that the multimodal phenotypic data collected in the field has a reference range corresponding to the current growth status and insect population density, thereby reducing the situation where phenotypic changes within the range of single-factor stress of the target insect are directly used for disease risk assessment.

[0017] (2) Based on the phenotypic changes of each modality between adjacent acquisition times, this invention determines the excess amplitude of the phenotypic change relative to the boundary of the corresponding single-factor phenotypic change interval, and uses the excess amplitude to form the boundary crossing amount of each modality. The boundary crossing amount directly reflects the degree to which the current phenotypic change exceeds the range of single-factor phenotypic changes of the target insect, thus further limiting the information used for disease risk assessment from the original phenotypic abnormality to phenotypic changes that exceed the corresponding reference range.

[0018] (3) This invention performs time-series fusion of the out-of-bounds quantities of different modalities according to the collection time to form multimodal phenotypic out-of-bounds temporal features, and determines the disease risk of each operating area based on the multimodal phenotypic out-of-bounds temporal features, and then determines the variable application parameters based on the disease risk. Thus, the variable application parameters correspond to the disease risk after being constrained by the single-factor phenotypic change range of the target insect, reducing the interference of insect phenotypic abnormalities on the determination process of variable application parameters. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 A diagram illustrating the relationship between boundary crossings provided in this embodiment of the invention; Figure 3 The near-infrared phenotypic characteristic time-series curves provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1As shown, this embodiment provides a method for early identification of crop diseases and variable-rate application of pesticides based on multimodal phenotypic time-series fusion, including: Step 100: Obtain the growth period of the target crop, the population density of the target insects, and multimodal phenotypic data at multiple collection times within each operational area of ​​the target field; Step 200: Based on the growth period and insect population density, determine the range of single-factor phenotypic variation for each modality of the multimodal phenotypic data from the pre-established single-factor phenotypic reference data that are only subjected to target insect stress; Step 300: Determine the phenotypic changes of each modality between adjacent acquisition times based on the multimodal phenotypic data; Step 400: Determine the out-of-bounds amount for each mode based on the magnitude of the phenotypic change relative to the boundary of the corresponding single-factor phenotypic change interval. Step 500: Perform temporal fusion of the out-of-bounds amounts of different modalities according to the acquisition time to form multimodal phenotypic out-of-bounds temporal features; Step 600: Determine the disease risk of each work area based on the multimodal phenotypic out-of-bounds timing characteristics; Step 700: Determine the variable application parameters for each work area based on the disease risk; Step 800: Perform variable application of pesticides on the corresponding work area according to the variable application parameters.

[0024] In one specific implementation, chili peppers are used as the target crop and the green peach aphid as the target insect. The target field is divided into multiple operational areas according to the spatial range where the variable application parameters can be determined separately. The boundaries of the operational areas remain unchanged at different data collection times. Visible light images, spectral data, and thermal infrared data are spatially located and mapped to the corresponding operational areas. Each operational area generates a set of multimodal phenotypic data at one data collection time.

[0025] The population density of the green peach aphid was determined based on the number of green peach aphids per unit leaf or per unit plant. The same statistical objects were used in both the single-factor phenotypic reference data establishment phase and the target field detection phase. The population density of the corresponding work area was recorded simultaneously each time multimodal phenotypic data was obtained, and the phenotypic data and insect infestation data were observed over the same time period.

[0026] Multimodal phenotypic data includes visible light phenotypic data, spectral phenotypic data, and canopy temperature phenotypic data. These three modes respectively reflect the canopy color and texture of the target crop, leaf or canopy reflectivity, and canopy thermal state. Different modes are not required to have the same spatial resolution; after spatial localization, the operational area is used as the corresponding data unit.

[0027] In one specific implementation, the visible light image first separates the target crop canopy region.

[0028] Let the first The average values ​​of the red channel, green channel, and blue channel of the target crop pixels within the work area at each acquisition time are respectively... , and The color changes of the visible light modes are represented by green chromaticity coordinates.

[0029] The green chromaticity coordinates are calculated using the relative proportions of the three color channels. The same canopy segmentation range and color feature calculation methods are used in both the reference data establishment phase and the target field detection phase.

