CN114863296A

CN114863296A - Method and system for identifying and positioning wood damaged by pine wilt disease

Info

Publication number: CN114863296A
Application number: CN202210365057.6A
Authority: CN
Inventors: 张凝; 柴秀娟; 张文蓉; 夏雪; 张建华; 孙坦
Original assignee: Agricultural Information Institute of CAAS
Current assignee: Agricultural Information Institute of CAAS
Priority date: 2022-04-07
Filing date: 2022-04-07
Publication date: 2022-08-05

Abstract

The invention provides a method and a system for identifying and positioning wood harmed by pine wilt disease, wherein the method comprises the following steps: periodically acquiring original hyperspectral images of vegetation in a monitored area by using an aerial hyperspectral imaging sensor to form a multi-stage hyperspectral image; inputting the multi-stage hyperspectral images into a disease identification characteristic map model, and identifying the disease grade of each damaged wood in the monitored area and a corresponding disease identification characteristic map thereof; extracting the time sequence characteristics of each disease identification characteristic map, and performing time sequence prediction analysis on the time sequence characteristics to determine each hazard tree grade as the identification time of the early disease; extracting characteristic map data corresponding to the identification time of the hazard trees according to the disease identification characteristic maps corresponding to the disease grades to obtain a time-space-spectrum characteristic map data set for early identification of the hazard trees; and classifying the time-space-spectrum characteristic map data set so as to realize accurate identification and accurate positioning of early harmful trees in the original hyperspectral image to be positioned.

Description

Method and system for identifying and positioning wood damaged by pine wilt disease

Technical Field

The invention relates to the technical field of forest pest monitoring and prediction, in particular to a method and a system for identifying and positioning wood damaged by pine wilt disease.

Background

Pine nematode disease (PWD) is caused by infection of Pine nematodes (Pine nematodes), and is a serious forest disease that can cause Pine death in a short time, and is also called Pine nematode wilting disease, Pine wilting disease, and Pine wilt disease. Once the disease occurs, the pine can be rapidly died within 60-90 days after infection, the disease is rapidly spread, the phenomenon of large-area forest damage can be caused in 3-5 years, and the disease control difficulty is extremely high, so that the disease is called as the cancer of the pine.

Under the influence of factors such as climate change, acceleration of economic and trade exchanges and the like, the pine wood nematode disease area is rapidly expanded and spread, and because susceptible pine trees are difficult to cure, the early discovery of diseases becomes a basic premise for preventing and controlling the pine wood nematode disease, and the problem to be solved is also needed. Through the development of many years, the satellite remote sensing technology is widely applied and developed in the large-scale monitoring and predicting direction of forest diseases and insect pests due to the multispectral and multi-temporal characteristics of the satellite remote sensing technology. However, satellite-borne remote sensing technology still has many disadvantages in forest pest application: if the collected data is greatly influenced by weather conditions, the data collected in rainy days has large errors; the method is suitable for large-scale regional analysis, and is difficult to meet ideal precision requirements for small-scale analysis, especially for detection of pest and disease damage, and the sensitivity for detecting the early symptoms of plant pest and disease damage is insufficient. Therefore, how to accurately identify and accurately locate the early-stage damaged wood caused by the pine wilt disease is the key point for stopping the outbreak and spread of the disease.

Disclosure of Invention

The invention aims to provide a method and a system for identifying and positioning wood damaged by pine wilt disease, which can realize accurate identification and accurate positioning of early-stage damaged wood caused by pine wilt disease.

In order to overcome the defects in the prior art, the invention provides a method for identifying and positioning wood endangered by pine wilt disease, wherein the method comprises the following steps:

step 1, acquiring an original hyperspectral image of vegetation in a monitored area by using an aerial hyperspectral imaging sensor, wherein the vegetation in the monitored area comprises healthy vegetation and multi-grade hazard trees;

step 2, matching the corresponding disease grade for the original hyperspectral image of each damaged wood to obtain a damaged wood image with the disease grade;

step 3, training a feature map recognition model by taking the hazard tree image as training data, and taking the trained feature map recognition model as a disease recognition feature map model;

step 4, the aerial hyperspectral imaging sensor regularly acquires original hyperspectral images of vegetation in the monitored area to form a multi-stage hyperspectral image;

step 5, inputting the multi-stage hyperspectral images into the disease identification characteristic map model, and identifying the disease grade of each hazard tree in the monitored area and a corresponding disease identification characteristic map thereof;

step 6, extracting time sequence characteristics of each disease identification characteristic map, and performing time sequence prediction analysis on the time sequence characteristics to determine each hazard tree grade as the identification time of the disease early stage;

step 7, extracting characteristic map data corresponding to the identification time of the hazard trees according to the disease identification characteristic maps corresponding to the disease grades to obtain a time-space-spectrum characteristic map data set for early identification of the hazard trees;

and 8, classifying the time-space-spectrum characteristic spectrum data set to locate the early-stage damaged trees.

The method for identifying and positioning the wood endangered by the pine wilt disease, wherein the step 2 comprises the following steps of:

acquiring the positions of the hazard trees of different grades in the monitored area;

splicing, terrain correction and spectrum correction are carried out on the original hyperspectral images of the harmfuls of different grades in each period to obtain hyperspectral preprocessed images;

marking the hyperspectral preprocessed image according to the positioning of the hazard trees of different grades to obtain a time sequence-based pixel label data set of the hyperspectral image;

extracting single-tree crown image data of hazard trees of different grades from the pixel label data set of the hyperspectral image based on the time sequence;

and taking the single-wood crown image data of the hazard trees with different grades as hazard tree images representing different disease grades.

