A monitoring and early warning system and method for fruit tree diseases and insect pests

Abstract

The present application relates to the field of agricultural information technology, and specifically discloses a kind of monitoring and early warning system and method of fruit tree disease and insect pest, comprising: the color abnormality patch of crown layer digital image is extracted by segmentation and collection, and time series evolution profile is generated by cross-time phase tracking;Electronic account of plant protection operation is read to form pesticide deposition sequence;With time series evolution profile is analyzed, and liquid residue suspected patch is defined;Based on the color removal rate obtained by separation in corresponding time series evolution profile, combined with the decomposition and attenuation characteristics of pesticide, the disease suspicious spot is identified;The corresponding leaf wetting time length and crown temperature sequence of disease suspicious spot are obtained, standard area sequence is generated, and the coincidence measure value of disease suspicious spot is calculated;The disease spot to be warned is identified, and the early warning level is determined according to its distribution, and the disease and insect pest early warning information is generated and output;Pesticide spot false alarm is effectively filtered out, the failure of early warning system caused by frequent false alarm is avoided, and the reliability of output disease and insect pest early warning information is ensured.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, specifically to a monitoring and early warning system and method for fruit tree diseases and pests. Background Technology

[0002] In modern orchard disease continuous monitoring, the system needs to frequently photograph leaves based on machine vision and perform image recognition to output lesions. However, after orchards regularly spray protective fungicides such as Bordeaux mixture and mancozeb, the dried solution leaves light blue or grayish-white irregular spots on the leaf surface. The texture, color gradation, and edge features are highly similar to the very early symptoms of diseases such as brown spot and anthracnose. Existing recognition models are mostly trained using standard diseased leaf images from untreated experimental orchards or public datasets, resulting in a lack of ability to distinguish drug spots. The damage is significant under long-term continuous monitoring. After each spraying, the system misreports numerous pesticide spots as disease within a short period, leading to frequent false alarms from the early warning platform. Orchard managers gradually become accustomed to ignoring these warnings, thus drowning out information about genuine disease outbreaks. Furthermore, pesticide traces fade and evolve over time, exhibiting time-varying characteristics that further increase the difficulty of differentiation. Since system development and agricultural operations are separate processes, developers struggle to fully perceive the cumulative damage caused by these time-varying and fading interference factors to the identification system. This results in such frequent false alarms remaining hidden for a long time, severely hindering the practical implementation of automatic early warning technology for orchard diseases.

[0003] Therefore, the present invention provides a monitoring and early warning system and method for fruit tree diseases and pests. Summary of the Invention

[0004] The purpose of this invention is to provide a monitoring and early warning system and method for fruit tree diseases and pests to solve the aforementioned problems.

[0005] The objective of this invention can be achieved through the following technical solution: a method for monitoring and early warning of fruit tree diseases and pests, comprising the following: Multi-temporal digital images of the canopy are acquired, segmented, and color aberration patches are extracted. The color aberration patches are then tracked across time phases to extract their contours and color attributes, generating a temporal evolution contour. The electronic ledger of plant protection operations is read to construct pesticide deposition events and form pesticide deposition sequences; the temporal evolution profile of abnormal color patches is verified with the pesticide deposition sequence in terms of time domain, space and appearance to identify suspected pesticide residue patches; For suspected pesticide residue patches, the decolorization rate is separated from the temporal evolution profile. The decolorization rate is combined with the pesticide decomposition and decay characteristics to identify pesticide residue patches and suspected disease patches. Obtain the leaf surface wetting duration and canopy temperature sequence corresponding to suspected disease spots, input them into the disease symptom progression reference scale library, generate a standard area sequence, calculate the coincidence value between the temporal evolution profile of suspected disease spots and the standard area sequence, and store them in association. Based on the coincidence metric value of suspected disease spots, disease spots to be warned are identified, and the warning level is determined according to the distribution of the coincidence metric values ​​of all spots to be warned. Disease and pest warning information is generated and output.

[0006] Furthermore, the process of generating the temporal evolution profile is as follows: All color anomalous patches are tracked across time phases. The contour intersection-union ratio of color anomalous patches in adjacent time phases is calculated. The Hungarian algorithm is used to solve the pairing problem and obtain the cross-time phase record of each color anomalous patch. By recording color anomalous patches across time phases, the moment when the color anomalous patches were first detected was extracted and marked as the start time; the contour of the color anomalous patches in each time phase was extracted, and the physical area of ​​the color anomalous patches was obtained through a pre-calibrated object surface resolution, forming an area transition sequence. For color aberration patches, the average color value of each phase is extracted from the cross-temporal records to form the color fading process; the start time, area change sequence and color fading process of the color aberration patches are associated and stored to generate a temporal evolution profile.

[0007] Furthermore, the process of forming the drug deposition sequence is as follows: Read the electronic ledger of plant protection operations, extract the spraying completion time, the name of the active ingredient of the sprayed pesticide, and the coverage area of ​​each spraying operation, and construct it as a pesticide deposition event; Each pesticide deposition event is assigned a unique event number, and the spray completion time, active ingredient name, and coverage area are combined to form a pesticide deposition sequence.

[0008] Furthermore, the process of identifying suspected drug residue patches is as follows: The temporal evolution profile of the color anomalous patches is verified against the drug deposition sequence in terms of time, space, and appearance. If a color anomalous patch satisfies the temporal overlap, spatial location matching, and appearance characteristics with only one drug deposition event, then the color anomalous patch is marked as a suspected drug residue patch.

