Tobacco leaf nicotine content monitoring method and system based on unmanned aerial vehicle

Abstract

The invention discloses a tobacco leaf nicotine content monitoring method and system based on an unmanned aerial vehicle, and relates to the technical field of tobacco leaf nicotine content monitoring, and the method comprises the steps: obtaining a multispectral image covering a whole planting area by using the unmanned aerial vehicle carrying a multispectral sensor under a preset phenological period and weather conditions; preprocessing the image and extracting spectrum and texture features of the tobacco leaves; secondly, screening characteristic parameters sensitive to the nicotine content, calculating the contribution rate of the characteristic parameters by an entropy weight method, and constructing a diagnostic index for comprehensively representing the nicotine content; constructing an estimation model based on the index, and inverting the global nicotine content distribution; finally, the distribution is visualized to generate a distribution diagram, and support is provided for planting management decision making. The method depends on an unmanned aerial vehicle and a multispectral technology to realize global and efficient monitoring and evaluation of the nicotine content.

Description

Technical Field

[0001] This invention relates to the field of tobacco nicotine content monitoring technology, and in particular to a method and system for monitoring tobacco nicotine content based on unmanned aerial vehicles (UAVs). Background Technology

[0002] As an important economic crop, the nicotine content is a key indicator in the growth process of tobacco leaves. It not only affects the quality and characteristics of tobacco products, but is also closely related to the physiological characteristics of tobacco, such as its resistance to insects and diseases.

[0003] Traditional monitoring methods primarily rely on manual sampling and laboratory analysis. First, representative sample points are selected within the tobacco-growing area. These points are typically determined based on factors such as topography, soil type, and planting density to ensure representativeness. Then, professionals collect leaf samples from tobacco plants at specific growth stages, such as the vigorous growth phase and maturity phase, using specialized sampling tools like scissors and sampling bags, following standardized sampling methods. Sample collection generally follows certain rules, such as sampling from different parts of the plant and different leaf positions, to reduce errors caused by sample variations. In practice, to improve the accuracy of monitoring results, multiple samplings and repeated analyses are usually performed to reduce the impact of random errors. Simultaneously, laboratories regularly calibrate and maintain instruments and equipment to ensure the reliability of analytical results. Traditional manual sampling and laboratory analysis methods have several drawbacks in monitoring nicotine content in tobacco leaves. These include the time-consuming and labor-intensive process, uneven sampling due to traditional methods, and high costs associated with purchasing chemical reagents, standards, consumables, and acquiring and maintaining equipment. Summary of the Invention

[0004] This invention aims to address the problems of low efficiency, high cost, and insufficient representativeness in traditional methods for monitoring nicotine content in tobacco leaves. It provides a method and system for monitoring nicotine content in tobacco leaves using unmanned aerial vehicles (UAVs), enabling accurate assessment and prediction of nicotine content, generating intuitive monitoring maps, and providing strong support for farmers and agricultural decision-makers to formulate refined field management strategies, thereby helping to improve tobacco quality and the economic benefits of tobacco production.

[0005] In a first aspect, the present invention provides a method for monitoring the nicotine content of tobacco leaves using a drone, the method comprising the following steps: S1. Under preset phenological and weather conditions, multispectral images of the tobacco planting area are acquired using a multispectral sensor mounted on a drone, with the multispectral images covering the entire planting area. S2. The multispectral images are preprocessed and feature extracted to obtain the spectral and texture features of the tobacco leaves. S3. Feature filtering is performed based on the spectral and texture features to identify feature parameters sensitive to nicotine content in the tobacco leaves. The contribution rate of these feature parameters to nicotine content is calculated using the entropy weight method. A tobacco nicotine diagnostic index is constructed based on the contribution rate and the feature parameters, and this index is used to comprehensively characterize nicotine content. S4. An estimation model is constructed based on the tobacco nicotine diagnostic index, and the distribution of nicotine content in the entire tobacco leaf area is inverted using the estimation model. S5. The distribution of nicotine content in the entire tobacco leaf area is visualized to generate a distribution map to support planting management decisions.

