Method and system for identifying and locating wood damaged by pine wood nematode disease

Abstract

The present invention provides a method and system for identifying and locating trees damaged by pine wood nematode disease. The method comprises: using an aerial hyperspectral imaging sensor to periodically acquire original hyperspectral images of vegetation in a monitoring area to form multi-period hyperspectral images; inputting the multi-period hyperspectral images into a disease recognition feature map model to identify the disease level of each damaged tree in the monitoring area and its corresponding disease recognition feature map; extracting the time series characteristics of each disease recognition feature map, and performing time series prediction analysis on the map to determine the recognition time of each damaged tree level as an early stage of the disease; extracting feature map data corresponding to the recognition time of the damaged tree according to the disease recognition feature map corresponding to each disease level, and obtaining a time-space-spectrum feature map data set for early recognition of the damaged tree; and classifying the time-space-spectrum feature map data set to achieve accurate recognition and precise positioning of the early damaged tree in the original hyperspectral image to be positioned.

Description

Technical Field

[0001] The invention relates to the technical field of forest pest monitoring and prediction, and in particular to a method and system for identifying and locating wood damaged by pine wood nematode disease. Background Art

[0002] Pine wilt disease (PWD) is caused by the infection of pine wood nematodes (Bursaphelenchusxylophilus). It is a serious forest disease that can cause pine trees to die in a short period of time. It is also known as pine wilt disease, pine wilt disease, and pine wilt disease. Once the disease occurs, it can cause pine trees to die quickly within 60-90 days after infection. It spreads rapidly and can cause large-scale deforestation in 3-5 years. At the same time, the disease is extremely difficult to prevent and control, and is called the "cancer" of pine trees.

[0003] Under the influence of climate change, accelerated economic and trade exchanges and other factors, the pine wilt disease epidemic area has expanded and spread rapidly. Since the infected pine trees are difficult to cure, the early detection of the disease has become the basic premise for the prevention and control of pine wilt disease, and it is also a problem that needs to be solved urgently. After years of development, satellite remote sensing technology has been widely used and developed in the large-scale monitoring and prediction of forest pests and diseases due to its multi-spectral and multi-temporal characteristics. However, there are still many deficiencies in the application of satellite remote sensing technology in forest pests and diseases: for example, the collected data is greatly affected by weather conditions, and the data collected on rainy days will have large errors; it is suitable for large-scale regional analysis, but it is difficult to achieve the ideal accuracy requirements for small-scale analysis, especially for the detection of pests and diseases, and the sensitivity of the early symptoms of plant pests and diseases is insufficient. Therefore, how to accurately identify and precisely locate the early-stage damaged trees of pine wilt disease is the key to curbing the outbreak and spread of the disease. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for identifying and locating wood damaged by pine wilt disease, which can realize accurate identification and precise positioning of wood damaged by pine wilt disease in its early stage.

[0005] In view of the shortcomings of the prior art, the present invention provides a method for identifying and locating wood damaged by pine wood nematode disease, wherein the method comprises:

[0006] Step 1: using an aerial hyperspectral imaging sensor to obtain an original hyperspectral image of vegetation in a monitoring area, wherein the vegetation in the monitoring area includes healthy vegetation and multiple levels of hazardous trees;

[0007] Step 2: Match the original hyperspectral image of each damaged tree with the corresponding disease level to obtain a damaged tree image with the disease level;

[0008] Step 3: using the damaged wood image as training data, training a feature spectrum recognition model, and using the trained feature spectrum recognition model as a disease recognition feature spectrum model;

[0009] Step 4: The aerial hyperspectral imaging sensor periodically acquires original hyperspectral images of vegetation in the monitoring area to form multi-period hyperspectral images;

[0010] Step 5: input the multi-period hyperspectral image into the disease identification characteristic spectrum model to identify the disease level of each harmful tree in the monitoring area and its corresponding disease identification characteristic spectrum;

[0011] Step 6: extract the time series characteristics of each disease identification feature map, and perform time series prediction analysis on it to determine the early identification time of each damaged wood grade;

[0012] Step 7: extract the characteristic spectrum data corresponding to the identification time of the damaged tree according to the disease identification characteristic spectrum corresponding to each disease level, and obtain the time-space-spectrum characteristic spectrum data set for early identification of damaged trees;

[0013] Step 8: classify the temporal-spatial-spectral characteristic map dataset to locate early-stage damaged trees.

[0014] The method for identifying and locating wood damaged by pine wilt disease, wherein step 2 comprises:

[0015] Obtain the locations of hazardous trees of different levels within the monitoring area;

[0016] The original hyperspectral images of the damaged trees of different grades in each period are spliced, terrain corrected and spectrally corrected to obtain hyperspectral preprocessed images;

[0017] According to the positioning of the different levels of hazardous trees, the hyperspectral preprocessed image is labeled to obtain a pixel label data set of the hyperspectral image based on time series;

[0018] Extracting single tree crown image data of trees with different levels of damage from the pixel label data set of the time-series-based hyperspectral image;

[0019] The single tree crown image data of the trees with different levels of damage are used as damaged tree images representing different disease levels.

[0020] The method for identifying and locating wood damaged by pine wilt disease, wherein step 3 comprises:

[0021] Determine the physiological and biochemical parameters of damaged wood of different grades, including pigment content, water content, transpiration rate and photosynthetic index, etc.;

[0022] Analyzing the physiological and biochemical parameters of the damaged wood of different disease levels according to the damaged wood images representing different disease levels, and determining the optimal quantitative parameters representing the damaged wood of different disease levels;

[0023] Extracting a regional average spectrum of the original hyperspectral image;

[0024] According to the optimal quantitative parameters of the damaged wood representing different disease levels, the sensitive bands of the regional average spectrum are screened to obtain single-band image data of the sensitive bands;

[0025] Determining the spectral index of plant diseases according to the single-band image data;

[0026] Determining image geometric information according to the single-band images of different disease levels;

[0027] Analyze the correlation between the single-band image, the spectral index and the image geometric information according to the physiological and biochemical parameters characterizing the disease level, and obtain disease identification feature maps of different disease levels;

[0028] The damaged wood images representing different disease levels are used as input, and the disease identification feature maps of different disease levels are used as output to train the feature map recognition model.

[0029] The method for identifying and locating wood damaged by pine wilt disease, wherein step 8 comprises:

[0030] The optimal classification algorithm is used to extract the time-spectrum features and space-spectrum features of the time-space-spectrum feature map data set for early identification of harmful trees; the location of early harmful trees is obtained based on the time-spectrum features and the space-spectrum features.

[0031] The method for identifying and locating wood damaged by pine wilt disease, wherein step 8 comprises:

[0032] The time-space-spectrum characteristic map data set for early identification of harmful trees is classified based on the spectrum through the optimal classification algorithm to obtain the image to be located after pixel-by-pixel spectrum classification; the image to be located after pixel-by-pixel spectrum classification is segmented to obtain the location of the early harmful trees.