[0009] Furthermore, the method of triple verification in terms of time, space, and appearance is as follows: The starting point of the appearance of the drug deposition event is obtained, and the time when the spot can be detected is searched to form a drug spot detection time window. If the starting time of the color abnormal spot falls into the drug spot detection time window, then the color abnormal spot and the drug deposition event have temporal overlap. The sampling locations of abnormal color patches and the operational coverage of the pesticide deposition event were obtained. The sampling locations included the fruit tree rows and the inter-tree sections, and the operational coverage included the range of the treated fruit tree rows and the range of the treated inter-tree sections. If the location of the abnormal color patch in the row of fruit trees falls within the range of the drug-treated fruit tree row, and the location of the abnormal color patch in the inter-tree section falls within the range of the drug-treated inter-tree section, then the sampling location of the abnormal color patch is determined to match the spatial location of the drug deposition event. Obtain the initial color attributes of the color-abnormal patches and the reference data of the dry patch color of the reagent deposition event, calculate the Euclidean distance. If the Euclidean distance falls within the typical color distribution range, the color-abnormal patches are consistent with the appearance characteristics of the reagent deposition event.

[0010] Furthermore, the process of identifying pesticide residue spots and suspected disease spots is as follows: For suspected drug residue patches, the decolorization rate was separated from the temporal evolution profile, and the upper and lower bounds of the decay rate fluctuation of the drug decomposition decay characteristics were extracted. If the decolorization rate of a suspected drug residue patch is greater than or equal to the lower limit of the decay rate fluctuation and less than or equal to the upper limit of the decay rate fluctuation, it is determined to be a drug residue patch. If the fading rate of suspected plaques with residual pesticide residue is less than the lower limit of the decay rate fluctuation or greater than the upper limit of the decay rate fluctuation, then the plaques are identified as suspected disease spots.

[0011] Furthermore, the decolorization rate is obtained by separation as follows: Obtain the color fading process in the temporal evolution profile corresponding to the suspected drug residue patches; for any two adjacent time phases in the color fading process, calculate the difference in the mean color value between the later time phase and the previous time phase, obtain the lightness difference ΔL, chromaticity a difference Δa and chromaticity b difference Δb, and calculate the color difference value. Obtain two corresponding acquisition times for adjacent time phases, and calculate the time difference between the two acquisition times as the interval between adjacent time phases; The decolorization rate is obtained by calculating the ratio of the color difference value to the interval duration.

[0012] Furthermore, the process of calculating the consistency metric between the temporal evolution profile of suspected disease spots and the standard area sequence is as follows: Based on the temporal evolution contours of suspected disease spots, area change sequences are extracted, and the Pearson correlation coefficient between the area change sequences and the standard area sequences is calculated as a measure of consistency.

[0013] Furthermore, the process of determining the warning level is as follows: Obtain the coincidence measurement values ​​of all disease spots to be warned, form a set of coincidence measurement values ​​to be warned, calculate the second quartile and the third quartile; and obtain the warning reference value; If the coincidence value of the disease spot to be warned is greater than or equal to the third quartile, a level one warning is set. If the coincidence value is greater than or equal to the second quartile and less than the third quartile, a level two warning is set. If the coincidence value is greater than or equal to the warning reference value and less than the second quartile, a level three warning is set.

[0014] A monitoring and early warning system for fruit tree diseases and pests includes the following: The temporal tracking module acquires multi-temporal digital images of the canopy, segments and extracts color aberration patches, performs cross-temporal tracking of color aberration patches to extract contours and color attributes, and generates temporal evolution contours. The triple verification module reads the electronic ledger of plant protection operations to construct pesticide deposition events and form a pesticide deposition sequence; it performs time-domain, spatial and appearance triple verification with the pesticide deposition sequence to identify suspected pesticide residue patches. Discoloration identification module: For suspected pesticide residue patches, the discoloration rate is separated from the time evolution profile. The discoloration rate is combined with the pesticide decomposition and decay characteristics to identify pesticide residue patches and suspected disease patches. Causality verification module: Obtain the leaf surface wetting duration and canopy temperature sequence corresponding to suspected disease spots, input them into the disease symptom development reference scale library, generate a standard area sequence, calculate the consistency metric between the temporal evolution profile of suspected disease spots and the standard area sequence, and store them in association. Early warning classification module: Identifies disease spots to be warned based on the coincidence measurement value of suspected disease spots, and classifies the early warning level according to the distribution of the coincidence measurement value of all spots to be warned, and generates and outputs disease and pest early warning information.

[0015] The beneficial effects of this invention are as follows: 1. This invention extracts color-abnormal patches by segmenting multi-temporal digital images of the canopy, tracks these patches across time phases to extract their contours and color attributes, and generates temporal evolution contours. It identifies color-abnormal patches and quantifies the temporal variation patterns of their area and color, providing crucial temporal data for distinguishing between pesticide residues and disease lesions. It also reads electronic records of plant protection operations to construct pesticide deposition events, forming pesticide deposition sequences. The temporal evolution contours of color-abnormal patches are then verified against the pesticide deposition sequences in terms of time, space, and appearance to identify suspected pesticide residue patches. Finally, it isolates suspected pesticide residue patches caused by agricultural operations from all color-abnormal patches, significantly narrowing the scope of subsequent disease analysis and reducing the workload of subsequent analysis.

[0016] 2. This invention separates the depigmentation rate from the temporal evolution profile of suspected pesticide residue patches. By combining the depigmentation rate with the pesticide decomposition and decay characteristics, pesticide residue patches and suspected disease patches are identified. The degree of agreement between the depigmentation rate and the pesticide decomposition and decay characteristics is used to eliminate interference from pesticide residues and prevent their encroachment on disease early warning resources. The invention obtains the leaf surface wetting duration and canopy temperature sequences corresponding to suspected disease patches, inputs them into a disease symptom progression reference library, generates a standard area sequence, and calculates and stores the agreement between the temporal evolution profile of suspected disease patches and the standard area sequence. The invention also evaluates the degree of matching between the actual development process of suspected disease patches and the standardized expansion relationship of specific diseases, effectively eliminating sporadic patches that do not conform to the disease development pattern. 3. This invention identifies disease spots to be warned by matching the coincidence measurement value of suspected disease spots, and classifies the warning level according to the distribution of the coincidence measurement value of all spots to be warned, and generates and outputs disease and pest warning information; it generates graded warning information that filters out false alarms of pesticide spots, preventing orchard managers from making erroneous results of false reporting, neglect, and missed prevention, and ensuring that the real disease information can be conveyed in a timely manner. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of the steps of a monitoring and early warning method for fruit tree diseases and pests according to the present invention; Figure 2 This is a logical diagram for defining suspected drug residue patches in this invention; Figure 3 This is a functional module diagram of a fruit tree disease and pest monitoring and early warning system according to the present invention. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] Example 1 like Figures 1-2 As shown, a method for monitoring and early warning of fruit tree diseases and pests includes: Step 1: Acquire multi-temporal digital images of the canopy, segment and extract color aberration patches, perform cross-temporal tracking of color aberration patches to extract contours and color attributes, and generate temporal evolution contours; Multiple fixed imaging devices are set up in the orchard. The fixed imaging devices take pictures of the canopy of the fruit trees at fixed time intervals to obtain digital images of the canopy with the acquisition time. The digital images of the canopy taken by all the fixed imaging devices at the acquisition time form a temporal phase. The digital images of the canopy at different temporal phases constitute a multi-temporal canopy image sequence. For each frame of the canopy digital image at each time phase, the trained leaf segmentation model is used to segment the leaf region and output a leaf mask image. In the leaf mask image, the leaf pixels are assigned a first-class label and the background pixels are assigned a second-class label. Understandably, the leaf segmentation model is a semantic segmentation model based on a deep fully convolutional network. It uses a residual network as the feature extraction backbone and is trained using a dataset containing tens of thousands of orchard canopy images and corresponding leaf pixel annotations. The leaf mask image is multiplied pixel by pixel with the canopy digital image to obtain a leaf local image that retains only the leaf region. Each spatially connected leaf region in the leaf local image is taken as a leaf instance and assigned an independent leaf instance number. For each leaf instance, it is input into a pre-built color anomalous patch segmentation network. The color anomalous patch segmentation network classifies each pixel in the leaf instance and outputs a binary anomalous patch map. Among them, the color abnormality patch segmentation network is a fully convolutional network with an encoder-decoder structure, and its training dataset contains manually finely annotated pesticide residue and early lesion regions in the leaf area. In the binary abnormal patch image, pixels whose color differs from that of normal green tissue are marked as abnormal pixels, while normal green tissue is marked as background. Connectivity analysis is performed on the binary abnormal patch map to extract each connected component composed of abnormal pixels. Each connected component is taken as a color abnormal patch, and the contour coordinate set of the color abnormal patch is obtained. Calculate the geometric centroid coordinates of the color anomalous patch based on the contour coordinate set, and use the geometric centroid coordinates and the contour coordinate set as the position and contour of the color anomalous patch. Based on the time phase of the first detection of color aberration patches, extract all aberration pixels in the color aberration patches in CIE L. a b The value of lightness L, the value of chromaticity a, and the value of chromaticity b in the color space are calculated, and the average value of lightness L_init, the average value of chromaticity a_init, and the average value of chromaticity b_init for all abnormal pixels are calculated respectively. The average lightness value L_init, the average chromaticity value a_init, and the average chromaticity value b_init are used as the initial color attributes of the color aberration patches; Perform cross-temporal tracking of all color anomalous patches in the current time phase and all color anomalous patches in the next adjacent time phase: Calculate the contour intersection-union ratio between each color anomalous patch in the current time phase and each color anomalous patch in the next time phase. Construct a cost matrix using the contour intersection-union ratio, where the cost is set as the reciprocal of the contour intersection-union ratio. Solve the patch pairing relationship that minimizes the total cost using the Hungarian algorithm. Color anomalous patches that are successfully paired and have a contour intersection-union ratio greater than zero are identified as the continuation of the same physical patch in the two time phases. When a certain color aberration patch does not have any color aberration patch with an intersection ratio greater than zero with its outline in the next time phase, mark the color aberration patch as terminated after its current time phase; When a certain color aberration patch does not have any color aberration patch with an intersection ratio greater than zero in the previous time phase, it is marked as a newly appearing color aberration patch. Color anomalous patches identified as the same physical patch in different time phases are connected in sequence to obtain a cross-time phase record of a color anomalous patch. The acquisition time corresponding to the first time phase in the cross-time phase record is marked as the start time, and the time phase in which the color anomalous patch last appeared is marked as the end time. For cross-temporal recordings of color abnormal patches, the contours of the color abnormal patches in each temporal phase are extracted. Using the object resolution of the pre-calibrated fixed imaging device in the canopy plane, the number of pixels enclosed by the contours is converted into the physical area on the leaf surface in square millimeters, forming a sequence of area changes of color abnormal patches. For the cross-temporal records of color aberration patches, the average color value of the color aberration patches in each temporal phase is extracted, and the average color value is arranged according to the time series to form the color fading process of the color aberration patches. The average color value is composed of the average lightness value L_init, the average chromaticity value a_init, and the average chromaticity value b_init. The start time, area change sequence, and color fading process of the color abnormality patches are associated and stored to generate the temporal evolution profile of the color abnormality patches. Step 2: Read the electronic ledger of plant protection operations to construct pesticide deposition events and form a pesticide deposition sequence; verify the temporal evolution outline of the abnormal color patches with the pesticide deposition sequence in terms of time, space and appearance to identify suspected pesticide residue patches; Read the electronic ledger of plant protection operations, extract the spraying completion time, the name of the active ingredient of the sprayed pesticide, and the coverage area of ​​each spraying operation from the electronic ledger of plant protection operations, and construct it as an independent pesticide deposition event. The scope of the operation must at least include the range of rows of fruit trees receiving the medicine and the range of sections between the trees receiving the medicine. Each pesticide deposition event is assigned a unique event number, and the spray completion time, active ingredient name, and coverage area are combined to form a pesticide deposition sequence. For each drug deposition event, access the pre-built drug property database, use the active ingredient name carried by the drug deposition event to search the drug property database, and obtain the reference time for drying spots and reference data for the color of the dried spots corresponding to the active ingredient name; It should be noted that the pesticide characteristic database stores reference time for drying and reference data for the color of dried spots for each active ingredient. The reference time for drying is the average time required for multiple repeated tests to completely dry the pesticide solution from a liquid state to form a visible spot area after it has been evenly sprayed onto the surface of healthy canopy leaves under standard climatic conditions. The reference data for the color of dried spots includes multiple batches of imaging measurements of the spot area after drying and recording them in CIE L. a b The color mean values ​​L_ref, a_ref, and b_ref ​​obtained statistically in the color space, and the typical color distribution range constituted by the spatial distribution range of multiple batches of color samples in the three-dimensional color space; The time when the pesticide deposition event was completed was calculated by summing the spraying completion time and the drying and spot formation reference time; The time of stain appearance, name of active ingredient, coverage area, and reference data of dry stain color are used as extended attributes of the drug deposition event and stored in association with the event number of the drug deposition event. Based on the temporal evolution contour of each color aberration patch generated in step one, the initial color attributes of the color aberration patches are extracted. Obtain the device number of the fixed imaging device corresponding to each color abnormality patch at the time of its first detection. Based on the correspondence between the fixed imaging device and the fruit trees in the orchard recorded during system deployment, determine the position of the fruit tree row to which the color abnormality patch belongs. At the same time, based on the geometric centroid coordinates of the color abnormality patch at the time of its first detection, combined with the object resolution of the fixed imaging device, calculate the position of the color abnormality patch in the inter-tree section along the row direction within the fruit tree row. The locations of fruit tree rows and inter-tree sections were used as the sampling locations for abnormal color patches; S201. The process of performing time-domain correlation analysis between the start time of each color abnormality patch and the appearance time of the stain in each drug deposition event is as follows: Obtain the spot appearance time of the drug deposition event, and use the spot appearance time as the starting point to search along the time axis in the forward direction for the sampling time with the smallest time interval from the spot appearance time, which is taken as the spot detection time. Use the spot appearance time as the starting point and the spot detection time as the ending point as the spot detection time window for the corresponding drug deposition event. Among them, the sampling time refers to each acquisition time when the fixed imaging device captures digital images of the canopy at a fixed time period; If the start time of an abnormal color patch falls within the drug deposit event's drug spot detection time window, then the abnormal color patch and the drug deposit event are determined to have temporal overlap. S202. The process of analyzing whether each color abnormality patch matches each drug deposition event spatially is as follows: Obtain the sampling location of each color abnormality patch and the operational coverage area of ​​each reagent deposition event; Spatial intersection is performed between the sampling location of the abnormal color patch and the operational coverage of the pesticide deposition event. If the location of the abnormal color patch in the fruit tree row falls within the range of the treated fruit tree row, and the location of the abnormal color patch in the inter-tree section falls within the range of the treated inter-tree section, then the sampling location of the abnormal color patch is determined to match the spatial location of the pesticide deposition event. S203. The process of analyzing whether each color abnormality patch matches the appearance characteristics with each drug deposition event is as follows: Obtain the initial color attributes of each color-abnormal patch and the dry spot color reference data of the drug deposition event; Calculate the Euclidean distance between the initial color attributes of the color abnormality patch and the color mean of the drug deposition event. If the Euclidean distance falls within the typical color distribution range, it is determined that the initial color attributes of the color abnormality patch are consistent with the typical appearance characteristics of the drug drying spots in the drug deposition event. If a color anomalous patch satisfies temporal overlap, spatial location matching, and appearance characteristics with only one drug deposition event, then the color anomalous patch is marked as a suspected drug residue patch, and the drug deposition event corresponding to each suspected drug residue patch is associated. The technical solution of this invention is as follows: Multi-temporal digital images of the canopy are acquired, segmented, and color-abnormal patches are extracted. These patches are then tracked across time phases to extract their contours and color attributes, generating a temporal evolution contour. Color-abnormal patches are identified, and the changes in their area and color over time are quantified, providing crucial temporal data for distinguishing between pesticide residues and disease lesions. Electronic records of plant protection operations are read to construct pesticide deposition events, forming a pesticide deposition sequence. The temporal evolution contours of the color-abnormal patches are verified against the pesticide deposition sequence in terms of time, space, and appearance, identifying suspected pesticide residue patches. Suspected pesticide residue patches resulting from agricultural operations are removed from all color-abnormal patches, significantly narrowing the scope of subsequent disease analysis and reducing the workload of subsequent analysis.

[0021] Example 2 Please see Figure 1 As shown, a method for monitoring and early warning of fruit tree diseases and pests includes: Step 3: For suspected pesticide residue patches, the decolorization rate is obtained from the temporal evolution profile. The decolorization rate is combined with the pesticide decomposition and decay characteristics to identify pesticide residue patches and suspected disease patches. For each suspected drug residue patch marked, the color fading process in the temporal evolution contour corresponding to the suspected drug residue patch is obtained; For any two adjacent time phases in the color fading process, calculate the difference in the mean color value between the later time phase and the earlier time phase to obtain the lightness difference ΔL, the chromaticity a difference Δa, and the chromaticity b difference Δb. Based on the lightness difference ΔL, chromaticity a difference Δa, and chromaticity b difference Δb, the color difference ΔE between adjacent time phases is calculated. The formula for calculating the color difference is: ; Obtain two corresponding acquisition times for adjacent time phases, and calculate the time difference between the two acquisition times as the interval between adjacent time phases; The color difference value is calculated as a ratio to the interval length to obtain the color removal rate; The arithmetic mean of the corresponding decolorization rates of all adjacent patches suspected of drug residue was obtained to get the decolorization rate of the suspected drug residue patch. Obtain drug deposition events associated with suspected drug residue patches and extract the names of active ingredients recorded in the drug deposition events; Based on the name of the active ingredient, access a pre-built drug property database and retrieve the drug decomposition and decay characteristics corresponding to the name of the active ingredient in the drug property database; It should be noted that the decomposition and decay characteristics of the agent include a reference value for decay rate, an upper limit for decay rate fluctuation, and a lower limit for decay rate fluctuation. The reference value for decay rate is obtained by repeatedly measuring the decay process of the color of the dry spots over time after spraying the drug solution corresponding to the active ingredient onto the surface of healthy canopy leaves under standard climatic conditions to form dry spots, and calculating the statistical mean of the rate of change of color difference value. The upper limit for decay rate fluctuation is the maximum value among the samples of the rate of change of color difference value measured repeatedly, and the lower limit for decay rate fluctuation is the minimum value among the samples of the rate of change of color difference value measured repeatedly. The decolorization rate of suspected drug residue patches was compared with the upper and lower bounds of the decay rate fluctuation in the drug decomposition and decay characteristics: If the decolorization rate of the suspected drug residue plaque is greater than or equal to the lower limit of the decay rate fluctuation and less than or equal to the upper limit of the decay rate fluctuation, then the decolorization rate of the suspected drug residue plaque is determined to be consistent with the drug decomposition and decay characteristics, and the suspected drug residue plaque is determined to be a drug residue plaque. If the fading rate of suspected pesticide residue patches is less than the lower limit of the decay rate fluctuation or greater than the upper limit of the decay rate fluctuation, then the fading rate of suspected pesticide residue patches is determined to be inconsistent with the pesticide decomposition and decay characteristics, and the suspected pesticide residue patches are determined to be suspected disease patches. The above judgment process is performed on each suspected pesticide residue patch marked in step two until all suspected pesticide residue patches are respectively judged as pesticide residue patches or suspected disease patches; Step 4: Obtain the leaf surface wetting duration and canopy temperature sequence corresponding to suspected disease spots, input them into the disease symptom progression reference scale library, generate a standard area sequence, calculate the consistency metric between the temporal evolution profile of suspected disease spots and the standard area sequence, and store them in association. Multiple microclimate monitoring nodes are deployed in the orchard, and each microclimate monitoring node includes a leaf humidity sensor and a canopy temperature sensor. The microclimate monitoring node and the fixed imaging device use the same fixed time period to collect data. At each collection moment, the leaf surface wetness duration and the canopy temperature are acquired once, generating a microclimate monitoring data sequence containing the collection moment. The deployment locations of the microclimate monitoring nodes and the installation locations of the fixed imaging devices ensure that the canopy area monitored by each microclimate monitoring node corresponds to the range of fruit tree rows and inter-tree sections captured by at least one fixed imaging device. The starting time is extracted based on the temporal evolution contour of suspected disease spots, and the corresponding collection position of the suspected disease spot in step two is extracted. Based on the collection location of suspected disease spots, and combined with the spatial mapping relationship between fixed imaging devices, microclimate monitoring nodes and fruit trees, the microclimate monitoring nodes that spatially match the suspected disease spots are determined. From the microclimate monitoring data sequence of the matched microclimate monitoring nodes, extract the total leaf surface wet duration and canopy temperature between the start and end times of the collection time when the suspected disease spot is located. The extracted leaf surface wetting durations were arranged chronologically to form a leaf surface wetting duration sequence, and the extracted canopy temperatures were also arranged chronologically to form a canopy temperature sequence; and the area change sequence was extracted from the temporal evolution profile of suspected disease spots. A disease symptom progression reference scale library is constructed within the system. The construction process of the disease symptom progression reference scale library is as follows: For each infectious disease of fruit trees that requires key prevention and control, the corresponding pathogens were obtained, and the pathogens were inoculated onto healthy canopy leaves. Multiple sets of different constant temperature levels and multiple sets of different leaf surface moisture duration levels were set to form a combination of temperature and leaf surface moisture duration experimental conditions. Under each experimental combination, after inoculating the healthy canopy leaves with pathogens, the healthy canopy leaves were continuously photographed and the lesion area was measured using a fixed imaging device. The complete growth process of the lesions was recorded, and based on the physical area of ​​the lesions at each time point, a lesion area growth time sequence sample corresponding to each experimental combination was formed. Using nonlinear regression, with temperature and leaf wet duration as driving variables, the time series samples of lesion area growth under various combinations of conditions were fitted to obtain the mathematical relationship function between lesion area increment and temperature and leaf wet duration. The functional form and coefficients of the mathematical relation function are used as the standardized extended relation of infectious diseases, and the disease name of the infectious disease is used as the index to store it in the disease symptom process reference scale library. Based on the current phenological stage of the fruit trees and historical disease occurrence records, determine the names of the infectious diseases that need to be monitored at present. Use the disease names to search in the disease symptom progress reference scale library to obtain the standardized extended relationship corresponding to the disease name. The leaf wetting duration and canopy temperature of each phase in the leaf wetting duration sequence and canopy temperature sequence of suspected disease spots are input into the mathematical relation function of the standardized extended relation phase by phase to obtain the increase of disease spot area in adjacent phases. The physical area of ​​the first phase in the area change sequence is used as the initial lesion area. The lesion area increments corresponding to each phase are accumulated sequentially to calculate the theoretical lesion area of ​​each subsequent phase. After sorting, a standard area sequence corresponding one-to-one with the phases of the area change sequence is obtained. Calculate the Pearson correlation coefficient between the area change series and the standard area series; The Pearson correlation coefficient was used as the consistency metric for suspected disease spots to match infectious diseases, and the corresponding consistency metric was associated with the leaf instance number of the suspected disease spot and stored. Step 5: Identify disease spots to be warned based on the coincidence measurement value of suspected disease spots, and determine the warning level according to the distribution of the coincidence measurement value of all disease spots to be warned, and generate and output disease and pest warning information; Obtain all pesticide residue spots identified in step three. For each pesticide residue spot, extract the leaf instance number corresponding to the pesticide residue spot in step one and generate a pesticide residue spot masking list. Among them, the list of drug residue spots is uploaded to the system, and no image alarm record is generated for any drug residue spots corresponding to the list of drug residue spots. Obtain all suspected disease spots, and extract the leaf instance number corresponding to the suspected disease spot in step one, as well as the matching metric value associated with the leaf instance number; Obtain the names of the infectious diseases that need to be monitored in step four, search the disease symptom progress reference scale library using the disease name, and extract the warning reference value corresponding to the disease name from the disease symptom progress reference scale library. Among them, the warning reference value is generated synchronously during the construction of the disease symptom progress reference scale library, and its generation method is as follows: Under the combined experimental conditions of multiple temperature and leaf surface moisture duration, a control group of healthy canopy leaves that were not inoculated with pathogens was set up, and digital images of the control canopy were acquired using a fixed imaging device. The reference canopy digital image was input into the color aberration patch segmentation network for segmentation and extraction, resulting in several color aberration patches; It is understandable that due to non-infectious factors such as natural physiological spots, minor physical damage, local reflection differences caused by changes in light angle, and pesticide residues on the surface of healthy canopy leaves, some abnormally colored patches will be extracted. Following the cross-temporal tracking method in step one, corresponding temporal evolution contours are generated based on the aforementioned color aberration patches, and the area change sequences of each color aberration patch in the control group are extracted from the temporal evolution contours. The area change sequence of each control group abnormal color patch was compared with the standard area sequence generated by the standardized expansion relationship under the same temperature and leaf moisture duration conditions. The Pearson correlation coefficient was calculated. All the calculated Pearson correlation coefficients were used to form a control correlation coefficient set. The maximum value in the control correlation coefficient set was taken as the warning reference value of the infectious disease. The warning reference value was associated with the disease name and stored in the disease symptom progress reference scale library. The coincidence metric of each suspected disease spot is compared with the warning reference value of the corresponding infectious disease; If the coincidence value of a suspected disease spot is greater than or equal to the warning reference value, the suspected disease spot will be marked as a disease spot awaiting warning. If the coincidence value of a suspected disease spot is less than the warning reference value, the suspected disease spot is retained as a color abnormality spot and no further treatment is carried out. Obtain the coincidence measurement values ​​of all disease spots to be warned, form a coincidence measurement value set to be warned, calculate the first quartile, second quartile, third quartile and fourth quartile of the coincidence measurement value set to be warned, and divide the coincidence measurement values ​​of disease spots to be warned into three warning intervals with the second quartile and third quartile as the dividing line. Among them, disease spots with a coincidence value greater than or equal to the third quartile are classified into the first warning interval, disease spots with a coincidence value greater than or equal to the second quartile and less than the third quartile are classified into the second warning interval, and disease spots with a coincidence value greater than or equal to the warning reference value and less than the second quartile are classified into the third warning interval. The warning level corresponding to the disease spots to be warned that are classified into the first warning interval is set as Level 1 warning, the warning level corresponding to the disease spots to be warned that are classified into the second warning interval is set as Level 2 warning, and the warning level corresponding to the disease spots to be warned that are classified into the third warning interval is set as Level 3 warning. For each disease spot with a set warning level, the leaf instance number of the disease spot and the corresponding start time are extracted as the earliest time of disease detection, and the collection location corresponding to the disease spot is obtained. The leaf instance number of the disease spot to be warned, the warning level, the collection location, the coincidence value, and the earliest time of disease discovery are combined into a disease and pest warning information record. All pest and disease warning information records corresponding to the disease spots to be warned are arranged in order of warning level from high to low. Among them, the pest and disease warning information records of level 1 warning are listed first, followed by level 2 warning, and level 3 warning is listed last. Within the same warning level, the pest and disease warning information records are arranged in order of matching value from large to small. The arranged pest and disease early warning information is recorded and encapsulated into pest and disease early warning information with false alarms removed by pesticide spots, and the pest and disease early warning information with false alarms removed by pesticide spots is output to the early warning display interface of the orchard management platform. The technical solution of this invention is as follows: For suspected pesticide residue patches, the decolorization rate is separated from the temporal evolution profile. The decolorization rate is combined with the pesticide decomposition and decay characteristics to identify pesticide residue patches and suspected disease patches. The correlation between the decolorization rate and the pesticide decomposition and decay characteristics is used to eliminate the interference of pesticide patches on operations and prevent pesticide residues from encroaching on disease early warning resources. The leaf surface wetting duration and canopy temperature sequences corresponding to suspected disease patches are obtained and input into a disease symptom progression reference scale library to generate a standard area sequence. The temporal evolution profile of suspected disease patches is then calculated in relation to the standard area sequence. The system associates and stores the coincidence metrics of the area sequence; assesses the degree of matching between the actual development process of suspected disease spots and the standardized expansion relationship of specific diseases, effectively eliminating sporadic spots that do not conform to the disease development pattern; identifies disease spots to be warned based on the coincidence metrics of suspected disease spots, and classifies the warning level according to the distribution of the coincidence metrics of all spots to be warned, generating and outputting disease and pest warning information; and generates graded warning information that filters out false alarms from pesticide spots, preventing orchard managers from making erroneous results of false reporting, neglect, and missed prevention, ensuring that real disease information can be conveyed in a timely manner.

[0022] Example 3 Please see Figure 3 As shown, a monitoring and early warning system for fruit tree diseases and pests includes the following modules: The temporal tracking module acquires multi-temporal digital images of the canopy, segments and extracts color aberration patches, performs cross-temporal tracking of color aberration patches to extract contours and color attributes, and generates temporal evolution contours. The triple verification module reads the electronic ledger of plant protection operations to construct pesticide deposition events and form a pesticide deposition sequence; it performs time-domain, spatial and appearance triple verification with the pesticide deposition sequence to identify suspected pesticide residue patches. Discoloration identification module: For suspected pesticide residue patches, the discoloration rate is separated from the time evolution profile. The discoloration rate is combined with the pesticide decomposition and decay characteristics to identify pesticide residue patches and suspected disease patches. Causality verification module: Obtain the leaf surface wetting duration and canopy temperature sequence corresponding to suspected disease spots, input them into the disease symptom development reference scale library, generate a standard area sequence, calculate the consistency metric between the temporal evolution profile of suspected disease spots and the standard area sequence, and store them in association. Early warning classification module: Identifies disease spots to be warned based on the coincidence measurement value of suspected disease spots, and classifies the early warning level according to the distribution of the coincidence measurement value of all spots to be warned, and generates and outputs disease and pest early warning information.

[0023] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for monitoring and early warning of fruit tree diseases and pests, characterized in that: Including the following: Multi-temporal digital images of the canopy are acquired, segmented, and color aberration patches are extracted. The color aberration patches are then tracked across time phases to extract their contours and color attributes, generating a temporal evolution contour. The electronic ledger of plant protection operations is read to construct pesticide deposition events and form pesticide deposition sequences; the temporal evolution profile of abnormal color patches is verified with the pesticide deposition sequence in terms of time domain, space and appearance to identify suspected pesticide residue patches; For suspected pesticide residue patches, the decolorization rate is separated from the temporal evolution profile. The decolorization rate is combined with the pesticide decomposition and decay characteristics to identify pesticide residue patches and suspected disease patches. Obtain the leaf surface wetting duration and canopy temperature sequence corresponding to suspected disease spots, input them into the disease symptom progression reference scale library, generate a standard area sequence, calculate the coincidence value between the temporal evolution profile of suspected disease spots and the standard area sequence, and store them in association. Based on the coincidence metric value of suspected disease spots, disease spots to be warned are identified, and the warning level is determined according to the distribution of the coincidence metric values ​​of all spots to be warned. Disease and pest warning information is generated and output.

2. The method for monitoring and early warning of fruit tree diseases and pests according to claim 1, characterized in that: The process of generating a time series evolution profile is as follows: All color anomalous patches are tracked across time phases. The contour intersection-union ratio of color anomalous patches in adjacent time phases is calculated. The Hungarian algorithm is used to solve the pairing problem and obtain the cross-time phase record of each color anomalous patch. By recording color anomalous patches across time phases, the moment when the color anomalous patches were first detected was extracted and marked as the start time; the contour of the color anomalous patches in each time phase was extracted, and the physical area of ​​the color anomalous patches was obtained through a pre-calibrated object surface resolution, forming an area transition sequence. For color aberration patches, the average color value of each phase is extracted from the cross-temporal records to form the color fading process; the start time, area change sequence and color fading process of the color aberration patches are associated and stored to generate a temporal evolution profile.

3. The method for monitoring and early warning of fruit tree diseases and pests according to claim 1, characterized in that: The process of forming a drug deposition sequence is as follows: Read the electronic ledger of plant protection operations, extract the spraying completion time, the name of the active ingredient of the sprayed pesticide, and the coverage area of ​​each spraying operation, and construct it as a pesticide deposition event; Each pesticide deposition event is assigned a unique event number, and the spray completion time, active ingredient name, and coverage area are combined to form a pesticide deposition sequence.

4. The method for monitoring and early warning of fruit tree diseases and pests according to claim 1, characterized in that: The process of identifying suspected drug residue patches is as follows: The temporal evolution profile of the color anomalous patches is verified against the drug deposition sequence in terms of time, space, and appearance. If a color anomalous patch satisfies the temporal overlap, spatial location matching, and appearance characteristics with only one drug deposition event, then the color anomalous patch is marked as a suspected drug residue patch.

5. The method for monitoring and early warning of fruit tree diseases and pests according to claim 4, characterized in that: The method for triple verification in terms of time, space, and appearance is as follows: The starting point of the appearance of the drug deposition event is obtained, and the time when the spot can be detected is searched to form a drug spot detection time window. If the starting time of the color abnormal spot falls into the drug spot detection time window, then the color abnormal spot and the drug deposition event have temporal overlap. The sampling locations of abnormal color patches and the operational coverage of the pesticide deposition event were obtained. The sampling locations included the fruit tree rows and the inter-tree sections, and the operational coverage included the range of the treated fruit tree rows and the range of the treated inter-tree sections. If the location of the abnormal color patch in the row of fruit trees falls within the range of the drug-treated fruit tree row, and the location of the abnormal color patch in the inter-tree section falls within the range of the drug-treated inter-tree section, then the sampling location of the abnormal color patch is determined to match the spatial location of the drug deposition event. Obtain the initial color attributes of the color-abnormal patches and the reference data of the dry patch color of the reagent deposition event, calculate the Euclidean distance. If the Euclidean distance falls within the typical color distribution range, the color-abnormal patches are consistent with the appearance characteristics of the reagent deposition event.

6. The method for monitoring and early warning of fruit tree diseases and pests according to claim 1, characterized in that: The process of identifying pesticide residue spots and suspected disease spots is as follows: For suspected drug residue patches, the decolorization rate was separated from the temporal evolution profile, and the upper and lower bounds of the decay rate fluctuation of the drug decomposition decay characteristics were extracted. If the decolorization rate of a suspected drug residue patch is greater than or equal to the lower limit of the decay rate fluctuation and less than or equal to the upper limit of the decay rate fluctuation, it is determined to be a drug residue patch. If the fading rate of suspected plaques with residual pesticide residue is less than the lower limit of the decay rate fluctuation or greater than the upper limit of the decay rate fluctuation, then the plaques are identified as suspected disease spots.

7. The method for monitoring and early warning of fruit tree diseases and pests according to claim 6, characterized in that: The method for obtaining the decolorization rate by separation is as follows: Obtain the color fading process in the temporal evolution profile corresponding to the suspected drug residue patches; for any two adjacent time phases in the color fading process, calculate the difference in the mean color value between the later time phase and the previous time phase, obtain the lightness difference ΔL, chromaticity a difference Δa and chromaticity b difference Δb, and calculate the color difference value. Obtain two corresponding acquisition times for adjacent time phases, and calculate the time difference between the two acquisition times as the interval between adjacent time phases; The decolorization rate is obtained by calculating the ratio of the color difference value to the interval duration.

8. The method for monitoring and early warning of fruit tree diseases and pests according to claim 1, characterized in that: The process of calculating the consistency metric between the temporal evolution profile of suspected disease spots and the standard area sequence is as follows: Based on the temporal evolution contours of suspected disease spots, area change sequences are extracted, and the Pearson correlation coefficient between the area change sequences and the standard area sequences is calculated as a measure of consistency.

9. A method for monitoring and early warning of fruit tree diseases and pests according to claim 1, characterized in that: The process of determining the warning level is as follows: Obtain the coincidence measurement values ​​of all disease spots to be warned, form a set of coincidence measurement values ​​to be warned, calculate the second quartile and the third quartile; and obtain the warning reference value; If the coincidence value of the disease spot to be warned is greater than or equal to the third quartile, a level one warning is set. If the coincidence value is greater than or equal to the second quartile and less than the third quartile, a level two warning is set. If the coincidence value is greater than or equal to the warning reference value and less than the second quartile, a level three warning is set.

10. A monitoring and early warning system for fruit tree diseases and pests, used to implement the monitoring and early warning method for fruit tree diseases and pests as described in any one of claims 1-9, characterized in that: Including the following: The temporal tracking module acquires multi-temporal digital images of the canopy, segments and extracts color aberration patches, performs cross-temporal tracking of color aberration patches to extract contours and color attributes, and generates temporal evolution contours. The triple verification module reads the electronic ledger of plant protection operations to construct pesticide deposition events and form a pesticide deposition sequence; it performs time-domain, spatial and appearance triple verification with the pesticide deposition sequence to identify suspected pesticide residue patches. Discoloration identification module: For suspected pesticide residue patches, the discoloration rate is separated from the time evolution profile. The discoloration rate is combined with the pesticide decomposition and decay characteristics to identify pesticide residue patches and suspected disease patches. Causality verification module: Obtain the leaf surface wetting duration and canopy temperature sequence corresponding to suspected disease spots, input them into the disease symptom development reference scale library, generate a standard area sequence, calculate the consistency metric between the temporal evolution profile of suspected disease spots and the standard area sequence, and store them in association. Early warning classification module: Identifies disease spots to be warned based on the coincidence measurement value of suspected disease spots, and classifies the early warning level according to the distribution of the coincidence measurement value of all spots to be warned, and generates and outputs disease and pest early warning information.

Images