[0006] Preferably, step S2 includes: S21. Correcting and stitching the multispectral image to eliminate image distortion and displacement; S22. Denoising the corrected and stitched multispectral image using a filtering algorithm to obtain a denoised multispectral image; S23. Performing color correction on the denoised multispectral image to obtain the true surface reflectance; S24. Extracting the spectral features based on the true surface reflectance, wherein the spectral features include reflectance information of multiple bands; S25. Extracting the texture features using an image analysis algorithm, wherein the texture features characterize the differences in the surface structure of tobacco leaves.

[0007] In this step, the multispectral image is first corrected and stitched to eliminate deformation and displacement, then denoising is performed using a filtering algorithm, followed by color correction to obtain the true surface reflectance. Based on this, spectral features containing reflectance information of multiple bands are extracted based on the true surface reflectance, and texture features characterizing the differences in the surface structure of tobacco leaves are extracted using image analysis algorithms.

[0008] Preferably, step S3 includes: S31. Quantifying the correlation between the spectral features and texture features and nicotine content using a correlation analysis method; S32. Screening highly correlated feature parameters based on the correlation, the feature parameters including reflectance in a specific band and texture index; S33. Optimizing the feature parameters using a redundancy removal algorithm to obtain a simplified feature combination, the simplified feature combination including feature parameters such as red band, red edge band, near-infrared band, contrast, and coarseness; S34. Standardizing the feature parameters in the simplified feature combination to obtain standardized feature parameters; S35. Calculating the contribution rate of the standardized feature parameters to nicotine content using the entropy weight method, and weighting the standardized feature parameters according to the contribution rate to construct the tobacco nicotine diagnostic index; S36. Optimizing the tobacco nicotine diagnostic index using a nonlinear transformation method, the nonlinear transformation method being used to improve the sensitivity of the tobacco nicotine diagnostic index to changes.

[0009] In this step, the correlation between spectral and texture features and nicotine content is first quantified through correlation analysis. Based on the correlation, highly correlated feature parameters containing specific band reflectance and texture indicators are selected. Then, a simplified feature combination containing red light, red edge, near-infrared bands, contrast, and coarseness is obtained through a redundant information removal algorithm. After standardization, the contribution rate of each parameter is calculated using the entropy weight method, and a tobacco nicotine diagnostic index is constructed by weighting. Finally, the index is optimized through nonlinear transformation to improve its sensitivity to changes.

[0010] Preferably, step S35 includes: determining the contribution rates (a, b, c, d, e) of the red light band, red edge band, near-red band, contrast, and coarseness to the nicotine content of tobacco leaves using the entropy weight method, respectively; and then constructing a nicotine diagnostic index for tobacco leaves according to the formula: Among them, TLNDI is the nicotine diagnostic index for tobacco leaves, and R', RE', NIR', CON' and COA' are the standardized red light band, red edge band, near-infrared band, contrast and coarseness, respectively.

[0011] Preferably, step S4 includes: S41. Constructing the estimation model using regression analysis, the estimation model mapping the relationship between the tobacco nicotine diagnostic index and the actual nicotine content; S42. Evaluating the performance of the estimation model according to preset evaluation indicators, the evaluation indicators including the model's coefficient of determination, root mean square error, and normalized root mean square error; S43. Adjusting the parameters of the estimation model based on the performance evaluation results to obtain an optimized estimation model; S44. Applying the optimized estimation model to the global image processing to output the global distribution of tobacco nicotine content.

[0012] Preferably, step S5 includes: S51. Grading the distribution of nicotine content in the tobacco leaves across the entire region according to a preset grading standard to obtain grading results; S52. Combining the grading results with geographical data of the planting area using geographic information processing technology to generate the distribution map; S53. Rendering the distribution map to highlight the differences between different grading areas; S54. Outputting the distribution map to a user terminal, which is used for display and management decision support.

[0013] Secondly, this application provides a monitoring system for nicotine content in tobacco leaves based on unmanned aerial vehicles (UAVs), the system comprising: The multispectral image acquisition module acquires multispectral images of the tobacco planting area using a multispectral sensor mounted on a drone under preset phenological and weather conditions. The multispectral images cover the entire planting area. The feature extraction module preprocesses and extracts features from the multispectral image to obtain the spectral reflectance and texture features of the tobacco leaves; The tobacco nicotine diagnostic index construction module performs feature screening based on the spectral and texture features to select feature parameters that are sensitive to the nicotine content of tobacco leaves. It then uses the entropy weight method to calculate the contribution rate of the feature parameters to the nicotine content and constructs the tobacco nicotine diagnostic index based on the contribution rate and the feature parameters. The tobacco nicotine diagnostic index is used to comprehensively characterize the nicotine content. The inversion module constructs an estimation model based on the tobacco nicotine diagnostic index and uses the estimation model to invert the distribution of nicotine content in tobacco leaves across the entire region. The visualization module visualizes the distribution of nicotine content in tobacco leaves across the entire region, generating a distribution map to support planting management decisions.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the method for monitoring the nicotine content of tobacco leaves based on UAV imagery as described above.

