Online detection and prevention method applied to forestry diseases and insect pests

Abstract

The invention provides an online detection and prevention method applied to forestry diseases and insect pests, and relates to the field of data processing. The method comprises the following steps: performing multiple polling on a to-be-detected forest region according to the same polling rule to obtain hyperspectral images of a detection region and a contrast region in the to-be-detected forest region; wherein the contrast area is an area which only contains pinus species and is subjected to prevention and control of monochamus alternatus hope larvae, and the detection area is other areas except the contrast area; constructing a reference wave band curve of the pinus species based on the hyperspectral image of the contrast area; comparing the hyperspectral curve of each pixel point in the hyperspectral image of the detection area with the reference wave band curve, and identifying an abnormal area; calculating an insect pest possibility index of each abnormal region, and determining an insect pest region in each abnormal region based on time sequence data of the insect pest possibility index; and generating an early warning index of each pest and disease damage area, and outputting early warning information of each pest and disease damage area based on the early warning index. Accurate pest and disease damage areas can be obtained and early warning can be performed.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an online detection and control method for forestry pests and diseases. Background Technology

[0002] Forest pests and diseases refer to diseases and insect pests that harm forests and trees, causing serious impacts on forest ecosystems and forestry production. Timely detection and control of forest pests and diseases can limit damage, protecting forest ecological balance and ensuring sustainable forestry economic development. The pine sawyer beetle larvae typically bore into the phloem and xylem of tree trunks, carrying pine wilt nematodes that invade the pine tree, disrupting water and nutrient transport and ultimately causing the entire tree to die. In the detection and control of these forest pests and diseases, monitoring pine sawyer beetle larvae allows for the application of pesticides to affected trees during the larval stage, thus controlling both pine sawyer beetle and pine wilt nematode diseases.

[0003] In related technologies, the above-mentioned monitoring of pine sawyer beetle larval disease can be achieved as follows: hyperspectral detection results of trees in forest areas are obtained using a hyperspectral camera mounted on an unmanned aerial vehicle (UAV); a fixed model is constructed to extract the spectral features of diseased pine areas, thereby achieving intelligent identification of pine sawyer beetle larval disease. However, during UAV operation, external factors may cause errors in the hyperspectral detection results obtained by the UAV; moreover, the tree species in forest areas are quite diverse, and the reflectance spectral characteristics of different tree species differ. There may be overlapping areas between the spectrum of a healthy tree and the spectrum of a tree infected with pine sawyer beetle larvae, leading to misjudgment as an infected tree. Therefore, the above method cannot accurately identify pine sawyer beetle larval disease. Summary of the Invention

[0004] To address the issues of inaccurate identification of pine sawyer beetle larval disease in forests, such as errors in hyperspectral detection results due to attitude changes caused by external factors during drone operation, and the diverse tree species and varying reflectance spectral characteristics of different species leading to potential overlap between the spectra of a healthy tree and those of trees infested with pine sawyer beetle larvae, this invention provides an online detection and control method for forestry pests and diseases. The specific technical solution adopted is as follows:

[0005] Multiple inspections of the forest area to be inspected were conducted using the same inspection rules to obtain hyperspectral images of the detection area and the control area in the forest area to be inspected; wherein, the control area is the area in the forest area to be inspected that contains only pine species and has been controlled for pine sawyer beetle larvae, and the detection area is all other areas in the forest area to be inspected except for the control area;

[0006] A reference band curve for the pine species is constructed based on the hyperspectral image of the control region;

[0007] The hyperspectral curves of each pixel in the hyperspectral image of the detection area are compared with the reference band curve to identify abnormal areas in the detection area.

[0008] Calculate the pest probability index of the presence of the pine sawyer beetle larvae in each of the abnormal regions, and determine the pest-infested areas in each of the abnormal regions based on the time series data of the pest probability index corresponding to each of the abnormal regions.

[0009] An early warning index is generated for each of the pest and disease areas, and early warning information for each pest and disease area is output based on the early warning index; wherein, the early warning index is used to indicate the degree of pest spread in the pest and disease area.

[0010] Furthermore, the step of conducting multiple inspections of the forest area to be inspected using the same inspection rules to obtain hyperspectral images of the detection area and the control area within the forest area to be inspected includes:

[0011] Determine the testing cycle;

[0012] Within each detection cycle, the detection area is inspected multiple times at fixed time points to acquire the hyperspectral image of the detection area; wherein, the hyperspectral images of the detection area acquired at adjacent time points during a single inspection have overlapping areas.

[0013] Within each detection cycle, the control area is inspected multiple times at the fixed time point to acquire the hyperspectral image of the control area; wherein, the hyperspectral images of the control area acquired at adjacent times during a single inspection have overlapping areas.

[0014] Furthermore, after acquiring the hyperspectral images of the detection area and the control area in the forest area to be inspected, the method further includes:

[0015] For each hyperspectral image of the detection area and the control area obtained in a single inspection, the tree regions in the hyperspectral images are labeled based on a semantic recognition algorithm;

[0016] For each single-band grayscale image in the hyperspectral images, edge points of the tree region are obtained based on an edge detection algorithm;

[0017] The feature points in the edge points are determined based on the curvature and grayscale gradient magnitude of the edge points;

[0018] Based on the feature points, the hyperspectral images obtained at different times during a single inspection are stitched together to obtain a complete image of the detection area and a complete image of the control area.

[0019] Furthermore, determining the feature points in the edge points based on the curvature and grayscale gradient magnitude of the edge points includes:

[0020] Calculate the curvature of each edge point, the gray-level gradient magnitude of each edge point, and the average gray-level gradient magnitude of all edge points;

[0021] The identifiability index of each edge point is determined based on the curvature of each edge point, the gray-level gradient magnitude, and the average gray-level gradient magnitude of all edge points, and the edge points whose identifiability index is greater than a first preset threshold are marked as identifiable points.

[0022] A first frequency at which each edge point is marked as an identifiable point in the grayscale image of all bands of the hyperspectral image is determined, and a second frequency at which all the edge points are marked as identifiable points in the grayscale image of all bands of the hyperspectral image;

[0023] The ratio of the first frequency to the second frequency is determined. When the ratio of the first frequency to the second frequency is greater than a second preset threshold, the edge point corresponding to the first frequency is marked as a feature point.

[0024] Furthermore, the construction of the reference band curve for the pine species based on the hyperspectral image of the control region includes:

[0025] For the complete image of the control area, calculate the extreme ratio of reflectance of each pixel in the tree area in each band, and the extreme ratio of the slope of the hyperspectral curve of all pixels.

[0026] The band consistency index for each band is determined based on the extreme ratio of reflectivity and the extreme ratio of slope.

[0027] Bands whose band consistency index is greater than the third preset threshold are marked, and bands that are marked in every inspection are identified as similar bands.

[0028] The average reflectance of the pixels in all the tree areas in the control area in the similar band is determined as the normal reflectance reference value of the similar band;

[0029] Arrange the normal reflectance reference values ​​of each similar band in wavelength order, plot the wavelength-reflectance curve, and obtain the reference band curve.

[0030] Furthermore, the step of comparing the hyperspectral curves of each pixel in the hyperspectral image of the detection area with the reference band curve to identify abnormal regions in the detection area includes:

[0031] Identify the sensitive wavelength bands for detection;

[0032] For the complete image of the detection area determined by a single inspection, calculate the first tangent slope and first reflectance of each pixel in the tree region of the complete image of the detection area in the detection sensitive band of the corresponding hyperspectral curve;

[0033] Obtain the second tangent slope and second reflectance of the detection sensitive band in the reference band curve;

[0034] The probability of an anomaly of the corresponding pixel is determined based on the first tangent slope, the first reflectivity, the second tangent slope, and the second reflectivity.

[0035] When the probability of an anomaly is greater than a fourth preset threshold, the corresponding pixel is marked as an anomaly point, and the anomaly region is determined based on the anomaly point.

[0036] Furthermore, the calculation of the pest infestation probability index of the pine sawyer beetle larvae in each of the abnormal regions, and the determination of the pest-infested areas in each of the abnormal regions based on the time-series data of the pest infestation probability index corresponding to each of the abnormal regions, includes:

[0037] For each detection cycle, the minimum number of times each pixel in each abnormal region is marked as the detected abnormal point, the first average of the abnormal probability of all pixels in each abnormal region in multiple inspections, and the second average of the number of times the pixels in all abnormal regions are marked as the detected abnormal point are obtained.

[0038] The pest probability index of the presence of the pine sawyer beetle larvae in each of the abnormal areas is determined based on the minimum number of occurrences, the first average value, and the second average value.

[0039] Obtain time-series data of the pest probability index, and determine the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio of the abnormal region of the time-series data;

[0040] The pest and disease area characteristic index corresponding to the abnormal area is determined based on the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio.

[0041] When the characteristic index of the disease and pest area is greater than the fifth preset threshold, the corresponding abnormal area is identified as the disease and pest area.

[0042] Furthermore, determining the effective continuous difference sequence length of the time-series data, the mean of the positive difference values, and the area expansion ratio of the abnormal region includes:

[0043] The time series data of the pest probability index is subjected to first-order difference, and the values ​​of the first-order difference result greater than 0 are marked. If there are consecutive differences that are marked, the number of consecutive marks is determined as the length of the effective consecutive difference sequence.

[0044] The mean of the positive difference values ​​is obtained by calculating the average of the first-order difference results that are positive.

[0045] The area of ​​the first region corresponding to the first detection cycle in which the abnormal region first appears, and the area of ​​the second region corresponding to the target detection cycle of the abnormal region are obtained, and the area expansion ratio is determined based on the area of ​​the first region and the area of ​​the second region.

[0046] Furthermore, the generation of early warning indices for each of the aforementioned pest and disease areas includes:

[0047] Determine the first pest and disease area characteristic index corresponding to the first appearance of the pest and disease area, the second pest and disease area characteristic index corresponding to the pest and disease area in the target detection period, and the duration from the first appearance of the pest and disease area to the target detection period.

[0048] The early warning index for each of the pest and disease areas is determined based on the first pest and disease area characteristic index, the second pest and disease area characteristic index, and the duration.

[0049] Furthermore, the step of outputting early warning information for each of the pest and disease areas based on the early warning index includes:

[0050] During the detection period, the early warning information for each of the pest and disease areas is output in descending order of the early warning index. The early warning information includes the regional location of the corresponding pest and disease area, the pest transmission path, and prevention and control recommendations.

[0051] This invention may have some or all of the following beneficial effects:

[0052] In the online detection and control method for forest pests and diseases provided by this invention, the forest area to be inspected is inspected multiple times according to the same inspection rules to obtain hyperspectral images of the detection area and the control area in the forest area to be inspected. The control area is the area in the forest area to be inspected that contains only pine species and has been controlled for pine sawyer beetle larvae. The detection area is all other areas in the forest area to be inspected except for the control area. A reference band curve for pine species is constructed based on the hyperspectral image of the control area. The hyperspectral curve of each pixel in the hyperspectral image of the detection area is compared with the reference band curve to identify abnormal areas in the detection area. The pest probability index of pine sawyer beetle larvae in each abnormal area is calculated, and the pest and disease areas in each abnormal area are determined based on the time series data of the pest probability index corresponding to each abnormal area. An early warning index for each pest and disease area is generated, and early warning information for each pest and disease area is output based on the early warning index. The early warning index is used to indicate the degree of pest spread in the pest and disease area. This invention constructs reference band curves for pine trees using hyperspectral images of control areas obtained through multiple inspections of the forest area under test. By comparing the hyperspectral curves of each pixel in the hyperspectral image of the test area with the reference band curves, abnormal areas in the test area are preliminarily identified. Based on the changing trends of pest and disease characteristics over time, the affected areas can be determined, allowing for the elimination of cases where healthy trees are mistakenly identified, thus improving the accuracy of identifying trees infested with pine sawyer beetle larvae. Furthermore, this invention can output early warning information for each pest and disease area based on an early warning index, enabling staff to implement control measures according to the severity of the damage, thereby improving the effectiveness of pest and disease control.

[0053] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0054] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0055] Figure 1 A flowchart is shown of an online detection and control method for forestry pests and diseases according to an exemplary embodiment of the present disclosure;

[0056] Figure 2 This diagram illustrates a hyperspectral detection result obtained at a specific location during a single inspection in an online detection and control method for forestry pests and diseases according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0057] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation methods, structures, features, and effects of the online detection and control method for forestry pests and diseases proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0059] The specific solution of the online detection and control method for forestry pests and diseases provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0060] Please see Figure 1 It illustrates a flowchart of an online detection and control method for forestry pests and diseases provided in an embodiment of the present invention, as shown below. Figure 1 As shown, this online detection and control method for forestry pests and diseases specifically includes the following steps:

[0061] S110: Conduct multiple inspections of the forest area to be inspected using the same inspection rules to obtain hyperspectral images of the detection area and the control area in the forest area to be inspected; wherein, the control area is the area in the forest area to be inspected that contains only pine species and has been controlled for pine sawyer beetle larvae, and the detection area is all other areas in the forest area to be inspected except for the control area.

[0062] S120: Construct reference band curves for pine species based on hyperspectral images of the control region;

[0063] S130: Compare the hyperspectral curves of each pixel in the hyperspectral image of the detection area with the reference band curve to identify abnormal areas in the detection area;

[0064] S140: Calculate the pest probability index of pine sawyer beetle larvae in each abnormal area, and determine the pest-infested areas in each abnormal area based on the time series data of the pest probability index corresponding to each abnormal area.

[0065] S150: Generate early warning indices for each pest and disease area, and output early warning information for each pest and disease area based on the early warning indices; wherein, the early warning index is used to indicate the degree of pest spread in the pest and disease area.

[0066] This invention constructs reference band curves for pine trees using hyperspectral images of control areas obtained through multiple inspections of the forest area under test. By comparing the hyperspectral curves of each pixel in the hyperspectral image of the test area with the reference band curves, abnormal areas in the test area are preliminarily identified. Based on the changing trends of pest and disease characteristics over time, the affected areas can be determined, allowing for the elimination of cases where healthy trees are mistakenly identified, thus improving the accuracy of identifying trees infested with pine sawyer beetle larvae. Furthermore, this invention can output early warning information for each pest and disease area based on an early warning index, enabling staff to implement control measures according to the severity of the damage, thereby improving the effectiveness of pest and disease control.

[0067] The following is a detailed explanation of each step in the above-mentioned online detection and control method for forestry pests and diseases:

[0068] In step S110, the forest area to be inspected is inspected multiple times using the same inspection rules to obtain hyperspectral images of the detection area and the control area in the forest area to be inspected; wherein, the control area is the area in the forest area to be inspected that contains only pine species and has been controlled for pine sawyer beetle larvae, and the detection area is all other areas in the forest area to be inspected except for the control area.

[0069] In this embodiment of the application, the forest area to be inspected is a forest area for monitoring the larval disease of the pine sawyer beetle based on the online detection and control method for forest pests and diseases provided in this application.

[0070] In this embodiment, the aforementioned control area serves as a reference standard for the detection area. By selecting a region of healthy trees and processing its hyperspectral images, data such as the spectral characteristics of normal trees can be obtained, providing a clear reference for the detection area. By comparing the spectral data of the detection area with that of the control area, it is possible to determine whether the trees in the detection area have health problems or other abnormalities. Furthermore, the aforementioned control area can also eliminate interference from irrelevant factors. By setting the same environmental conditions for the data acquisition process of the detection area and the control area (i.e., acquiring hyperspectral images of the control area and the detection area in the forest area to be tested using the same inspection rules), interference from irrelevant factors such as drone flight altitude, lighting conditions, and atmospheric environment can be eliminated to a certain extent, making the detection results more accurately reflect the actual condition of the trees.

[0071] Since the online detection and control method for forestry pests and diseases proposed in this application mainly targets the pine sawyer beetle larvae disease, and the damage caused by the pine sawyer beetle larvae disease mainly affects pine species, the selection method for the control area can be as follows: A small area containing only pine species is selected in the forest area to be tested, and pine sawyer beetle larvae control is carried out in advance to obtain a healthy tree area as a control area, providing a reference for the detection area. The control of pine sawyer beetle larvae can be achieved as follows: A suitable concentration of Beauveria bassiana spore suspension is evenly sprayed onto the selected small area containing only pine species to avoid the impact of the pine sawyer beetle larvae disease on the trees.

[0072] In this embodiment, the detection area is the area in the forest area to be tested that requires monitoring for pine sawyer beetle larvae. For example, other areas in the forest area to be tested besides the control area can be designated as the detection area.

[0073] In this embodiment, to eliminate interference from irrelevant factors such as UAV flight altitude, lighting conditions, and atmospheric environment, the detection area and control area need to be inspected according to the same inspection rules. For example, the above-mentioned multiple inspections of the forest area to be inspected using the same inspection rules to obtain hyperspectral images of the detection area and control area can be achieved as follows: Determine the detection cycle; within each detection cycle, conduct multiple inspections of the detection area at fixed time points to obtain hyperspectral images of the detection area; wherein, hyperspectral images of the detection area obtained at adjacent times during a single inspection have overlapping areas; within each detection cycle, conduct multiple inspections of the control area at fixed time points to obtain hyperspectral images of the control area; wherein, hyperspectral images of the control area obtained at adjacent times during a single inspection have overlapping areas.

[0074] Specifically, in a practical detection scenario, hyperspectral images of the detection area can be obtained through the following process: A drone-mounted hyperspectral camera is used to conduct multiple inspections of the detection area at a fixed altitude along a fixed path to acquire hyperspectral images of the area. During this inspection process, a one-day inspection cycle can be used. Within each cycle, inspections are conducted at fixed times (e.g., between 11:00 AM and 1:00 PM), with multiple inspections per day (at least three). During each inspection, the drone-mounted hyperspectral camera acquires hyperspectral images of different locations at a uniform speed and perpendicular to the canopy, ensuring that there are overlapping areas in the hyperspectral images acquired at adjacent times during the inspection.

[0075] To eliminate interference from irrelevant factors, the control area also needs to be inspected using the same inspection rules. That is, it also needs to be inspected at fixed times each day (e.g., between 11:00 AM and 1:00 PM), with multiple inspections per day (at least 3 times). During each inspection, a drone-mounted hyperspectral camera is used to acquire hyperspectral images of different locations in the control area at a uniform speed and perpendicular to the canopy, ensuring that there are overlapping areas in the hyperspectral images acquired at adjacent times. Preferably, to avoid interference from the white film formed by the Beauveria bassiana suspension on the spectral detection results, this embodiment requires multiple inspections of the control area within 12-24 hours after spraying the Beauveria bassiana suspension.

[0076] After acquiring the hyperspectral images of the detection and control areas, the acquired data can be further stored and preprocessed. This preprocessing includes spectral radiometric correction and geometric correction of the hyperspectral images. Spectral radiometric correction eliminates the influence of sensor characteristics and atmospheric conditions on spectral radiometric measurements, ensuring data accurately reflects the spectral radiometric characteristics of ground objects. Geometric correction eliminates geometric distortions caused by various factors during hyperspectral image acquisition, ensuring the location of ground objects in the hyperspectral image matches their actual geographical location.

[0077] In this embodiment, the hyperspectral image is optical image data containing hundreds of continuous narrow bands, covering the visible light (400-700nm), near-infrared (700-1300nm), and short-wave infrared (1300-2500nm) ranges. Each pixel corresponds to a complete spectral curve, which can reflect the fine spectral characteristics of ground objects, such as... Figure 2 The image shown is a schematic diagram of the hyperspectral detection results obtained at a certain location during a single inspection.

[0078] Since the forest area to be tested is usually large, the hyperspectral image obtained by the UAV in a single operation typically only contains a portion of the forest area. Therefore, in order to obtain complete hyperspectral images of the detection area and the control area, this embodiment of the application can utilize the complementary information of the overlapping area between two hyperspectral images at adjacent times during a single inspection to stitch together images obtained by the UAV at different locations to obtain complete hyperspectral images of the detection area and the control area. For example, this process can be implemented as follows: For each hyperspectral image of the detection area and the control area obtained in a single inspection, tree regions in the hyperspectral images are labeled based on a semantic recognition algorithm; for each single-band grayscale image in each hyperspectral image, edge points of the tree regions are obtained based on an edge detection algorithm; feature points in the edge points are determined based on the curvature and grayscale gradient magnitude of the edge points; and the hyperspectral images obtained at different times during a single inspection are stitched together based on the feature points to obtain complete images of the detection area and the control area.

[0079] The above-mentioned determination of feature points among edge points based on the curvature and gray-level gradient magnitude of edge points can be achieved as follows: Calculate the curvature of each edge point, the gray-level gradient magnitude of each edge point, and the average gray-level gradient magnitude of all edge points; determine the identifiability index of each edge point based on the curvature, gray-level gradient magnitude, and average gray-level gradient magnitude of all edge points, and mark edge points with identifiability indices greater than a first preset threshold as identifiable points; determine the first frequency at which each edge point is marked as an identifiable point in the gray-level image of all bands of the hyperspectral image, and the second frequency at which all edge points are marked as identifiable points in the gray-level image of all bands of the hyperspectral image; determine the ratio of the first frequency to the second frequency, and when the ratio of the first frequency to the second frequency is greater than a second preset threshold, mark the edge point corresponding to the first frequency as a feature point.

[0080] The process of selecting feature points and image stitching described above will be explained in detail below in a specific embodiment:

[0081] Taking the k-th hyperspectral image obtained in a single inspection as an example, the process of determining feature points described above can be implemented as follows:

[0082] S1-1: Labeling tree regions in hyperspectral images based on semantic recognition algorithms.

[0083] In this embodiment, the hyperspectral image is segmented, and a semantic recognition algorithm is used to label the tree regions in the hyperspectral image, while masks are applied to the remaining regions. The semantic recognition algorithm can be implemented using a 3D-UNet model (a three-dimensional U-shaped network model), and the labeling results for the tree regions can be mapped to the grayscale images of all bands in the hyperspectral image.

[0084] S2-1: For any single-band grayscale image of the hyperspectral image, obtain the edge points in the above-mentioned tree region based on the edge detection algorithm.

[0085] In this application embodiment, for example, the Canny operator can be used to obtain the edge points of tree regions in any single-band grayscale image.

[0086] S3-1: Determine feature points in edge points based on the curvature and gray-level gradient magnitude of edge points.

[0087] In this embodiment, the aforementioned feature points are used to stitch together hyperspectral images from a single inspection, and need to reflect the morphological and spectral features of the tree region in the current image. For example, the aforementioned morphological features are related to the curvature of edge points; the greater the curvature of an edge point, the greater the likelihood that it is a turning point or a point of abrupt texture change at the edge of a tree region. The aforementioned spectral features are related to the gray-level gradient amplitude of edge points; that is, the greater the gray-level gradient amplitude in the gray-level image corresponding to the current band, the greater the likelihood that an edge point is a ground feature edge point, and the higher the identifiability of that point. Therefore, feature points in edge points can be determined based on the curvature and gray-level gradient amplitude of edge points. Specifically, in this embodiment, the identifiability index of each edge point can be calculated using the curvature and gray-level gradient amplitude of the edge points. Taking the i-th edge point in the j-th tree region of the current single-band gray-level image as an example, the aforementioned identifiability index can be calculated using the following formula:

[0088]

[0089] Where, δ i,j c represents the discrimination index of the i-th edge point in the j-th tree region of the current single-band grayscale image. i,j G represents the curvature of that edge point. i,j This represents the grayscale gradient magnitude at the edge point. This represents the gray-level gradient magnitude g of all edge points in the current single-band grayscale image. i,j The mean; The larger the value, the greater the change in reflectance of the edge point relative to other edge points in the current single-band grayscale image. The sigmoid function is a normalization function with a range of (0,1).

[0090] After calculating the identifiability index of each edge point in the current single-band grayscale image using the above formula, the identifiable points among each edge point can be determined based on this identifiability index. For example, δ can be... i,j Edge points with a value greater than 0.5 (where 0.5 is the value of the first preset threshold mentioned above) are marked as identifiable points in the current single-band grayscale image.

[0091] To improve the stability of selected feature points, embodiments of this application can determine feature points by the frequency with which each edge point is marked as an identifiable point in the single-band grayscale image of all bands of the aforementioned hyperspectral image, thereby improving the stability of feature points. The higher the frequency with which an edge point is marked as an identifiable point in all bands, the more stable the edge point is, and the greater the probability that it is a feature point. For example, it can be... (Here, 0.5 is the value of the second preset threshold mentioned above) identifiable points are feature points, where n i,j (That is, the first frequency mentioned above) represents the frequency at which the i-th edge point is marked as a identifiable point. (That is, the second frequency mentioned above) represents the frequency at which all edge points are marked as identifiable points.

[0092] Specifically, assuming the hyperspectral image has 10 bands, in the single-band grayscale image corresponding to each band, the discrimination index of each edge point is determined based on the above steps. If the i-th edge point is in the single-band grayscale image corresponding to 7 bands, the discrimination index is calculated by the above steps. i,j If the value is greater than 0.5 (meaning the i-th edge point is marked as identifiable in 7 bands), then the above n i,j The value is 7; if we statistically analyze the marking of all edge points across the 10 bands, we find that the total number of times each edge point is marked as an identifiable point across the 10 bands is 500. Therefore, the above... At this time, the above Since 0.014 < 0.5, the stability of the i-th edge point as a feature point is relatively low, and it cannot be selected as a feature point.

[0093] S4-1: Based on the feature points determined in steps S1 to S3 above, the hyperspectral images at adjacent times during a single inspection are stitched together to obtain a complete image of the detection area or a complete image of the control area corresponding to that inspection process.

[0094] In this embodiment of the application, for example, the above-described image stitching process based on feature points can be implemented using the SuperGlue algorithm. This algorithm is a deep learning algorithm for image feature matching, used to match feature points in two images to determine feature points representing the same physical location in different images. After matching the extracted feature points using the SuperGlue algorithm and finding the corresponding feature points in the hyperspectral images at different times, the transformation matrix between the images can be calculated based on the matched feature points, and the hyperspectral images at different times during this inspection process can be stitched together based on the calculated transformation matrix.

[0095] In step S120, a reference band curve for pine species is constructed based on the hyperspectral image of the control region.

[0096] In this embodiment, the aforementioned reference band curve serves as a baseline curve characterizing the spectral characteristics of healthy Pinus species, providing a reliable comparison standard for pest and disease detection. This reference band curve is generated by statistically analyzing the hyperspectral data of healthy pine trees, selecting bands with high consistency, and calculating their average reflectance. The hyperspectral data of the healthy pine trees can be a complete hyperspectral image of the control area generated based on the above steps (i.e., the complete image of the control area).

[0097] For example, the above-mentioned construction of reference band curves for pine species based on hyperspectral images of the control area can be achieved as follows: For a complete image of the control area, calculate the extreme reflectance ratio of each pixel in each band and the extreme slope ratio of the hyperspectral curves of all pixels; determine the band consistency index of each band based on the extreme reflectance ratio and the extreme slope ratio; mark the bands with a band consistency index greater than a third preset threshold, and determine the bands marked in each inspection as similar bands; determine the average reflectance of pixels in the similar bands in all tree areas of the control area as the normal reflectance reference value of the similar band; arrange the normal reflectance reference values ​​of each similar band in wavelength order, draw a wavelength-reflectance curve, and obtain the reference band curve.

[0098] In the embodiments of this application, the aforementioned similar bands refer to bands with highly consistent reflectance and spectral morphology selected by analyzing the hyperspectral data of healthy pine trees in the control area. These bands can stably characterize the optical properties of healthy pine trees and provide a reliable benchmark for subsequent detection of pine sawyer beetle larvae in the detection area.

[0099] In the above steps, this application embodiment protects the pine trees in the control area from the pine sawyer beetle larvae disease. Furthermore, the hyperspectral curves of different pine species exhibit a high degree of overlap in their overall spectral morphology. Therefore, based on the consistency analysis of the hyperspectral detection of pine trees in the control area, the normal reflectance reference values ​​of pine trees in similar bands under the test forest environment can be determined, and the reference band curve can be determined based on these normal reflectance reference values. Specifically, taking the complete image of the control area obtained from a single inspection and stitching as an example, the process of marking similar bands and determining the reference band curve is explained in detail:

[0100] S1-2: Calculate the ratio of the extreme reflectance values ​​of each pixel in the tree region of the complete image of the above-mentioned control area in each band.

[0101] In this embodiment, the aforementioned reflectance extreme value ratio is an index used to characterize the similarity of reflectance of tree region pixels in the complete image of the comparison region in corresponding bands. Specifically, taking the z-th band of the complete image of the comparison region as an example, the aforementioned reflectance extreme value ratio a of the tree region pixels in the complete image of the comparison region in the z-th band is... z It can be calculated using the following formula:

[0102]

[0103] Where, r z,min r is the minimum reflectance value of the hyperspectral curves corresponding to all pixels in the tree region of the complete image of the above-mentioned control area in the z-th band. z,max The maximum reflectance value in the z-th band of the hyperspectral curve corresponding to all pixels in the tree region of the complete image of the above-mentioned control area; the reflectance extreme value ratio a z The larger the band size, the more similar the reflectance of all trees in the z-th band.

[0104] S2-2: Calculate the ratio of the extreme slopes of the hyperspectral curves of all pixels in each band.

[0105] In this embodiment, the aforementioned slope extreme value ratio is an index used to characterize the similarity of the reflectance variation trends of pixels in the complete image of the comparison region in the corresponding band. Specifically, taking the z-th band of the complete image of the comparison region as an example, the slope extreme value ratio b of the pixels in the complete image of the comparison region in the z-th band is... z It can be calculated using the following formula:

[0106]

[0107] Where, k z,min Let k be the minimum slope of the tangent at the z-th band in the hyperspectral curves corresponding to all pixels in the complete image of the control region. z,max This represents the maximum value of the tangent slope at the z-th band in the hyperspectral curves corresponding to all pixels in the complete image of the control region.

[0108] S3-2: Determine the band consistency index for each band based on the above extreme reflectance ratio and extreme slope ratio.

[0109] In this embodiment, the aforementioned band consistency index is used to evaluate the overall stability of each band by combining the aforementioned reflectance extreme value ratio and slope extreme value ratio. Specifically, taking the z-th band of the complete image of the comparison region as an example, the band consistency index u of the z-th band... z It can be calculated using the following formula:

[0110] u z =premnmx(a z×b z )

[0111] Among them, u z is the band consistency index for the z-th band, used to represent the consistency of reflectance of all tree region pixels in the z-th band in the complete image of the current control area; premnmx is the normalization function with a value range of [-1,1].

[0112] S4-2: Based on the above band consistency index, determine the similar bands in the complete image of the control area.

[0113] In this embodiment, the determination of similar bands in the complete image of the comparison region based on the aforementioned band consistency index can be achieved as follows: bands with a band consistency index greater than a third preset threshold are marked, and bands marked in each inspection are determined as similar bands. Specifically, taking the z-th band of the complete image of the comparison region as an example, u z Bands with a normalization result greater than 0.5 (where 0.5 is the value of the third preset threshold mentioned above) are marked. If the z-th band is marked in every inspection process, the z-th band is marked as a similar band to the pine trees in the forest area to be inspected, and the average reflectance of the pixels in the z-th band in all tree areas in the control area is taken as the normal reflectance reference value of the z-th band.

[0114] S5-2: Determine the reference band curve based on the normal reflectance reference values ​​of each similar band.

[0115] In this embodiment of the application, the normal reflectance reference values ​​of all similar bands are plotted at the corresponding positions in the wavelength (horizontal axis)-reflectance (vertical axis) coordinate system to obtain the reference band curves of the pine trees in the forest area to be inspected.

[0116] In step S130, the hyperspectral curves of each pixel in the hyperspectral image of the detection area are compared with the reference band curve to identify abnormal areas in the detection area.

[0117] In this embodiment, the aforementioned abnormal area refers to the region within the detection area where pine sawyer beetle larvae may be present. Because pine sawyer beetle larvae damage the internal structure of pine trees, causing imbalances in water metabolism and decreased photosynthesis, the hyperspectral curves of the trees change. Furthermore, due to different growing environments, the spectral curves of different trees also vary. The aforementioned reference band curve serves as a baseline curve characterizing the spectral features of healthy pine trees in the tested forest area. Therefore, by comparing the differences between the hyperspectral curves of pine trees within the detection area and the aforementioned reference band curve, the likelihood of pine sawyer beetle larvae infestation in the corresponding trees can be analyzed, and the aforementioned abnormal area can be identified accordingly.

[0118] For example, the above-mentioned comparison of the hyperspectral curves of each pixel in the hyperspectral image of the detection area with the reference band curve to identify abnormal regions in the detection area can be achieved as follows: determine the detection sensitive band; for the complete image of the detection area determined by a single inspection, calculate the first tangent slope and first reflectance of each pixel in the tree area of ​​the complete image of the detection area in the corresponding detection sensitive band of the hyperspectral curve; obtain the second tangent slope and second reflectance of the detection sensitive band in the reference band curve; determine the abnormal probability of the corresponding pixel based on the first tangent slope, first reflectance, second tangent slope, and second reflectance; when the abnormal probability is greater than a fourth preset threshold, mark the corresponding pixel as a detection abnormal point, and determine the abnormal region based on the detection abnormal point.

[0119] In this embodiment, the aforementioned sensitive detection band refers to a specific wavelength range in which the reflectance of vegetation undergoes a significant and detectable change when it is affected by pests and diseases. That is, the band with the most significant difference in reflectance when comparing the normal reference spectrum of healthy vegetation with the spectrum of pest-infested vegetation. Since the pine sawyer beetle larvae bore into pine trees, reducing chlorophyll in pine leaves, the reflectance in the near-infrared band shows a downward trend. Therefore, for pine species affected by pine sawyer beetle larvae, the aforementioned sensitive detection band is concentrated in the near-infrared band (700nm-1300nm).

[0120] The process of determining the abnormal region described above will be explained in detail below in a specific embodiment:

[0121] In this embodiment, the reference band curve determined based on step S120 is used as a reference benchmark. The hyperspectral curves corresponding to pixels at different locations within the detection area are analyzed. The greater the difference between a hyperspectral curve and the normal reference band curve, the greater the likelihood of an anomaly at the location of the pixel corresponding to that hyperspectral curve. Specifically, taking a single pixel in the tree region of the complete image of the detection area obtained in a single inspection process as an example, if the reference band curve contains a near-infrared band, the corresponding band (i.e., the band in the reference band curve with wavelengths in the range of 700nm-1300nm) is marked as a detection-sensitive band. Taking the i-th pixel in the tree region of the complete image of the detection area as an example, the anomaly probability v of each pixel is... i It can be calculated using the following formula:

[0122]

[0123] Among them, v i This indicates the probability that the i-th pixel in the tree region of the complete image contains an anomaly. i,z' This indicates that the i-th pixel is at the z-th position. 'The reflectance of each detection-sensitive band (i.e., the first reflectance mentioned above), k i,z' The hyperspectral curve corresponding to the i-th pixel at the z-th position ' The tangent slope of each detection sensitive band (i.e., the first tangent slope mentioned above), r z' k represents the reflectance of the z'-th detection sensitive band in the reference band curve (i.e., the second reflectance mentioned above). z' Indicates the z-th band in the reference band curve ' The tangent slope of each detection sensitive band (i.e., the second tangent slope mentioned above); n z' This indicates the total number of sensitive bands detected.

[0124] In the above formula, The larger the value, the more likely the i-th pixel is to be at the z-th position. ' The reflectance of the z-th sensitive band in the detection band is compared to that of the reference band curve. ' The smaller the reflectance reference value for each sensitive wavelength band, the better; |k i,z' -k z' The larger the | value, the more significant the hyperspectral curve corresponding to the i-th pixel at the z-th pixel. ' The greater the difference between the shape of the morphology at the detection sensitive band and the shape of the reference band curve, the greater the possibility that the ith pixel has a detection anomaly. If the hyperspectral curve of the ith pixel has a smaller reflectance in the detection sensitive band compared to the reference band curve, and the slope of the tangent at the corresponding position of the ith pixel differs more from that of the detection sensitive band, the greater the possibility that the ith pixel has a detection anomaly.

[0125] After calculating the probability of anomalies for each pixel in the tree region of the complete image of the detection area using the above formula, the detection anomaly points can be determined based on the probability of anomalies. Specifically, the probability of anomalies for each pixel can be normalized to (0,1), and pixels with a normalization result greater than 0.5 (where 0.5 is the value of the fourth preset threshold mentioned above) can be marked as detection anomalies.

[0126] In this embodiment of the application, after determining the abnormal points in each pixel of the tree region of the complete image of the detection area through the above process, abnormal regions in the tree region of the complete image of the detection area can also be determined based on the above-mentioned abnormal points.

[0127] For example, the above-mentioned identification of abnormal regions can be achieved as follows:

[0128] S1-3: Extract the connected components containing all detected outlier points as the outlier detection regions.

[0129] Because drones are affected by outdoor environmental interference and airflow when flying outdoors, their attitude may be unstable, which may lead to changes in the incident angle of hyperspectral detection, resulting in inaccurate hyperspectral detection results. Ultimately, this can cause the detected abnormal areas to be different in multiple inspections.

[0130] To address this issue, after extracting the connected components containing all detected outliers as outlier regions, the following steps are required to determine the aforementioned outlier regions based on these outlier regions:

[0131] S2-3: Match the complete images of the detection area obtained from each inspection in a single detection cycle based on the selected feature points. In step S110 above, when stitching the hyperspectral images obtained from a single inspection, it is necessary to ensure that the image size of the complete images of the detection area obtained from each stitching is the same.

[0132] S3-3: Create an empty image with an initial grayscale value of 0 for all pixels. Mark all the pixels in the above-mentioned abnormal detection areas in the empty image. Each time a pixel is marked, the grayscale value of the corresponding pixel is incremented by 1.

[0133] S4-3: Use threshold segmentation to obtain the foreground region, which is then used as the abnormal region in the current detection cycle.

[0134] In this embodiment, the threshold can be selected based on the tolerance for false positives and the detection requirements. For example, if three inspections are performed within a detection cycle and the threshold is set to 2, a pixel must be marked as an abnormal point in at least two detections before it can be segmented into the foreground region.

[0135] In step S140, the pest probability index of pine sawyer beetle larvae in each abnormal area is calculated, and the pest-infested areas in each abnormal area are determined based on the time series data of the pest probability index corresponding to each abnormal area.

[0136] Because in actual testing, there are cases where the hyperspectral detection results of other healthy tree species are similar to those of pine trees affected by pine sawyer beetle larvae, healthy tree species may be mistakenly identified as areas affected by pine sawyer beetle larvae during the process of identifying abnormal areas. Therefore, in this embodiment, after identifying abnormal areas through the above process, it is necessary to further distinguish the above situations based on the trend of the degree of abnormality of each abnormal area changing over time, and determine the diseased areas within the abnormal areas.

[0137] In this embodiment of the application, the above-mentioned pest probability index is used to characterize the probability that the corresponding abnormal area contains pine sawyer beetle larvae in the current detection period. Based on the pest probability index, the abnormal area can be distinguished as a healthy tree species or an area suffering from pest infestation by pine sawyer beetle larvae.

[0138] During the development of pine sawyer beetle larval infestations, the degree of abnormality in the aforementioned abnormal areas may fluctuate somewhat in the early stages of larval feeding. However, as the larvae penetrate deeper into the xylem, the degree of abnormality increases continuously and significantly. Meanwhile, the healthy trees of other species in the misidentified abnormal areas remain relatively stable, with their corresponding spectral curves showing minimal fluctuations at different times. In other words, the healthy trees of other species in the misidentified abnormal areas maintain a relatively consistent degree of abnormality over a longer period. Based on the above analysis, the infested areas of the pine sawyer beetle larvae can be determined by the consistency of the degree of abnormality in each abnormal area.

[0139] For example, the above calculation of the pest probability index of pine sawyer beetle larvae in each abnormal area, and the determination of the pest-infested area in each abnormal area based on the time series data of the pest probability index corresponding to each abnormal area, can be achieved as follows: For each detection cycle, obtain the minimum number of times each pixel in each abnormal area is marked as an abnormal point, the first average value of the abnormal probability of all pixels in each abnormal area in multiple inspections, and the second average value of the number of times pixels in all abnormal areas are marked as abnormal points; determine the pest probability index of pine sawyer beetle larvae in each abnormal area based on the minimum number of times, the first average value, and the second average value; obtain the time series data of the pest probability index, and determine the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio of the abnormal area; determine the pest-infested area feature index corresponding to the abnormal area based on the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio; when the pest-infested area feature index is greater than a fifth preset threshold, the corresponding abnormal area is determined as a pest-infested area.

[0140] The determination of the effective continuous difference sequence length, the mean of positive difference values, and the area expansion ratio of abnormal regions for the aforementioned time-series data can be achieved as follows: First-order difference is performed on the time-series data of the pest probability index, and values ​​greater than 0 in the first-order difference result are marked. If continuous differences are marked, the number of continuous marks is determined as the effective continuous difference sequence length. The average value of the first-order difference results with positive results is calculated to obtain the mean of positive difference values. The area of ​​the first region corresponding to the first detection period of the abnormal region and the area of ​​the second region corresponding to the abnormal region in the target detection period are obtained, and the area expansion ratio is determined based on the area of ​​the first region and the area of ​​the second region.

[0141] The process of determining the pest and disease area described above will be explained in detail below in a specific embodiment:

[0142] S1-4: Based on feature points, match the complete images of the detection regions corresponding to adjacent detection cycles, and obtain the abnormal region pair with the largest overlapping area in the adjacent detection cycle.

[0143] In this embodiment of the application, taking the Tth period and the (T-1)th period as examples, the above process can be implemented as follows: a graph neural network-based matching method matches feature point pairs in the Tth period and the (T-1)th period, and outputs the coordinates of the matched point pairs; the Random Sample Consensus (RANSAC) algorithm is used to estimate the affine transformation matrix, and the Tth period image is aligned to the coordinate system of the (T-1)th period; all abnormal region pairs corresponding to the abnormal regions in the (T-1)th period and the Tth period are determined, and the overlap area between the two abnormal regions in each abnormal region pair is calculated; the abnormal region pair with the largest overlap area among all abnormal region pairs in the (T-1)th period and the Tth period is obtained.

[0144] S2-4: Calculate the pest infestation probability index of pine sawyer beetle larvae in each abnormal area.

[0145] In this embodiment of the application, taking the j-th abnormal region as an example, for a single detection cycle, the more times each pixel in the j-th abnormal region is marked as an abnormal point, the higher the probability v of each pixel being abnormal during multiple inspections in that detection cycle. i The larger the value, the greater the probability that the j-th abnormal area contains pine sawyer beetle larvae. Therefore, the pest probability index of the j-th abnormal area can be calculated using the following formula:

[0146]

[0147] Where, q j This represents the probability that the j-th abnormal region contains pine sawyer beetle larvae within the current detection period (i.e., the aforementioned pest probability index), n j,min This represents the minimum number of times each pixel in the j-th abnormal region is marked as an abnormal point. This represents the average number of times each pixel in all abnormal regions was marked as an abnormal point (i.e., the second average mentioned above). This represents the average probability of anomalies for all pixels in the j-th anomaly region during each inspection process. The average value (i.e., the first average value mentioned above); The larger the value, the more times the pixels in the j-th abnormal region are marked.

[0148] Specifically, if the j-th abnormal region contains 100 pixels, and in the current detection cycle's three inspections, 20 pixels are marked as abnormal three times, 40 pixels are marked as abnormal once, and 40 pixels are marked as abnormal twice, then the above n j,min =1; If the detection area has 3 abnormal regions, each containing 50, 100, and 150 pixels respectively, and the total number of times each pixel in each abnormal region is marked as an abnormal point is 200, 300, and 450 times respectively, then the above... If the j-th abnormal region contains 100 pixels, then in the three inspections of the current detection cycle, the average probability of 20 pixels being abnormal is... The mean anomaly probability of 80 pixels is 0.7. If it is 0.6, then

[0149] S3-4: Determine the pest and disease areas in each abnormal area based on the time series data of the pest probability index corresponding to each abnormal area.

[0150] In this embodiment of the application, taking the j-th abnormal region as an example, if from the first appearance of the j-th abnormal region up to the T-th period, the abnormal development of the j-th abnormal region satisfies the following conditions: (1) the pest probability index q corresponding to the j-th abnormal region j (1) Continue to grow; (2) The pest probability index increases significantly and for a long time; (3) The area of ​​the jth abnormal region expands significantly; then it is determined that the jth abnormal region is consistent with the development trend of the pine sawyer beetle larvae pest, and the jth abnormal region is likely to be a pest area of ​​the pine sawyer beetle larvae.

[0151] Specifically, taking the j-th abnormal area appearing in the first detection cycle and continuing until the third detection cycle as an example, based on the above judgment criteria, the pest probability index q is determined. j The effective continuous difference sequence length, mean of positive difference values, and area expansion of outlier regions in the time series data are as follows: Obtain the time series data of the pest probability index for the j-th outlier region from the first to the third detection period. Assuming the pest probability indices for the first to third detection periods are 0.6, 0.7, and 0.8 respectively, then the above pest probability index time series data is q. j={0.6, 0.7, 0.8|i=1, 2, 3}; Performing a first-order difference on the time-series data of the pest probability index for the j-th abnormal region from the first to the third detection period yields: From the first to the second detection period, the pest probability index increases from 0.6 to 0.7, with a first-order difference of 0.1; From the second to the third detection period, the pest probability index increases from 0.7 to 0.8, with a first-order difference of 0.1. Therefore, the effective continuous difference sequence length of the above pest probability index time-series data is 2; The first-order differences with positive values ​​include those from the first to the second detection period and from the second to the third detection period, both of which are 0.1. Therefore, the mean of the above positive difference values ​​is (0.1+0.1) / 2 = 0.1; Assuming the initial area of ​​the j-th abnormal region is 200 pixels, and the area up to the third detection period is 350 pixels, then the area expansion ratio is 350 / 200 = 1.75.

[0152] In this embodiment of the application, the pest probability index q in the abnormal area j After determining the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio of the abnormal region using the time series, specifically, taking the j-th abnormal region as an example, the above determination of the pest and disease area characteristic index corresponding to the abnormal region based on the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio can be achieved through the following formula:

[0153]

[0154] Among them, w j,T The index representing the region where the j-th abnormal area conforms to the pest and disease characteristics of the pine sawyer beetle larvae in the T-th detection period (i.e., the aforementioned pest and disease region characteristic index), n j The insect pest probability index q represents the j-th abnormal region. j The effective continuous difference sequence length of the time series (i.e., the effective continuous difference sequence length mentioned above), n T This indicates the number of cycles up to the current testing cycle. This represents the mean of all differences greater than 0 (i.e., the mean of the positive differences mentioned above); The larger the value, the greater the probability index q of pest infestation corresponding to the j-th abnormal region. j The longer the period of sustained growth; j,T This represents the area (in pixels) of the j-th abnormal region in the T-th detection period (i.e., the area of ​​the second region mentioned above), s j,1 This represents the area of ​​the j-th abnormal region when it first appears (i.e., the area of ​​the first region mentioned above). The larger the value, the greater the expansion of the j-th abnormal region.

[0155] Specifically, the effective continuous difference sequence length n calculated above is... j 2, the mean of the positive differences The area expansion ratio is 0.1. Taking 1.75 as an example, the number of periods n corresponding to the current detection period from the first detection period to the third detection period is... T =3, then the pest and disease regional characteristic index w of the j-th abnormal region is... j,T = (2×0.1) / 3×1.75≈0.12.

[0156] After obtaining the pest and disease regional characteristic index w corresponding to the abnormal area j,T Next, the regional characteristic index of pests and diseases can be normalized, and the result of the normalization can be used to determine whether an abnormal area is a pest and disease area. Specifically, taking the j-th abnormal area as an example, after calculating the regional characteristic index w of the j-th abnormal area... j,T w j,T Normalize to (0,1). If the normalization result is greater than 0.5 (where 0.5 is the value of the fifth preset threshold mentioned above), then the j-th abnormal region is identified as the pest and disease area of ​​the pine sawyer beetle larvae.

[0157] In step S150, an early warning index is generated for each pest and disease area, and early warning information for each pest and disease area is output based on the early warning index; wherein, the early warning index is used to indicate the degree of pest spread in the pest and disease area.

[0158] In this embodiment, the aforementioned early warning index is used to indicate the degree of pest spread in areas infested by pine sawyer beetle larvae. Alarm information for infested areas can be set based on this early warning index. For example, if, up to the current moment, the faster the degree to which a pest-affected area matches the characteristics of pine sawyer beetle larvae, the greater the degree and speed of tree borer damage in that area, and the more necessary it is to take action.

[0159] In this embodiment, the aforementioned early warning information includes the regional location of the pest-affected area, the pest transmission path, and control recommendations. Taking a pest-affected area of ​​a pine sawyer beetle larva in the Tth cycle as an example, the pest transmission path can be obtained by the following method: marking the coordinate center point of the pest-affected area in the Tth cycle, with the transmission path being the line connecting the coordinate center point of the (T-1)th cycle to the coordinate center point of the Tth cycle; repeating the above steps to obtain the historical transmission path of the pest-affected area.

[0160] For example, the above-mentioned generation of early warning indices for each pest and disease area can be achieved as follows: determining the first pest and disease area characteristic index corresponding to the initial appearance of the pest and disease area, the second pest and disease area characteristic index corresponding to the pest and disease area in the target detection period, and the duration from the initial appearance of the pest and disease area to the target detection period; determining the early warning index for each pest and disease area based on the first pest and disease area characteristic index, the second pest and disease area characteristic index, and the duration. Specifically, if the j-th abnormal area is identified as a pest and disease area, taking the j-th pest and disease area as an example, the above-mentioned early warning index can be calculated using the following formula:

[0161]

[0162] Where, p j,T w represents the early warning index required for the j-th pest and disease area during the T-th detection cycle (i.e., the target detection cycle mentioned above). j,1 This represents w (also known as the characteristic index of the first pest-affected area) when the j-th pest-affected area first appears. j,T This indicates the index (also known as the second pest and disease area characteristic index) that represents the j-th pest and disease area in the T-th detection cycle, which corresponds to the pest and disease area characteristics of the pine sawyer beetle larvae. j This represents the duration from the first appearance of the disease / pest in the j-th area to the T-th cycle (i.e., the aforementioned duration).

[0163] In this embodiment of the application, after determining the early warning index of each pest and disease area, the above-mentioned output of early warning information for each pest and disease area based on the early warning index can be achieved as follows: within the current detection period, the early warning information for each pest and disease area is output in descending order of the early warning index, wherein the early warning information includes the regional location of the corresponding pest and disease area, the pest transmission path, and control recommendations. Specifically, the above-calculated index can be normalized to (0,1), and P can be used according to the current period. j,T Output the location information of the corresponding pest and disease area, the transmission path, and prevention and control suggestions in descending order.

[0164] This invention constructs reference band curves for pine trees using hyperspectral images of control areas obtained through multiple inspections of the forest area under test. By comparing the hyperspectral curves of each pixel in the hyperspectral image of the test area with the reference band curves, abnormal areas in the test area are preliminarily identified. Based on the changing trends of pest and disease characteristics over time, the affected areas can be determined, allowing for the elimination of cases where healthy trees are mistakenly identified, thus improving the accuracy of identifying trees infested with pine sawyer beetle larvae. Furthermore, this invention can output early warning information for each pest and disease area based on an early warning index, enabling staff to implement control measures according to the severity of the damage, thereby improving the effectiveness of pest and disease control.

[0165] The foregoing primarily describes the solutions provided by the embodiments of the present invention from a methodological perspective. To achieve the aforementioned functions, it includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0166] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0167] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for online detection and control of forest pests and diseases, characterized in that, The method includes: Multiple inspections of the forest area to be inspected were conducted using the same inspection rules to obtain hyperspectral images of the detection area and the control area in the forest area to be inspected; wherein, the control area is the area in the forest area to be inspected that contains only pine species and has been controlled for pine sawyer beetle larvae, and the detection area is all other areas in the forest area to be inspected except for the control area; A reference band curve for the pine species is constructed based on the hyperspectral image of the control region; The hyperspectral curves of each pixel in the hyperspectral image of the detection area are compared with the reference band curve to identify abnormal areas in the detection area. Calculate the pest probability index of the presence of the pine sawyer beetle larvae in each of the abnormal regions, and determine the pest-infested areas in each of the abnormal regions based on the time series data of the pest probability index corresponding to each of the abnormal regions. An early warning index is generated for each of the pest and disease areas, and early warning information for each pest and disease area is output based on the early warning index; wherein, the early warning index is used to indicate the degree of pest spread in the pest and disease area.

2. The online detection and control method for forestry pests and diseases according to claim 1, characterized in that, The process of conducting multiple inspections of the forest area to be inspected using the same inspection rules, and acquiring hyperspectral images of the detection area and the control area within the forest area to be inspected, includes: Determine the testing cycle; Within each detection cycle, the detection area is inspected multiple times at fixed time points to acquire the hyperspectral image of the detection area; wherein, the hyperspectral images of the detection area acquired at adjacent time points during a single inspection have overlapping areas. Within each detection cycle, the control area is inspected multiple times at the fixed time point to acquire the hyperspectral image of the control area; wherein, the hyperspectral images of the control area acquired at adjacent times during a single inspection have overlapping areas.

3. The online detection and control method for forestry pests and diseases according to claim 2, characterized in that, After acquiring the hyperspectral images of the detection area and the control area in the forest area to be inspected, the method further includes: For each hyperspectral image of the detection area and the control area obtained in a single inspection, the tree regions in the hyperspectral images are labeled based on a semantic recognition algorithm; For each single-band grayscale image in the hyperspectral images, edge points of the tree region are obtained based on an edge detection algorithm; The feature points in the edge points are determined based on the curvature and grayscale gradient magnitude of the edge points; Based on the feature points, the hyperspectral images obtained at different times during a single inspection are stitched together to obtain a complete image of the detection area and a complete image of the control area.

4. The online detection and control method for forestry pests and diseases according to claim 3, characterized in that, The step of determining the feature points in the edge points based on the curvature and grayscale gradient magnitude of the edge points includes: Calculate the curvature of each edge point, the gray-level gradient magnitude of each edge point, and the average gray-level gradient magnitude of all edge points; The identifiability index of each edge point is determined based on the curvature of each edge point, the gray-level gradient magnitude, and the average gray-level gradient magnitude of all edge points, and the edge points whose identifiability index is greater than a first preset threshold are marked as identifiable points. A first frequency at which each edge point is marked as an identifiable point in the grayscale image of all bands of the hyperspectral image is determined, and a second frequency at which all the edge points are marked as identifiable points in the grayscale image of all bands of the hyperspectral image; The ratio of the first frequency to the second frequency is determined. When the ratio of the first frequency to the second frequency is greater than a second preset threshold, the edge point corresponding to the first frequency is marked as a feature point.

5. The online detection and control method for forestry pests and diseases according to claim 3, characterized in that, The construction of the reference band curve for the pine species based on the hyperspectral image of the control region includes: For the complete image of the control area, calculate the extreme ratio of reflectance of each pixel in the tree area in each band, and the extreme ratio of the slope of the hyperspectral curve of all pixels. The band consistency index for each band is determined based on the extreme ratio of reflectivity and the extreme ratio of slope. Bands whose band consistency index is greater than the third preset threshold are marked, and bands that are marked in every inspection are identified as similar bands. The average reflectance of the pixels in all the tree areas in the control area in the similar band is determined as the normal reflectance reference value of the similar band; Arrange the normal reflectance reference values ​​of each similar band in wavelength order, plot the wavelength-reflectance curve, and obtain the reference band curve.

6. The online detection and control method for forestry pests and diseases according to claim 5, characterized in that, The step of comparing the hyperspectral curves of each pixel in the hyperspectral image of the detection area with the reference band curve to identify abnormal regions in the detection area includes: Identify the sensitive wavelength bands for detection; For the complete image of the detection area determined by a single inspection, calculate the first tangent slope and first reflectance of each pixel in the tree region of the complete image of the detection area in the detection sensitive band of the corresponding hyperspectral curve; Obtain the second tangent slope and second reflectivity of the detection sensitive band in the reference band curve; The probability of an anomaly of the corresponding pixel is determined based on the first tangent slope, the first reflectivity, the second tangent slope, and the second reflectivity. When the probability of an anomaly is greater than a fourth preset threshold, the corresponding pixel is marked as an anomaly point, and the anomaly region is determined based on the anomaly point.

7. The online detection and control method for forestry pests and diseases according to claim 6, characterized in that, The calculation of the pest infestation probability index of the pine sawyer beetle larvae in each of the abnormal regions, and the determination of the pest-infested areas in each of the abnormal regions based on the time-series data of the pest infestation probability index corresponding to each of the abnormal regions, includes: For each detection cycle, the minimum number of times each pixel in each abnormal region is marked as the detected abnormal point, the first average of the abnormal probability of all pixels in each abnormal region in multiple inspections, and the second average of the number of times the pixels in all abnormal regions are marked as the detected abnormal point are obtained. The pest probability index of the presence of the pine sawyer beetle larvae in each of the abnormal areas is determined based on the minimum number of occurrences, the first average value, and the second average value. Obtain time-series data of the pest probability index, and determine the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio of the abnormal region of the time-series data; The pest and disease area characteristic index corresponding to the abnormal area is determined based on the effective continuous difference sequence length, the mean of the positive difference values, and the area expansion ratio. When the characteristic index of the disease and pest area is greater than the fifth preset threshold, the corresponding abnormal area is identified as the disease and pest area.

8. The online detection and control method for forestry pests and diseases according to claim 7, characterized in that, Determining the effective continuous difference sequence length, the mean of positive difference values, and the area expansion ratio of the outlier region of the time series data includes: The time series data of the pest probability index is subjected to first-order difference, and the values ​​of the first-order difference result greater than 0 are marked. If there are consecutive differences that are marked, the number of consecutive marks is determined as the length of the effective consecutive difference sequence. The mean of the positive difference values ​​is obtained by calculating the average of the first-order difference results that are positive. The area of ​​the first region corresponding to the first detection cycle in which the abnormal region first appears, and the area of ​​the second region corresponding to the target detection cycle of the abnormal region are obtained, and the area expansion ratio is determined based on the area of ​​the first region and the area of ​​the second region.

9. The online detection and control method for forestry pests and diseases according to claim 7, characterized in that, The generation of early warning indices for each of the aforementioned pest and disease areas includes: Determine the first pest and disease area characteristic index corresponding to the first appearance of the pest and disease area, the second pest and disease area characteristic index corresponding to the pest and disease area in the target detection period, and the duration from the first appearance of the pest and disease area to the target detection period. The early warning index for each pest and disease area is determined based on the first pest and disease area characteristic index, the second pest and disease area characteristic index, and the duration.

10. The online detection and control method for forestry pests and diseases according to claim 9, characterized in that, The step of outputting early warning information for each of the pest and disease areas based on the early warning index includes: During the detection period, the early warning information for each of the pest and disease areas is output in descending order of the early warning index. The early warning information includes the regional location of the corresponding pest and disease area, the pest transmission path, and prevention and control recommendations.

Images