[0033] The present invention also proposes a system for identifying and locating wood damaged by pine wood nematode disease, wherein the method comprises:

[0034] The original image acquisition module is used to acquire the original hyperspectral image of the vegetation in the monitoring area through the aerial hyperspectral imaging sensor, wherein the vegetation in the monitoring area includes healthy vegetation and multiple levels of harmful trees;

[0035] A preprocessing module is used to match the corresponding disease level for the original hyperspectral image of each damaged tree to obtain a damaged tree image with the disease level;

[0036] A training module, used to train a feature spectrum recognition model using the damaged wood image as training data, and use the trained feature spectrum recognition model as a disease recognition feature spectrum model;

[0037] A multi-period original hyperspectral image acquisition module is used to periodically acquire original hyperspectral images of vegetation in the monitoring area through the aerial hyperspectral imaging sensor to form multi-period hyperspectral images;

[0038] A disease identification characteristic spectrum acquisition module is used to input the multi-period hyperspectral image into the disease identification characteristic spectrum model to identify the disease level of each harmful tree in the monitoring area and its corresponding disease identification characteristic spectrum;

[0039] The time series feature determination module is used to extract the time series features of each disease identification feature map, and perform time series prediction analysis on it to determine the early identification time of each damaged wood grade;

[0040] A characteristic spectrum data set determination module is used to extract characteristic spectrum data corresponding to the identification time of the damaged tree according to the disease identification characteristic spectrum corresponding to each disease level, and obtain a time-space-spectrum characteristic spectrum data set for early identification of damaged trees;

[0041] A positioning module is used to classify the temporal-spatial-spectral characteristic map data set to locate early-stage damaged trees.

[0042] In the system for identifying and locating wood damaged by pine wilt disease, the pretreatment module is used to:

[0043] Obtain the locations of hazardous trees of different levels within the monitoring area;

[0044] The original hyperspectral images of the damaged trees of different grades in each period are spliced, terrain corrected and spectrally corrected to obtain hyperspectral preprocessed images;

[0045] According to the positioning of the different levels of hazardous trees, the hyperspectral preprocessed image is labeled to obtain a pixel label data set of the hyperspectral image based on time series;

[0046] Extracting single tree crown image data of trees with different levels of damage from the pixel label data set of the time-series-based hyperspectral image;

[0047] The single tree crown image data of the trees with different levels of damage are used as damaged tree images representing different disease levels.

[0048] In the system for identifying and locating wood damaged by pine wilt disease, the training module is used to:

[0049] Determine the physiological and biochemical parameters of damaged wood of different grades, including pigment content, water content, transpiration rate and photosynthetic index, etc.;

[0050] Analyzing the physiological and biochemical parameters of the damaged wood of different disease levels according to the damaged wood images representing different disease levels, and determining the optimal quantitative parameters representing the damaged wood of different disease levels;

[0051] Extracting a regional average spectrum of the original hyperspectral image;

[0052] According to the optimal quantitative parameters of the damaged wood representing different disease levels, the sensitive bands of the regional average spectrum are screened to obtain single-band image data of the sensitive bands;

[0053] Determining the spectral index of plant diseases according to the single-band image data;

[0054] Determining image geometric information according to the single-band images of different disease levels;

[0055] Analyze the correlation between the single-band image, the spectral index and the image geometric information according to the physiological and biochemical parameters characterizing the disease level, and obtain disease identification feature maps of different disease levels;

[0056] The damaged wood images representing different disease levels are used as input, and the disease identification feature maps of different disease levels are used as output to train the feature map recognition model.

[0057] In the system for identifying and locating wood damaged by pine wilt disease, step 8 comprises:

[0058] The optimal classification algorithm is used to extract the time-spectrum features and space-spectrum features of the time-space-spectrum feature map data set for early identification of harmful trees; the location of early harmful trees is obtained based on the time-spectrum features and the space-spectrum features.

[0059] In the system for identifying and locating wood damaged by pine wilt disease, the positioning module is used to:

[0060] The time-space-spectrum characteristic map data set for early identification of harmful trees is classified based on the spectrum through the optimal classification algorithm to obtain the image to be located after pixel-by-pixel spectrum classification; the image to be located after pixel-by-pixel spectrum classification is segmented to obtain the location of the early harmful trees.

[0061] It can be seen from the above scheme that the advantages of the present invention are:

[0062] The present invention provides a method for identifying and locating wood damaged by pine wood nematode disease, comprising: using a hyperspectral imaging sensor carried by an unmanned aerial vehicle to obtain original hyperspectral images of vegetation in a monitoring area; the vegetation in the monitoring area includes healthy vegetation and damaged wood of different grades; preprocessing the original hyperspectral images of damaged wood of different grades to obtain damaged wood images representing different disease grades; constructing a disease identification feature map model based on a machine learning algorithm; using the damaged wood images representing different disease grades as input, training the disease identification feature map model to obtain a trained disease identification feature map model; using the hyperspectral imaging sensor carried by the unmanned aerial vehicle to obtain multiple original hyperspectral images of vegetation in the monitoring area at set time intervals; and preprocessing the multiple hyperspectral images of damaged wood of different grades to obtain damaged wood images representing different disease grades. The original hyperspectral images of vegetation in the monitoring area are input into the trained disease recognition feature map model, and the disease recognition feature maps of different disease levels are output; based on the disease recognition feature maps of different disease levels, the time series characteristics corresponding to the feature maps of harmful trees infected by pathogens are determined; based on the time series characteristics, the time series prediction and analysis method is used to determine the time point for early identification of harmful trees; the time point is the time interval when the harmful tree level is in the early stage of the disease; according to the disease recognition feature maps of different disease levels, the feature map data corresponding to the time point for early identification of harmful trees are extracted to obtain the time-space-spectrum feature map data set for early identification of harmful trees; based on the time-space-spectrum feature map data set for early identification of harmful trees, the optimal classification algorithm is used to locate early harmful trees. The present invention obtains the feature map for early identification of harmful trees by determining the time point for early identification of harmful trees, and determines the optimal classification algorithm by using the feature map for early identification of harmful trees; using the optimal classification algorithm, the early harmful trees in the original hyperspectral image to be located can be accurately identified and precisely located. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flow chart of the method for identifying and locating wood damaged by pine wood nematode disease provided by the present invention;

[0064] Figure 2 A system block diagram for identifying and locating wood damaged by pine wood nematode disease provided by the present invention.

[0065] Explanation of symbols:

[0066] Original image acquisition module—1, preprocessing module—2, construction module—3, training module—4, acquisition of multiple periods of original hyperspectral images—5, disease identification feature map acquisition module—6, time series feature determination module—7, early identification time point determination module—8, feature map data set determination module—9, positioning module—10. DETAILED DESCRIPTION

[0067] To achieve the above object, the present invention provides the following solutions:

[0068] A method for identifying and locating wood damaged by pine wood nematode disease, the method comprising:

[0069] The hyperspectral imaging sensor carried by the UAV is used to obtain the original hyperspectral image of the vegetation in the monitoring area; the vegetation in the monitoring area includes healthy vegetation and different levels of harmful trees;

[0070] Preprocessing the original hyperspectral images of the damaged trees of different grades to obtain damaged tree images representing different disease grades;

[0071] Construct a disease identification feature map model based on machine learning algorithms;

[0072] Taking the damaged wood images representing different disease levels as input, training the disease recognition feature map model to obtain a trained disease recognition feature map model;

[0073] The hyperspectral imaging sensor carried by the UAV is used to obtain the original hyperspectral images of vegetation in the monitoring area at set time intervals for multiple periods;

[0074] Inputting the original hyperspectral images of vegetation in the multi-period monitoring area into the trained disease identification feature map model, and outputting disease identification feature maps of different disease levels corresponding to each damaged tree;

[0075] Based on the disease identification characteristic maps of different disease levels, determining the time series characteristics corresponding to the characteristic maps of the wood damaged by the pathogen infection;

[0076] Based on the time series characteristics, a time series prediction and analysis method is used to determine the time point for early identification of damaged trees; the time point is the time interval when the damaged tree level is in the early stage of the disease;

[0077] According to the disease identification characteristic maps of different disease levels, the characteristic map data corresponding to the time point of early identification of the damaged trees are extracted to obtain a time-space-spectrum characteristic map data set for early identification of the damaged trees;

[0078] Based on the temporal-spatial-spectral characteristic map data set for early identification of harmful trees, the optimal classification algorithm is used to locate early harmful trees. Each pixel of the image corresponds to a geographic coordinate. After classification, the geographic coordinates of the pixels corresponding to different categories are the positioning. The optimal classification algorithm is the algorithm with the highest positioning accuracy obtained by evaluating multiple classification algorithms using overall classification accuracy, average classification accuracy, Kappa coefficient and T-test as evaluation indicators.