[0015] Fourthly, this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for monitoring nicotine content in tobacco leaves based on UAV imagery as described above.

[0016] The present invention provides a method and system for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles (UAVs). This involves acquiring multispectral images covering the entire planting area using a UAV equipped with a multispectral sensor under preset phenological and weather conditions; preprocessing the images and extracting spectral and texture features of the tobacco leaves; then screening feature parameters sensitive to nicotine content, calculating their contribution rate using the entropy weight method, and constructing a comprehensive diagnostic index characterizing nicotine content; furthermore, building an estimation model based on this index to invert the distribution of nicotine content across the entire area; and finally, visualizing the distribution to generate a distribution map for planting management decisions. This technology supports the use of multispectral sensors on drones to rapidly scan tobacco growing areas, significantly shortening the monitoring cycle and changing the time-consuming and labor-intensive nature of traditional manual monitoring. It enables real-time monitoring and timely decision-making. Utilizing low-altitude remote sensing by drones to acquire high-resolution multispectral images, combined with advanced feature extraction and screening technologies, it comprehensively captures information on tobacco growth status, overcoming the representativeness limitations of traditional methods due to limited sampling points. This more accurately reflects the actual distribution of nicotine content in tobacco leaves. It reduces reliance on large quantities of chemical reagents, specialized experimental equipment, and manpower, while avoiding the potential environmental pollution from organic solvents in traditional methods, making nicotine content monitoring more economical and environmentally friendly, and easier to promote and apply in large-scale tobacco cultivation. Furthermore, by constructing a nicotine diagnostic index and estimation model, it achieves accurate assessment and prediction of nicotine content, generating intuitive monitoring maps. This provides strong support for farmers and agricultural decision-makers to formulate refined field management strategies, helping to improve tobacco quality and the economic benefits of tobacco production. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles (UAVs) provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the framework of the UAV-based tobacco nicotine content monitoring system provided in an embodiment of the present invention; Figure 3 This is a monitoring map of nicotine grade in tobacco leaves based on multispectral imagery from drones; Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] 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.

[0019] Reference Figure 1 This invention provides a method for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles (UAVs), comprising the following steps: Step S1: Under preset phenological and weather conditions, use a multispectral sensor mounted on a drone to acquire multispectral images of the tobacco planting area, with the multispectral images covering the entire planting area. In this embodiment, UAV multispectral imagery was acquired during a time period with suitable weather conditions (sunny, cloudless, and wind speed below level 3) and stable tobacco leaf growth and light levels (between 10:00 AM and 2:00 PM). The flight altitude was set between 30 and 50 meters to ensure image acquisition accuracy. The forward overlap rate of the UAV flight was set to 80%, and the lateral overlap rate was set to 80%. The multispectral sensor included red (R), red-edge (RE), and near-infrared (NIR) bands.

[0020] The acquired images are then corrected and stitched using stitching software (such as Pix4D or Agisoft Metashape). Next, the stitched high-resolution images undergo preprocessing, including noise removal and color correction, to improve the accuracy of subsequent recognition algorithms and obtain multispectral images showing the true reflectance of surface objects. These images will be displayed in the geographic coordinate system GCS_WGS_1984 and the projected coordinate system UTM_Zone_50N, with a spatial resolution of 0.05m.

[0021] While acquiring spectral imagery, agronomic parameters of tobacco leaves will also be obtained. Crop plants from a representative unit of land area will be collected and subjected to blanching and drying in an oven. Subsequently, the plant leaves will be sent to a laboratory for nicotine content measurement. This step aims to accurately assess the nicotine content of tobacco plants, providing calibration data support for the drone-based nicotine content monitoring model.