The method for identifying and positioning the wood endangered by the pine wilt disease, wherein the step 3 comprises the following steps:

measuring physiological and biochemical parameters of the hazard wood of different grades, wherein the physiological and biochemical parameters comprise pigment content, water content, transpiration rate, photosynthetic index and the like;

analyzing physiological and biochemical parameters of the harmful wood with different grades according to the harmful wood images representing different disease grades, and determining optimal quantitative parameters of the harmful wood representing different disease grades;

extracting an area average spectrum of the original hyperspectral image;

screening the sensitive wave band of the area average spectrum according to the optimal quantitative parameters of the hazard trees representing different disease grades to obtain single-wave-band image data of the sensitive wave band;

determining the spectral index of the plant disease according to the single-waveband image data;

determining image geometric information according to the single-waveband images with different disease grades;

analyzing the correlation among the single-waveband image, the spectral index and the image geometric information according to the physiological and biochemical parameters representing the disease grades to obtain disease identification characteristic maps of different disease grades;

and training the characteristic map recognition model by taking the harmful wood images representing different disease grades as input and the disease recognition characteristic maps representing different disease grades as output.

The method for identifying and positioning the wood endangered by the pine wilt disease, wherein the step 8 comprises the following steps:

extracting time-spectrum features and space-spectrum features of the hazard tree early identification time-space-spectrum feature spectrum data set by adopting an optimal classification algorithm; and obtaining the location of the early-stage hazard wood according to the time-spectrum feature and the space-spectrum feature.

classifying the hazard tree early-stage identification time-space-spectrum characteristic spectrum data set based on spectrum through an optimal classification algorithm to obtain an image to be positioned after pixel-by-pixel spectrum classification; and carrying out image segmentation on the image to be positioned after the pixel-by-pixel spectral classification to obtain the positioning of the early-stage damaged wood.

The invention also provides a system for identifying and positioning the wood endangered by the pine wilt disease, wherein the method comprises the following steps:

the system comprises an original image acquisition module, a hyperspectral imaging module and a monitoring module, wherein the original image acquisition module is used for acquiring an original hyperspectral image of vegetation in a monitored area through an aerial hyperspectral imaging sensor, and the vegetation in the monitored area comprises healthy vegetation and multi-grade harmful trees;

the preprocessing module is used for matching the corresponding disease grade for the original hyperspectral image of each damaged wood to obtain a damaged wood image with the disease grade;

the training module is used for training a feature map recognition model by taking the hazard tree image as training data, and taking the trained feature map recognition model as a disease recognition feature map model;

the multi-stage original hyperspectral image acquisition module is used for periodically acquiring original hyperspectral images of vegetation in the monitored area through the aerial hyperspectral imaging sensor to form multi-stage hyperspectral images;

the disease identification characteristic map acquisition module is used for inputting the multi-stage hyperspectral images into the disease identification characteristic map model and identifying the disease grade of each damaged wood in the monitored area and the corresponding disease identification characteristic map;

the time sequence characteristic determining module is used for extracting the time sequence characteristics of the disease identification characteristic maps, carrying out time sequence prediction analysis on the time sequence characteristics and determining the damage wood grades as the early identification time of the diseases;

the characteristic map data set determining module is used for extracting characteristic map data corresponding to the identification time of the hazard tree according to the disease identification characteristic map corresponding to each disease grade to obtain a time-space-spectrum characteristic map data set for early identification of the hazard tree;

and the positioning module is used for classifying the time-space-spectrum characteristic spectrum data set so as to position the early-stage hazard trees.

The system for identifying and locating wood endangered by pine wilt disease, wherein the preprocessing module is used for:

The system for identifying and positioning the wood endangered by the pine wilt disease, wherein the training module is used for:

extracting the area average spectrum of the original hyperspectral image;

The system for identifying and locating wood endangered by pine wilt disease, wherein the step 8 comprises:

extracting time-spectrum features and space-spectrum features of the hazard tree early-stage identification time-space-spectrum feature spectrum data set by adopting an optimal classification algorithm; and obtaining the location of the early-stage hazard wood according to the time-spectrum feature and the space-spectrum feature.

The system for identifying and positioning the wood endangered by the pine wilt disease, wherein the positioning module is used for:

According to the scheme, the invention has the advantages that:

the invention provides a method for identifying and positioning wood harmed by pine wilt disease, which comprises the following steps: acquiring an original hyperspectral image of vegetation in a monitored area by using a hyperspectral imaging sensor carried by an unmanned aerial vehicle; vegetation in the monitored area comprises healthy vegetation and harmful trees with different grades; preprocessing original hyperspectral images of the harmfulness trees of different grades to obtain harmfulness tree images representing different disease grades; constructing a disease identification characteristic map model based on a machine learning algorithm; training a disease recognition characteristic map model by taking harmful wood images representing different disease grades as input to obtain a trained disease recognition characteristic map model; acquiring original hyperspectral images of vegetation in a multi-period monitoring area according to a set time interval by using a hyperspectral imaging sensor carried by an unmanned aerial vehicle; inputting original hyperspectral images of vegetation in a multi-stage monitoring area into a trained disease identification characteristic map model, and outputting disease identification characteristic maps of different disease grades; determining time sequence characteristics corresponding to the characteristic spectrum of the pathogen infected harmful wood based on the disease identification characteristic spectrums of different disease grades; determining the time point of early identification of the hazard trees by adopting a time sequence prediction analysis method based on time sequence characteristics; the time point is a time interval with the grade of the damaged wood being the early disease stage; according to the disease identification characteristic maps of different disease grades, extracting characteristic map data corresponding to the time points of early identification of the hazard trees to obtain a time-space-spectrum characteristic map data set of early identification of the hazard trees; and (3) positioning early-stage damaged trees by adopting an optimal classification algorithm based on the time-space-spectrum characteristic spectrum data set for early-stage identification of damaged trees. According to the method, the early identification time point of the damaged wood is determined to obtain the early identification feature map of the damaged wood, and the optimal classification algorithm is determined by utilizing the early identification feature map of the damaged wood; by utilizing an optimal classification algorithm, accurate identification and accurate positioning of early harmful trees in the original hyperspectral image to be positioned can be realized.