[0079] Optionally, the preprocessing of the original hyperspectral images of the damaged trees of different disease levels to obtain damaged tree images representing different disease levels specifically includes:

[0080] Obtain the location of hazardous trees of different levels in the monitoring area;

[0081] The original hyperspectral images of the damaged trees of different grades in each period are spliced, terrain corrected and spectrally corrected to obtain hyperspectral preprocessed images;

[0082] According to the positioning of the different levels of hazardous trees, the hyperspectral preprocessed image is labeled to obtain a pixel label data set of the hyperspectral image based on time series;

[0083] Extracting single tree crown image data of trees with different levels of damage from the pixel label data set of the time-series-based hyperspectral image;

[0084] The single tree crown image data of the trees with different levels of damage are used as damaged tree images representing different disease levels.

[0085] Optionally, the taking the damaged wood images representing different disease levels as input, training the disease identification feature map model to obtain a trained disease identification feature map model specifically includes:

[0086] The physiological and biochemical parameters of the wood with different levels of damage are measured in the laboratory; the physiological and biochemical parameters of the wood with different levels of damage include pigment content, water content, transpiration rate and stomatal conductance, etc.;

[0087] Analyzing the physiological and biochemical parameters of the damaged wood of different disease levels according to the damaged wood images representing different disease levels, and determining the optimal quantitative parameters representing the damaged wood of different disease levels;

[0088] Extracting a regional average spectrum of the original hyperspectral image;

[0089] According to the optimal quantitative parameters of the damaged wood representing different disease levels, the sensitive bands of the regional average spectrum are screened to obtain single-band image data of the sensitive bands;

[0090] Determining the spectral index of plant diseases according to the single-band image data;

[0091] Determining image geometric information according to the single-band images of different disease levels;

[0092] Analyze the correlation between the single-band image, the spectral index and the image geometric information according to the physiological and biochemical parameters characterizing the disease level, and obtain disease identification feature maps of different disease levels;

[0093] The disease identification feature map model is trained by taking the damaged wood images representing different disease levels as input and taking the disease identification feature maps of different disease levels as output.

[0094] Optionally, the method of locating early-stage harmful trees based on the temporal-spatial-spectral characteristic map dataset for early identification of harmful trees adopts an optimal classification algorithm; specifically includes:

[0095] The optimal classification algorithm is used to extract the time-spectrum characteristics of the time-space-spectrum characteristic spectrum data set for early identification of harmful trees;

[0096] The optimal classification algorithm is used to extract the space-spectrum features of the time-space-spectrum feature map data set for early identification of harmful trees;

[0097] The location of early damaged trees is obtained according to the time-spectrum characteristics and the space-spectrum characteristics.

[0098] Optionally, based on the time-space-spectrum characteristic map data set for early identification of harmful trees, an optimal classification algorithm is used to locate early harmful trees; the optimal classification algorithm is an algorithm with the highest positioning accuracy obtained by evaluating multiple classification algorithms using overall classification accuracy, average classification accuracy, Kappa coefficient and T-test as evaluation indicators, specifically including:

[0099] Classify the time-space-spectrum characteristic map data set for early identification of harmful trees based on the spectrum by using an optimal classification algorithm to obtain an image to be located after pixel-by-pixel spectrum classification;

[0100] The image to be located after the pixel-by-pixel spectral classification is subjected to image segmentation to obtain the location of the early-stage damaged trees.

[0101] In order to make the above features and effects of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings.

[0102] The purpose of the present invention is to provide a method and system for identifying and locating wood damaged by pine wilt disease, which can realize accurate identification and precise positioning of wood damaged by pine wilt disease in its early stage.

[0103] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0104] like Figure 1 As shown, the present invention provides a method for identifying and locating wood damaged by pine wood nematode disease, comprising:

[0105] Step S1: using the hyperspectral imaging sensor carried by the UAV to obtain the original hyperspectral image of the vegetation in the monitoring area; the vegetation in the monitoring area includes healthy vegetation and different levels of harmful trees.

[0106] Specifically, a hyperspectral imaging sensor mounted on an unmanned aerial vehicle is used to obtain an original hyperspectral image of plants covering the study area once, or to obtain original hyperspectral images of plants covering the study area multiple times at a certain time interval; the time interval is 5-7 days.

[0107] Furthermore, taking the key epidemic areas of pine wood nematode disease in the past two years as an example, Guangdong, Henan, Liaoning and other regions were identified as the main test areas, and 20 square continuous monitoring plots of 30m×30m were established, with a spacing of 30m between plots, to carry out basic surveys; 10-20 single trees (the sample trees include all the damaged trees in the plots) were selected as experimental standard trees in each plot using the five-point method, and numbered (year-tree number) and subsequent surveys and analyses were carried out. At the same time, new damaged trees in the plots were added to the survey scope and numbered every year based on the incidence situation; relying on the life cycle of pine wood nematodes and pine beetles and the main stages of pine wood nematode disease, the first survey was started every spring when the temperature reached above 15℃, and the specific time thereafter was determined by the specific time of the reproduction cycle of pine wood nematode larvae in the test area until the pine beetles entered the wintering period. Each survey includes ground data survey and corresponding drone data acquisition.

[0108] Step S2: preprocessing the original hyperspectral images of damaged trees of different grades to obtain damaged tree images representing different disease grades.

[0109] S2 specifically includes:

[0110] Step S21: obtaining the locations of hazardous trees of different levels in the monitoring area.

[0111] Specifically, while obtaining the original hyperspectral images of plants covering the study area, a ground survey is conducted on the damage level of pine wood nematode-damaged trees in the monitoring area to determine the damage level of the damaged trees and the location of the damaged trees.

[0112] Step S22: stitching, terrain correction and spectral correction are performed on the original hyperspectral images of the damaged trees of different grades in each period to obtain a hyperspectral preprocessed image. Specifically, the hyperspectral preprocessed image is a drone hyperspectral time series data set.

[0113] Step S23: labeling the hyperspectral preprocessed image according to the location of the harmful trees at different levels, and obtaining a pixel label data set of the hyperspectral image based on the time series.

[0114] Specifically, according to the location of the harmful trees, the image annotation software is used to classify and annotate the harmful trees of different grades on the original hyperspectral images acquired continuously for multiple times. The classification basis is shown in Table 1, forming a pixel label dataset based on the time series hyperspectral images.

[0115] Table 1 Classification standards for wood damaged by pine wilt disease

[0116]

[0117]

[0118] As a specific implementation method of this embodiment, the ground survey is carried out with a single tree as the survey unit. Obtain the leaf loss / leaf withering rate data of all test samples (randomly cut standard branches from the upper, middle and lower layers in the four directions of east, west, south and north, and bring them back to the room to calculate the leaf loss / leaf withering rate); for each damaged tree in the sample plot, according to the tree growth status, pine needle color and resin flow changes, the pine trees with different degrees of damage are divided into five stages, and the specific standards are shown in Table 1 above; in addition, obtain the ASD spectral data, chlorophyll content data, transpiration rate and water content data of the needles with different degrees of discoloration of the sample trees; GPS coordinate information of the sample plot and damaged trees, forest structure and age, canopy density, single tree crown width, etc.

[0119] As a specific implementation of this embodiment, near-ground airborne remote sensing image data is obtained using an eight-rotor drone, which is equipped with a high-definition digital camera and a hyperspectral imager (UHD185 or other). Clear and windless weather is selected between 10:00-14:00 at noon. According to the terrain, vegetation conditions and coverage area of ​​the test plot, the flight altitude is set to 50-100m. Hyperspectral images and synchronized high-definition digital images covering all plots are obtained. Black and white standard plates are placed in the flight area to provide radiation calibration parameters for hyperspectral images. PhotoScan is used to calibrate hyperspectral data and high-definition digital images, and they are spliced ​​and ENVI5.3 is used to implement radiation correction.

[0120] Step S24: extracting the single tree crown image data of trees with different levels of damage from the pixel label dataset of the time-series-based hyperspectral image. Specifically, extracting the single tree crown image data of healthy, beginning infection, early, middle and late death stages from the drone hyperspectral time series dataset and the pixel label dataset of the time-series-based hyperspectral image.

[0121] Step S25: using the single tree crown image data of trees with different levels of damage as damaged tree images representing different disease levels.

[0122] Step S3: Construct a disease identification feature map model based on a machine learning algorithm.

[0123] Step S4: using damaged wood images representing different disease levels as input, training the disease recognition feature map model to obtain a trained disease recognition feature map model.

[0124] Step S41: measuring the physiological and biochemical parameters of the damaged wood of different grades in the laboratory; the physiological and biochemical parameters of the damaged wood of different grades include pigment content, water content, transpiration rate and stomatal conductance, etc.

[0125] Step S42: Analyze the physiological and biochemical parameters of the damaged trees of different disease levels according to the damaged trees images representing different disease levels, and determine the optimal quantitative parameters representing the damaged trees of different disease levels.

[0126] Specifically, the images of damaged trees representing different disease levels are single tree crown image data of damaged trees of different levels; based on the single tree crown image data of damaged trees of different levels, for different stages of disease occurrence, that is, when the disease is at different levels, the physiological and biochemical parameters of damaged trees of different levels measured in the laboratory are used to analyze the temporal variation characteristics of needle physiological and biochemical parameters such as chlorophyll content, water content, transpiration rate, photosynthetic rate, and carotenoids, and the single tree leaf loss / leaf withering rate data are combined to determine the optimal quantitative parameters that can characterize the disease level at different stages of the disease.

[0127] Step S43: extracting the regional average spectrum of the original hyperspectral image.

[0128] Step S44: According to the optimal quantitative parameters of the damaged trees representing different disease levels, the sensitive bands of the regional average spectrum are screened to obtain single-band image data of the sensitive bands.

[0129] Specifically, based on the regional average spectrum, the optimal quantitative parameters determined in each disease period are taken as dependent variables, the sensitive bands of the regional average spectrum are screened, and the single-band images of each sensitive band are extracted; further, the screening methods are inter-class instability index (ISIC), principal component analysis (PCA) or continuous projection transformation (SPA).

[0130] Step S45: Determine the spectral index of plant diseases based on the single-band image data.

[0131] Specifically, based on single-band image data, effective spectral indices for disease monitoring such as the Normalized Difference Vegetation Index (NDVI), the Difference Vegetation Index (DVI), and the Transformed Chlorophyll Absorption Reflectance Index (TCARI) are constructed.

[0132] Step S46: Determine image geometric information based on single-band images of different disease levels.

[0133] Specifically, based on the single-band image data, color space transformation, pattern recognition and other technologies are used to analyze the canopy geometric information such as texture, roughness, shape, etc. of the image data corresponding to different disease stages.

[0134] Step S47: Analyze the correlation between the single-band image, spectral index and image geometric information according to the physiological and biochemical parameters characterizing the disease level, and obtain disease identification feature maps of different disease levels.

[0135] Specifically, for different disease stages, the specific quantitative characterization parameters of the stage are used as the dependent variable; when the characteristics are obvious in the late stage of the disease, the rate of leaf loss / leaf death is used as the dependent variable; the correlation of the screened sensitive bands, characteristic spectral indexes, image geometry information, texture information, brightness information, and chromaticity information is comprehensively analyzed to obtain disease identification feature maps of different disease levels; further, the comprehensive analysis method is a statistical method similar to correlation analysis.

[0136] Step S48: using damaged wood images representing different disease levels as input and disease identification feature maps of different disease levels as output, training the disease identification feature map model.

[0137] Step S5: using the hyperspectral imaging sensor carried by the drone to obtain the original hyperspectral images of the vegetation in the multi-period monitoring area at set time intervals; specifically, the set time interval is 5-7 days; in actual applications, the original hyperspectral images of the vegetation in the multi-period monitoring area obtained are obtained by the hyperspectral imaging sensor carried by the drone during the monitoring of the vegetation in the monitoring area, and it is necessary to identify and locate the early harmful trees on the original hyperspectral images of the vegetation in the multi-period monitoring area.

[0138] Step S6: input the original hyperspectral images of vegetation in the multi-period monitoring area into the trained disease identification feature map model, and output disease identification feature maps of different disease levels.

[0139] Step S7: Based on the disease identification characteristic maps of different disease levels, determine the time series characteristics corresponding to the characteristic maps of the wood infected by the pathogen.

[0140] Specifically, based on the disease identification characteristic maps of different disease levels obtained at different disease periods, the disease identification characteristic maps of different disease levels in the overall process of disease occurrence are obtained, thereby determining the temporal characteristics corresponding to the characteristic maps of wood infected by pathogens.

[0141] Step S8: Based on the time series characteristics, the time series prediction and analysis method is used to determine the time point for early identification of harmful trees. The time point is the time interval when the harmful tree level is in the early stage of the disease. Since the time span from infection to early symptoms is universal, the characteristic map model (S2-S4) corresponding to the harmful trees in each period is first obtained by combining the field test data method, and then the model is used to analyze the entire image map obtained at a certain time (the image obtained in S5). Finally, according to the characteristics corresponding to the determined early time points, it is clear at what stage each tree in the entire image is in, that is, the harmful trees in the early stage are determined at the same time.

[0142] Specifically, the time series prediction and analysis method is the original autoregressive moving average (ARMA) model or the summed autoregressive moving average (ARIMA) model. Furthermore, the time series prediction and analysis method is used to construct the time series model of each physiological and biochemical parameter, and the characteristic physiological and biochemical parameters of the harmful wood that can meet the monitoring needs of the entire disease occurrence period are determined based on the accuracy of the disease level prediction; the characteristic physiological and biochemical parameters of the harmful wood are single parameters or combined parameters. Taking the characteristic physiological and biochemical parameters as the dependent variable and the disease occurrence process obtained by continuous collection as the time axis, the atlas data of the characteristic physiological and biochemical parameters at different disease levels on the time axis are obtained. On the time axis, the correlation analysis method is applied to determine the time point corresponding to the balance point of the evaluation index, which is the time point for early identification of diseased wood.

[0143] Step S9: According to the disease identification feature maps of different disease levels, the feature map data corresponding to the time point of early identification of the damaged tree is extracted to obtain a time-space-spectrum feature map data set for early identification of the damaged tree, so as to clarify the specific features, and then use the clear features for accurate classification. The "space" in the feature map refers to the spatial feature, such as the position of the target tree relative to a certain tree, or relative to the position of two trees, or relative to a certain relationship between grids.

[0144] Specifically, on the time axis, the atlas data within the time period of early identification of diseased wood is the characteristic atlas for early identification of damaged wood.

[0145] Step S10: Based on the temporal-spatial-spectral characteristic map data set for early identification of harmful trees, an optimal classification algorithm is used to locate early harmful trees; the optimal classification algorithm is an algorithm with the highest positioning accuracy obtained by evaluating multiple classification algorithms using overall classification accuracy, average classification accuracy, Kappa coefficient and T-test as evaluation indicators.

[0146] Specifically, there are two methods for locating early-stage damage trees. The first method specifically includes:

[0147] Step 201: extracting the time-spectrum features of the time-space-spectrum feature map dataset for early identification of harmful trees using an optimal classification algorithm.

[0148] Step 202: using an optimal classification algorithm to extract the spatial-spectral features of the temporal-spatial-spectral feature map dataset for early identification of harmful trees.

[0149] Step 203: The location of early-stage damaged trees is obtained according to the time-spectrum characteristics and the space-spectrum characteristics.

[0150] The second method specifically includes:

[0151] Step 301: Classify the time-space-spectrum characteristic map data set for early identification of harmful trees based on the spectrum using an optimal classification algorithm to obtain an image to be located after pixel-by-pixel spectrum classification.

[0152] Step 302: performing image segmentation on the image to be located after the pixel-by-pixel spectral classification to obtain the location of the early-stage damaged trees.

[0153] Among them, in the first method, the process of determining the optimal classification algorithm specifically includes:

[0154] Step 2011: respectively extracting the time-spectrum features of the time-space-spectrum feature map dataset for early identification of harmful trees by using empirical mode decomposition and multi-scale signal decomposition, and obtaining empirical mode decomposition time-spectrum features and multi-scale signal decomposition time-spectrum features.

[0155] Step 2012: spectral gradient enhancement, generalized kernel support vector machine classification and sparse matrix extraction are used to extract the space-spectrum features of the time-space-spectrum feature map dataset for early identification of harmful trees, and spectral gradient enhancement space-spectrum features, generalized kernel support vector machine classification space-spectrum features and sparse matrix space-spectrum features are obtained.

[0156] Step 2013: According to the time-spectrum characteristics of the empirical mode decomposition and the space-spectrum characteristics of the spectral gradient enhancement, the positioning of the early-stage damaged trees of the first classification algorithm is obtained; the first classification algorithm includes the empirical mode decomposition and the spectral gradient enhancement.

[0157] Step 2014: According to the time-spectrum features of the empirical mode decomposition and the space-spectrum features of the generalized kernel support vector machine classification, the positioning of the early-stage damaged trees of the second classification algorithm is obtained; the second classification algorithm includes the empirical mode decomposition and the generalized kernel support vector machine classification.

[0158] Step 2015: According to the time-spectrum characteristics of multi-scale signal decomposition and the space-spectrum characteristics of sparse matrix, the early-stage harmful tree positioning of the third classification algorithm is obtained; the third classification algorithm includes multi-scale signal decomposition and sparse matrix.

[0159] Step 2016: Based on the early location of the harmful trees in the harmful trees early identification feature map, multiple classification algorithms are evaluated using overall classification accuracy, average classification accuracy, Kappa coefficient and T-test as evaluation indicators to determine the optimal classification algorithm.

[0160] In the second method, the process of determining the optimal classification algorithm specifically includes:

[0161] Step 3011: Use support vector machine and spectral angle mapping respectively to classify the time-space-spectral feature map dataset for early identification of harmful trees based on the spectrum, and obtain support vector machine harmful tree early positioning spectral image and spectral angle mapping harmful tree early positioning spectral image.

[0162] Step 3012: Use edge filtering to segment the support vector machine early-stage location spectrum image of hazardous trees to obtain the early-stage location of hazardous trees of the first classification algorithm; the first classification algorithm includes edge filtering and support vector machine.

[0163] Step 3013: Use the random walk model to segment the spectral image of early damaged wood location using the support vector machine to obtain the location of early damaged wood using the second classification algorithm; the second classification algorithm includes the random walk model and the support vector machine.

[0164] Step 3014: Use edge filtering to segment the spectral image of early-stage location of damaged trees using spectral angle mapping to obtain the location of early-stage damaged trees using the third classification algorithm; the third classification algorithm includes edge filtering and spectral angle mapping.

[0165] Step 3015: Use the random walk model to segment the spectral image of early-stage damaged trees using the spectral angle mapping to obtain the early-stage damaged trees positioning of the fourth classification algorithm; the fourth classification algorithm includes the random walk model and the spectral angle mapping.

[0166] Step 3016: Based on the positioning of the early stage of harmful trees in the early stage identification feature map of harmful trees, multiple classification algorithms are evaluated with overall classification accuracy, average classification accuracy, Kappa coefficient and T-test as evaluation indicators to obtain the optimal classification algorithm.

[0167] In the existing representative hyperspectral data "space-spectrum" fusion algorithms (GSA, PCA, SFIM, CNMF, etc.), the present invention uses visual inspection and quantitative indicators formed by parameters such as algorithm time consumption (Time), spatial feature retention ability (PSNR, RMSE), spectral feature retention ability (SAM) and comprehensive fusion effect (ERGAS) as measurement standards to determine the optimal "space-spectrum" fusion algorithm suitable for the early identification time point of pine wood nematode disease, and on the basis of the existing algorithm, adds time series analysis to form a targeted "time-space-spectrum" optimization fusion algorithm, optimizes and proposes a specific algorithm suitable for the identification of pine wood nematode disease-infected wood.

[0168] Therefore, the present invention forms a positioning and identification framework for pine wood nematode-damaged trees by clarifying the disease identification characteristic maps of different disease levels of pine wood nematode-damaged trees, the time points for early identification of diseased trees, and a classification and identification method that integrates time and space spectra. The recognition system classifies and identifies the data acquired and transmitted by real-time images carried by drones, thereby realizing automatic positioning and identification of early-stage damaged trees.

[0169] like Figure 2 As shown, the present invention provides a system for identifying and locating wood damaged by pine wood nematode disease, the system comprising:

[0170] The original image acquisition module 1 is used to obtain the original hyperspectral image of the vegetation in the monitoring area using the hyperspectral imaging sensor carried by the UAV; the plants in the monitoring area include healthy vegetation and harmful trees of different levels.

[0171] The preprocessing module 2 is used to preprocess the original hyperspectral images of damaged wood of different grades to obtain damaged wood images representing different disease grades.

[0172] Construction module 3 is used to construct a disease identification feature map model based on a machine learning algorithm.

[0173] The training module 4 is used to train the disease recognition feature map model by taking the damaged wood images representing different disease levels as input, and obtain a trained disease recognition feature map model.

[0174] Acquiring multiple periods of original hyperspectral images 5 is used to obtain multiple periods of original hyperspectral images of vegetation in the monitoring area at set time intervals using a hyperspectral imaging sensor carried by the unmanned aerial vehicle.

[0175] The disease identification characteristic map acquisition module 6 is used to input the original hyperspectral images of vegetation in the multi-period monitoring area into the trained disease identification characteristic map model, and output disease identification characteristic maps of different disease levels.

[0176] The time series feature determination module 7 is used to determine the time series features corresponding to the characteristic maps of the wood damaged by the pathogen infection based on the disease identification characteristic maps of different disease levels.

[0177] The early identification time point determination module 8 is used to determine the time point for early identification of damaged trees based on time series characteristics and using a time series prediction and analysis method; the time point is the time interval when the damaged tree level is in the early stage of the disease.

[0178] The characteristic map data set determination module 9 is used to extract the characteristic map data corresponding to the time point of early identification of damaged trees according to the disease identification characteristic maps of different disease levels, and obtain the time-space-spectrum characteristic map data set for early identification of damaged trees.

[0179] The positioning module 10 is used to locate early-stage harmful trees based on the temporal-spatial-spectral characteristic map data set for early identification of harmful trees, using the optimal classification algorithm; the optimal classification algorithm is the algorithm with the highest positioning accuracy obtained by evaluating multiple classification algorithms using overall classification accuracy, average classification accuracy, Kappa coefficient and T-test as evaluation indicators.

[0180] Wherein, the preprocessing module 2 includes:

[0181] The acquisition submodule is used to obtain the location of hazardous trees of different levels in the monitoring area.

[0182] The hyperspectral preprocessing image determination submodule is used to perform splicing, terrain correction and spectral correction on the original hyperspectral images of damaged trees of different stages and grades to obtain hyperspectral preprocessing images.

[0183] The annotation submodule is used to annotate the hyperspectral preprocessed images according to the location of different levels of harmful trees, and obtain a pixel label dataset of the hyperspectral image based on time series.

[0184] The extraction submodule is used to extract the single tree crown image data of trees with different levels of damage according to the pixel label dataset of the time-series-based hyperspectral image.

[0185] A submodule is determined, which is used to use the single tree crown image data of trees with different levels of damage as damaged tree images representing different disease levels.

[0186] Among them, training module 4 includes:

[0187] The determination submodule is used to determine the physiological and biochemical parameters of different grades of damaged wood through the laboratory; the physiological and biochemical parameters of different grades of damaged wood include pigment content, water content, transpiration rate and photosynthetic index.

[0188] The optimal quantitative parameter acquisition submodule is used to analyze the physiological and biochemical parameters of damaged wood of different disease levels based on the damaged wood images representing different disease levels, and determine the optimal quantitative parameters representing the damaged wood of different disease levels.

[0189] The average spectrum extraction submodule is used to extract the regional average spectrum of the original hyperspectral image.

[0190] The single-band image data determination submodule is used to screen the sensitive bands of the regional average spectrum according to the optimal quantitative parameters of the damaged wood with different disease levels, and obtain the single-band image data of the sensitive bands.

[0191] The spectral index determination submodule is used to determine the spectral index of plant diseases based on single-band image data.

[0192] The image geometric information determination submodule is used to determine the image geometric information based on the single-band images of different disease levels.

[0193] The analysis submodule is used to analyze the correlation between single-band images, spectral indexes and image geometric information according to the physiological and biochemical parameters that characterize the disease level, and obtain disease identification feature maps of different disease levels.

[0194] The feature map model training submodule is used to train the disease identification feature map model by taking images of damaged wood representing different disease levels as input and taking disease identification feature maps of different disease levels as output.

[0195] Among them, the positioning module is composed of two types, the first type includes:

[0196] The submodule for extracting time-spectrum features is used to extract the time-spectrum features of the time-space-spectrum feature map dataset for early identification of harmful trees using the optimal classification algorithm.

[0197] The submodule for extracting spatial-spectral features is used to extract the spatial-spectral features of the temporal-spatial-spectral feature map dataset for early identification of harmful trees using the optimal classification algorithm.

[0198] The positioning submodule is used to obtain the location of early-damaged trees based on time-spectrum characteristics and space-spectrum characteristics.

[0199] The second composition includes:

[0200] Based on the spectral classification submodule, it is used to classify the time-space-spectral characteristic map data set for early identification of harmful trees based on the spectrum through the optimal classification algorithm, and obtain the image to be located after pixel-by-pixel spectral classification.

[0201] The positioning submodule is used to perform image segmentation on the image to be positioned after pixel-by-pixel spectral classification to obtain the location of the early-stage damaged trees.

[0202] The overall design concept of the method for identifying and locating wood damaged by pine wood nematode disease provided by the present invention is as follows:

[0203] From the invasion of pine wood nematodes to the complete death of trees within 60-90 days, the host vegetation changes from no characteristic traits to all needles turning yellow-brown or reddish brown. The entire process of the pine wood nematode disease is accompanied by changes in tree structure, chlorophyll, water content and other indicators. During the entire change process, the image characteristics shown on the drone hyperspectral image are also different. The present invention comprehensively considers the multi-index map characteristics represented by the image of each susceptible stage, determines the sensitive parameters and indicative characteristic maps for early identification of the disease, and analyzes the time series characteristics of subtle changes on this basis to determine the best time point for early identification of the disease. Utilizing the advantages of time, space and spectral resolution presented by drone hyperspectral data, on the basis of studying the effective fusion method of "time-space-spectrum", through image segmentation, machine learning and other technical means, the time, space and spectral information of the image data are comprehensively utilized to realize the automatic extraction and positioning of pine wood nematode-infected wood, providing technical support for the proposal and implementation of timely and effective prevention and control measures.

[0204] The method for identifying and locating wood damaged by pine wood nematode disease provided by the present invention systematically integrates the theoretical and technical advantages of multiple disciplines such as forest protection, spatial statistical analysis, and image processing. Based on the life cycle of pine wood nematodes and the mechanism of their damage to host vegetation, the method takes unmanned aerial vehicle remote sensing technology as the core, analyzes the image and spectral response mechanism suitable for pine wood nematode disease identification on the basis of existing spectral feature recognition and "space-spectrum" fusion algorithm, optimizes existing classification and image segmentation methods, and constructs a non-contact system for disease identification. Fully utilizing the experimental conditions of fixed sample plots and multi-year cooperative forest farms, on the basis of the qualitative determination of the degree of disease in the traditional field and the quantitative determination of physiological and biochemical parameters such as transpiration rate and chlorophyll content, the method simultaneously utilizes unmanned aerial vehicle remote sensing technology and mathematical correlation analysis, and utilizes parameter inversion, model construction, correlation analysis and other technologies to convert the qualitative degree of disease into a quantitative physiological and biochemical parameter evaluation standard, combines traditional field surveys with remote sensing quantitative inversion, and guides disease prevention and control from a scientific and quantitative perspective. We fully draw on the existing commonly used pine wood nematode disease investigation methods, conduct field disease investigations at three scales: fixed sample plots, single trees and conifers, combine conifer ground spectra and drone hyperspectral data to form a multi-source data set, and apply airborne hyperspectral data to fixed-point and timed monitoring of pine wood nematode disease; at the same time, we explore the optimization algorithm for disease monitoring based on the original "space-spectrum" classification algorithm, add time series analysis, and propose a "time-space-spectrum" joint classification method to expand the application scope of hyperspectral image classification technology to support forest disaster monitoring.

[0205] In addition, the present invention takes pine wilt disease as the research object, targets the unique biological characteristics of pine wilt disease, comprehensively utilizes the advantages of multidisciplinary cross-disciplinary studies such as remote sensing, forestry, and forest protection, and uses drone-borne hyperspectral data and high-definition digital images, combined with ground survey data, to conduct investigations and identification studies on diseased trees, aiming to seek non-contact and efficient means for early identification and positioning of pine wilt disease-infected trees, and solve the key issues of timing, positioning, and data determination in disease monitoring. It has the following advantages:

[0206] (1) By analyzing the disease characteristics of pine wilt disease, we studied targeted data fusion and spatial spectrum feature extraction algorithms, and proposed a solution for data acquisition and effective information extraction that meets the needs of small-scale drone monitoring of pine wilt disease.

[0207] (2) Based on the time series characteristic data, the time problem of early disease identification of pine wood nematode disease was solved by studying the "time-space-spectrum" fusion algorithm and corresponding graph characteristics of UAV remote sensing data.

[0208] (3) Research the methods of “spatial spectrum feature extraction” and “spatial spectrum classification framework construction” to realize the automatic identification of damaged trees and solve the problem of accurate positioning of early disease identification, thereby providing a basis for timely and effective disaster prevention and control.

[0209] The present invention constructs disease identification feature maps based on UAV hyperspectral data, determines the time point for early disease identification, and constructs an early damaged wood automatic positioning system based on feature map extraction and space-spectrum classification based on the "time-space-spectrum" fusion algorithm, ultimately realizing the effective application of UAV imaging hyperspectral data in the identification of pine wood nematode disease.

[0210] The following is a system embodiment corresponding to the above method embodiment. This embodiment can be implemented in conjunction with the above embodiment. The relevant technical details mentioned in the above embodiment are still valid in this embodiment. In order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiment.

[0211] The present invention also proposes a system for identifying and locating wood damaged by pine wood nematode disease, wherein the method comprises:

[0212] The original image acquisition module is used to acquire the original hyperspectral image of the vegetation in the monitoring area through the aerial hyperspectral imaging sensor, wherein the vegetation in the monitoring area includes healthy vegetation and multiple levels of harmful trees;

[0213] A preprocessing module is used to match the corresponding disease level for the original hyperspectral image of each damaged tree to obtain a damaged tree image with the disease level;

[0214] A training module, used to train a feature spectrum recognition model using the damaged wood image as training data, and use the trained feature spectrum recognition model as a disease recognition feature spectrum model;

[0215] A multi-period original hyperspectral image acquisition module is used to periodically acquire original hyperspectral images of vegetation in the monitoring area through the aerial hyperspectral imaging sensor to form multi-period hyperspectral images;

[0216] A disease identification characteristic spectrum acquisition module is used to input the multi-period hyperspectral image into the disease identification characteristic spectrum model to identify the disease level of each harmful tree in the monitoring area and its corresponding disease identification characteristic spectrum;

[0217] The time series feature determination module is used to extract the time series features of each disease identification feature map, and perform time series prediction analysis on it to determine the early identification time of each damaged wood grade;

[0218] A characteristic spectrum data set determination module is used to extract characteristic spectrum data corresponding to the identification time of the damaged tree according to the disease identification characteristic spectrum corresponding to each disease level, and obtain a time-space-spectrum characteristic spectrum data set for early identification of damaged trees;

[0219] A positioning module is used to classify the temporal-spatial-spectral characteristic map data set to locate early-stage damaged trees.

[0220] In the system for identifying and locating wood damaged by pine wilt disease, the pretreatment module is used to:

[0221] Obtain the locations of hazardous trees of different levels within the monitoring area;

[0222] The original hyperspectral images of the damaged trees of different grades in each period are spliced, terrain corrected and spectrally corrected to obtain hyperspectral preprocessed images;

[0223] According to the positioning of the different levels of hazardous trees, the hyperspectral preprocessed image is labeled to obtain a pixel label data set of the hyperspectral image based on time series;

[0224] Extracting single tree crown image data of trees with different levels of damage from the pixel label data set of the time-series-based hyperspectral image;

[0225] The single tree crown image data of the trees with different levels of damage are used as damaged tree images representing different disease levels.

[0226] In the system for identifying and locating wood damaged by pine wilt disease, the training module is used to:

[0227] Determine the physiological and biochemical parameters of damaged wood of different grades, including pigment content, water content, transpiration rate and photosynthetic index, etc.;

[0228] Analyzing the physiological and biochemical parameters of the damaged wood of different disease levels according to the damaged wood images representing different disease levels, and determining the optimal quantitative parameters representing the damaged wood of different disease levels;

[0229] Extracting a regional average spectrum of the original hyperspectral image;

[0230] According to the optimal quantitative parameters of the damaged wood representing different disease levels, the sensitive bands of the regional average spectrum are screened to obtain single-band image data of the sensitive bands;

[0231] Determining the spectral index of plant diseases according to the single-band image data;

[0232] Determining image geometric information according to the single-band images of different disease levels;

[0233] Analyze the correlation between the single-band image, the spectral index and the image geometric information according to the physiological and biochemical parameters characterizing the disease level, and obtain disease identification feature maps of different disease levels;

[0234] The damaged wood images representing different disease levels are used as input, and the disease identification feature maps of different disease levels are used as output to train the feature map recognition model.

[0235] In the system for identifying and locating wood damaged by pine wilt disease, step 8 comprises:

[0236] The optimal classification algorithm is used to extract the time-spectrum features and space-spectrum features of the time-space-spectrum feature map data set for early identification of harmful trees; the location of early harmful trees is obtained based on the time-spectrum features and the space-spectrum features.

[0237] In the system for identifying and locating wood damaged by pine wilt disease, the positioning module is used to:

[0238] The time-space-spectrum characteristic map data set for early identification of harmful trees is classified based on the spectrum through the optimal classification algorithm to obtain the image to be located after pixel-by-pixel spectrum classification; the image to be located after pixel-by-pixel spectrum classification is segmented to obtain the location of the early harmful trees.

[0239] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0240] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for identifying and locating wood damaged by pine wood nematode disease, characterized in that: The method comprises: Step 1: using an aerial hyperspectral imaging sensor to obtain an original hyperspectral image of vegetation in a monitoring area, wherein the vegetation in the monitoring area includes healthy vegetation and multiple levels of hazardous trees; Step 2: Match the original hyperspectral image of each damaged tree with the corresponding disease level to obtain a damaged tree image with the disease level; Step 3: using the damaged wood image as training data, training a feature spectrum recognition model, and using the trained feature spectrum recognition model as a disease recognition feature spectrum model; Step 4: The aerial hyperspectral imaging sensor periodically acquires original hyperspectral images of vegetation in the monitoring area to form multi-period hyperspectral images; Step 5: input the multi-period hyperspectral image into the disease identification characteristic spectrum model to identify the disease level of each harmful tree in the monitoring area and its corresponding disease identification characteristic spectrum; Step 6: extract the time series characteristics of each disease identification feature map, and perform time series prediction analysis on it to determine the early identification time of each damaged wood grade; Step 7: extract the characteristic spectrum data corresponding to the identification time of the damaged tree according to the disease identification characteristic spectrum corresponding to each disease level, and obtain the time-space-spectrum characteristic spectrum data set for early identification of damaged trees; Step 8: classifying the temporal-spatial-spectral characteristic map dataset to locate early-stage damaged trees; The step 3 comprises: determining the physiological and biochemical parameters of the damaged wood of different grades; Analyze the physiological and biochemical parameters of the damaged wood of different disease levels according to the images of the damaged wood of different disease levels, and determine the optimal quantitative parameters of the damaged wood of different disease levels; Extracting a regional average spectrum of the original hyperspectral image; According to the optimal quantitative parameters of the damaged wood representing different disease levels, the sensitive bands of the regional average spectrum are screened to obtain single-band image data of the sensitive bands; Determining the spectral index of plant diseases according to the single-band image data; Determining image geometric information according to the single-band images of different disease levels; The correlation between the single-band image, the spectral index and the image geometric information is analyzed according to the physiological and biochemical parameters characterizing the disease level to obtain disease identification feature maps of different disease levels.

2. The method for identifying and locating wood damaged by pine wilt disease according to claim 1, characterized in that: The step 2 comprises: Obtain the locations of hazardous trees of different levels within the monitoring area; The original hyperspectral images of the damaged trees of different grades in each period are spliced, terrain corrected and spectrally corrected to obtain hyperspectral preprocessed images; According to the positioning of the different levels of hazardous trees, the hyperspectral preprocessed image is labeled to obtain a pixel label data set of the hyperspectral image based on time series; Extracting single tree crown image data of trees with different levels of damage from the pixel label data set of the time-series-based hyperspectral image; The single tree crown image data of the trees with different levels of damage are used as damaged tree images representing different disease levels.

3. The method for identifying and locating wood damaged by pine wilt disease according to claim 1, characterized in that: The physiological and biochemical parameters in step 3 include pigment content, water content, transpiration rate and photosynthetic index; The step 3 comprises: taking the damaged wood images representing different disease levels as input and taking the disease identification feature maps of different disease levels as output, training the feature map recognition model.

4. The method for identifying and locating wood damaged by pine wilt disease according to claim 1, characterized in that: The step 8 comprises: The optimal classification algorithm is used to extract the time-spectrum features and space-spectrum features of the time-space-spectrum feature map data set for early identification of harmful trees; the location of early harmful trees is obtained based on the time-spectrum features and the space-spectrum features.

5. The method for identifying and locating wood damaged by pine wood nematode disease according to claim 1, characterized in that: The step 8 comprises: The time-space-spectrum characteristic map data set for early identification of harmful trees is classified based on the spectrum through the optimal classification algorithm to obtain the image to be located after pixel-by-pixel spectrum classification; the image to be located after pixel-by-pixel spectrum classification is segmented to obtain the location of the early harmful trees.

6. A system for identifying and locating wood damaged by pine wilt disease, characterized in that: The system comprises: The original image acquisition module is used to acquire the original hyperspectral image of the vegetation in the monitoring area through the aerial hyperspectral imaging sensor, wherein the vegetation in the monitoring area includes healthy vegetation and multiple levels of harmful trees; A preprocessing module is used to match the corresponding disease level for the original hyperspectral image of each damaged tree to obtain a damaged tree image with the disease level; A training module, used to train a feature spectrum recognition model using the damaged wood image as training data, and use the trained feature spectrum recognition model as a disease recognition feature spectrum model; A multi-period original hyperspectral image acquisition module is used to periodically acquire original hyperspectral images of vegetation in the monitoring area through the aerial hyperspectral imaging sensor to form multi-period hyperspectral images; A disease identification characteristic spectrum acquisition module is used to input the multi-period hyperspectral image into the disease identification characteristic spectrum model to identify the disease level of each harmful tree in the monitoring area and its corresponding disease identification characteristic spectrum; The time series feature determination module is used to extract the time series features of each disease identification feature map, and perform time series prediction analysis on it to determine the early identification time of each damaged wood grade; A characteristic spectrum data set determination module is used to extract characteristic spectrum data corresponding to the identification time of the damaged tree according to the disease identification characteristic spectrum corresponding to each disease level, and obtain a time-space-spectrum characteristic spectrum data set for early identification of damaged trees; A positioning module, used for classifying the temporal-spatial-spectral characteristic map data set to locate early-stage damaged trees; The training module is also used to: determine the physiological and biochemical parameters of different grades of harmful wood; Analyze the physiological and biochemical parameters of the damaged wood of different disease levels according to the images of the damaged wood of different disease levels, and determine the optimal quantitative parameters of the damaged wood of different disease levels; Extracting a regional average spectrum of the original hyperspectral image; According to the optimal quantitative parameters of the damaged wood representing different disease levels, the sensitive bands of the regional average spectrum are screened to obtain single-band image data of the sensitive bands; Determining the spectral index of plant diseases according to the single-band image data; Determining image geometric information according to the single-band images of different disease levels; The correlation between the single-band image, the spectral index and the image geometric information is analyzed according to the physiological and biochemical parameters characterizing the disease level to obtain disease identification feature maps of different disease levels.

7. The system for identifying and locating wood damaged by pine wilt disease according to claim 6, characterized in that: The pre-processing module is used for: Obtain the locations of hazardous trees of different levels within the monitoring area; The original hyperspectral images of the damaged trees of different grades in each period are spliced, terrain corrected and spectrally corrected to obtain hyperspectral preprocessed images; According to the positioning of the different levels of hazardous trees, the hyperspectral preprocessed image is labeled to obtain a pixel label data set of the hyperspectral image based on time series; Extracting single tree crown image data of trees with different levels of damage from the pixel label data set of the time-series-based hyperspectral image; The single tree crown image data of the trees with different levels of damage are used as damaged tree images representing different disease levels.

8. The system for identifying and locating wood damaged by pine wilt disease according to claim 6, characterized in that: The physiological and biochemical parameters in the training module include pigment content, water content, transpiration rate and photosynthetic index, etc.; The training module is also used to: use the damaged wood images representing different disease levels as input and the disease identification feature maps of different disease levels as output to train the feature map recognition model.

9. The system for identifying and locating wood damaged by pine wilt disease according to claim 6, characterized in that: The step 8 comprises: The optimal classification algorithm is used to extract the time-spectrum features and space-spectrum features of the time-space-spectrum feature map data set for early identification of harmful trees; the location of early harmful trees is obtained based on the time-spectrum features and the space-spectrum features.

10. The system for identifying and locating wood damaged by pine wilt disease according to claim 6, characterized in that: The positioning module is used for: The time-space-spectrum characteristic map data set for early identification of harmful trees is classified based on the spectrum through the optimal classification algorithm to obtain the image to be located after pixel-by-pixel spectrum classification; the image to be located after pixel-by-pixel spectrum classification is segmented to obtain the location of the early harmful trees.

Images