[0022] Specifically, taking flue-cured tobacco cultivation in the Huang-Huai tobacco region as an example, nicotine content enters a rapid accumulation and stabilization phase from the late vigorous growth stage to maturity (60-90 days after transplanting). This stage is the most strongly correlated with the physiological state of tobacco leaves and nicotine content, making it the core phenological period for monitoring. If the rosette stage (30 days after transplanting) is chosen, the nicotine accumulation is minimal, and the spectral signal difference is not significant. If the process is delayed until after harvest, it will not provide sufficient time for decision-making in field regulation. Therefore, it is necessary to accurately target the late vigorous growth stage (such as mid-to-late July) as the predetermined time period.

[0023] The practical standard for suitable weather conditions is to meet the core requirements of "cloudless clear skies + light winds". Specific parameters are: solar radiation intensity ≥ 8000 lux, cloud cover ≤ 10%, wind speed ≤ 3 m / s, and to avoid strong midday sunlight (10:00-14:00) and the morning dew period (before 8:00). For example, in the tobacco-growing areas of Yunnan, choosing 9:00-10:00 or 15:00-16:00 for flight is ideal, as the angle of sunlight is moderate, and there is no glare or shadow interference on the tobacco leaf surface. This avoids uneven image brightness caused by cloud cover or blurry images caused by strong winds.

[0024] The configuration and operation examples of drones and sensors, equipment selection: DJI Matrice 350 RTK drone was selected, equipped with MicaSense Altum multispectral sensor (including red, red edge and near-infrared bands, spectral resolution of 10nm, spatial resolution of 10cm). This sensor can accurately capture the reflectance changes of tobacco leaves in the red edge band (730nm) - this band is closely related to the chlorophyll content related to nicotine synthesis.

[0025] The flight operation targeted a contiguous 100-mu (approximately 6.7 hectares) tobacco field, planning a zigzag flight path with a flight altitude of 80 meters, a speed of 5 m / s, a forward overlap of 80%, and a lateral overlap of 70%. GPS accuracy was calibrated using ground control points (GCPs) before flight, and the quality of image transmission was monitored in real time during the flight. The resulting multispectral images completely covered the tobacco field boundaries, and the overlapping areas of adjacent images ensured seamless stitching. Individual images clearly revealed the morphological outline of single tobacco leaves, providing high-resolution foundational data for subsequent feature extraction.

[0026] By employing UAV high-altitude remote sensing technology, information on tobacco plants can be quickly acquired. Compared with traditional manual inspection, this significantly shortens the monitoring cycle and improves monitoring efficiency. In addition, high-resolution images are collected based on multispectral sensors to obtain data on the tobacco planting area with full coverage.

[0027] Step S2: Preprocess and extract features from the multispectral image to obtain the spectral and texture features of the tobacco leaves; In this embodiment, the multispectral image data of the aforementioned 100-mu contiguous tobacco field (acquired by a DJI Matrice 350RTK drone equipped with a MicaSense Altum sensor) is corrected and stitched together to eliminate image distortion and displacement; the corrected and stitched multispectral image is then denoised using a filtering algorithm to obtain a denoised multispectral image; color correction is performed on the denoised multispectral image to obtain the true surface reflectance; spectral features are extracted based on the true surface reflectance, and these spectral features include reflectance information in multiple bands; texture features are extracted using an image analysis algorithm, and these texture features characterize the differences in the surface structure of the tobacco leaves.

[0028] Specifically, Agisoft Metashape professional image processing software was used, and the steps were as follows: the factory parameters of the MicaSense Altum sensor (focal length 6.0mm, pixel size 3.75μm) were imported, and the software automatically corrected the image distortion caused by lens distortion; five black and white targets (30cm on each side) with GPS coordinates were placed at the four corners and the center of the tobacco field, the target positions were marked in the image, the image coordinates were aligned with the real geographic coordinates, the displacement caused by GPS drift was eliminated, and the planar accuracy error was controlled within 5cm after correction; Based on the "feature point matching algorithm", the software identifies the overlapping areas of adjacent images (80% in the heading direction and 70% in the side direction). The software automatically calculates the coordinates of the corresponding points and completes the stitching to generate an orthophoto map (DOM) covering 100 acres of tobacco fields. There are no stitching faults and the edges completely match the boundaries of the tobacco fields.

[0029] Denoising through filtering: The image contains sensor electronic noise (random grayscale fluctuations) and high-frequency noise caused by flight vibrations. Filtering is needed to retain tobacco leaf features while removing interference. A bilateral filtering algorithm (balancing denoising and edge preservation) was selected, with parameters set in ENVI software as follows: spatial domain standard deviation 5px, grayscale domain standard deviation 10. Post-processing comparison shows that the "granular" noise in the red-edge band (730nm) of the original image has significantly disappeared, the boundary between tobacco leaves and soil is clearer, and the internal texture details of the tobacco leaves are not blurred. Traditional mean filtering can easily lead to the loss of texture information, while bilateral filtering can accurately balance denoising effect and feature preservation requirements.

[0030] Color Correction: To obtain the true surface reflectance, the grayscale values ​​of the original image are affected by solar intensity and atmospheric scattering. They need to be converted to the true surface reflectance to reflect the physiological characteristics of tobacco leaves. The correction is based on the standard reflector data provided with the MicaSense Altum sensor. Specifically, the FLAASH model is used, inputting the meteorological parameters during flight (temperature 25℃, humidity 60%, air pressure 101kPa) to eliminate the attenuation of the spectral signal by water vapor and aerosols in the atmosphere; the standard reflector image (with a known reflectance of 50%) provided with the sensor is imported, and a linear relationship between the image grayscale value and reflectance is established (reflectance = 0.0003 × grayscale value + 0.02). The soil reflectance in the corrected red band (650nm) remained stable at 20%-25%, consistent with the typical reflectance range of soil in this region. The reflectance of tobacco leaves in the near-infrared band (850nm) reached 60%-70%, which is consistent with the spectral characteristics of tobacco leaves during the vigorous growth period, proving that the reflectance data is true and reliable.

[0031] Spectral Feature Extraction: This involves capturing multi-band reflectance information. Specifically, it extracts target and band selection based on corrected true reflectance, extracting key band information related to nicotine content, with a focus on the five core bands of the MicaSense Altum sensor. The extraction method and data format utilize the "Zoning Statistics" tool in ArcGIS software, using the tobacco plant as the smallest unit (identifying individual tobacco leaf regions through image segmentation technology). The average reflectance of each band for each unit is extracted as follows: Red band (650nm): reflectance range 3%-8%, negatively correlated with chlorophyll content, indirectly related to nicotine synthesis; Red edge band (730nm): reflectance range 15%-25%, sensitive to tobacco leaf senescence, which is synchronized with nicotine accumulation; Near-infrared band (850nm): reflectance range 60%-70%, reflecting the integrity of tobacco leaf cell structure and affecting nicotine storage capacity. After extraction, a structured data table is formed consisting of "tobacco plant ID - red band reflectance - red edge band reflectance - near-infrared band reflectance", which provides a basis for subsequent correlation analysis.

[0032] Texture Feature Extraction: Characterizing differences in tobacco leaf surface structure, the selection of extraction targets and algorithms, and the correlation between texture features such as wrinkle density and smoothness of tobacco leaf surfaces and nicotine content (tobacco leaves with high nicotine content are generally more porous) are addressed by using the Gray-Level Co-occurrence Matrix (GLCM) algorithm to extract core texture indicators. Practical parameters and extraction results are also presented.

[0033] In ENVI software, GLCM parameters were set as follows: window size 5×5, grayscale level 256, step size 1, and orientation 0°, 45°, 90°, and 135°. Two key indicators were extracted: contrast, which reflects the degree of light and dark difference on the surface of tobacco leaves. The upper leaves with higher nicotine content have a contrast of 80-100, while the lower leaves only have a contrast of 40-60; and texture, which characterizes the size of texture units. The texture value of tobacco leaves with sufficient nicotine accumulation is 1.2-1.5, while that of low-nicotine tobacco leaves with a compact structure is 0.8-1.0. After extraction, the texture indicators were associated with the spectral feature data of the corresponding tobacco plants to form a complete tobacco leaf feature dataset.

[0034] Step S3: Based on the spectral and texture features, feature screening is performed to select feature parameters that are sensitive to the nicotine content of tobacco leaves. The contribution rate of the feature parameters to the nicotine content is calculated using the entropy weight method. