Method and equipment for early-stage asymptomatic diagnosis of tobacco black shank

Abstract

The invention belongs to the technical field of tobacco black shank diagnosis, and particularly relates to a tobacco black shank early-stage asymptomatic diagnosis method and equipment, firstly, after wave band screening, 437 nm, 530 nm, 649 nm and 710 nm are finally determined as main sensitive wave bands of a tobacco black shank early-stage asymptomatic stage; and then combining the Dataset 1 and the Dataset 2, and constructing a Logistic regression model based on the four wavebands to obtain an early-stage symptomless diagnosis index of the tobacco black shank. Therefore, the method shows extremely high sensitivity to weak physiological changes in the early stage of black shank infection based on wave bands, and effective distinguishing between healthy plants and infected plants can be achieved by means of leaf spectrum details even if macroscopic decay symptoms do not appear in root systems and stem bases.

Description

Technical Field

[0001] This invention relates to the field of tobacco black shank diagnosis technology, specifically to a method and device for early asymptomatic diagnosis of tobacco black shank. Background Technology

[0002] Tobacco Black Shank Disease (TBS) is a highly destructive soil-borne disease caused by the plant pathogen Phytophthora nicotianae. Widely distributed in major tobacco-growing regions worldwide, it is a key factor affecting tobacco yield and quality. The disease primarily infects the roots and base of the stem, leading to root rot, blackening of the lower stem, and yellowing and eventual withering of the leaves, ultimately causing the entire plant to wilt and die. It has a short course, spreads rapidly, and recurs frequently. Outbreaks are particularly severe in hot and humid environments, often resulting in yield losses of 30%–80% in tobacco fields, and in severe cases, can destroy the entire field.

[0003] Currently, the control strategies for tobacco black shank mainly rely on soil-based pesticide treatment and planting disease-resistant varieties. However, these measures are usually implemented only after symptoms are obvious, potentially missing the optimal control window. Furthermore, current tobacco disease monitoring largely relies on manual field inspections and symptom observation. These methods are not only time-consuming and labor-intensive but also highly subjective, lack real-time accuracy, and pose a risk of cross-contamination through sampling. Therefore, achieving early, non-invasive, and rapid detection before visible symptoms appear on tobacco plants is a crucial scientific issue that urgently needs to be addressed in the field of tobacco disease monitoring and control.

[0004] Unlike traditional machine vision methods, hyperspectral imaging technology can capture changes in plant reflectance across different spectral bands, thereby reflecting subtle physiological and biochemical changes in plants during the early stages of pathogen infection, thus opening a new window for early disease diagnosis. For example, patent CN113435252A proposes a tobacco pest and disease early warning and monitoring system based on hyperspectral and remote sensing data, which achieves efficient early identification by selecting sensitive bands and establishing models; patent CN112697723A develops a method and system for predicting tobacco yield based on UAV hyperspectral imagery, and its spectral-UAV monitoring process is also of reference value for early asymptomatic identification of tobacco black shank; in addition, the tobacco disease identification and control system proposed in patent CN113962258A provides a technical approach for large-area, non-destructive, and rapid monitoring.

[0005] However, the high price and complex data processing requirements of current hyperspectral equipment limit its widespread application in large-scale field monitoring. Therefore, future research urgently needs to combine UAV hyperspectral remote sensing platforms with machine learning algorithms to achieve large-scale, rapid, and non-contact monitoring of tobacco fields while ensuring diagnostic accuracy. By extracting key spectral bands for the early stages of tobacco black shank and constructing predictive models, it is possible to shift disease monitoring from "symptom identification" to "potential lesion identification," thus providing solid technical support for the precise control of tobacco black shank. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] A method for early asymptomatic diagnosis of tobacco black shank includes the following specific steps:

[0008] S1, Training Data Acquisition: Acquire spectral data and standardized samples of tobacco black shank in the early asymptomatic stage to support subsequent modeling and analysis. The training data acquisition work is carried out by artificial inoculation experiment combined with hyperspectral acquisition and data preprocessing to obtain two datasets containing infected and healthy samples.

[0009] S2, Feature band selection for tobacco black shank: Based on the spectral data preprocessed by standard normal variable transformation, a competitive adaptive reweighted sampling algorithm was used for screening. First, a partial least squares regression model was constructed with 80% of the samples. The importance of bands was evaluated by the absolute value weight of the regression coefficients. Low-weight bands were gradually eliminated by combining the exponential decay function. The performance of candidate bands was evaluated by 10-fold cross-validation. Then, after 70 iterations, 60 and 77 feature bands were selected from the two datasets respectively. The intersection of the two datasets was taken as the final feature input for early asymptomatic identification of tobacco black shank.

[0010] S3, Early Asymptomatic Diagnostic Index for Tobacco Black Shank: First, based on the results of feature band screening, the sensitive bands for the early asymptomatic stage of tobacco black shank are determined; then, the two datasets are merged, and a Logistic regression model is constructed based on the sensitive bands to finally obtain the early asymptomatic diagnostic index for tobacco black shank.

[0011] As a preferred embodiment of the method for early asymptomatic diagnosis of tobacco black shank according to the present invention, the specific steps of S1 are as follows:

[0012] S11, Hyperspectral Data Acquisition:

[0013] S111, in an open-air experimental field, select Yunyan 116 tobacco plants with uniform growth and in the seedling stage to carry out an artificial inoculation experiment of tobacco blackleg;

[0014] S112, after inoculation, the plants were first allowed to continue to grow and be managed under natural field conditions. Then, on the 2nd and 4th day after inoculation, UAV hyperspectral data was collected. At the same time, UAV hyperspectral images were collected on the 2nd and 4th day after inoculation.

[0015] S113. After image acquisition, whiteboard correction was first performed to eliminate ambient light differences. Then, the tobacco plant area was manually labeled using Python's Labelme module to remove background and high-noise parts. Afterward, reflectance data of all bands in each labeled area were extracted by programming, and their average value was calculated as the spectral characteristics of a single sample. Among them, the data collected on the second day after vaccination was defined as Dataset 1, containing 495 infected samples and 320 healthy samples, and the data on the fourth day was defined as Dataset 2, containing 508 infected samples and 410 healthy samples.

[0016] S12, Data Preprocessing:

[0017] The standard normal transformation of spectral data is calculated using the following formula:

[0018]

[0019] in, n is the number of wavelengths. The wavelength reflectance is used; after standard normal transformation, the mean of the sample is 0 and the standard deviation is 1, thus ensuring comparability between different wavelength bands.

[0020] As a preferred embodiment of the method for early asymptomatic diagnosis of tobacco black shank disease described in this invention, the inoculation method adopts the stem wound inoculation method: firstly, a shallow wound is cut at the base of the tobacco seedling stem about 1-2 cm from the ground surface using scissors disinfected with alcohol; then, the pre-prepared Phytophthora nicotianae inoculum is inserted into the wound and covered with fine, moist soil to maintain local humidity and promote the colonization of the pathogen in the stem tissue.

[0021] As a preferred embodiment of the early asymptomatic diagnosis method for tobacco black shank disease described in this invention, the two samplings in S112 are both arranged between 11:00 and 12:00 noon on a sunny and breezy day to reduce the impact of changes in light.

[0022] As a preferred embodiment of the method for early asymptomatic diagnosis of tobacco black shank disease described in this invention, in step S112, the flight altitude of the UAV is set to 30m, which can ensure that the image resolution of the leaf and stem base area meets the analysis requirements, thereby achieving high-precision spectral acquisition of early asymptomatic plants.

[0023] As a preferred embodiment of the method for early asymptomatic diagnosis of tobacco black shank according to the present invention, the specific steps of S2 are as follows:

[0024] S21. Based on the spectral data preprocessed by standard normal variable transformation, a competitive adaptive reweighted sampling algorithm is used to screen out redundant or irrelevant bands and extract key feature bands for the early diagnosis of tobacco black shank. The specific process is as follows: 80% of the samples are randomly selected to establish a partial least squares regression model, the regression coefficients of each band are calculated, and the importance of the bands is evaluated by their absolute value weights; bands with lower weights are gradually eliminated using an exponential decay function to obtain a subset of candidate bands; the predictive performance of the candidate band set is then evaluated by 10-fold cross-validation; after 70 iterations, the band set with the best prediction accuracy is selected as the final feature bands.

[0025] S22, the results of the competitive adaptive reweighted sampling algorithm show that Dataset 1 retains 60 bands and Dataset 2 retains 77 bands, with a total of 38 bands. Then, a second round of competitive adaptive reweighted sampling algorithm optimization and screening is performed on these 38 bands. Finally, the intersection of the two screening bands is taken as the final feature input for the early asymptomatic identification of tobacco black shank, which is used for subsequent modeling and analysis.

[0026] As a preferred embodiment of the method for early asymptomatic diagnosis of tobacco black shank according to the present invention, the specific steps of S3 are as follows:

[0027] After band screening, S31 was finally determined to be 437 nm, 530 nm, 649 nm and 710 nm as the main sensitive bands in the early asymptomatic stage of tobacco black shank.

[0028] S32. After merging Dataset 1 and Dataset 2, a Logistic regression model is constructed based on these four bands to obtain the early asymptomatic diagnostic index for tobacco black shank.

[0029] A diagnostic device for early asymptomatic tobacco black shank includes four narrowband DT filters for acquiring reflected light in corresponding wavelengths and an integrated data processing chip to perform real-time calculation of wavelength reflectance, preprocessing, and automatic calculation of the early asymptomatic diagnostic index for tobacco black shank, thereby quickly outputting the infection status result of tobacco black shank.

[0030] Compared with existing technologies:

[0031] 1. Highly efficient identification of early spectral characteristics of tobacco black shank

[0032] This invention, through systematic analysis of the spectral response characteristics of tobacco leaves, successfully screened four key reflection bands: 437 nm, 530 nm, 649 nm, and 710 nm. These bands exhibit extremely high sensitivity to the subtle physiological changes in the early stages of black shank infection. Even before visible rot symptoms appear in the roots and stem base, healthy and infected plants can be effectively distinguished based on the details of the leaf spectrum. Experiments have shown that the diagnostic accuracy of this characteristic band combination is consistently above 83%, meeting the accuracy requirements for early field screening. Its application contributes to the "early detection and early prevention" of tobacco black shank, reducing the spread of the disease in field irrigation and soil, and improving control efficiency.

[0033] 2. Construct a concise and practical diagnostic index, BSEDI

[0034] Based on characteristic bands, a normalized early diagnostic index for tobacco black shank disease was established. This index, by quantifying the reflectance differences between key bands, can intuitively reflect the health status of tobacco plants: when the early diagnostic index value is greater than 0, the plant is determined to be infected with black shank disease; when the early diagnostic index value is less than 0, the plant is determined to be healthy. This method does not require complex spectral analysis or professional knowledge; operators can quickly make judgments based on the system's automatic output results, making it very suitable for grassroots agricultural technicians to promote and use in field conditions.

[0035] 3. Develop lightweight multispectral detection equipment to achieve efficient field monitoring.

[0036] Using 437 nm, 530 nm, 649 nm, and 710 nm as core wavelength bands, this invention designs and prototypes a dedicated multispectral detection device with a multi-lens single-camera structure. Equipped with a built-in processing chip, this device can perform real-time pixel-level reflectance preprocessing and calculate the early diagnostic index for tobacco black shank, automatically outputting a field disease distribution map. Compared to traditional hyperspectral imaging systems, this device maintains detection accuracy while reducing manufacturing costs by over 60%. It also boasts advantages such as small size, low power consumption, and high portability, making it easy to install on drones or mobile monitoring platforms. This device enables large-scale, real-time, and non-destructive monitoring of tobacco fields, providing solid technical support for regionalized early warning and precise control of tobacco black shank. Attached Figure Description

[0037] Figure 1 This is a graph showing the independent test accuracy of each model in this invention;

[0038] Figure 2 This is a graph showing the accuracy of the TMV early asymptomatic diagnosis index in two datasets. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0040] This invention provides a method for early asymptomatic diagnosis of tobacco black shank disease. Please refer to [link / reference]. Figures 1-2 The specific steps are as follows:

[0041] S1, Training Data Acquisition: Acquire spectral data and standardized samples of tobacco black shank in the early asymptomatic stage to support subsequent modeling and analysis. The training data acquisition work is carried out by artificial inoculation experiment combined with hyperspectral acquisition and data preprocessing to obtain two datasets containing infected and healthy samples.

[0042] The specific steps of S1 are as follows:

[0043] S11, Hyperspectral Data Acquisition:

[0044] S111, In an open-field experimental field, tobacco seedlings of Yunyan 116 with uniform growth were selected to conduct an artificial inoculation experiment for tobacco black shank disease (TBS). The inoculation method adopted was stem wound inoculation: First, a shallow wound was cut at the base of the tobacco seedling stem about 1-2 cm from the ground surface using scissors that had been disinfected with alcohol. Then, the pre-prepared Phytophthora nicotianae inoculum was inserted into the wound and covered with fine, moist soil to maintain local humidity and promote the colonization of the pathogen in the stem tissue.

[0045] After inoculation with S112, the plants were initially allowed to continue growing and managing under natural field conditions, maintaining a moist but not waterlogged field. Based on the typical course of black shank disease, the pathogen enters an incubation period approximately 4–6 days after infection, and around day 5, slight browning may appear at the base of some stems. To capture the spectral changes during the early asymptomatic stage, hyperspectral data were collected by drone on days 2 and 4 after inoculation. Combined with the typical disease progression of tobacco black shank, the pathogen enters an incubation period approximately 4–6 days after infection, and slight browning appears at the base of the stem around day 5. To capture the spectral changes of plants in the latent stage, UAV hyperspectral image acquisition was conducted on the 2nd and 4th days after inoculation. The hyperspectral camera had a spectral resolution of 3.5 nm. Both acquisitions were scheduled between 11:00 and 12:00 noon on a sunny and breezy day to minimize the impact of light variations. The UAV flight altitude was set at 30 m to ensure that the image resolution of the leaf and stem base areas met the analysis requirements, thus achieving high-precision spectral acquisition of early asymptomatic plants.

[0046] S113. After image acquisition, whiteboard correction was first performed to eliminate ambient light differences. Then, the tobacco plant area was manually labeled using Python's Labelme module to remove background and high-noise parts. Afterward, reflectance data of all bands in each labeled area were extracted by programming, and their average value was calculated as the spectral characteristics of a single sample. Among them, the data collected on the second day after vaccination was defined as Dataset 1, containing 495 infected samples and 320 healthy samples, and the data on the fourth day was defined as Dataset 2, containing 508 infected samples and 410 healthy samples.

[0047] S12, Data Preprocessing:

[0048] After whiteboard calibration, each sample contains reflectance information for 178 bands. To reduce light scattering and baseline drift effects, this invention performs Standard Normal Variation (SNV) processing on the spectral data, the calculation formula of which is as follows:

[0049]

[0050] in, n is the number of wavelengths. The wavelength reflectance is used; after standard normal transformation, the mean of the sample is 0 and the standard deviation is 1, thus ensuring comparability between different wavelength bands.

[0051] S2, TBS Feature Band Selection: Based on the spectral data preprocessed by standard normal variable transformation, a competitive adaptive reweighted sampling algorithm is used for screening. First, a partial least squares regression model is constructed with 80% of the samples. The importance of bands is evaluated by the absolute value weight of the regression coefficients. Low-weight bands are gradually eliminated by combining the exponential decay function. The performance of candidate bands is evaluated by 10-fold cross-validation. Then, after 70 iterations, 60 and 77 feature bands are selected from the two datasets respectively. The intersection of the two datasets is taken as the final feature input for TBS early asymptomatic identification.

[0052] The specific steps of S2 are as follows:

[0053] S21. Based on the spectral data preprocessed by standard normal variable transformation, a Competitive Adaptive Reweighted Sampling (CARS) algorithm is used to screen out redundant or irrelevant bands and extract key feature bands for early diagnosis of TBS. The specific process is as follows: 80% of the samples are randomly selected to establish a partial least squares regression (PLS) model, the regression coefficients of each band are calculated, and the importance of the bands is evaluated by their absolute value weights; bands with lower weights are gradually eliminated using an exponential decay function to obtain a subset of candidate bands; then, the predictive performance of the candidate band set is evaluated by 10-fold cross-validation; after 70 iterations, the band set with the best prediction accuracy is selected as the final feature bands.

[0054] S22, the results of the competitive adaptive reweighted sampling algorithm show that Dataset 1 retains 60 bands and Dataset 2 retains 77 bands, with a total of 38 bands; then a second round of CARS optimization screening is performed on these 38 bands, and finally the intersection of the two screening bands is taken as the final feature input for TBS early asymptomatic identification, which is used for subsequent modeling and analysis.

[0055] S3, Early Asymptomatic Diagnostic Index for Tobacco Black Shank: First, based on the results of feature band screening, the sensitive bands for the early asymptomatic stage of TBS are determined; then, the two datasets are merged, and a Logistic regression model is constructed based on the sensitive bands to finally obtain the early asymptomatic diagnostic index for tobacco black shank.

[0056] The specific steps of S3 are as follows:

[0057] After band screening, S31 was finally determined to be 437 nm, 530 nm, 649 nm and 710 nm as the main sensitive bands in the early asymptomatic stage of TBS.

[0058] S32, after merging Dataset 1 and Dataset 2, a Logistic Regression Model is constructed based on these four bands to obtain the Black Shank Early Detection Index (BSEDI) as follows. This index is used to distinguish between healthy and potentially infected plants, providing quantitative basis for model training and subsequent automated monitoring.

[0059] .

[0060] A diagnostic device for early asymptomatic TBS is specifically designed and prototyped around four core wavelength bands: 437 nm, 530 nm, 649 nm, and 710 nm. The device includes four narrowband DT filters for acquiring reflected light in the corresponding wavelength bands and integrates a data processing chip to perform real-time calculation of band reflectance, preprocessing, and automatic calculation of the early asymptomatic TBS diagnostic index, thereby rapidly outputting the TBS infection status result.

[0061] Based on the above, the following performance evaluation is conducted:

[0062] After screening using the CARS algorithm, Dataset 1 and Dataset 2 yielded 60 and 77 feature bands, respectively, defined as subsets A and B. To evaluate the model's generalization and predictive capabilities, a cross-testing strategy was adopted: training with Dataset 1 and testing with Dataset 2; then training with Dataset 2 and testing with Dataset 1. Based on this, Support Vector Classification (SVC), Partial Least Squares Discriminant Analysis (PLS-DA), K-Nearest Neighbor (KNN), Random Forest (RF), XGBoost, and Decision Tree (DT) were introduced to construct classification models to compare the performance of different algorithms in early asymptomatic diagnosis.

[0063] The prediction results of each model on the independent test set are shown in Table 1. Without band selection, the average accuracy is approximately 74.95%; after CARS selection, the average accuracy increases to 75.98%. Although the prediction accuracy has improved slightly, there is still room for further improvement in the overall accuracy.

[0064]

[0065] Based on the spectral response characteristics of tobacco black shank, this invention extracted four sensitive feature bands (437 nm, 530 nm, 649 nm, and 710 nm) from hyperspectral data and constructed an early asymptomatic diagnostic model based on these bands. To evaluate the discriminative ability of the feature bands, 11 commonly used vegetation indices were selected for comparison (see Table 2). The classification performance results of various models on independent test sets are as follows: Figure 1 As shown.

[0066] Under the experimental setup of training set Dataset 1 and test set Dataset 2, the overall classification performance of models built based on feature bands showed significant differences. Among them, the classification accuracies of PLS-DA, KNN, RF, and XGBoost were all above 92%, at 94.44%, 92.48%, 93.14%, and 92.81% respectively, with an average accuracy exceeding 90%. The SVC model, however, had a relatively low accuracy of only 55.34%. Regarding vegetation indices, EVI performed best among the five models, with an average accuracy of 88.21%. MSAVI, PVI, and RDVI had average accuracies between 83% and 84%. DVI, due to its lower recognition rate (70.70%) in the KNN model, had an average accuracy of 83.94%. Traditional vegetation indices such as RVI and NDVI had an average accuracy of approximately 82%, slightly lower than EVI overall.

[0067] When Dataset 2 was used as the training set and Dataset 1 as the test set, the feature bands still showed good recognition ability in PLS-DA, KNN, RF, and XGBoost models, with accuracies of 87.73%, 88.22%, 87.61%, and 86.75%, respectively, averaging over 87%. The accuracy of the SVC model remained at a lower level (60.86%). The classification performance of each vegetation index was generally stable. The average accuracy of EVI, DVI, and NDGI (not used as a feature band) was 81.43%, 81.71%, and 79.68%, respectively, concentrated in the range of 78%–82%, which is basically consistent with the results of the previous experimental scenario.

[0068] The combined results from the two training-test scenarios show that the selected feature bands achieve an average recognition accuracy of 89% in the PLS-DA, KNN, RF, and XGBoost models, significantly outperforming most vegetation indices (average 78%–82%). This indicates that the selected bands can effectively characterize the spectral response features of tobacco black shank in its early stages. Regarding the stability of the classification algorithms, RF and XGBoost achieved recognition accuracies of 93.14% / 87.61% and 92.81% / 86.75% based on feature bands in the two experimental scenarios, respectively. They also maintained high diagnostic performance for most vegetation indices, demonstrating that these two algorithms have good reliability and applicability in achieving early, asymptomatic identification of tobacco black shank.

[0069] Table 2 Eleven General Vegetation Indices

[0070]

[0071] Finally, based on the four core characteristic bands (437 nm, 530 nm, 649 nm, and 710 nm) selected through screening, the Tobacco Black Shank Early Diagnostic Index (BSEDI) was constructed, as shown in Figure 2. This index quantifies the health status of tobacco plants by measuring changes in leaf spectral reflectance characteristics at the pathological level: when the index value is greater than 0, it indicates that the tobacco plant may be infected with black shank; when the index value is less than 0, the plant is considered to be in a healthy state.

[0072] In the test results of two independent datasets, the BSEDI index demonstrated high stability and reliability, with diagnostic accuracy rates of 83.19% for Dataset 1 and 83.77% for Dataset 2, maintaining an overall range of 80%–85%. Scatter plot results show that infected and healthy samples are clearly separated in the index coordinate space, with clear cluster boundaries and only a small number of samples showing overlap (i.e., healthy sample index > 0 or infected sample index ≤ 0), reflecting the randomness brought about by individual differences and spectral fluctuations in actual field conditions.

[0073] Comprehensive analysis results show that the BSEDI index has a good ability to identify early latent infection of tobacco black shank, and can provide reliable spectral signal criteria when the plant has not yet shown obvious symptoms, providing effective technical support for non-destructive, rapid and visual monitoring in the field.

[0074] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for early asymptomatic diagnosis of tobacco black shank, characterized in that, The specific steps are as follows: S1, Training Data Acquisition: Acquire spectral data and standardized samples of tobacco black shank in the early asymptomatic stage to support subsequent modeling and analysis. The training data acquisition work is carried out by artificial inoculation experiment combined with hyperspectral acquisition and data preprocessing to obtain two datasets containing infected and healthy samples. S2, Feature band selection for tobacco black shank: Based on the spectral data preprocessed by standard normal variable transformation, a competitive adaptive reweighted sampling algorithm was used for screening. First, a partial least squares regression model was constructed with 80% of the samples. The importance of bands was evaluated by the absolute value weight of the regression coefficients. Low-weight bands were gradually eliminated by combining the exponential decay function. The performance of candidate bands was evaluated by 10-fold cross-validation. Then, after 70 iterations, 60 and 77 feature bands were selected from the two datasets respectively. The intersection of the two datasets was taken as the final feature input for early asymptomatic identification of tobacco black shank. S3, Early Asymptomatic Diagnostic Index for Tobacco Black Shank: First, based on the results of feature band screening, the sensitive bands for the early asymptomatic stage of tobacco black shank are determined; then, the two datasets are merged, and a Logistic regression model is constructed based on the sensitive bands to finally obtain the early asymptomatic diagnostic index for tobacco black shank.

2. The method for early asymptomatic diagnosis of tobacco black shank according to claim 1, characterized in that, The specific steps of S1 are as follows: S11, Hyperspectral Data Acquisition: S111, in an open-air experimental field, select Yunyan 116 tobacco plants with uniform growth and in the seedling stage to carry out an artificial inoculation experiment of tobacco blackleg; S112, after inoculation, the plants were first allowed to continue to grow and be managed under natural field conditions. Then, on the 2nd and 4th day after inoculation, UAV hyperspectral data was collected. At the same time, UAV hyperspectral images were collected on the 2nd and 4th day after inoculation. S113. After image acquisition, whiteboard correction was first performed to eliminate ambient light differences. Then, the tobacco plant area was manually labeled using Python's Labelme module to remove background and high-noise parts. Afterward, reflectance data of all bands in each labeled area were extracted by programming, and their average value was calculated as the spectral characteristics of a single sample. Among them, the data collected on the second day after vaccination was defined as Dataset 1, containing 495 infected samples and 320 healthy samples, and the data on the fourth day was defined as Dataset 2, containing 508 infected samples and 410 healthy samples. S12, Data Preprocessing: The standard normal transformation of spectral data is calculated using the following formula: in, n is the number of wavelengths. The wavelength reflectance is used; after standard normal transformation, the mean of the sample is 0 and the standard deviation is 1, thus ensuring comparability between different wavelength bands.

3. The method for early asymptomatic diagnosis of tobacco black shank according to claim 2, characterized in that, The inoculation method adopted is stem wound inoculation: First, use scissors disinfected with alcohol to make a shallow wound about 1-2 cm from the ground at the base of the tobacco seedling stem. Then, insert the pre-prepared Phytophthora nicotianae inoculation body into the wound and cover it with fine, moist soil to maintain local humidity and promote the colonization of the pathogen in the stem tissue.

4. The method for early asymptomatic diagnosis of tobacco black shank according to claim 2, characterized in that, In S112, both data collections were scheduled between 11:00 and 12:00 noon on a sunny and breezy day to minimize the impact of changes in light intensity.

5. A method for early asymptomatic diagnosis of tobacco black shank according to claim 3, characterized in that, The UAV flight altitude in S112 is set to 30m, which ensures that the image resolution of the leaf and stem base area meets the analysis requirements, thereby achieving high-precision spectral acquisition of early asymptomatic plants.

6. The method for early asymptomatic diagnosis of tobacco black shank according to claim 2, characterized in that, The specific steps of S2 are as follows: S21. Based on the spectral data preprocessed by standard normal variable transformation, a competitive adaptive reweighted sampling algorithm is used to screen out redundant or irrelevant bands and extract key feature bands for the early diagnosis of tobacco black shank. The specific process is as follows: 80% of the samples are randomly selected to establish a partial least squares regression model, the regression coefficients of each band are calculated, and the importance of the bands is evaluated by their absolute value weights; bands with lower weights are gradually eliminated using an exponential decay function to obtain a subset of candidate bands; the predictive performance of the candidate band set is then evaluated by 10-fold cross-validation; after 70 iterations, the band set with the best prediction accuracy is selected as the final feature bands. S22, the results of the competitive adaptive reweighted sampling algorithm show that Dataset 1 retains 60 bands and Dataset 2 retains 77 bands, with a total of 38 bands. Then, a second round of competitive adaptive reweighted sampling algorithm optimization and screening is performed on these 38 bands. Finally, the intersection of the two screening bands is taken as the final feature input for the early asymptomatic identification of tobacco black shank, which is used for subsequent modeling and analysis.

7. A method for early asymptomatic diagnosis of tobacco black shank according to claim 6, characterized in that, The specific steps of S3 are as follows: After band screening, S31 was finally determined to be 437 nm, 530 nm, 649 nm and 710 nm as the main sensitive bands in the early asymptomatic stage of tobacco black shank. S32. After merging Dataset 1 and Dataset 2, a Logistic regression model is constructed based on these four bands to obtain the early asymptomatic diagnostic index for tobacco black shank.

8. A diagnostic device for early asymptomatic tobacco black shank disease, characterized in that, It includes four narrowband DT filters for acquiring reflected light in the corresponding bands, and integrates a data processing chip to perform real-time band reflectivity calculation, preprocessing, and automatic calculation of the early asymptomatic diagnostic index for tobacco black shank, thereby quickly outputting the infection status results of tobacco black shank.

Images