[0030] Spectral data reflectance correction was performed using white and dark references. Let the original spectral response, dark reference response, white reference response, and known reflectance of the white reference plate at the i-th acquisition time for any wavelength be represented by the corresponding parameters in the following formula. The corrected reflectance is: In the formula, This indicates the corrected reflectance; Represents the original spectral response; Indicates the dark reference response; Indicates a clear reference response; This represents the known reflectivity of the white reference plate at the corresponding wavelength. The above calculation is used when the white reference response and the dark reference response are not equal. The same characteristic band is used at different acquisition times.

[0031] When the sensitive bands corresponding to the target insect or target disease have been determined, spectral phenotypic characteristics are formed by using the reflectance of the sensitive band or the ratio between the reflectance of the bands.

[0032] To illustrate the calculation process of spectral phenotypic characteristics, Table 1 lists a set of near-infrared reflectance data for pepper leaves under the stress of the green peach aphid at different observation times. Both near-infrared bands use corrected reflectance.

[0033] Table 1. Near-infrared reflectance data of pepper leaves under the stress of green peach aphids.

[0034] The spectral phenotypic characteristics corresponding to the two near-infrared bands are represented by the reflectance ratio. Let the reflectances at 816.92 nm and 896.01 nm at the i-th acquisition time be the corresponding band reflectances, then the spectral phenotypic characteristic values ​​are: Substituting the reflectance values ​​from Table 1 into the above formula, we obtain the spectral phenotypic characteristic values ​​corresponding to each acquisition time: After using the reflectance ratio, no artificial fusion weights need to be set for the two bands. Table 1 is used to illustrate the numerical calculation relationship of spectral characteristics. The single-factor phenotypic variation range is still established based on the reference plant data within the same growth stage and the same preset insect population density range.

[0035] Canopy temperature phenotypic data are derived from the temperature difference between the canopy and the ambient air temperature at the time of acquisition. After extracting the target crop canopy region from the thermal infrared image, the average canopy temperature of the corresponding work area is calculated, and the ambient air temperature at the time of acquisition is simultaneously obtained. The canopy temperature phenotypic characteristic values ​​are: In the formula, Indicates the first The average canopy temperature of the work area at each sampling time This represents the ambient air temperature at the same sampling time. Soil, roads, and non-target vegetation areas are not included in the average canopy temperature calculation. After adopting the canopy ambient temperature difference, the changes in ambient air temperature and the changes in the thermal state of the target crop itself on different sampling dates are recorded separately.

[0036] In this embodiment, green chromaticity coordinates, near-infrared reflectance ratio, and canopy ambient temperature difference are used as phenotypic feature values ​​for the visible light mode, spectral mode, and canopy temperature mode, respectively. When using different phenotypic features for other crops or other diseases, the same feature calculation rules are maintained between the data establishment stage and the target field detection stage.

[0037] Single-factor phenotypic reference data were established before the target field testing. Reference plants of the same variety as the target crop in the target field were selected, and these plants were subjected to the same water, nutrient, and cultivation conditions, and tested for the target disease. Reference plants with negative disease test results were then entered into the single-factor stress experiment for the target insect.

[0038] Multiple preset growth stages were set according to the growth process of the target crop, and multiple preset insect population density intervals were set for each preset growth stage. Multiple reference plants were set for each preset insect population density interval, and multimodal phenotypic data were obtained for each reference plant at more than three collection times.

[0039] When the target insect population changed during the experiment, the insect population density was recalculated at each collection time. Only data from consecutive adjacent collection times falling within the same preset insect population density range were used to establish the reference phenotypic change for that range. Once the insect population density shifted to another preset insect population density range, the newly formed phenotypic change was assigned to the new preset insect population density range.

[0040] In one specific implementation, when the target insect is a vector for the target disease, the single-factor stress experiment uses an insect source with a negative target pathogen detection result. At the end of the experiment, the control plants are tested for the target disease again, and the data corresponding to the control plants with negative disease detection results are retained in the single-factor phenotypic reference data.

[0041] Data acquired due to sensor failure, significant canopy obstruction, spatial positioning errors, or measurements exceeding the sensor's effective range will not be included in the calculation of reference phenotypic changes. Data acquired normally but located at the edge of the reference sample distribution will still be retained.

[0042] Let the first The reference plant was in the first The data collection time corresponds to the first... The reference phenotypic characteristic value of each modality is The change in the reference phenotype formed at adjacent sampling times is: In the formula, This indicates the reference plant number. Indicates the modal number. This indicates the data collection time number. The phenotypic change is attributed to the next data collection time.

[0043] Within the same preset growth period and the same preset insect population density range, there are a total of Each effective reference plant has [number] plants, and each reference plant has [number] plants. At the effective data acquisition time, the first The set of reference phenotypic changes corresponding to each modality is represented as follows: No. The lower and upper boundaries of the intervals for each modality are taken as the minimum and maximum values ​​in the set, respectively: The single-factor phenotypic variation ranges corresponding to the current reproductive period and the current preset insect population density range are as follows: The interval boundaries are directly determined by the changes in the effective reference phenotypes, without setting additional standard deviation multiples or modal weights. When the number of reference plants or the number of sampling times are increased, the newly obtained changes in the effective reference phenotypes are directly added to the corresponding set, and the rules for determining the interval boundaries remain unchanged.

[0044] The spectral data in Table 1 further illustrate the calculation of phenotypic changes between adjacent acquisition times. The first acquisition time only forms a phenotypic baseline, and the difference between the second and previous acquisition times is calculated starting from the second acquisition time.

[0045] Table 2 Near-infrared phenotypic characteristics and changes between adjacent acquisition times

[0046] The first acquisition time in Table 2 has no previous acquisition time, so phenotypic changes are not calculated. The same time assignment rules are applied to the visible light mode and the canopy temperature mode. The time-series curves plotted based on the near-infrared phenotypic characteristics listed in Table 2 are as follows: Figure 3 As shown.

[0047] Figure 3 This is a time-series curve of near-infrared phenotypic characteristics provided in an embodiment of the present invention. The horizontal axis represents the time after aphid migration, in days, and the vertical axis represents the near-infrared phenotypic characteristics. The five data points on the curve correspond to the 2nd, 4th, 7th, 9th, and 14th days after aphid transfer, respectively, and represent the corresponding near-infrared phenotypic characteristics. The values ​​are 0.954545, 0.958904, 0.952381, 0.925373 and 0.962963, respectively. Adjacent data points are connected by a broken line to represent the relationship between near-infrared phenotypic features and acquisition time.

[0048] For the target field, let the first... The work area in the first The data collection time corresponds to the first... Phenotypic features of each modality are The phenotypic change between adjacent data collection times is: According to the The current reproductive stage and insect population density of each work area, obtaining the corresponding conditions for the first... The range of single-factor phenotypic variation for each modality. This range will be referred to as the lower boundary of the range. and the upper boundary of the interval .

[0049] When the actual phenotypic change falls within the corresponding single-factor phenotypic change range, the out-of-bounds value is taken as zero. When the actual phenotypic change is less than the lower boundary of the interval, the out-of-bounds amount is determined by the difference between the lower boundary of the interval and the actual phenotypic change. When the actual phenotypic change is greater than the upper boundary of the interval, the amount of exceeding the boundary is determined by the difference between the actual phenotypic change and the upper boundary of the interval: The above three positional relationships cover any valid phenotypic change. The out-of-bounds value is always not less than zero. A zero out-of-bounds value indicates that the current phenotypic change is still within the single-factor reference range of the target insect corresponding to the same growth stage and the same insect population density range; an out-of-bounds value greater than zero only records the amplitude exceeding the nearest interval boundary, and does not directly determine the occurrence of disease based on a single out-of-bounds value.

[0050] Figure 2 The diagram for determining the out-of-bounds amount provided in this embodiment of the invention shows that the horizontal axis represents the time of data collection and the vertical axis represents the phenotypic change. Hollow circles represent the actual phenotypic change values ​​corresponding to each data collection time, and broken lines connecting adjacent hollow circles represent the change of the actual phenotypic change with the data collection time. Indicates the first The first work area The mode in the th ... The actual phenotypic change corresponding to each data collection time. Indicates the first [year] under the current growth period and insect population density conditions. The upper boundary of the single-factor phenotypic variation interval corresponding to each modality. This represents the lower boundary of the single-factor phenotypic variation range; the actual phenotypic variation is higher than... At that time, the collection point was reached. The vertical distance represents the corresponding out-of-bounds amount. The actual phenotypic change was lower than At that time, the collection point was reached. The vertical distance represents the corresponding out-of-bounds amount. When the actual phenotypic change is between the upper and lower boundaries of the interval, the out-of-bounds distance shown in the figure is not formed.

[0051] The out-of-bounds quantities of different modalities have different dimensions. Before each modality enters the time series fusion, it is standardized according to the width of the corresponding single-factor phenotypic variation interval. The interval width of each mode is: When the interval width is greater than zero, the first... The first work area, the first The mode in the th ... The standardized out-of-bounds value at each acquisition time is: The standardized out-of-bounds value represents the proportion of the out-of-bounds magnitude to the width of the single-factor phenotypic variation range of the current mode. Each mode uses its own reference range to complete the scaling transformation, and no artificial weights are set between the visible light mode, the spectral mode, and the canopy temperature mode.

[0052] When the upper boundary of an interval equals the lower boundary, the interval width is zero, and standardized division is not performed. The interval boundary is redefined after adding valid reference data for the same growth period and insect population density interval; if the interval width is still zero after adding valid reference data, the current phenotypic feature does not participate in the multimodal time series fusion under the corresponding conditions.

[0053] Let the number of modes participating in the fusion be . , No. The work area in the first The multimodal out-of-bounds features corresponding to each acquisition time point are: The modes are arranged in a fixed order. When using the three modes of visible light, spectrum, and canopy temperature, the green chromaticity coordinates, near-infrared reflectance ratio, and standardized out-of-bounds values ​​corresponding to the canopy ambient temperature difference are arranged in sequence.

[0054] No. Each work area has At each data collection time point, the out-of-bounds temporal features of the multimodal phenotype are represented as follows: The first acquisition time is used to form the phenotypic feature baseline, and the multimodal out-of-bounds features are formed starting from the second acquisition time. The temporal features simultaneously retain the out-of-bounds states of different modalities at the same acquisition time, as well as the changes in the out-of-bounds states over time.

[0055] Furthermore, the number of consecutive out-of-bounds occurrences for a single mode can be recorded. When the standardized out-of-bounds value is greater than zero, the corresponding out-of-bounds state is recorded as 1; when the standardized out-of-bounds value is equal to zero, the corresponding out-of-bounds state is recorded as 0. The number of consecutive boundary violations is recursively calculated based on the number of consecutive boundary violations at the previous data collection time. If the data collection time is still in a boundary violation state: If no boundary violation occurs at the current data collection time: The number of consecutive out-of-bounds occurrences before the second data collection time is set to zero. When the next data collection time re-enters the single-factor phenotypic change range after an isolated out-of-bounds occurrence, the number of consecutive out-of-bounds occurrences returns to zero; when out-of-bounds occurrences occur at multiple consecutive data collection times for the same modality, the number of consecutive out-of-bounds occurrences increases with each data collection time.

[0056] The number of modes that exceed the limit at the same acquisition time can also be expressed as: In the formula, the indicator function is defined when the condition is met. The value is set to 1 if the condition is not met, and to 0 if the condition is not met. The number of out-of-bounds modes is directly derived from the standardized out-of-bounds values ​​of each mode, without setting mode weights.

[0057] In one specific implementation, the disease identification model employs a random forest model. Training samples undergo the same data processing procedure as the target field to generate multimodal phenotypic out-of-bounds temporal features. Healthy samples are labeled as category 0, and samples confirmed to have the target disease are labeled as categories 1 through 2 according to disease severity, from low to high. .

[0058] Let random forest include The first decision tree, the The number of decision tree votes obtained for each category is: , No. The voting probabilities for each category are: The sum of the probabilities of voting in each category satisfies: The voting probability for the health category is Disease occurrence probability for: The probability of disease occurrence is directly determined by the proportion of total votes received by the non-healthy category. When the value is greater than zero, the normalized severity of the disease Calculated based on the conditional voting distribution for the non-health category: When the value equals zero, the severity of the disease is taken as zero. The sum of the voting probabilities for non-healthy categories equals the disease occurrence probability; therefore, the normalized disease severity value ranges from zero to one. Both the disease occurrence probability and disease severity are retained. In early samples where a high disease severity has not yet been achieved, the corresponding disease occurrence probability information is still retained.

[0059] Disease risk is determined by a combination of pre-established disease occurrence probability ranges and disease severity ranges. Training and validation samples are used to determine the probability and severity ranges corresponding to each risk level. When the disease occurrence probability increases, the disease risk level does not decrease; within the same disease occurrence probability range, when the disease severity increases, the disease risk level does not decrease.

[0060] Step 700 determines the variable application parameters for the corresponding work area based on the disease risk level. In one specific implementation, the variable application parameters are the spray volume per unit area. The operating speed of the spraying equipment is Effective spray width is The required spray flow rate for the corresponding work area is: In the formula, The unit is L / ha. The unit is km / h. The unit is m. The unit is L / min. The constant 600 comes from the conversion between area and time units and is not an empirical correction factor. The values ​​of the variable application parameters are limited to the allowable application range of the corresponding pesticide for the target disease.

[0061] After the spraying equipment enters the corresponding work area, it performs spraying according to the variable application parameters associated with the current work area. When the operating speed changes, the spray flow rate is recalculated based on the current operating speed; when the spray volume per unit area remains constant, changes in the operating speed do not change the amount of pesticide applied per unit area in the corresponding work area.

[0062] In this embodiment, the original multimodal phenotypic data is not directly used to determine variable application parameters. Field phenotypic changes are first compared with the range of single-factor phenotypic changes of the target insect under the same growth stage and insect population density range. Changes within the reference range correspond to zero out-of-bounds values, while those exceeding the reference range are classified as positive out-of-bounds values ​​based on their distance from the interval boundary. Different modalities are then standardized using their respective single-factor phenotypic change range widths, and multimodal phenotypic out-of-bounds temporal characteristics are formed according to the collection time. Disease risk is based on these temporal characteristics; common phenotypic changes that can be explained by single-factor stress in the target insect will not directly increase the variable application parameters for the corresponding operating area simply due to changes in absolute values.

[0063] like Figure 4 As shown, this embodiment also provides a crop disease early identification and variable application system based on multimodal phenotypic time-series fusion, used to execute the aforementioned crop disease early identification and variable application method based on multimodal phenotypic time-series fusion.

[0064] The system includes a data acquisition unit, a reference interval determination unit, a phenotypic change determination unit, an out-of-bounds determination unit, a time-series fusion unit, a disease risk determination unit, a variable application parameter determination unit, and a variable application control unit.

[0065] The data acquisition unit is used to acquire the growth period of the target crop, the insect population density of the target insect, and multimodal phenotypic data at multiple collection times in each working area of ​​the target field, and sends the growth period, the insect population density, and the multimodal phenotypic data to the corresponding data processing unit.

[0066] The reference interval determination unit is used to determine the single-factor phenotypic variation intervals corresponding to each modality of the multimodal phenotypic data from the pre-established single-factor phenotypic reference data that are only subjected to the target insect stress, based on the reproductive period and the insect population density, and to send the single-factor phenotypic variation intervals to the boundary determination unit.

[0067] The phenotypic change determination unit is used to determine the phenotypic change of each mode between adjacent acquisition times based on the multimodal phenotypic data, and send the phenotypic change to the out-of-bounds determination unit.

[0068] The out-of-bounds determination unit is used to compare the phenotypic change amount of each modality with the corresponding single-factor phenotypic change interval, determine the out-of-bounds amount of each modality based on the excess amplitude of the phenotypic change amount relative to the boundary of the corresponding interval, and send the out-of-bounds amount to the time-series fusion unit.

[0069] The time-series fusion unit is used to perform time-series fusion of the out-of-bounds quantities of different modalities according to the acquisition time, forming multimodal phenotypic out-of-bounds time-series features corresponding to each working area, and sending the multimodal phenotypic out-of-bounds time-series features to the disease risk determination unit.

[0070] The disease risk determination unit is used to determine the disease risk of the corresponding work area based on the multimodal phenotypic out-of-bounds temporal characteristics of each work area, and send the disease risk to the variable application parameter determination unit.

[0071] The variable application parameter determination unit is used to determine the variable application parameters corresponding to each work area based on the disease risk, and send the variable application parameters to the variable application control unit.

[0072] The variable application control unit is used to control the variable application of pesticides in the corresponding work area according to the variable application parameters corresponding to each work area.

[0073] In one specific implementation, the data acquisition unit, the reference interval determination unit, the phenotypic change determination unit, the boundary crossing determination unit, the time-series fusion unit, the disease risk determination unit, and the variable application parameter determination unit are implemented by a data processing device executing corresponding data processing programs. The variable application control unit generates application control information for the corresponding work area based on the variable application parameters and controls the spraying execution mechanism to complete the application according to the corresponding variable application parameters.

[0074] When each unit completes processing according to the above data transfer relationships, the single-factor phenotypic change interval is used to define the range of phenotypic changes corresponding to single-factor stress on the target insect under the current growth stage and insect population density conditions. The out-of-bounds amount is used to characterize the amplitude of the actual phenotypic change exceeding the boundary of the corresponding interval. The multimodal phenotypic out-of-bounds temporal characteristics retain the changes of different modal out-of-bounds states over the collection time. Disease risk is determined based on the multimodal phenotypic out-of-bounds temporal characteristics and serves as the basis for determining the pesticide application parameters for the corresponding operating area.

[0075] This embodiment selects corresponding single-factor phenotypic change intervals as references based on the growth stage of the target crop and the population density of the target insect. This ensures that phenotypic changes caused by single-factor stress of the target insect under different growth stages and population densities have corresponding judgment ranges. Furthermore, based on multimodal phenotypic data from multiple collection times, the amount of phenotypic change between adjacent collection times is determined. The magnitude of the phenotypic change exceeding the boundary of the corresponding interval is used to form an out-of-bounds quantity. Phenotypic changes within the single-factor phenotypic change interval have zero out-of-bounds quantities, thereby reducing the direct involvement of phenotypic changes caused by target insect stress in disease risk assessment. After the out-of-bounds quantities of different modalities are used to form multimodal phenotypic out-of-bounds temporal characteristics according to collection time, the determination of disease risk not only relies on local anomalies at a certain collection time but also utilizes the changes in out-of-bounds states of different modalities over time, thus preserving phenotypic anomaly information continuously formed during the early disease development process. Finally, based on the disease risk of each operating area, variable application parameters are determined to ensure that the application intensity corresponds to the disease risk after being constrained by the range of single-factor phenotypic changes of the target insect, thereby reducing the possibility that abnormal insect phenotypic changes will cause an excessively high disease risk and lead to an increase in application parameters.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0077] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion, characterized in that, include: Acquire the growth period of the target crop, the population density of the target insect, and multimodal phenotypic data at multiple collection times within each operational area of ​​the target field; Based on the reproductive period and the insect population density, the range of single-factor phenotypic variation corresponding to each modality of the multimodal phenotypic data is determined from the pre-established single-factor phenotypic reference data that are only subjected to the target insect stress. The phenotypic changes of each modality between adjacent acquisition times are determined based on the multimodal phenotypic data; The out-of-bounds amount for each mode is determined based on the magnitude of the phenotypic change relative to the boundary of the corresponding single-factor phenotypic change interval. The out-of-bounds amounts of different modalities are time-series fused according to the collection time to form multimodal phenotypic out-of-bounds temporal features; The disease risk of each work area is determined based on the cross-boundary timing characteristics of the multimodal phenotypes. Determine the variable application parameters for each work area based on the aforementioned disease risk; Variable application of pesticides is carried out in the corresponding work area according to the variable application parameters.

2. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 1, characterized in that, The multimodal phenotypic data includes visible light phenotypic data, spectral phenotypic data, and canopy temperature phenotypic data.

3. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic temporal fusion according to claim 1, characterized in that, The pre-established single-factor phenotypic reference data, which are only subjected to the target insect stress, are established according to the following steps: Select reference plants that are the same variety as the target crop in the target field and have negative disease test results; The reference plants are set according to multiple preset growth periods; Under each of the preset growth stages, the reference plant is subjected to single-factor stress from the target insect according to multiple preset insect population density ranges; During the single-factor stress, multimodal phenotypic data of the reference plant were acquired at multiple sampling times; The multimodal phenotypic data of the reference plants are collected according to the preset growth period and the preset insect population density range to form the single-factor phenotypic reference data.

4. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 3, characterized in that, The range of single-factor phenotypic variation is determined according to the following steps: From the single-factor phenotypic reference data corresponding to the same preset reproductive period and the same preset insect population density range, determine the reference phenotypic feature values ​​of each modality at multiple collection times; The reference phenotypic change of each mode is determined based on the reference phenotypic feature values ​​at adjacent acquisition times; The minimum value among the multiple reference phenotypic changes corresponding to the same modality is determined as the lower boundary of the interval, and the maximum value is determined as the upper boundary of the interval; The range between the lower boundary and the upper boundary of the interval is defined as the single-factor phenotypic change range under the corresponding preset reproductive period and the corresponding preset insect population density range.

5. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 4, characterized in that, Based on the stated reproductive period and insect population density, the range of single-factor phenotypic variation corresponding to each modality of the multimodal phenotypic data is determined from pre-established single-factor phenotypic reference data only subjected to the target insect stress, including: Determine the preset reproductive period corresponding to the reproductive period from among the multiple preset reproductive periods; Determine a preset insect population density range that includes the insect population density from a plurality of preset insect population density ranges; From the multiple single-factor phenotypic change intervals that correspond to both the preset reproductive period and the preset insect population density interval, the single-factor phenotypic change interval corresponding to each modality is determined.

6. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 1, characterized in that, Based on the multimodal phenotypic data, the phenotypic changes of each modality between adjacent acquisition times are determined, including: The phenotypic feature values ​​corresponding to each modality are determined from the multimodal phenotypic data at each acquisition time. The phenotypic feature values ​​corresponding to the same modality are arranged according to the collection time. Calculate the difference between the phenotypic feature value at the later acquisition time and the phenotypic feature value at the previous acquisition time in adjacent acquisition times, and determine the difference as the phenotypic change of the mode corresponding to the later acquisition time.

7. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 6, characterized in that, The out-of-bounds amount for each mode is determined based on the magnitude of the phenotypic change relative to the boundary of the corresponding single-factor phenotypic change interval, including: Determine the lower and upper boundaries of the interval for the single-factor phenotypic variation of the corresponding modality; When the phenotypic change is between the lower boundary and the upper boundary of the interval, the out-of-bounds amount of the corresponding mode is determined to be zero; When the phenotypic change is less than the lower boundary of the interval, the difference between the lower boundary of the interval and the phenotypic change is determined as the out-of-bounds amount of the corresponding mode. When the phenotypic change is greater than the upper boundary of the interval, the difference between the phenotypic change and the upper boundary of the interval is determined as the out-of-bounds amount of the corresponding mode.

8. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 7, characterized in that, The out-of-bounds amounts of different modalities are time-series fused according to the acquisition time to form multimodal phenotypic out-of-bounds temporal features, including: Determine the interval width of the single-factor phenotypic variation interval corresponding to each modality; The ratio between the out-of-bounds amount of each mode and the corresponding interval width is determined as the standardized out-of-bounds amount; The standardized out-of-bounds quantities corresponding to the same acquisition time are arranged according to the preset modal arrangement order to form the multimodal out-of-bounds features corresponding to the acquisition time. The multimodal out-of-bounds features corresponding to multiple acquisition times are arranged according to the acquisition time to form the multimodal phenotypic out-of-bounds temporal features.

9. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 8, characterized in that, Based on the multimodal phenotypic out-of-bounds timing characteristics, the disease risk of each work area is determined, including: The multimodal phenotypic out-of-bounds temporal features corresponding to each work area are input into the disease identification model. The disease identification model is trained from the multimodal phenotypic out-of-bounds temporal features corresponding to training samples with disease occurrence status labels and disease severity labels. The disease identification model is used to determine the probability of disease occurrence and the severity of disease in each work area. Based on the preset disease risk assessment rules, the probability of disease occurrence, and the severity of disease, the disease risk of the corresponding work area is determined.

10. The method for early identification and variable-rate application of crop diseases based on multimodal phenotypic time-series fusion according to claim 9, characterized in that, Based on the aforementioned disease risk, determine the variable application parameters for each work area, including: The disease risk of each work area is matched with multiple preset disease risk intervals to determine the disease risk level of the corresponding work area; Based on the pre-established correspondence between the disease risk level and the variable application parameters, the variable application parameters corresponding to the disease risk level are determined.

Images