Method for early and accurate identification of pinewood nematode disease epidemic wood

Abstract

The application provides a method for early accurate identification of pine wilt disease, relates to the technical field of processing and analysis of image data, sample chlorophyll and water content data, spectral data and the like, and comprises the following steps: based on obtained unmanned aerial vehicle hyperspectral remote sensing data, chlorophyll content data of standard needle leaf samples, water content data and ASD hyperspectral data, sensitivity analysis is carried out by using a SPA algorithm in combination with physiological characteristic parameters such as classic vegetation index, pine pigment data and water content data, optimal pigment and water content index data are screened, and water content data-pigment data collaborative indexes are adopted, a plurality of indexes and a plurality of data including image data, sample chlorophyll and water content data, spectral data and the like are utilized, and the data are processed and analyzed, so that early accurate identification of pine wilt disease is realized.

Description

Technical Field

[0001] This application relates to the technical field of early and accurate identification of pests and diseases, specifically to the technical field of processing and analyzing image data, sample chlorophyll and moisture content data, spectral data, etc., and more specifically to a method for early and accurate identification of pine wilt disease-infected trees using the aforementioned data, including image data, sample chlorophyll and moisture content data, spectral data, etc. Background Technology

[0002] Pine wilt disease is a forestry pest that causes pine trees to gradually change color in the needles, lose resin flow, and eventually wither and die within a few months. Studies have shown that pine wilt disease has become one of the most dangerous forest biological disasters in my country, severely damaging pine forest resources, endangering forest ecosystem security, and significantly impacting import and export trade. Currently, there is no effective method to completely eradicate this disease. Timely removal of infected trees after identification remains the most effective way to control the disease. Therefore, in the current context, solving the problem of accurate monitoring of pine wilt disease, focusing on developing scientific and effective technical methods for monitoring infected trees in the early stages, and constructing a comprehensive pest monitoring and identification system are crucial for improving the overall control capacity of pine wilt disease. In particular, accurate early identification of infected trees is essential for taking timely measures to improve the overall control capacity of pine wilt disease.

[0003] Previous studies, based on the analysis of ground-based hyperspectral data, have shown that when pine trees are infected by pine wilt nematodes and undergo color changes, they exhibit typical spectral characteristics such as higher red band reflectance than green band reflectance, decreased slope, and a blue shift at the red edge. This laid the theoretical foundation for remote sensing monitoring of pine wilt nematode-infected trees. UAV remote sensing imagery offers higher spatiotemporal resolution than satellite remote sensing imagery and is more efficient than traditional ground-based surveys; therefore, UAV remote sensing technology is widely used in current pine wilt nematode monitoring. Methodologically, monitoring methods are mainly divided into two categories: those based on classical machine learning and those based on deep learning. The former involves constructing data features through various methods such as building spectral feature indices and vegetation indices, and extracting texture information, followed by selecting classical classifiers such as SVM, random forests, and conditional random fields to identify infected trees. The latter involves constructing a deep learning model based on convolutional neural networks, training the model on standard samples, and ultimately extracting and identifying pine wilt nematode-infected trees.

[0004] However, current methods for monitoring pine wilt disease-infected trees using UAV remote sensing mainly focus on improving the accuracy of disease identification by optimizing the model network structure and improving samples. They do not fully leverage the advantages of sensitive spectral features in disease identification, resulting in low accuracy in identifying the diseased stage of trees. At the same time, the monitoring relies more on indicators that characterize the pigment content of the pine canopy, with less application of data such as canopy moisture content.

[0005] Therefore, this invention combines the pathogenic mechanism of pine wilt disease with spectral characteristic index data analysis, providing a feasible technology for the accurate monitoring of pine wilt disease-infected trees based on UAV hyperspectral remote sensing. Specifically, this invention utilizes multiple indicators and data, including image data, sample chlorophyll and water content data, and spectral data, to achieve early and accurate identification of pine wilt disease-infected trees. Summary of the Invention

[0006] The purpose of this invention is to provide a method for early and accurate identification of pine wilt disease-infected trees. By utilizing multiple indicators and data, including image data, sample chlorophyll and water content data, spectral data, etc., and processing and analyzing these data, the method can achieve early and accurate identification of pine wilt disease-infected trees, and further improve the accuracy of identifying pine wilt disease-infected trees based on high-resolution remote sensing data.

[0007] The technical solution adopted in this invention is as follows:

[0008] S1. Acquire high-resolution UAV hyperspectral remote sensing images of the detection area, as well as chlorophyll content data, water content data, and ASD hyperspectral data of standard conifer samples.

[0009] S2. Screening of sensitive bands for pine wilt disease based on the SPA band selection algorithm;

[0010] Furthermore, step S2, based on the SPA band selection algorithm, filters sensitive bands for pine wilt disease, specifically including the following steps:

[0011] S21. Spectral data preprocessing: Outlier removal and mean normalization are performed on the raw ASD hyperspectral data, and the sample spectral data are summarized and organized according to different disease susceptibility.

[0012] S22. Sensitive band selection: Based on the SPA algorithm, variables that retain the original information to the greatest extent are selected through projection transformation operation, and the optimal sensitive band is extracted by band iteration method.

[0013] S3. Construct spectral indices based on differential normalization, and conduct correlation analysis with classical vegetation indices and physiological characteristic parameters such as pine pigment data and water content data to screen the best pigment and water content index data.

[0014] Furthermore, step S3 involves constructing spectral indices using differential normalization, performing correlation analysis with classical vegetation indices and physiological characteristic parameters such as pine pigment data and water content data, and selecting the optimal pigment and water content index data. This process specifically includes the following steps:

[0015] S31. Construction of candidate index set: Based on the sensitive bands selected in step 2, construct the spectral index DVI using the differential normalization method; select the classic vegetation index used for monitoring pine wilt disease, and combine it with the newly constructed spectral index to form the candidate index set.

[0016] S32. Optimal Spectral Index Screening: Combining field surveys and UAV hyperspectral remote sensing imagery, the correlation between each spectral index in the candidate index set and pine pigment and water content was analyzed one by one. Based on the correlation coefficient R value, the spectral indices CI and WASCOSBNDI were selected as the optimal pigment and water content indices for modeling analysis.

[0017] Furthermore, the formula for constructing the difference normalization index in step S31 is as follows:

[0018]

[0019] In the formula, B1 and B2 represent two different bands in the extracted sensitive bands;

[0020] The expressions for the spectral indices CI and WASCOSBNDI in step S32 are as follows:

[0021] CI=(ρ 850 -ρ 710 ) / (ρ 850 +ρ 680 );

[0022] WASCOSBNDI=(ρ 531 -ρ 570 ) / (ρ 531 +ρ 570 );

[0023] In the formula: ρ850, ρ710, ρ680, ρ531, and ρ570 represent the reflectance values ​​of the spectrum at the 850nm, 710nm, 680nm, 531nm, and 570nm wavelength bands, respectively.

[0024] S4. Construct an algorithm for monitoring and identifying the diseased stage of pine wood nematode-infected trees, and identify pine wood nematode-infected trees.

[0025] Furthermore, in step S4, an algorithm for monitoring and identifying the diseased stage of pine wood nematode-infected trees is constructed, and the pine wood nematode-infected trees are identified. Specific steps include:

[0026] After selecting the optimal pigment and moisture content indicators, the best classification threshold is obtained by measuring the inter-class separability based on the single indicator of pigment or moisture content to identify infected trees at different disease stages. The optimal combination threshold is obtained by using linear discriminant analysis based on the two-dimensional features of the pigment-moisture content synergistic indicator to identify infected trees at different disease stages. The optimal monitoring and identification algorithm for disease stages of pine wood nematode diseased standing trees is established by comparing the identification accuracy of infected trees with different indicators, with a focus on identifying infected trees in the early stage of infection.

[0027] CI is used to construct the Normalized Difference Vegetation Index. For vegetation, CI is based on the reflectance of red and green light bands, and its expression is:

[0028] CI for vegetation can also be related to chlorophyll content; it is constructed by the ratio or difference of reflectance in specific wavelength bands, based on formulas for near-infrared, red, and blue light.

[0029] In the formula, NIR represents near-infrared light, R represents red light, and B represents blue light.

[0030] In WASCOS BNDI, WASCOS can also be based on moisture content and spectral correlation; BNDI is correlated with biomass, near-infrared, red light, difference, and normalization, and its expression is:

[0031]

[0032] In the formula: WC is the moisture content, S is the spectrum, ST is a structural feature, BM is the biomass, NIR is the near-infrared, and R is the red light.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] This invention identifies the disease stage of pine wilt disease-infected trees based on UAV hyperspectral remote sensing image data and other data such as chlorophyll data, water content data, and spectral data. It uses high-resolution remote sensing image data combined with other data such as chlorophyll data, water content data, and spectral data to identify trees at different disease stages. It enhances sensitive features based on the synergistic index of pigment data and water content data, thereby enabling monitoring of disease stages of infected trees based on UAV hyperspectral remote sensing and focusing on identifying early-stage infected trees, achieving early and accurate identification of pine wilt disease-infected trees.