Drawings

FIG. 1 is a flow chart of a method for identifying and locating wood damaged by pine wilt disease provided by the present invention;

fig. 2 is a block diagram of a system for identifying and locating wood endangered by pine wilt disease provided by the invention.

Description of the symbols:

the method comprises an original image acquisition module-1, a preprocessing module-2, a construction module-3, a training module-4, a multi-period original hyperspectral image acquisition module-5, a disease identification characteristic map acquisition module-6, a time sequence characteristic determination module-7, an early identification time point determination module-8, a characteristic map data set determination module-9 and a positioning module-10.

Detailed Description

In order to achieve the purpose, the invention provides the following scheme:

a method of identifying and locating a wood endangered by pine wilt disease, the method comprising:

acquiring an original hyperspectral image of vegetation in a monitored area by using a hyperspectral imaging sensor carried by an unmanned aerial vehicle; the vegetation in the monitored area comprises healthy vegetation and harmful trees with different grades;

preprocessing the original hyperspectral images of the harmfulness trees with different grades to obtain harmfulness tree images representing different disease grades;

constructing a disease identification characteristic map model based on a machine learning algorithm;

training the disease recognition characteristic map model by taking the harmful wood images representing different disease grades as input to obtain a trained disease recognition characteristic map model;

acquiring original hyperspectral images of vegetation in a multi-period monitoring area according to a set time interval by using a hyperspectral imaging sensor carried by an unmanned aerial vehicle;

inputting original hyperspectral images of vegetation in a multi-stage monitoring area into the trained disease identification characteristic map model, and outputting disease identification characteristic maps of different disease grades corresponding to each harming tree;

determining the time sequence characteristics corresponding to the characteristic spectrum of the pathogen infected harmful wood based on the disease identification characteristic spectrums of different disease grades;

determining the time point of early identification of the hazard trees by adopting a time sequence prediction analysis method based on the time sequence characteristics; the time point is a time interval with the grade of the damaged wood being the early disease stage;

according to the disease identification characteristic maps of different disease grades, extracting characteristic map data corresponding to the time points of early identification of the hazard trees to obtain a time-space-spectrum characteristic map data set of early identification of the hazard trees;

based on the early stage identification time-space-spectrum characteristic spectrum data set of the hazard trees, positioning the early stage hazard trees by adopting an optimal classification algorithm, wherein each pixel of the image corresponds to a geographic coordinate, and after classification is finished, the geographic coordinates of the pixels corresponding to different classes are positioning; the optimal classification algorithm is an algorithm with the highest positioning precision, which is obtained by evaluating a plurality of classification algorithms by taking the overall classification precision, the average classification precision, the Kappa coefficient and the T-test as evaluation indexes.

Optionally, the preprocessing is performed on the original hyperspectral images of the damaged wood of different grades to obtain damaged wood images representing different disease grades, and specifically includes:

obtaining the positioning of the hazard trees of different grades in the monitored area;

and taking the single-tree crown image data of the harmful trees with different grades as harmful wood images representing different disease grades.

Optionally, the training of the disease recognition feature map model by using the harmful wood image representing different disease grades as input to obtain a trained disease recognition feature map model specifically includes:

measuring physiological and biochemical parameters of the hazard wood with different grades through a laboratory; the physiological and biochemical parameters of the hazard wood with different grades comprise pigment content, water content, transpiration rate, porosity conductivity and the like;

extracting the area average spectrum of the original hyperspectral image;

and training the disease recognition characteristic map model by taking the harmful wood images representing different disease grades as input and the disease recognition characteristic maps representing different disease grades as output.

Optionally, based on the time-space-spectrum characteristic spectrum data set for early identification of the dangerous wood, an optimal classification algorithm is adopted to locate the early dangerous wood; the method specifically comprises the following steps:

extracting time-spectrum features of the time-space-spectrum feature spectrum data set for early identification of the hazard trees by adopting an optimal classification algorithm;

extracting the space-spectrum characteristics of the time-space-spectrum characteristic spectrum data set for early identification of the hazard trees by adopting an optimal classification algorithm;

and obtaining the location of the early-stage hazard wood according to the time-spectrum feature and the space-spectrum feature.

Optionally, based on the time-space-spectrum characteristic spectrum data set for early identification of the dangerous wood, an optimal classification algorithm is adopted to locate the early dangerous wood; the optimal classification algorithm is an algorithm with the highest positioning precision, which is obtained by evaluating a plurality of classification algorithms by taking the overall classification precision, the average classification precision, the Kappa coefficient and the T-test as evaluation indexes, and specifically comprises the following steps:

classifying the hazard tree early-stage identification time-space-spectrum characteristic spectrum data set based on spectrum through an optimal classification algorithm to obtain an image to be positioned after pixel-by-pixel spectrum classification;

and carrying out image segmentation on the image to be positioned after the pixel-by-pixel spectral classification to obtain the positioning of the early-stage damaged wood.

In order to make the aforementioned features and effects of the present invention more comprehensible, embodiments accompanied with figures are described in detail below.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

As shown in fig. 1, the method for identifying and locating wood endangered by pine wilt disease provided by the invention comprises the following steps:

step S1: acquiring an original hyperspectral image of vegetation in a monitored area by using a hyperspectral imaging sensor carried by an unmanned aerial vehicle; the vegetation in the monitored area includes healthy vegetation and different grades of hazardous wood.

Specifically, an unmanned aerial vehicle is used for carrying a hyperspectral imaging sensor to acquire an original hyperspectral image of a plant covering a range of a research area for one time or continuously and repeatedly acquiring the original hyperspectral image of the plant covering the range of the research area at a certain time interval; the time interval is 5-7 days.

Further, taking a key epidemic area in which the pine wood nematode disease occurs in the last two years as an example, taking areas such as Guangdong, Henan, Liaoning and the like to determine areas containing damaged wood and healthy wood at the same time as main test areas, establishing 20 square continuous monitoring sample plots of 30m multiplied by 30m, wherein the interval between the sample plots is 30m, and developing basic investigation; selecting 10-20 single trees (the sample trees comprise all damaged trees in the sample plot) as experimental standard trees for numbering (year-tree number) in each sample plot by a five-point method, carrying out follow-up investigation and analysis, and adding newly added damaged trees in the sample plot into an investigation range and numbering according to the disease occurrence condition every year; based on the life cycle of the pine wood nematodes and the monochamus alternatus and the main stage of the occurrence of diseases caused by the pine wood nematodes, the first investigation is started when the temperature reaches above 15 ℃ in spring every year, and the specific time is determined by the specific time of the breeding cycle of larvae of the pine wood nematodes in the test until the monochamus alternatus enters the wintering stage. Each survey comprises a ground data survey and corresponding unmanned aerial vehicle data acquisition.

Step S2: and preprocessing the original hyperspectral images of the harmfulness trees with different grades to obtain the harmfulness tree images representing different disease grades.

S2 specifically includes:

step S21: and obtaining positioning of the hazard trees of different grades in the monitored area.

Specifically, while an original hyperspectral image of a plant covering a research area is obtained, ground investigation is conducted on the damage level of the pine wood nematode harming wood in a monitored area, and the damage level of the harmed wood and the location of the harmed wood are determined.

Step S22: and splicing, terrain correction and spectrum correction are carried out on the original hyperspectral images of the harmfulness trees of different grades in each period to obtain a hyperspectral preprocessed image. Specifically, the hyperspectral preprocessed image is a hyperspectral time sequence dataset of the unmanned aerial vehicle.

Step S23: and (4) marking the hyperspectral preprocessed image according to positioning of hazard trees of different grades to obtain a time sequence-based pixel label data set of the hyperspectral image.

Specifically, according to the positioning of the hazard trees, on the original hyperspectral images which are continuously obtained for multiple times, the hazard trees with different grades are classified and labeled through image labeling software, and the classification basis is as shown in table 1, so that a pixel label data set based on the time sequence hyperspectral images is formed.

TABLE 1 pine wood nematode affected wood grading Standard

As a specific implementation manner of this embodiment, the ground survey is developed by taking a single tree as a survey unit. Acquiring the data of the leaf loss/withered leaf rate of all test sample trees (standard branches are randomly cut from the upper layer, the middle layer and the lower layer according to the east direction, the west direction, the south direction and the north direction respectively, and are brought back to the room for calculating the leaf loss/withered leaf rate); dividing pine trees with different damage degrees into five stages according to the tree growth condition, the pine needle color and the pine resin flow change of each damaged wood in the sample plot, wherein the specific standard is shown in the table 1; in addition, ASD spectral data, chlorophyll content data, transpiration rate and water content data of the needle leaves with different color change degrees of the sample wood are obtained; GPS coordinate information of sample plot and damaged trees, forest stand structure and forest age, canopy intensity, single-tree crown and the like.

As a specific implementation manner of this embodiment, the near-field airborne remote sensing image data is acquired by using an eight-rotor unmanned aerial vehicle, and synchronously carrying a high-definition digital camera and a hyperspectral imager (UHD185 or others), and the weather is selected from clear and calm weather, and is performed at 10:00-14:00 noon. And setting the flying height to be 50-100m according to the terrain, vegetation condition and coverage area of the test sample plot. And acquiring hyperspectral images and synchronous high-definition digital images covering all sample areas. A black-and-white standard plate is placed in the flight area to provide radiation calibration parameters for the hyperspectral image. The calibration of high-spectrum data and high-definition digital images is realized by using PhotoSacan, and the radiation correction is realized by splicing and using ENVI 5.3.

Step S24: and extracting single tree canopy image data of the hazard trees of different grades from the pixel label data set of the hyperspectral image based on the time sequence. Specifically, single-tree canopy image data of health, infection beginning, early stage, middle stage and late stage death are extracted from a hyperspectral time sequence data set of the unmanned aerial vehicle and a pixel label data set of a hyperspectral image based on time sequence.

Step S25: and taking the single-wood crown image data of the harmful wood with different grades as the harmful wood image representing different disease grades.

Step S3: and constructing a disease identification characteristic map model based on a machine learning algorithm.

Step S4: and taking the images representing the harmful wood with different disease grades as input, and training the disease recognition characteristic map model to obtain the trained disease recognition characteristic map model.

Step S41: measuring physiological and biochemical parameters of the harmful wood with different grades through a laboratory; the physiological and biochemical parameters of the hazard wood with different grades comprise pigment content, water content, transpiration rate, porosity conductivity and the like.

Step S42: analyzing physiological and biochemical parameters of the harmful wood with different grades according to the harmful wood images representing different disease grades, and determining optimal quantitative parameters of the harmful wood representing different disease grades.

Specifically, the harmful wood images representing different disease grades are single-wood crown image data of the harmful wood of different grades; based on single-tree canopy image data of harmful trees with different grades, aiming at different stages of disease occurrence, namely diseases in different grades, corresponding to physiological and biochemical parameters of the harmful trees with different grades, measured in a laboratory, analyzing time sequence change characteristics of physiological and biochemical parameters of needle leaves such as chlorophyll content, water content, transpiration rate, photosynthetic rate, carotenoid and the like, integrating single-tree leaf loss/withered leaf rate data, and determining optimal quantitative parameters capable of representing disease grades in different periods of diseases.

Step S43: and extracting the area average spectrum of the original hyperspectral image.

