CN114387528A

CN114387528A - Pine nematode disease monitoring space-air-ground integrated monitoring method

Info

Publication number: CN114387528A
Application number: CN202111642152.8A
Authority: CN
Inventors: 陈小华; 吴利平; 丁丽霞; 李伟明; 茹磊; 季卓; 周通
Original assignee: Zhejiang Tongchuang Space Technology Co ltd
Current assignee: Zhejiang Tongchuang Space Technology Co ltd
Priority date: 2021-12-29
Filing date: 2021-12-29
Publication date: 2022-04-22

Abstract

The invention discloses a pine wood nematode disease monitoring space-air ground integrated monitoring method, which adopts the technical scheme that: s1, extracting the infected dying trees through the unmanned aerial vehicle image, extracting plaques of the infected dying trees through the satellite image, and carrying out on-site marking, positioning and measuring on the dying trees through ground investigation; s2, respectively obtaining high spatial resolution satellite remote sensing data and unmanned aerial vehicle remote sensing data as data sources, and combining ground survey data and forest resource second class survey data to jointly form a data set of a monitoring area; s3, establishing a sample of the epidemic wood according to the data, wherein the sample comprises the spectral characteristics, the feature texture characteristics and the geometric characteristics of the pine wood nematode disease withered and dead tree; and S4, confirming the infected dead tree plaques by satellite remote sensing according to the samples, comprehensively applying an image enhancement and image classification method, verifying, and establishing the spatial distribution information and the position information of the pine wood nematode dead tree. The unmanned aerial vehicle remote sensing data acquisition system has the advantages of high epidemic area screening efficiency, capability of realizing large-area high-efficiency monitoring, convenience in unmanned aerial vehicle remote sensing data acquisition, high identification precision and guarantee for precision verification.

Description

Pine nematode disease monitoring space-air-ground integrated monitoring method

Technical Field

The invention relates to the technical field of pine wood nematode research, in particular to a pine wood nematode disease monitoring space-air-ground integrated monitoring method.

Background

Pine wood nematodes are the most serious and devastating foreign invasive species of pine species, and can cause the death of pine trees about 40 days the fastest and all deaths within 2-3 months after the pine tree is infected with diseases. Since the early discovery in China in 1982, the method is rapidly diffused, and 3 months in 2021 have spread to 723 county-level administrative districts in 17 provinces (cities and municipalities) in China, wherein the epidemic districts in Zhejiang province reach 70 counties (cities and municipalities), and the prevention and control situation is very severe.

The major reason is that active prevention and control efforts with continuously increased investment and gradually tightened measures cannot completely obstruct the spread of epidemic diseases, and most importantly, timely, accurate and detailed epidemic situation monitoring information is lacked, so that the prevention and control work cannot be dynamically and accurately guided. At present, the general survey information once a year is mainly obtained by means of artificial ground typical survey, the frequency is low, the precision is not high, and more importantly, the spatial position information of epidemic trees is not available. Therefore, a series of problems that a treatment plan deviates from the actual situation, the implementation efficiency is low, the supervision cannot be in place and the like are caused, and the final result is that the prevention and control effect is unsatisfactory under many conditions.

The pine wilt disease dead tree remote sensing monitoring is mainly based on the change of crown color, has typical characteristics in visible light and multispectral remote sensing images, and is based on the satellite remote sensing of the chameleon pine. At present, the monitoring of the epidemic situation of the pine wood nematode disease by the image with medium and low resolution ratio is poor due to the discreteness of the space distribution of infected trees. Aiming at high-resolution images, the pine wood nematode disease monitoring is mainly carried out by utilizing domestic sub-meter GF-2 and BJ-2 satellite data in China, and research and practice show that a pine wood nematode disease area or infected wood is extracted by using methods such as supervision classification, object-oriented CART decision tree classification and the like in combination with topographic data, so that a certain effect is achieved. The remote sensing data can meet the red symptom typical of coniferous trees and the identification of big trees (or trees) with the crown diameter of more than 5m, and the identification of single-plant epidemic trees is difficult; the method is suitable for rapidly and macroscopically monitoring the spatial distribution condition of the epidemic situation, and can provide scientific basis for subsequent detailed investigation and epidemic situation determination.

In addition, the unmanned aerial vehicle remote sensing technology developed in recent years brings a solution for the rapid positioning and total amount estimation of the discolored pine trees in China. However, the existing unmanned aerial vehicle image extraction of the scattered single color-changing pine still stays at the visual interpretation level, and the working scheme of completely depending on manual visual interpretation of the color-changing pine has low efficiency and strong subjectivity. Although the efficiency of manual visual interpretation is improved to a certain extent by the existing classification and extraction algorithms of remote sensing data such as a support vector machine, an object-oriented method and a neural network, the image calculation requirements of ultra-high spatial resolution large-area unmanned aerial vehicles of GB or even TB orders of magnitude are difficult to meet in terms of time complexity.

How to distinguish the discolored pine trees from other red broad-leaved trees, sparse vegetation, bare soil and the like is still a difficult problem facing the analysis based on visible light and multispectral data at present. The appearance of hyperspectrum provides a solution for the situation of homomorphism and heteromorphism or homomorphism of foreign matters on a high-resolution image, but still faces the dilemma of large high-dimensional data volume and strong correlation between wave bands. The hyperspectral remote sensing data has continuous ground object spectrum information, and by utilizing the advantage, airborne hyperspectral remote sensing is adopted to identify the damage degree of the pine wilt disease, and the result is obviously superior to multispectral images. However, as the spatial resolution of airborne hyperspectrum is very high and the data is subject to the difference of local view field conditions, illumination conditions and canopy structure during data acquisition, the difference of canopy brightness values of the same tree is very obvious, and certain interference is brought to the determination of the color-changing pine tree pixels. And the hyperspectral remote sensing image acquisition cost is high, the data processing difficulty is large, the hyperspectral remote sensing image acquisition is limited to a small-area research stage at present, and the hyperspectral remote sensing image acquisition method is not popularized in a large area.

In a word, with diversification of remote sensing platforms, continuous improvement of quality of remote sensing data and continuous progress of classification technology, the theory and technology of monitoring the pine wood nematode disease are rapidly developed. The method is based on pixel classification, an object-oriented classification method, a traditional classification method, a deep learning identification method and a satellite remote sensing data source to a multi-platform multi-spatial resolution data source, and provides a foundation for accurate classification and identification. However, these methods are still in a single technology utilization level in a practical level, the respective advantages of satellite remote sensing and unmanned aerial vehicle remote sensing are not fully exerted, a set of complete production-oriented application technology system is not formed, and the cost and the efficiency are still main problems.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides the integrated monitoring method for monitoring the space and the ground by the pine wilt disease, which has the advantages of high efficiency of screening an affected area, realization of large-area high-efficiency monitoring, convenience in acquisition of remote sensing data of an unmanned aerial vehicle, high identification precision and guarantee for precision verification.

In order to achieve the purpose, the invention is realized by the following technical scheme: the integrated monitoring method for monitoring the space and the ground by the pine wilt disease comprises the following steps:

s1, selecting area positions, firstly extracting the infected withered trees through unmanned aerial vehicle images, extracting plaques of the infected withered trees through satellite images, and carrying out on-site marking, positioning and measuring on the withered trees through ground investigation;

s2, respectively obtaining high spatial resolution satellite remote sensing data and unmanned aerial vehicle remote sensing data as data sources, and combining ground survey data and forest resource second class survey data to jointly form a data set of a monitoring area;

s3, establishing a sample of the epidemic wood according to the data, wherein the sample comprises the spectral characteristics, the feature texture characteristics and the geometric characteristics of the pine wood nematode disease withered and dead tree;

and S4, confirming the infected dead tree plaques by satellite remote sensing according to the samples, comprehensively applying an image enhancement and image classification method, verifying, and establishing the spatial distribution information and the position information of the pine wood nematode dead tree.

Preferably: the high spatial resolution satellite remote sensing data comprises 50cm spatial resolution panchromatic waveband (470 and 830nm) data and 2m spatial resolution multispectral data of blue, green, red and near infrared wavebands.

Preferably: the unmanned aerial vehicle remote sensing data comprises a plurality of areas, wherein one area is selected from each area for 1km²-2km²The area as verify the district, shoot the selected area with unmanned aerial vehicle respectively, unmanned aerial vehicle remote sensing data acquisition user equipment carries on FC6310R aerial photography appearance for the many rotor unmanned aerial vehicle of Xinjiang spirit 4 PRO RTK.

Preferably: the ground survey data comprises dead trees in partial subordination of one or more selected areas in the area shot by the unmanned aerial vehicle, and the dead trees are marked, positioned and measured by a handheld GPS in the field to complete modeling and thematic analysis;

data measured in the field includes: the method comprises the following steps of establishing an image feature library of the epidemic trees through field investigation, wherein the image feature library comprises the characteristics of color, shape, texture, pattern and spatial distribution.

Preferably: the second class survey data of the forest resources comprise space information of administrative area boundary lines and class boundary lines of all regions, and attribute information of administrative area names, class numbers, class areas, ground classes and tree species of all levels.

Preferably: and the unmanned aerial vehicle remote sensing data analysis is used for extracting the number of infected and died trees, selecting pine trees infected with the pine wood nematode disease according to the visual characteristics of the epidemic trees on the image, recording and determining the distribution result of the epidemic trees. Unmanned aerial vehicle image spatial resolution is very high, and the epidemic wood presents yellow on unmanned aerial vehicle's true color image in the early stage of onset, and middle and later stage presents obvious reddish brown or tan, has obvious difference with the green of healthy trees. Pine trees whose disease duration is half a year or longer have fallen off the coniferous leaves, and the branches appear white or off-white on the image. According to the two characteristics of the shape, the pattern and the color of the tree, after man-machine interaction interpretation is carried out on the epidemic trees in the whole research area, a primary result is formed. And (4) further investigating the forest epidemic which is judged and is in question in the field, and checking and verifying the result to finally obtain an accurate forest epidemic distribution result on the unmanned aerial vehicle image in the research area.

Preferably: the method comprises the steps that high spatial resolution satellite remote sensing data analysis is used for extracting disease-infected dead tree plaques, on the basis of satellite remote sensing images, epidemic wood plaques are firstly identified by using an HSV threshold method, then pine forest distribution areas are extracted by using second-class investigation background data of forest resources, hollow areas of forests are removed through a segmentation algorithm, epidemic wood strain digital models are established by using the area of the plaques and the number of the epidemic wood strains, the number of the epidemic wood strains of all the plaques is estimated by using the strain digital models, and position coordinates of each strain are extracted; the segmentation algorithm is to gradually merge the single image element into a larger object from the top until the set segmentation scale (f) is met;

the partition scale (f) consists of four parameters, respectively spectral heterogeneity (h)_color) Shape heterogeneity (h)_shape) Weight of spectral information (w)_color) And shape information weight (w)_shape) The sum of the weights of the spectral feature and the shape feature is 1 (i.e., w)_color+w_shape＝1)，f＝w×h_color+(1-w)×h_shape；

Spectral heterogeneity (h)_color) Not only the number of pixels constituting the object, but also the standard deviation of each band: is the standard deviation, root, of the values of the pixels within the objectCalculating according to pixel values of the composition objects, wherein n is the number of pixels;

in addition, shape heterogeneity (h)_shape) By compactness (h)_compact) And smoothness (h)_smooth) The smoothness is used for optimizing the smoothness of the boundary of the segmentation object, so that the edge can be prevented from being broken; the compactness is used for optimizing the compactness of the segmentation object, and the sum of the weights of the two indexes is also 1 (namely w)_compact+w_smooth＝1)；

h_shape＝w_compact×h_compact+(1-w_compact)×h_smooth；

On the basis of image preprocessing, segmenting a P satellite image by means of eCG development Developer software, carrying out multi-scale segmentation on the image, wherein the segmentation range is 20-150, carrying out quantitative evaluation on a multi-scale segmentation result, finding an optimal segmentation scale, and selecting the optimal segmentation scale through visual evaluation;

after the optimal segmentation scale is quantitatively evaluated, feature variables, including spectra, textures and geometric feature variables, of each object in the object layer under the optimal segmentation scale, which are contained in the P satellite image, are derived through an eCoginization Developer, and various indexes are calculated and obtained through the original wave band of the image;

RGB in the high spatial resolution satellite remote sensing data is represented by colors of red, green and blue channels, HSV conversion image color enhancement is carried out on multispectral data of 3 wave bands of red, green and blue, an HSV color model is converted, epidemic trees and healthy trees are obviously distinguished on values of H wave bands, and the threshold value of a distinguisher is found, so that patches of the epidemic trees can be automatically identified;

the RGB conversion HSV formula is as follows:

V＝max(R，G，B)

If H＜0 then H＝H+360.On output 0≤V≤1，0≤S≤1，0≤H≤360.

after the plague wood plaques are identified, cutting the pine forest range by combining with second-class survey data of forest resources, removing non-pine forest regions and obtaining plague wood plaques of the pine forest regions;

by a local geometric correction method, satellite data is used as reference during correction, geometric correction is carried out on the result extracted by the unmanned aerial vehicle, so that the data extraction result of the unmanned aerial vehicle is used as reference data to be matched and superposed with the satellite data for analysis, and based on the images of the unmanned aerial vehicle, the satellite images and respective interpretation and identification results, the positions of the epidemic trees are accurately judged by comprehensively using methods of contrastive analysis and logical inference through the sizes, colors, spatial distribution characteristics, patterns and position relations with surrounding ground objects of the two epidemic trees at the same positions, so as to obtain a geometric fine correction result, and the positions of the epidemic trees interpreted by the inorganic images correspond to the satellite image identification results one by one;

obtaining new satellite epidemic wood point vector data in a man-machine interaction mode through unmanned aerial vehicle image interpretation results and satellite pattern spot data, wherein the data is used for modeling of epidemic wood identification strain number;

identifying epidemic wood plaques with different sizes based on satellite images, calculating the quantity of the epidemic wood, and obtaining the quantity of the epidemic wood through a relation model of the area and the strain quantity, wherein the model form is as follows:

y＝ax+b

or

y＝ax²+bx+c

Wherein y is the number of epidemic wood strains, x is the area of the plaque, and the number of the epidemic wood strains of all plaques is solved by the model formula.

Preferably: extracting partial areas in a research area, carrying out unmanned aerial vehicle shooting in global or local sampling areas within one week from the satellite image acquisition day, and acquiring a withered tree position through manual interpretation of the unmanned aerial vehicle image so as to verify the accuracy of the satellite image identification result;

unmanned aerial vehicle images randomly distributed in a plurality of verification areas obtain the number of plants in each area and the position of each withered tree in each area through manual interpretation results, and the number of plants and the position of each withered tree in each area are used for verification evaluation of position accuracy and plant number accuracy respectively.

Preferably: verification of plant number precision based on unmanned aerial vehicle image, plant number precision p_nExpressed as a function of the strain error rate:

p_n＝1-|E_n|

wherein E is_nAs the error rate of the number of plants, n as the number of identified plants, m as the number of verified plants, the following table i as the area number, n_iM is the number of identified strains in the i-th region_iThe number of verification strains in the i-th region, and b the number of regions. E_nCan have positive and negative, when E_nIf the value is greater than 0, the identification value is greater than the verification value; when E is_nIf < 0, the identification value is less than the verification value; obtaining verification information of the number of withered and dead trees after manual ground investigation and treatment removal work is completed, and verifying the accuracy of the recognition result;

the accuracy error of the single plant position adopts the Euclidean distance between the position (xd, yd) of the identification point and the position (xt, yt) of the verification point

Indicating that for the validation region, the overall position error (Ep) is expressed as the arithmetic mean of the individual position errors:

wherein the content of the first and second substances,

x-coordinate value representing the i-th strain of identification data,

the x-coordinate value of the i-th strain of the verification data,

a y coordinate value indicating the i-th strain of identification data,

y coordinate value representing the i-th plant of the verification data; by mean value E_pFor primary reference, both maximum and minimum values are considered.

Preferably: and separating the vacant areas (including forest edges) in the forest of the pine forest class from the forest by adopting an image classification method so as to pick the areas which are easy to be confused with the ill and dead trees from the satellite image recognition result.

The invention has the following beneficial effects:

the method has the advantages of high efficiency of screening the epidemic area, large-area and high-efficiency monitoring, and the identification precision is kept above 80% (the error is below 20%) by using a remote sensing technology. Secondly, the withered and dead trees are identified based on satellite remote sensing, the position coordinates of each tree are extracted, the information is obtained quickly and thoroughly, and efficiency and cost are both considered.

Drawings

FIG. 1 is a schematic diagram of a precision verification sampling region in accordance with the present invention;

FIG. 2 is a flow chart of a multi-scale segmentation algorithm of the present invention;

FIG. 3 is a schematic diagram of a patch identification result on a satellite remote sensing image according to the present invention;

FIG. 4 is a distribution diagram of hollow areas of pine forest and forest in the North China brocade street according to the present invention;

FIG. 5 is a distribution diagram of epidemic wood plaques extracted by satellite remote sensing in North-China brocade streets in the invention;

FIG. 6 is a diagram of the local geometric fine correction of the log position by the image interpretation of the log by the unmanned aerial vehicle in the invention;

FIG. 7 is a patch diagram of the unmanned aerial vehicle of the present invention;

FIG. 8 shows the recognition and interpretation results of the unmanned aerial vehicle shadow in the present invention;

FIG. 9 is a distribution diagram of a local class verification log in the invention.

Detailed Description

The first embodiment,

The integrated monitoring method for monitoring the space and the ground for the pine wilt disease in figures 1-9 comprises the following steps:

the high spatial resolution satellite remote sensing data comprises 50cm spatial resolution panchromatic waveband (470 and 830nm) data and 2m spatial resolution multispectral data of blue, green, red and near infrared wavebands.

The relevant technical parameters of the satellite remote sensing data with high spatial resolution are shown in the following table;

the unmanned aerial vehicle remote sensing data comprises a plurality of areas, wherein one area is selected from each area for 1km²-2km²The area of (1) is used as a verification area, referring to the attached figure 1, the Shandong village, the Jinma village and the Villa residence committee with more pine forest distribution and serious epidemic situation are selected as verification villages in the north-brocade street of the temporary security area in the embodiment, and each village is selected to be 1km²-2km²The area about is as verifying the district, shoots the regional unmanned aerial vehicle that uses of selecting respectively, and unmanned aerial vehicle remote sensing data acquisition user equipment carries on FC6310R aerial photography appearance for the many rotor unmanned aerial vehicle of Xinjiang spirit 4 PRO RTK.

The unmanned aerial vehicle image technical parameters and the area in the verification area are shown in the following table:

and shooting a visible light image by adopting an unmanned aerial vehicle in the area, wherein the spatial resolution of the orthoscopic image is 3cm-7 cm.

The ground survey data comprises dead trees in partial subordination of one or more selected areas in the area shot by the unmanned aerial vehicle, and the dead trees are marked, positioned and measured by a handheld GPS in the field to complete modeling and thematic analysis;

The second class survey data of the forest resources comprise space information of administrative area boundary lines and class boundary lines of all regions, and attribute information of administrative area names, class numbers, class areas, ground classes and tree species of all levels.

And the unmanned aerial vehicle remote sensing data analysis is used for extracting the number of infected and died trees, selecting pine trees infected with the pine wood nematode disease according to the visual characteristics of the epidemic trees on the image, recording and determining the distribution result of the epidemic trees. Unmanned aerial vehicle image spatial resolution is very high, and the epidemic wood presents yellow on unmanned aerial vehicle's true color image in the early stage of onset, and middle and later stage presents obvious reddish brown or tan, has obvious difference with the green of healthy trees. Pine trees whose disease duration is half a year or longer have fallen off the coniferous leaves, and the branches appear white or off-white on the image. According to the two characteristics of the shape, the pattern and the color of the tree, after man-machine interaction interpretation is carried out on the epidemic trees in the whole research area, a primary result is formed. And (4) further investigating the forest epidemic which is judged and is in question in the field, and checking and verifying the result to finally obtain an accurate forest epidemic distribution result on the unmanned aerial vehicle image in the research area.

In the embodiment, the remote sensing data source of the unmanned aerial vehicle selects the North brocade street in the Ministry district of Hangzhou city, the administrative district area is 81.54Km2(122310 mu), and the pine tree species mainly including Pinus massoniana are the main tree species of the North brocade street. 689 piny shifts in one or more in the region, the area is 2343.3hm2(35150 mu), and the piny area accounts for 28.28% of the administrative area.

Referring to the attached figure 1, in order to effectively verify the identification result of the satellite remote sensing image machine, 1 verification sampling area (verification area for short) is respectively arranged in the upper eastern village, the golden village, the dragon village and the villa residence committee, unmanned aerial vehicle image shooting is carried out on the area within one week after the satellite image shooting, and the verification is taken as basic data based on the satellite image identification precision.

The method comprises the steps that high spatial resolution satellite remote sensing data analysis is used for extracting disease-infected dead tree plaques, on the basis of satellite remote sensing images, epidemic wood plaques are firstly identified by using an HSV threshold method, then pine forest distribution areas are extracted by using second-class investigation background data of forest resources, hollow areas of forests are removed through a segmentation algorithm, epidemic wood strain digital models are established by using the area of the plaques and the number of the epidemic wood strains, the number of the epidemic wood strains of all the plaques is estimated by using the strain digital models, and position coordinates of each strain are extracted; the segmentation algorithm is to gradually merge the single image element into a larger object from the top until the set segmentation scale (f) is met;

the partition scale (f) consists of four parameters, respectively spectral heterogeneity (h)_color) Shape heterogeneity (h)_shape) Weight of spectral information (w)_color) And shape information weight (w)_shape) The sum of the weights of the spectral feature and the shape feature is 1 (i.e., w)_color+w_shape＝1)，f＝w×h_color+(1-w)×h_shape(ii) a Spectral heterogeneity (h)_color) Not only the number of pixels constituting the object, but also the standard deviation of each band: the standard deviation is the standard deviation of the pixel values in the object, and is obtained by calculation according to the pixel values of the object, wherein n is the number of pixels;

h_shape＝w_compact×h_compact+(1-w_compact)×h_smooth；

On the basis of image preprocessing, segmenting a P satellite image by means of eCG development Developer software, performing multi-scale segmentation on the image, wherein the segmentation range is 20-150, and performing multi-scale segmentation on the P satellite image; it can be seen from the above algorithm that in the same region, as the segmentation scale increases, the number of segmented objects is less and less, and the number of objects directly affects the operation speed and the classification accuracy. When the segmentation scale is too low, the number of objects is greatly increased, and the operation speed is greatly reduced. On the contrary, when the segmentation scale is too high, the number of objects is reduced, which easily causes the segmentation of different features into one object, thereby reducing the classification accuracy. Therefore, it is important to quantitatively evaluate the multi-scale segmentation result and find the optimal segmentation scale. By visual evaluation, the optimum segmentation scale was selected to be 100.

at present, the texture extraction method mainly comprises four methods based on statistical description, wavelet transformation, application fractal theory and geostatistics. Among them, the statistical description based gray level co-occurrence matrix (GLCM) has been shown to play an important role in vegetation classification, especially the homogenization of the gray level co-occurrence matrix (GLCM _ HOMO). Therefore, the present embodiment selects the gray level co-occurrence matrix algorithm to extract the texture information of the object.

The spectral characteristics are the most main characteristics of remote sensing image classification, and the spectral characteristic spectra of healthy trees and epidemic trees have large difference in blue, green, red and near infrared bands, so that the spectral characteristic variables of the object have the average value and standard deviation of each band and the normalized vegetation index in the embodiment.

The geometric attributes mainly describe the shape and size of the objects, and the overall effect is lower than that of spectral and texture characteristics when the geometric attributes are used for analyzing wetland characteristics of a smaller scale or a region with less human activity intervention, which can be interpreted as small co-dependency among the objects to a certain extent.

Specific spectral, texture, geometric features, etc. variables can be found in the following table:

for RGB in the satellite remote sensing data with high spatial resolution, wherein RGB is the color representing three channels of red, green and blue,

the standard of obtaining various colors by changing three color channels of red (R), green (G) and blue (B) and superposing the three color channels with each other almost comprises all colors which can be perceived by human vision, and is one of the most widely used color systems at present.

HSV: HSV (Hue, Saturation) is a color space created by a.r. smith in 1978 based on the intuitive nature of color, and is also known as the hexagonal pyramid Model (Hexcone Model) HSV color Model that can be transformed from the RGB Model. The parameters of the colors in this model are: hue (H), saturation (S), lightness (V). The value range of H is 0 to 360 degrees, wherein 0 degree represents red, and higher degree represents closer to green. S is the saturation represented by a percentage, wherein the larger the percentage is, the more saturated the color is, i.e., the higher the saturation is when the S value is large, the darker the color of the image is, and the lower the saturation is when the S value is small, and the lighter the color of the image is. V is also expressed as a percentage, 0% representing black and 100% white.

Referring to the attached figure 3, the color characteristics of the epidemic wood on the satellite remote sensing image with the resolution of 0.5 m are similar to those of the unmanned aerial vehicle, only the colors are dark, HSV conversion image color enhancement is carried out on multispectral data of 3 wave bands of red, green and blue, an HSV color model is converted, the epidemic wood and the healthy tree are obviously distinguished on the value of an H wave band, and the threshold value of a distinguisher is found, so that the plaque of the epidemic wood can be automatically identified.

The RGB conversion HSV formula is as follows:

V＝max(R，G，B)

If H＜0 then H＝H+360.On output 0≤V≤1，0≤S≤1，0≤H≤360.

when the satellite image extracts the log patches and the unmanned aerial vehicle image interpretation results are subjected to superposition analysis, the spatial reference basis of the log positions of the two is required to be consistent, and neither of the two can have any deformation, in the embodiment, the unmanned aerial vehicle image log spot vector result of the same slope with consistent deviation direction is matched with satellite data by a local geometric correction method, the satellite data is used as reference during correction, the unmanned aerial vehicle extracted result is subjected to geometric correction by taking the satellite data as reference data so as to be matched and subjected to superposition analysis with the satellite data, and based on the unmanned aerial vehicle image, the satellite image and respective interpretation identification results, the log positions are accurately judged by comprehensively using the methods of comparative analysis and logic inference through the size, the color, the spatial distribution characteristics and the patterns of the log at the same position and the position relation with surrounding ground objects, obtaining a geometric fine correction result, and realizing one-to-one correspondence between the position of the epidemic wood read by the inorganic image and the satellite image identification result;

the unmanned aerial vehicle image interpretation log data and the satellite image identification log data are superposed, so that after geometric correction, most of the unmanned aerial vehicle image interpretation log data fall into the log plaques identified by the satellite image, but few parts of the log plaques do not have corresponding points on the unmanned aerial vehicle data. After the unmanned aerial vehicle image data and the satellite image data are observed and compared, the part of the pattern spot is found to contain deciduous epidemic trees, red-leaf healthy broad-leaved trees, wastelands in forests and under-forest epidemic trees with some crowns covered to a greater extent. Through the description of the attribute addition of the epidemic wood plaque extracted from the satellite image, missing under-forest epidemic trees are supplemented during unmanned aerial vehicle image interpretation, and new unmanned aerial vehicle epidemic wood interpretation point vector data are generated at the moment and used for checking.

Most of the epidemic wood plaques extracted by the satellite images are one in one, but many plaques have 2 strains or more, and even some plague woods in a connected mode have 7-8 strains. Part of fallen leaves are mixed in the connected 2 or more strains of epidemic trees. In order to check and analyze the accuracy of unmanned aerial vehicle image recognition, the patterns of the two conditions are combined with the unmanned aerial vehicle image interpretation result, the condition of the epidemic trees corresponding to the patterns is subjected to supplementary interpretation, and the attribute is recorded. Finally, the obtained result patches of satellite data automatic identification all record attributes, wherein the attributes comprise 5 conditions, namely epidemic trees, forest epidemic trees, fallen leaf epidemic trees, broad leaf trees and open space barren grasses. The patch data is used for evaluating the position accuracy evaluation of the epidemic wood recognition. Meanwhile, a new satellite epidemic wood point vector data is obtained by referring to the unmanned aerial vehicle image interpretation result and the satellite pattern spot data in a man-machine interaction mode. This data was used for modeling the number of vaccine identification strains.

y＝ax+b

or

y＝ax²+bx+c

In the embodiment, the model is established in units of villages, the modeling samples (plaques) are extracted randomly in units of villages as a whole, the number of the sampled units of each whole is not less than 2% of the total amount (the number of plaques), and the number of the lowest sample units is not less than 200.

For all sample cells (plaques in the drawing), the number of epidemic trees per sample cell was visually interpreted based on the drone image. And establishing a linear or polynomial relation model between the number of the plaque interpretation strains and the area of the plaque interpretation strains by taking the interpretation result as a true value. When the decision coefficient r2 of the model is larger than 0.60, the model is effective, and the number of epidemic wood strains of all plaques is solved by the model.

Extracting partial areas in a research area, carrying out unmanned aerial vehicle shooting in global or local sampling areas within one week from the satellite image acquisition day, and acquiring a withered tree position through manual interpretation of the unmanned aerial vehicle image so as to verify the accuracy of the satellite image identification result;

unmanned aerial vehicle images randomly distributed in a plurality of verification areas, in the embodiment, unmanned aerial vehicle images randomly distributed in a plurality of verification areas (A, B, C, D) of the north brocade street are manually interpreted to obtain the number of plants in each area and the position of each withered tree in each area, and the number of plants and the position of each withered tree in each area are respectively used for verification evaluation of position precision and plant number precision.

Verification of plant number precision based on unmanned aerial vehicle image, plant number precision p_nExpressed as a function of the strain error rate:

p_n＝1-|E_n|

wherein E is_nAs the error rate of the number of strains, n as the number of identified strains, m as the number of verified strains, table i belowIs a region number, n_iM is the number of identified strains in the i-th region_iThe number of verification strains in the i-th region, and b the number of regions. E_nCan have positive and negative, when E_nIf the value is greater than 0, the identification value is greater than the verification value; when E is_nIf < 0, the identification value is less than the verification value; obtaining verification information of the number of withered and dead trees after manual ground investigation and treatment removal work is completed, and verifying the accuracy of the recognition result;

wherein the content of the first and second substances,

x-coordinate value representing the i-th strain of identification data,

the x-coordinate value of the i-th strain of the verification data,

a y coordinate value indicating the i-th strain of identification data,

And separating the vacant areas (including forest edges) in the forest of the pine forest class from the forest by adopting an image classification method so as to pick the areas which are easy to be confused with the ill and dead trees from the satellite image recognition result.

With reference to FIG. 4, a log is produced by the above methodThe classification results showed 2146 pieces of forest land, total area 185.04hm2(2776 mu), average area per piece 862m²。

Referring to fig. 5, in this embodiment, an epidemic wood plaque distribution map is extracted based on a satellite remote sensing image, and plaque distribution in spatial distribution of an epidemic wood obtained after H threshold processing is raster data, so that extraction of epidemic wood coordinate position information and plant tree information is facilitated, and results are vectorized. After the forest hollow land is removed, vector data are checked against unmanned aerial vehicle images, and then the patch with the area smaller than 1 square meter is basically removed without epidemic trees. The area and center point of each plaque are calculated. The number of the epidemic wood spots on the streets in North-jin is 11310, and the total area of the epidemic wood spots is 180860m2(217.3 mu).

Referring to fig. 5-6, in this embodiment, because the unmanned aerial vehicle image and the satellite image have different correction accuracies and the unmanned aerial vehicle image has a large change in terrain height difference, the identification result (dead tree patch) based on the satellite image and the interpretation result (dead tree single plant) based on the unmanned aerial vehicle image must be accurately registered before modeling the number of epidemic trees. Selecting the north 56hm of Shandongcun²And (5) carrying out partitioned fine correction on the areas with consistent small terrains by the pine forest by adopting a spline function model. By selecting 10 small shifts in the north of Shandong village and adopting a local geometric correction method, the positions of withered trees interpreted based on the images of the unmanned aerial vehicles are geometrically and precisely corrected by referring to the images of the unmanned aerial vehicles and the satellite images. And finally obtaining 1385 patches completely corresponding to satellite image identification patches according to the unmanned aerial vehicle image interpretation result, wherein the number of the patches is 935.

The relationship between the epidemic wood plaque and the corresponding strain number is shown in the following table;

ID	small class number	Plaque area (m)²)	Tree planting
				1	5	14.0	2
2	5	3.3	1
				3	5	12.0	1
4	1	3.1	1
				5	5	8.4	1
6	5	60.8	3
				7	5	16.8	1
8	1	4.3	1
				......	......	......	......
934	47	45.0	2
				935	47	20.0	1

And (3) judging and reading the result (strain) by using the unmanned aerial vehicle image after geometric fine correction, determining the number of strains of the epidemic trees in each patch by contrasting the satellite image identification result (patch), and finally obtaining a patch information determination table with the patch number, the patch area and the number of strains of the epidemic trees as main contents. Based on the table, a relational model of the area and the number of plants is established.

The lower graph shows a linear model of Shandong village, which shows the relationship between the area of the plague wood plaque and the number of the strains, and y is 0.0206x +0.785 (y-number of strains, x-area), and the coefficient is determined at 0.6422. Based on the model, the number of epidemic wood strains of each plaque is estimated by using the plaque data automatically identified and extracted by the satellite.

And estimating the number of the epidemic wood strains in the North-jin street by using the extracted epidemic wood plaque data according to a relation model of the area of the epidemic wood plaques and the number of the strains. According to statistics of data of second-class survey shifts of forest resources, the area of a Jinbei street administrative region is 8019.3hm2(120289 mu), 689 pine shift groups exist, and the total area of the pine shift groups is 35150 mu.

According to the patch identification result, through epidemic area extraction, interference analysis and epidemic tree strain number modeling estimation, 12814 strains of pine wood nematode disease dead trees are found in the north brocade street, and the average of 0.11 strains per mu is calculated according to the total area of the administrative area; the average of 0.37 plants per mu is calculated according to the actual area of the pine forest (epidemic disease), and the specific table is shown in the following table.

In this embodiment, the precision evaluation is performed in a typical sampling manner, and the sample characteristic value is an effective estimation value of the total characteristic value represented by the sample characteristic value. The total number accuracy (accuracy) and the position accuracy of the chinpei street are represented by the mean values of the number accuracy (accuracy) and the position accuracy of 4 sample areas drawn by the streets of the upper eastern village, the golden village and the western villa.

The evaluation of the plant number accuracy was carried out by using the degree of difference between the plant number of the machine recognition result and the plant number of the visual interpretation result of the evaluator. According to the following formula:

p_n＝1-|E_n|

and respectively calculating the error of the plant number and the precision of the plant number.

In the accuracy evaluation of the number of plants, identification and interpretation data are acquired by adopting a typical sampling mode. Considering the representativeness and stability of the samples, A, B, C, D total pine-containing sample units (small shifts) were drawn for all four validation areas, 217.7 hectares (3266 acres) of the actual pine forest small shift area, and 46 small shifts were counted. The conclusion is that the identification precision of the number of dead plants in the north brocade street is as follows: 80.2%, see in particular the table below.

The positional accuracy evaluation was evaluated by the degree to which the machine recognition position was deviated from the position visually recognized by the evaluator. In this embodiment, in a typical (serious) area, a minor class with a medium area size, namely the 15 th minor class of the dong-village stone bridge, is extracted as a sample for testing the position accuracy of the satellite image machine recognition result, and the global position error is analyzed by calculating the mean value of the position errors according to the formula 3.8.

Referring to the attached figure 8, the inspection is performed by adopting a visual interpretation method, an inspector performs withered tree interpretation on a 15 th class area on a satellite image, corrects the withered tree area by referring to a synchronous unmanned aerial vehicle image, and finally performs comparative analysis on the corrected result and the 15 th class machine recognition result to be inspected, wherein the analysis is performed by adopting an ArcMap neighborhood analysis tool.

The 15 th minor group common satellite image identifies 105 withered trees, and the result of artificial comprehensive interpretation is 107. And determining pairs of withered tree points corresponding to the machine identification and manual interpretation results pairwise according to the minimum distance, and counting the characteristic number according to the distance of each point pair so as to reflect the position error. The results were: the average (position error) was 3.16 meters, the minimum was 0.79 meters, and the maximum was 8.05 meters, as detailed in the table below.

And (3) further analyzing interference factors influencing the identification precision of the epidemic trees by using the monitoring result of the 10 minor shifts near the Shijiu of Shandong village, and further clearly identifying the actual type composition of the withered trees (the epidemic trees) in the identification result. The total area of the region was 56.5hm2(848 mu), and the wood strain 1466 was identified. The result of the unmanned aerial vehicle is used as auxiliary data, satellite monitoring results are further interpreted and subdivided in types, and the fact that interference factors such as various withered and dead trees (epidemic trees) 1385 strains, broad-leaved trees, forest species open space and the like cause misjudgment of 81 strains is confirmed;

the overall precision of the satellite image machine recognition result is 86.5% based on the visual interpretation result of the inspector. The recognition accuracy of 8 of 10 shifts is more than 94.0 percent, and the recognition accuracy of 1 shift is less than 50 percent. The proportion of the satellite image interpretation recognition result types in each small shift is counted and referred to the following table.

The reason why the number of 47-class minor confirmed plants is much less than the number of identified plants is that the minor broad-leaved trees account for the absolute majority and the deciduous trees are more in variety through field comparison, and the satellite images shot in 23 days in 10 months have part of broad-leaved yellows, so that the automatic identification is carried out by taking the part of the broad-leaved yellowed broad-leaved trees as dead trees for statistics, which also indicates that other discolored trees, particularly the discolored broad-leaved trees, are the main interference sources for remote sensing identification of the pine wilt disease.

7 subordinates are selected in Shandong village, the satellite remote sensing monitoring result is verified through unmanned aerial vehicle image man-machine interaction interpretation, the result is that the monitoring precision of the number of the infected and withered trees is larger than 74.4%, the monitoring precision of the number of the plants of the subordinates with more than 7 grades is larger than 67%, and the verification result of the local satellite monitoring subordinates in Shandong village is shown in the following table.

15 subordinates are selected from the Jinma village to conduct unmanned aerial vehicle image man-machine interaction interpretation, the result is used for verifying satellite remote sensing monitoring results, the overall monitoring precision of the number of trees suffering from blight is 81.9%, the precision difference of each subordinate is large, the precision difference is from 25.0% to 96.9%, but the number of the subordinates with the precision exceeding 60.0% still accounts for most. The area of the small shift with larger error is mostly smaller. The verification results of the local monitoring shifts of the Kinmura satellite are shown in the following table.

The technology solves the bottleneck problem of low efficiency of the pine wilt disease monitoring faced by the current technology of utilizing remote sensing and geographic information systems, and the identification precision is kept above 80% (the error is below 20%) by utilizing the remote sensing technology. Secondly, the withered and dead trees are identified based on satellite remote sensing, the position coordinates of each tree are extracted, the information is obtained quickly and thoroughly, and efficiency and cost are both considered. And thirdly, the technical system fully utilizes tool software in the aspects of the existing geographic information system, remote sensing and the like, and ensures the process flow of the working procedure and the consistency of the output result.

Example two

In the second embodiment, in the pine forest class, the forest coverage condition is checked from the satellite remote sensing image, and it can be found that some areas in the class area are not covered by the forest, such as the hollow land, the wasteland and the bare land of the forest margin, the spectral characteristics of the land features are very similar to those of the epidemic trees, and in order to reduce the interference of the land features on the extraction of the epidemic trees, the second embodiment further eliminates the non-forest coverage areas. Classifying images of the pine forest class area by adopting an object-oriented random forest classification method, and distinguishing a forest coverage area from types of wastelands, forest margin bare lands and the like, so as to eliminate non-forest areas in the pine forest class and eliminate interference factors for accurate extraction of epidemic trees; the random forest model is a novel machine learning algorithm based on decision trees, M new training sets are extracted from an original data set, 2/3 of which the number is about that of the original data set are extracted, and M attributes are extracted randomly in the new training sets to produce the decision trees. Finally, the prediction results of the N decision trees are aggregated, the category of a new sample is determined by adopting a voting mode, and the internal error can be estimated by using 1/3 data which are not extracted in each sampling. In many machine learning algorithms, random forests have 3 features and advantages: the classification performance is excellent, and big data can be processed under the condition of not performing feature selection and deletion; secondly, manual intervention is little, data preprocessing is usually not needed, and used characteristics can be determined according to data; thirdly, the operation speed is fast, and the parallelization processing is easy to be carried out.

EXAMPLE III

The experimental results of the method are selected from Jinhua Dong district and Jinhua Yongkang district in Jinhua City for general investigation of the pine wilt disease in 2020.

Firstly, remote sensing images are obtained in an aerial photography mode in the Jinhua east area of 658.19Km2 and the Yongkang area of the Jinhua Yongkang area of 1049Km 2. The image capturing time is 10 months and 31 days in 2020. The spatial resolution of the aerial image is 0.5 m, and the aerial image comprises four wave bands of red, green, blue and near infrared. The census results are shown in the following table.

And secondly, verifying the gold east region identification result.

And (3) extracting four towns of Jiangdong, Ling, Yuandong and Pinus densiflora as samples in the Jindong area, verifying general survey data based on the air-space-ground integration method by using the number of actual harvested trees in the treatment process of the area, and taking the verification result as the verification result of the Jindong area universe. The verification refers to the standing book record of the disease wood removal treatment result in the area from 1 month to 4 months in 2021, and the specific verification result is shown in the following table.

Generally, the error rate of the number of strains is 4.1%, which is expressed by that the number of identified strains is slightly more than that of the strains of the cleaning result, but the subentry difference is large and reaches-29.5% at most, and the number of identified strains is greatly lower than that of the actually cleaned strains, and the main reason is that the mountain in the remote mountain area of the rural area is high in forest density, and the crown of the epidemic trees is covered more. Also, it is stated from another aspect that the country increases the treatment power in remote mountainous areas. For the whole golden east region, the mean value and the extreme value of the error are within the range.

And thirdly, verifying the identification result of the Yongkang city.

The Yongkang city extracts two county level units of urban streets and economic development areas as samples, and verifies the identification result by adopting two methods.

The first method is based on actual cleaning of the number of the infected trees, and the error rate of the number of the infected trees in the same area is calculated by taking the standing account record of the disease tree removal result in the area from 1 month to 5 months in 2021 as the standard of the number of the infected trees. The specific verification results are shown in the following table.

In the table, the census data and the cleaning data of the eastern city streets and the economic development area are in certain difference, and the difference of the economic development area is large, but generally the census data is fluctuated up and down by taking the census data as a center, and the error rate is reduced along with the overall expansion.

Example four

The method comprises the steps of extracting two county level units of urban streets and an economic development area in Yongkang city as samples, verifying by adopting a second method, selecting mountain villages, lower villages and lower great road villages of the urban streets and the Ku pond based on unmanned aerial vehicle image local verification, judging withered trees by plants manually by adopting high-spatial-resolution unmanned aerial vehicle images, and judging the error rate of the number of plants based on aerial image recognition results, wherein specific verification results are shown in the following table.

The above-mentioned embodiments are only used for explaining the inventive concept of the present invention, and do not limit the protection of the claims of the present invention, and any insubstantial modifications of the present invention using this concept shall fall within the protection scope of the present invention.

Claims

1. The pine wood nematode disease monitoring space-air ground integrated monitoring method is characterized by comprising the following steps:

s1, extracting the infected dying trees through the unmanned aerial vehicle image, extracting plaques of the infected dying trees through the satellite image, and carrying out on-site marking, positioning and measuring on the dying trees through ground investigation;

2. The integrated monitoring method for monitoring the space and the ground by the pine wilt disease according to claim 1, characterized in that: the high spatial resolution satellite remote sensing data comprises 50cm spatial resolution panchromatic waveband (470 and 830nm) data and 2m spatial resolution multispectral data of blue, green, red and near infrared wavebands.

3. The integrated monitoring method for monitoring the space and the ground by the pine wilt disease according to claim 1, characterized in that: the unmanned aerial vehicle remote sensing data comprises a plurality of areas, wherein one area is selected from each area for 1km²-2km²The area as verify the district, shoot the selected area with unmanned aerial vehicle respectively, unmanned aerial vehicle remote sensing data acquisition user equipment carries on FC6310R aerial photography appearance for the many rotor unmanned aerial vehicle of Xinjiang spirit 4 PRO RTK.

4. The integrated monitoring method for monitoring the space and the ground by the pine wilt disease according to claim 1, characterized in that: the ground survey data comprises dead trees in partial subordination of one or more selected areas in the area shot by the unmanned aerial vehicle, and the dead trees are marked, positioned and measured by a handheld GPS in the field to complete modeling and thematic analysis;

5. The integrated monitoring method for monitoring the space and the ground by the pine wilt disease according to claim 1, characterized in that: the second class survey data of the forest resources comprise space information of administrative area boundary lines and class boundary lines of all regions, and attribute information of administrative area names, class numbers, class areas, ground classes and tree species of all levels.

6. The integrated monitoring method for monitoring the space and the ground by the pine wilt disease according to claim 1, characterized in that: and the unmanned aerial vehicle remote sensing data analysis is used for extracting the number of infected and died trees, selecting pine trees infected with the pine wood nematode disease according to the visual characteristics of the epidemic trees on the image, recording and determining the distribution result of the epidemic trees.

7. The integrated monitoring method for monitoring the space and the ground by the pine wilt disease according to claim 1, characterized in that: the method comprises the steps that high spatial resolution satellite remote sensing data analysis is used for extracting disease-infected dead tree plaques, on the basis of satellite remote sensing images, epidemic wood plaques are firstly identified by using an HSV threshold method, then pine forest distribution areas are extracted by using second-class investigation background data of forest resources, hollow areas of forests are removed through a segmentation algorithm, epidemic wood strain digital models are established by using the area of the plaques and the number of the epidemic wood strains, the number of the epidemic wood strains of all the plaques is estimated by using the strain digital models, and position coordinates of each strain are extracted; the segmentation algorithm is to gradually merge the single image element into a larger object from the top until the set segmentation scale (f) is met;

the partition scale (f) consists of four parameters, respectively spectral heterogeneity (h)_color) Shape heterogeneity (h)_shape) Weight of spectral information (w)_color) Andshape information weight (w)_shape) The sum of the weights of the spectral feature and the shape feature is 1 (i.e., w)_color+w_shape＝1)，f＝w×h_color+(1-w)×h_shape；

Spectral heterogeneity (h)_color) Not only the number of pixels constituting the object, but also the standard deviation of each band: the standard deviation is the standard deviation of the pixel values in the object, and is obtained by calculation according to the pixel values of the object, wherein n is the number of pixels;

h_shape＝w_compact×h_compact+(1-w_compact)×h_smooth；

the RGB conversion HSV formula is as follows:

V＝max(R，G，B)

If H＜0 then H＝H+360.On output 0≤V≤1，0≤S≤1，0≤H≤360.

y＝ax+b

or

y＝ax²+bx+c

8. The integrated monitoring method of pine wilt disease monitoring sky of claim 7, characterized in that: extracting partial areas, carrying out unmanned aerial vehicle shooting in global or local sampling areas within one week from the satellite image acquisition day, and acquiring withered tree positions through manual interpretation of unmanned aerial vehicle images so as to verify the accuracy of satellite image identification results;

9. The integrated monitoring method of pine wilt disease monitoring sky of claim 7, characterized in that: verification of plant number precision based on unmanned aerial vehicle image, plant number precision p_nExpressed as a function of the strain error rate:

p_n＝1-|E_n|

wherein the content of the first and second substances,

x-coordinate value representing the i-th strain of identification data,

the x-coordinate value of the i-th strain of the verification data,

a y coordinate value indicating the i-th strain of identification data,

10. The integrated monitoring method for monitoring the space and the ground by the pine wilt disease according to claim 1, characterized in that: and separating the vacant areas (including forest edges) in the forest of the pine forest class from the forest by adopting an image classification method so as to pick the areas which are easy to be confused with the ill and dead trees from the satellite image recognition result.