[0035] Compared with the traditional method of monitoring pine wilt disease by regular patrols by monitors, this invention can save a lot of manpower and resources, while covering a larger survey area and greatly improving work efficiency. Moreover, by using UAV hyperspectral remote sensing image data and other data such as sample chlorophyll data, water content data, spectral data, etc., to identify infected trees at an early stage, the accuracy of identification is improved and identification can be achieved as early as possible. Attached Figure Description

[0036] Figure 1 A flowchart illustrating a method for identifying the disease stage of pine wilt disease-infected trees based on UAV hyperspectral remote sensing imagery is shown.

[0037] Figure 2 A schematic diagram of the study area in Example 2 is shown.

[0038] Figure 3 A schematic diagram of different types of pine needle samples is shown.

[0039] Figure 4 A schematic diagram of the selection of different categories of pixel samples is shown.

[0040] Figure 5 The diagram shows the selection of sensitive bands and the evaluation of RMSE index for pine wilt disease.

[0041] Figure 6 A heatmap showing the correlation analysis between the pigment index set and water content index set constructed in the invention and pine pigments and water content is presented.

[0042] Figure 7 The correlation analysis diagram between the traditional pigment index set, the moisture content index set, and pine pigments and moisture content is shown.

[0043] Figure 8 The diagram shows the asynchronous changes in the optimal pigment index CI and the optimal moisture content index WASCOSBNDI at different stages of pine wilt disease.

[0044] Figure 9 The results of threshold segmentation based on CI, WASCOSBNDI, and CI-WASCOSBNDI metrics are shown in the figure.

[0045] Figure 10 The results of identifying the disease stage of pine wilt disease-infected trees based on CI, WASCOSBNDI, and CI-WASCOSBNDI indicators are shown in the figure. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1

[0048] like Figure 1 As shown in the figure, this embodiment discloses a method for early and accurate identification of pine wilt disease-infected trees, including the following steps:

[0049] S1. Acquire high-resolution UAV hyperspectral remote sensing image data, chlorophyll content data, water content data, and ASD hyperspectral data of standard conifer samples within the detection area;

[0050] S2. Screening of sensitive bands for pine wilt disease based on the SPA band selection algorithm;

[0051] S3. Construct spectral indices based on differential normalization, and conduct sensitivity analysis by combining existing classic vegetation indices with physiological characteristic parameters such as pine pigment and water content to screen the best pigment and water content index data.

[0052] S4. Construct an algorithm for monitoring and identifying the diseased stage of pine wood nematode-infected trees, and identify pine wood nematode-infected trees.

[0053] Furthermore, step S2, which involves screening for sensitive bands for pine wilt disease based on the SPA algorithm, specifically includes the following steps:

[0054] S21. Spectral data preprocessing: Outlier removal and mean normalization are performed on the raw ASD hyperspectral data, and the sample spectral data are summarized and organized according to different disease susceptibility.

[0055] S22. Sensitive band selection: Based on the SPA algorithm, variables that retain the original information to the greatest extent are selected through projection transformation operation, and the optimal sensitive band is extracted by band iteration method.

[0056] Furthermore, in step S3, a spectral index is constructed using differential normalization, and correlation analysis is performed with physiological characteristic parameters such as pine pigment data and water content data to screen for optimal pigment and water content index data. This specifically includes the following steps:

[0057] S31. Construction of candidate index set: Based on the sensitive bands selected in step 2, construct the spectral index DVI using the differential normalization method; select the classic vegetation index used for monitoring pine wilt disease, and combine it with the newly constructed index to form a candidate index set.

[0058] S32. Optimal Spectral Index Screening: Combining field surveys and UAV hyperspectral remote sensing imagery, the correlation between each spectral index in the candidate index set and pine pigment and water content was analyzed one by one. Based on the correlation coefficient R value, the spectral indices CI and WASCOSBNDI were selected as the optimal pigment and water content data indicators for modeling analysis.

[0059] Furthermore, the formula for constructing the difference normalization index in step S31 is as follows:

[0060]

[0061] In the formula, B1 and B2 represent two different bands in the extracted sensitive bands.

[0062] Furthermore, the expressions for the spectral indices CI and WASCOSBNDI in step S32 are as follows:

[0063] CI=(ρ 850 -ρ 710 ) / (ρ 850 +ρ 680 );

[0064] WASCOSBNDI=(ρ 531 -ρ 570 ) / (ρ 531 +ρ 570 );

[0065] In the formula: ρ850, ρ710, ρ680, ρ531, and ρ570 represent the reflectance values ​​of the spectrum at the 850nm, 710nm, 680nm, 531nm, and 570nm wavelength bands, respectively.

[0066] Further, in step S4, an algorithm for monitoring and identifying the diseased stage of pine wilt diseased trees is constructed, and pine wilt diseased trees are identified. Specific steps include:

[0067] After determining the optimal pigment data and moisture content data indicators, the best classification threshold is obtained by measuring the inter-class separability based on the single indicator of pigment data or moisture content data to identify infected trees at different disease stages. The optimal combination threshold is obtained by using linear discriminant analysis based on the two-dimensional features of the combined indicators of pigment data and moisture content data to identify the disease stage. The optimal algorithm for identifying the disease stage of pine wood nematode-infected trees is determined by comparing the identification accuracy of different indicators.

[0068] CI is used to construct the Normalized Difference Vegetation Index. For vegetation, CI is based on the reflectance of red and green light bands, and its expression is:

[0069] CI for vegetation can also be related to chlorophyll content; it is constructed by the ratio or difference of reflectance in specific wavelength bands, based on formulas for near-infrared, red, and blue light.

[0070] In the formula, NIR represents near-infrared light, R represents red light, and B represents blue light.

[0071] In WASCOS BNDI, WASCOS can also be based on moisture content and spectral correlation; BNDI is correlated with biomass, near-infrared, red light, difference, and normalization, and its expression is:

[0072]

[0073] In the formula: WC is the moisture content, S is the spectrum, ST is a structural feature, BM is the biomass, NIR is the near-infrared, and R is the red light.

[0074] Example 2

[0075] This example uses Fushun City, Liaoning Province as the study area.

[0076] Fushun City is located in the mid-latitudes (E124°13′06″, N41°56′15″). The study area has a temperate continental monsoon climate, characterized by hot and rainy summers and dry and low-rainfall winters. The average annual temperature is 6.8℃, and the average annual precipitation is 789.5 mm. The study area is a plain with a generally gentle and relatively flat terrain. The area is rich in forest resources, mainly consisting of artificially planted Korean pine (Pinus koraiensis Siebold & Zuccarini), Chinese pine (Pinus buliformis Carriere), and North China larch (Larix gmelinii var. principis-rupprechtii (Mayr) Pilger). The main soil types include brown soil, dark brown soil, and meadow soil. Due to incomplete control of infected trees, coupled with the presence of host plants (pine trees) and host insects (pine sawyer beetles) and a suitable living environment, pine wilt disease has not been completely eradicated in Fushun City.

[0077] This example selected two typical areas, the forest area near Dahuo Reservoir and Hada Forest Farm, for sample collection and physicochemical parameter determination. At the same time, standard pixel samples were selected for analysis of sensitive spectral characteristics and verification of the identification results of pine disease stages.

[0078] The experimental case in this embodiment was conducted in the study area from July 5, 2023 to October 10, 2023, as part of a field observation experiment on pine wilt disease.

[0079] In this embodiment, the hyperspectral imagery was acquired using a DJI M300 drone equipped with an X20P hyperspectral camera. The hyperspectral remote sensing imagery acquisition was conducted in a clear, windless environment from 11:00 AM to 2:00 AM. The experiment was set at a flight altitude of 110 meters, ultimately obtaining hyperspectral images with a spatial resolution of 2.73 centimeters and a spectral resolution of 4 nanometers, which meets the monitoring requirements for pixel-level index analysis.

[0080] The acquired remote sensing images were verified and found to cover the actual sampling survey area well. Based on the above operations, usable UAV hyperspectral remote sensing images were finally obtained for spectral index analysis.

[0081] The first step of this invention was data collection. On July 5, 2023, field investigations were conducted at two locations. Geographic coordinates were recorded at each location, and corresponding pine needle samples were collected. Key physicochemical parameters such as chlorophyll a / b ratio, xanthophyll content, and moisture content were measured. Since there is currently no unified standard for classifying the disease stage of pine wilt disease-infected trees, based on the field survey and a summary of existing classification systems, the obtained pine needle samples were divided into two types: healthy and discolored. A total of 173 valid sample spectra were obtained, including 75 healthy pine needle samples and 98 discolored pine needle samples. Figure 3 (a)-(c) show pine needles collected on-site. Figure 3 (d) shows the average spectral curve of the corresponding pine needle sample. The spectral and physicochemical parameters of the collected samples were integrated for subsequent analysis of sensitive bands for pine wilt disease.

[0082] Meanwhile, to explore the changes of typical indicators at different stages of pine wilt disease development, we constructed a pixel sample set, including three categories: healthy, infected with discoloration, and infected without discoloration. Among them, infected without discoloration pixels were selected using a retrospective approach: for a given pine tree, by comparing UAV remote sensing images taken in July, August, and October, if it appeared green in the July image but red or yellowish-green in August or October, these pine trees were defined as early-stage infected trees without discoloration, and infected without discoloration pixels were extracted for analysis. A total of 150 sample pixels (50 samples in each category) were labeled for monitoring indicator screening and subsequent identification of pine wilt disease infection stages.

[0083] Figure 4 The study showcases standard healthy pixel samples, diseased undiscolored pixel samples, and diseased discolored pixel samples, based on the results of field surveys at Hada Forest Farm and the characterization of pixels in hyperspectral images.

[0084] The second step of this invention is to extract sensitive bands for pine wilt disease based on the SPA band selection algorithm: (1) perform spectral data preprocessing and category classification; (2) extract sensitive bands for pine wilt disease based on the SPA algorithm. Data preprocessing is mainly to eliminate the influence of outliers and invalid bands on the extraction of sensitive bands, while the SPA algorithm-based selection aims to extract sensitive bands with the least redundancy and highest accuracy while considering time costs. Finally, the extracted sensitive bands are applied to the subsequent construction of the spectral index set.

[0085] like Figure 5 As shown, the distribution of sensitive bands extracted based on the SPA algorithm is as follows: the visible light region includes 545nm; the red-edge region includes 682nm, 735nm, and 762nm; the near-infrared region includes 864nm, 982nm, and 1052nm; and the short-wave infrared region includes 1124nm, 1240nm, 1245nm, 1350nm, 1352nm, 1441nm, 1443nm, and 1983nm. Overall, the RMSE value tends to optimize as the number of selected bands increases. Iterative calculations using the SPA algorithm show that the highest analysis accuracy is achieved when the final number of selected bands is 15, at which point the RMSE is 0.1320. Since the band range of UAV hyperspectral remote sensing imagery is 400-1000nm, we selected six bands—545nm, 682nm, 735nm, 762nm, 864nm, and 982nm—covering the visible light, red-edge, and near-infrared regions to construct the differentially normalized spectral index.

[0086] The third step of this invention is the screening of monitoring indicators and the determination of classification thresholds: (1) constructing a candidate indicator set; (2) screening the optimal pigment and water content indicators; and (3) determining the classification thresholds based on inter-class distance separability measurement and linear discriminant analysis. The purpose of indicator screening is to compare and analyze the newly constructed indicators with traditional indicators to identify the characteristic indicators most sensitive to changes in pine canopy pigment and water content. The classification thresholds are used to determine the criteria for different disease stages and to construct a disease stage identification model for pine wilt disease.

[0087] For the six sensitive bands extracted based on the SPA algorithm, we constructed 36 spectral indices using differential normalization, as shown in the following formula:

[0088]

[0089] In the formula, B1 and B2 represent two different bands in the extracted sensitive bands.

[0090] Subsequently, based on measured chlorophyll content data, water content data, and other key physicochemical parameters, we generated a heatmap through correlation analysis. Figure 6The DVI (682,735), DVI (682,864), and DVI (735,864) with the highest coefficients of determination were selected and added to the chlorophyll candidate index set, while DVI (545,982), DVI (682,982), and DVI (545,682) were selected and added to the water content candidate index set. Then, classic vegetation indices that have been proven effective in pine wilt disease monitoring studies were selected for physiological parameter correlation analysis. Figure 7 Three indicators, CI, NDVI, and REPI, were selected and added to the chlorophyll candidate indicator set. Three indicators, WASCOSBNDI, SOSNBDI, and NDWI, were selected and added to the water content candidate indicator set. The candidate traditional vegetation indices are shown in Tables 1 and 2.

[0091] Table 1. Vegetation indices characterizing pine pigments

[0092]

[0093]

[0094] Table 2. Vegetation indices representing the water content of pine trees.

[0095]

[0096] Trend analysis was performed on the constructed pigment index set (DVI(682,735), DVI(682,864), DVI(735,864), CI, NDVI, REPI) and the water content index set (DVI(545,682), DVI(545,982), DVI(682,982), SAPSNBDI, WASCOSBNDI, NDWI). The mean center of the box plots for each stage was connected, and the trend was simulated using the slope of the broken lines at different stages. In the early stages of pine wilt disease infection, the appearance color of pine trees generally does not change significantly. However, because pine wilt nematodes block the internal vascular bundles of pine trees, severely hindering water transport, changes in pine water content are more sensitive than pigment changes in the early stages. Based on the asynchronous change characteristics of the two indicators, the changes of each indicator in the candidate index set in the three stages of healthy-infected (no discoloration)-infected were analyzed. It was found that the pigment index CI and the water content index WASCOSBNDI can best simulate this physiological process. Figure 8 Among them, the CI index hardly changed during the transition from the healthy to the disease-pre-diseased stage, but dropped sharply in the middle and late stages of the disease. The WASCOSBNDI index had already changed significantly during the transition from the healthy to the disease-pre-diseased stage, and also showed a sharp downward trend in the middle and late stages of the disease. Therefore, these two indicators were selected for the subsequent construction of the disease stage monitoring and identification model.

[0097] Figure 9 This involves using labeled standard pixel samples to calculate corresponding index values ​​and then plotting a scatter plot. Based solely on the CI index for identifying the disease stage of infected trees, the classification threshold for healthy and infected trees without discoloration is 0.52, while the classification threshold for infected trees without discoloration and infected trees with discoloration is 0.67. Figure 9 a) Based solely on the WASCOSBNDI index for identifying the disease stage of infected trees, the classification threshold for healthy and infected trees without discoloration is 0.023, while the classification threshold for infected trees without discoloration and infected trees with discoloration is -0.048. Figure 9 b); Based on the 95% confidence interval ellipse drawn, it can be seen that there are significant differences in the index values ​​of the three different disease stages. Using LDA linear discriminant analysis, when the single-pixel index WASCOSBNDI > 0.019 and CI > 0.52, it is judged as a healthy state; when the single-pixel index WASCOSBNDI < 0.019 and CI > 0.52, it is judged as a diseased, non-discolored state; and when the single-pixel index WASCOSBNDI < 0.019 and CI < 0.52, it is judged as a diseased, discolored state. Figure 9 c).

[0098] The fourth step of this invention is the construction and accuracy evaluation of a model for identifying the susceptibility stage of pine wilt disease. Using the obtained threshold values ​​as the classification criteria for the susceptibility stage of infected trees, pixel-by-pixel classification is performed. Based on this, the pixel-level results are converted into target-level results using the following method: Within a single pine tree target, if only healthy pixels exist, it is classified as a healthy pine tree; if there are infected but undiscolored pixels and no infected but discolored pixels, it is classified as infected but undiscolored tree; if infected but discolored pixels exist, it is classified as infected but discolored tree. Based on this standard, an object-level model for identifying the susceptibility stage of pine wilt disease-infected trees is established. To further verify the model's identification effect, this invention evaluates the proposed susceptibility stage identification method.

[0099] This invention employs a standardized accuracy verification method to evaluate the identification results of infected trees. Based on the general method for evaluating the accuracy of binary classification, three basic indicators are statistically analyzed for the identification results based on different metrics: the number of correct detections (True Positive / TP), the number of false detections (Commission / False Positive / FP), and the number of missed detections (Omission / False Negative / FN). TP indicates that the model's identification of pine trees matches the field survey results; FP indicates that the model's identification of pine trees does not match the field survey results; and FN indicates that the actual ground survey results were not effectively identified by the model. After obtaining the corresponding statistical results, according to the general accuracy evaluation method, three accuracy indicators are selected to evaluate the model's identification effect on infected trees at the diseased stage: producer's accuracy (Recall), user's accuracy (Precision), and overall accuracy (F1 score).

[0100] This invention compares the identification effects of three indicators—CI, WASCOSBNDI, and CI-WASCOSBNDI—on the disease-affected stages of trees. The verification results are shown in Tables 3 and 4. Compared with identification based on single indicators such as CI and WASCOSBNDI, the synergistic indicator CI-WASCOSBNDI shows better identification accuracy, with an overall identification accuracy of 92.78%, which is significantly better than the WASCOSBNDI indicator (89.31%) and the CI indicator (81.82%).

[0101] Table 3. Accuracy Verification Table for CI Index-Based Recognition

[0102]

[0103] Table 4. Accuracy Verification Table Based on WASCOSBNDI Index Recognition

[0104]

[0105] Table 5. Accuracy Verification Table for CI-WASCOSBNDI Collaborative Index Recognition

[0106]

[0107] This invention further explored and analyzed the model's ability to identify early-stage infected trees. Combining field surveys and multi-temporal image comparisons, it was found that infected trees in the experimental area that did not change color were specifically divided into two categories: those that changed color by August and those that changed color by October. Table 6 shows the identification results of early-stage infected trees based on the CI-WASCOSBNDI synergistic index. The identification effect for infected trees that changed color by August was relatively ideal, with an identification accuracy of 83.3%; however, it was difficult to identify infected trees that changed color by October, with an identification accuracy of only 16.7%.

[0108] Table 6. Comparison of identification accuracy of infected but undiscolored diseased wood based on CI-WASCOSBNDI synergistic index.

[0109]

[0110] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for early and accurate identification of pine wilt-infected wood, characterized in that: Includes the following steps: S1. Acquire high-resolution UAV hyperspectral remote sensing image data, chlorophyll content data, water content data, and ASD hyperspectral data of standard conifer samples within the detection area; S2. Screening of sensitive bands for pine wilt disease based on the SPA band selection algorithm; S3. Construct spectral indices based on differential normalization, and perform sensitivity analysis by combining existing classic vegetation indices with pine pigment data, water content data, and physiological characteristic parameters to screen for optimal pigment and water content index data; specifically including the following steps: S31. Construction of Candidate Index Set: Based on the sensitive bands selected in step S2, the spectral index DVI is constructed using differential normalization. A classic vegetation index used for pine wilt disease monitoring is selected and combined with the newly constructed index to form the candidate index set. The formula for constructing the differential normalized spectral index is: ； In the formula, B1 and B2 represent two different bands in the extracted sensitive bands; S32. Optimal Spectral Index Selection: Combining field surveys and UAV hyperspectral remote sensing image data, the correlation between each spectral data point in the candidate index set and the pine pigment and moisture content data was analyzed one by one. Based on the correlation coefficient R value, the spectral indices CI and WASCOSBNDI were selected as the optimal pigment and moisture content indicators for modeling analysis. The expressions for the spectral indices CI and WASCOSBNDI are as follows: CI=(ρ 850 -r 710 ) / (ρ 850 +r 680 ); WASCOSBNDI=(ρ 531 -r 570 ) / (ρ 531 +r 570 ); In the formula: ρ850, ρ710, ρ680, ρ531, and ρ570 represent the reflectance values ​​of the spectrum at the 850nm, 710nm, 680nm, 531nm, and 570nm wavelength bands, respectively. S4. Construct an algorithm for monitoring and identifying pine wilt diseased trees at different disease stages, and identify pine wilt diseased trees to achieve early and accurate identification of pine wilt diseased trees. Specific steps include: After determining the optimal pigment and moisture content data indicators, the optimal classification threshold is obtained by measuring the inter-class separability based on the single pigment or moisture content data indicators to identify infected trees at different disease stages. The optimal combination threshold is obtained by using linear discriminant analysis based on the two-dimensional features of the pigment and moisture content data synergistic indicators to identify the disease stage. The optimal algorithm for identifying the disease stage of pine wood nematode-infected trees is determined by comparing the identification accuracy of different indicators.

2. The method for early and accurate identification of pine wilt-infected trees according to claim 1, characterized in that: Step S2 involves screening for sensitive bands for pine wilt disease based on the SPA algorithm, specifically including the following steps: S21. Spectral data preprocessing: Outlier removal and mean normalization are performed on the raw ASD hyperspectral data, and the sample spectral data are summarized and organized according to different disease susceptibility. S22. Sensitive band selection: Based on the SPA algorithm, variables that retain the original information to the greatest extent are selected through projection transformation operation, and the optimal sensitive band is extracted by band iteration method.

Images