Step S44: and screening the sensitive wave band of the average spectrum of the region according to the optimal quantitative parameters of the hazard trees representing different disease grades to obtain single-wave-band image data of the sensitive wave band.

Specifically, based on the regional average spectrum, the regional average spectrum is subjected to sensitive waveband screening by taking the optimal quantization parameter determined in each disease period as a dependent variable, and single waveband images of each sensitive waveband are extracted; further, the screening method is inter-class instability index (ISIC), Principal Component Analysis (PCA) or continuous projection transform (SPA).

Step S45: and determining the spectral index of the plant disease according to the single-waveband image data.

Specifically, according to single-waveband image data, effective spectral indexes for disease monitoring such as normalized vegetation index (NDVI), Difference Vegetation Index (DVI), conversion chlorophyll absorption and reflection index (TCARI) and the like are constructed.

Step S46: and determining the geometric information of the image according to the single-band images with different disease grades.

Specifically, according to the single-waveband image data, the canopy geometric information such as texture, roughness and shape of the image data corresponding to different disease stages is analyzed by using the technologies such as color space transformation and mode identification.

Step S47: and analyzing the relevance of the single-waveband image, the spectral index and the image geometric information according to physiological and biochemical parameters representing the disease grade to obtain the disease identification characteristic maps of different disease grades.

Specifically, specific quantitative characterization parameters in different disease periods are taken as dependent variables, when late-stage disease features are obvious, the loss/withered leaf rate is taken as the dependent variables, and correlations of the sensitive wave bands, the characteristic spectrum indexes, the image geometric information, the texture information, the brightness information and the chromaticity information obtained through comprehensive analysis and screening are obtained to obtain disease identification characteristic maps of different disease grades; further, the method of comprehensive analysis is a statistical method similar to correlation analysis.

Step S48: and training a disease recognition characteristic map model by taking harmful wood images representing different disease grades as input and disease recognition characteristic maps representing different disease grades as output.

Step S5: acquiring original hyperspectral images of vegetation in a multi-period monitoring area according to a set time interval by using a hyperspectral imaging sensor carried by an unmanned aerial vehicle; specifically, the set time interval is 5-7 days; in practical application, the obtained original hyperspectral image of the vegetation in the multi-period monitoring area is obtained through a hyperspectral imaging sensor carried by an unmanned aerial vehicle in the monitoring process of the vegetation in the monitoring area, and the obtained original hyperspectral image of the vegetation in the multi-period monitoring area needs to be identified and positioned for early damage trees.

Step S6: and inputting the original hyperspectral images of the vegetation in the multi-stage monitoring area into the trained disease identification characteristic map model, and outputting the disease identification characteristic maps with different disease grades.

Step S7: and determining the corresponding time sequence characteristics of the characteristic spectrum of the pathogen-infected harmful wood based on the disease identification characteristic spectrums of different disease grades.

Specifically, the disease identification characteristic maps of different disease grades in the whole disease occurrence process are obtained based on the disease identification characteristic maps of different disease grades acquired in different disease periods, so that the time sequence characteristics corresponding to the characteristic maps of the pathogen-infected harming wood are determined.

Step S8: and determining the time point of early identification of the hazard trees by adopting a time sequence prediction analysis method based on the time sequence characteristics. The time point is the time interval when the grade of the damaged wood is early disease. Because the time span infected with early symptoms has universality, a characteristic map model corresponding to the harmfulness trees in each period is firstly obtained by matching with a field test data method (S2-S4), then the whole image map obtained at a certain time is analyzed by using the model (the image obtained by S5), and finally according to the determined characteristics corresponding to the early time points, the stages of each tree on the whole image are determined, namely the harmfulness trees in the early period are determined at the same time.

Specifically, the time series prediction analysis method is an original autoregressive moving model (ARMA) or a summation autoregressive moving model (ARIMA) model. Further, a time sequence prediction analysis method is adopted to construct a time sequence model of each physiological and biochemical parameter, and the characteristic physiological and biochemical parameters of the harmfulwood, which can meet the monitoring requirement of the whole disease occurrence period, are determined according to the accuracy of disease grade prediction; the physiological and biochemical parameters of the characteristics of the hazard wood are single parameters or combined parameters. And determining a time point corresponding to a balance point of an evaluation index on the time axis by using the correlation analysis method, namely the time point of early identification of the disease wood.

Step S9: according to the disease identification characteristic maps of different disease grades, extracting characteristic map data corresponding to the time points of early identification of the hazard trees to obtain a time-space-spectrum characteristic map data set of early identification of the hazard trees so as to determine specific characteristics, and then accurately classifying by using the determined characteristics. Where "empty" in the feature map refers to a spatial feature, such as the location of the target tree relative to a tree, or relative to two trees, or relative to some relationship between grids.

Specifically, on the time axis, the map data in the time period of early identification of the diseased wood is the characteristic map of early identification of the diseased wood.

Step S10: based on the time-space-spectrum characteristic spectrum data set for early identification of the damaged trees, positioning the early damaged trees by adopting an optimal classification algorithm; the optimal classification algorithm is an algorithm with the highest positioning precision, which is obtained by evaluating a plurality of classification algorithms by taking the overall classification precision, the average classification precision, the Kappa coefficient and the T-test as evaluation indexes.

Specifically, there are two methods for locating the early hazard wood. The first method specifically comprises:

step 201: and extracting the time-spectrum features of the time-space-spectrum feature spectrum data set for early identification of the hazard trees by adopting an optimal classification algorithm.

Step 202: and extracting the space-spectrum characteristics of the time-space-spectrum characteristic spectrum data set for early identification of the hazard trees by adopting an optimal classification algorithm.

Step 203: and obtaining the location of the early-stage harmful wood according to the time-spectrum characteristic and the space-spectrum characteristic.

The second method specifically comprises the following steps:

step 301: and classifying the hazard tree early-stage identification time-space-spectrum characteristic spectrum data set based on the spectrum through an optimal classification algorithm to obtain an image to be positioned after pixel-by-pixel spectral classification.

Step 302: and carrying out image segmentation on the image to be positioned after the pixel-by-pixel spectral classification to obtain the positioning of the early-stage damaged wood.

In the first method, the process of determining the optimal classification algorithm specifically includes:

step 2011: and extracting time-spectrum characteristics of the time-space-spectrum characteristic spectrum data set for early identification of the hazard trees by adopting empirical mode decomposition and multi-scale signal decomposition respectively to obtain the time-spectrum characteristics of the empirical mode decomposition and the time-spectrum characteristics of the multi-scale signal decomposition.

Step 2012: and extracting the space-spectrum characteristics of the time-space-spectrum characteristic spectrum data set for early identification of the hazard trees by adopting spectral gradient enhancement, generalized kernel support vector machine classification and sparse matrix respectively to obtain spectral gradient enhancement space-spectrum characteristics, generalized kernel support vector machine classification space-spectrum characteristics and sparse matrix space-spectrum characteristics.

Step 2013: according to the empirical mode decomposition time-spectrum characteristic and the spectral gradient enhancement space-spectrum characteristic, positioning of the early-stage hazard trees of a first classification algorithm is obtained; the first classification algorithm includes empirical mode decomposition and spectral gradient enhancement.

Step 2014: according to the empirical mode decomposition time-spectrum feature and the generalized kernel support vector machine classification space-spectrum feature, the positioning of the early-stage hazard trees of the second classification algorithm is obtained; the second classification algorithm comprises empirical mode decomposition and generalized kernel support vector machine classification.

Step 2015: obtaining the positioning of the early hazard trees of a third classification algorithm according to the multi-scale signal decomposition time-spectrum characteristics and the sparse matrix space-spectrum characteristics; the third classification algorithm includes multi-scale signal decomposition and sparse matrices.

Step 2016: based on the positioning of the early stage of the hazard tree in the early stage identification feature map of the hazard tree, the overall classification precision, the average classification precision, the Kappa coefficient and the T-test are used as evaluation indexes to evaluate a plurality of classification algorithms respectively, and the optimal classification algorithm is determined.

In the second method, the process of determining the optimal classification algorithm specifically includes:

step 3011: and respectively adopting a support vector machine and a spectrum angle drawing to classify the time-space-spectrum characteristic spectrum data set for early identification of the hazard tree based on the spectrum, so as to obtain a support vector machine hazard tree early-stage positioning spectrum image and a spectrum angle drawing hazard tree early-stage positioning spectrum image.

Step 3012: the method comprises the steps of segmenting an early stage positioning spectral image of the support vector machine hazard tree by adopting edge filtering to obtain the positioning of the early stage hazard tree of a first classification algorithm; the first classification algorithm includes edge filtering and a support vector machine.

Step 3013: the random walking model is adopted to segment the early positioning spectral image of the support vector machine hazard tree to obtain the positioning of the early hazard tree of the second classification algorithm; the second classification algorithm comprises a random walk model and a support vector machine.

Step 3014: the spectral angle drawing hazard tree early positioning spectral image is segmented by adopting edge filtering to obtain the positioning of the hazard tree early by a third classification algorithm; the third classification algorithm includes edge filtering and spectral angle mapping.

Step 3015: a random walking model is adopted to segment the spectrum angle drawing hazard tree early positioning spectrum image to obtain the positioning of the hazard tree early by a fourth classification algorithm; the fourth classification algorithm includes a random walk model and a spectral angle chart.

Step 3016: based on the positioning of the early stage of the hazardous wood in the hazardous wood early stage identification characteristic spectrum, the overall classification precision, the average classification precision, the Kappa coefficient and the T-test are used as evaluation indexes to evaluate various classification algorithms to obtain the optimal classification algorithm.

In the present representative hyperspectral data space-spectrum fusion algorithm (GSA, PCA, SFIM, CNMF and the like), two major directions of quantitative indexes formed by parameters such as Time consumption (Time), space feature retention (PSNR, RMSE), spectral feature retention (SAM) and comprehensive fusion Effect (ERGAS) of a visual inspection and a dependent algorithm are used as measurement standards, an optimal space-spectrum fusion algorithm suitable for early pine wood nematode disease identification Time points is determined, Time sequence analysis is added on the basis of the existing algorithm, a targeted Time-space-spectrum optimization fusion algorithm is formed, and a specific algorithm suitable for pine wood nematode disease victim identification is optimized and provided.

Therefore, the invention forms a positioning and identifying frame of the pine wood nematode damage wood by defining the disease identifying characteristic spectrum of different disease grades of the pine wood nematode damage wood, the time point of early identification of the disease wood and a classification and identification method of fusion of space-time spectrum, integrates an identification system, and classifies and identifies the data carrying real-time image acquisition and transmission of the unmanned aerial vehicle, thereby realizing automatic positioning and identification of the early damage wood.

As shown in fig. 2, the present invention provides a system for identifying and locating wood endangered by pine wilt disease, which comprises:

the system comprises an original image acquisition module 1, a hyperspectral image acquisition module and a hyperspectral image acquisition module, wherein the original image acquisition module is used for acquiring an original hyperspectral image of vegetation in a monitored area by utilizing a hyperspectral imaging sensor carried by an unmanned aerial vehicle; plants in the monitored area include healthy vegetation and different grades of hazardous wood.

And the preprocessing module 2 is used for preprocessing the original hyperspectral images of the harmfulness trees with different grades to obtain the harmfulness tree images representing different disease grades.

And the construction module 3 is used for constructing a disease identification characteristic map model based on a machine learning algorithm.

And the training module 4 is used for training the disease recognition characteristic map model by taking the harmful wood images representing different disease grades as input so as to obtain the trained disease recognition characteristic map model.

And acquiring a multi-period original hyperspectral image 5, and acquiring an original hyperspectral image of vegetation in the multi-period monitoring area according to a set time interval by using a hyperspectral imaging sensor carried by the unmanned aerial vehicle.

And the disease identification characteristic map acquisition module 6 is used for inputting the original hyperspectral images of the vegetation in the multi-stage monitoring area into the trained disease identification characteristic map model and outputting the disease identification characteristic maps of different disease grades.

And the time sequence characteristic determining module 7 is used for determining time sequence characteristics corresponding to the characteristic map of the pathogen infected harmful wood based on the disease identification characteristic maps of different disease grades.

The early identification time point determining module 8 is used for determining the time point of early identification of the hazard trees by adopting a time sequence prediction analysis method based on time sequence characteristics; the time point is the time interval when the grade of the damaged wood is early disease.

And the characteristic map data set determining module 9 is used for extracting characteristic map data corresponding to the time point of early identification of the hazard tree according to the disease identification characteristic maps of different disease grades to obtain a time-space-spectrum characteristic map data set of early identification of the hazard tree.

The positioning module 10 is used for positioning early-stage damaged trees by adopting an optimal classification algorithm based on the time-space-spectrum characteristic spectrum data set for early-stage identification of damaged trees; the optimal classification algorithm is an algorithm with the highest positioning precision, which is obtained by evaluating a plurality of classification algorithms by taking the overall classification precision, the average classification precision, the Kappa coefficient and the T-test as evaluation indexes.

Wherein, preprocessing module 2 includes:

and the acquisition submodule is used for acquiring positioning of the hazard trees of different grades in the monitoring area.

And the hyperspectral preprocessed image determining submodule is used for splicing, terrain correcting and spectrum correcting the original hyperspectral images of the harmfulness trees of different grades in each period to obtain a hyperspectral preprocessed image.

And the labeling submodule is used for labeling the hyperspectral preprocessed image according to positioning of hazard trees of different grades to obtain a time sequence-based pixel label data set of the hyperspectral image.

And the extraction submodule is used for extracting single-tree canopy image data of the hazard trees of different grades according to the pixel label data set of the hyperspectral image based on the time sequence.

And the determining submodule is used for taking the single-wood crown image data of the harmful wood with different grades as the harmful wood image representing different disease grades.

Wherein, training module 4 includes:

the measuring submodule is used for measuring physiological and biochemical parameters of the harmful wood with different grades through a laboratory; the physiological and biochemical parameters of the hazard wood of different grades comprise pigment content, water content, transpiration rate and photosynthetic index.

And the optimal quantitative parameter obtaining submodule is used for analyzing physiological and biochemical parameters of the harmful wood with different grades according to the harmful wood images representing different disease grades and determining optimal quantitative parameters of the harmful wood representing different disease grades.

And the average spectrum extraction submodule is used for extracting the regional average spectrum of the original hyperspectral image.

And the single-waveband image data determining submodule is used for screening the sensitive waveband of the average spectrum of the region according to the optimal quantitative parameters of the hazard trees representing different disease grades to obtain the single-waveband image data of the sensitive waveband.

And the spectral index determining submodule is used for determining the spectral index of the plant disease according to the single-waveband image data.

And the image geometric information determining submodule is used for determining image geometric information according to the single-waveband images with different disease grades.

And the analysis submodule is used for analyzing the correlation among the single-waveband image, the spectral index and the image geometric information according to the physiological and biochemical parameters representing the disease grades to obtain the disease identification characteristic maps of different disease grades.

And the characteristic map model training submodule is used for training the disease recognition characteristic map model by taking the harmful wood images representing different disease grades as input and taking the disease recognition characteristic maps of different disease grades as output.

Wherein, the constitution of orientation module divides into two kinds, and first kind constitutes and includes:

and the time-spectrum feature extraction submodule is used for extracting the time-spectrum features of the time-space-spectrum feature spectrum data set for early identification of the hazard trees by adopting an optimal classification algorithm.

And the space-spectrum feature extraction submodule is used for extracting the space-spectrum features of the time-space-spectrum feature spectrum data set for early identification of the hazard trees by adopting an optimal classification algorithm.

And the positioning sub-module is used for obtaining the positioning of the early-stage hazard wood according to the time-spectrum characteristic and the space-spectrum characteristic.

The second configuration includes:

and the spectrum classification-based submodule is used for classifying the hazard tree early-stage identification time-space-spectrum characteristic spectrum data set based on the spectrum through an optimal classification algorithm to obtain an image to be positioned after pixel-by-pixel spectrum classification.

And the positioning submodule is used for carrying out image segmentation on the image to be positioned after the pixel-by-pixel spectral classification to obtain the positioning of the early-stage damaged wood.

The overall design idea of the method for identifying and positioning the wood endangered by the pine wilt disease provided by the invention is as follows:

within 60-90 days from the invasion of the pine wood nematodes to the complete withering of trees, the host vegetation of the pine wood nematodes is changed into yellow brown or red brown from uncharacterized traits to coniferous leaves, and the tree structure, chlorophyll, water content and other indexes are changed in the whole disease process of the pine wood nematode disease. In the whole change process, the image characteristics of the unmanned aerial vehicle on the hyperspectral image are different. The method comprehensively considers the multi-index map characteristics represented by the images at all disease-sensitive stages, determines the sensitive parameters and the indicative characteristic maps of early disease identification, and analyzes the time sequence characteristics of slight change on the basis to determine the optimal early disease identification time point. By utilizing the time, space and spectral resolution advantages of hyperspectral data of the unmanned aerial vehicle, on the basis of researching a time-space-spectrum effective fusion method, time, space and spectral information of image data are comprehensively utilized through technical means such as image segmentation and machine learning, automatic extraction and positioning of pine wood nematode damaged trees are achieved, and technical support is provided for timely and effective prevention and control measures.

The method and system for identifying and positioning the wood damaged by the pine wood nematode disease integrate theories and technical advantages of multiple subjects such as forest conservation science, spatial statistical analysis, image processing and the like, analyze an image and a spectral response mechanism suitable for identifying the pine wood nematode disease by taking an unmanned aerial vehicle remote sensing technology as a core on the basis of the existing map feature identification and 'empty-spectrum' fusion algorithm according to the life cycle of the pine wood nematode and the mechanism of causing damage to host vegetation, optimize the existing classification and image segmentation method, and construct a non-contact system for identifying the disease. The method fully utilizes the experimental conditions of a fixed sample plot and a perennial cooperative forest farm, synchronously utilizes the unmanned aerial vehicle remote sensing technology and mathematical correlation analysis, utilizes the technologies of parameter inversion, model construction, correlation analysis and the like on the basis of qualitative judgment of the traditional field disease degree and quantitative determination of physiological and biochemical parameters such as transpiration rate, chlorophyll content and the like, converts the qualitative disease degree into a quantitative physiological and biochemical parameter evaluation standard, combines the traditional field investigation with the remote sensing quantitative inversion, and guides disease prevention and control from the scientific and quantitative angle. By fully using the existing common pine wood nematode disease investigation method for reference, carrying out field on-site disease investigation from three scales of a fixed sample plot, a single wood and a needle, combining needle ground spectrum and unmanned aerial vehicle hyperspectral data to form a multi-source data set, and applying airborne hyperspectral data to fixed-point timing monitoring of pine wood nematode diseases; meanwhile, an optimization algorithm of disease monitoring is explored on the basis of the original space-spectrum classification algorithm, time sequence analysis is added, a time-space-spectrum combined classification method is provided, and the application range of the hyperspectral image classification technology supporting forest disaster monitoring is widened.

In addition, the pine wood nematode disease is taken as a research object, the advantages of crossing of multiple disciplines such as remote sensing, forestry and forest conservation are comprehensively utilized aiming at the specific biological characteristics of the pine wood nematode disease, the investigation and the recognition research of the damaged tree of the disease are developed by the unmanned aerial vehicle-mounted hyperspectral data and high-definition digital images and combining ground investigation data, the non-contact efficient means for early recognition and positioning of the damaged tree of the pine wood nematode disease is sought, and the key problem of timing, positioning and data fixing in disease monitoring is solved. Has the following advantages:

(1) by analyzing the disease characteristics of the pine wilt disease, a targeted data fusion and space spectrum feature extraction algorithm is researched, and a scheme for satisfying data acquisition and effective information extraction of small-scale unmanned aerial vehicle pine wilt disease monitoring is provided.

(2) On the basis of time sequence characteristic data, the time problem of early disease identification of the pine wood nematode disease is solved by researching a time-space-spectrum fusion algorithm and corresponding spectrum characteristics of remote sensing data of the unmanned aerial vehicle.

(3) The method of 'space spectrum feature extraction' and 'space spectrum classification framework construction' is researched, so that automatic identification of the damaged trees is realized, the problem of accurate positioning of early disease identification is solved, and a basis is provided for timely and effective disaster prevention and control.

The method is based on disease identification characteristic map construction of hyperspectrum of the unmanned aerial vehicle, determination of early disease identification time points, and development of an early damaged wood automatic positioning system based on characteristic map extraction and empty spectrum classification construction on the basis of a time-space-spectrum fusion algorithm, so that effective application of unmanned aerial vehicle imaging hyperspectral data in pine wood nematode disease identification is finally realized.

The following are system examples corresponding to the above method examples, and this embodiment can be implemented in cooperation with the above embodiments. The related technical details mentioned in the above embodiments are still valid in this embodiment, and are not described herein again in order to reduce repetition. Accordingly, the related-art details mentioned in the present embodiment can also be applied to the above-described embodiments.

extracting the area average spectrum of the original hyperspectral image;

extracting time-spectrum features and space-spectrum features of the hazard tree early identification time-space-spectrum feature spectrum data set by adopting an optimal classification algorithm; and obtaining the positioning of the early-stage damaged wood according to the time-spectrum characteristic and the space-spectrum characteristic.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims

1. A method of identifying and locating wood endangered by pine wilt disease, the method comprising:

step 3, training a feature map recognition model by taking the harmful wood image as training data, and taking the trained feature map recognition model as a disease recognition feature map model;

step 6, extracting time sequence characteristics of each disease identification characteristic map, and performing time sequence prediction analysis on the time sequence characteristics to determine each hazard tree grade as the identification time of the early disease;

2. The method of identifying and locating a wood nematode disease hazard of claim 1, wherein said step 2 comprises:

3. The method of identifying and locating a wood nematode disease hazard of claim 1, wherein said step 3 comprises:

extracting the area average spectrum of the original hyperspectral image;

4. The method of identifying and locating a wood nematode disease hazard of claim 1, wherein said step 8 comprises:

5. The method of identifying and locating a wood nematode disease hazard of claim 1, wherein said step 8 comprises:

6. A system for identifying and locating wood endangered by pine wilt disease, the method comprising:

7. The system for identification and location of bursaphelenchus xylophilus hazard wood of claim 1, wherein the preprocessing module is configured to:

8. The system of identification and location of pine wilt disease wood according to claim 1, wherein said training module is to:

extracting the area average spectrum of the original hyperspectral image;

9. The system for identification and location of pine wilt disease wood according to claim 1, wherein said step 8 comprises:

10. The system of identification and location of pine wilt disease wood according to claim 1, wherein said location module is to: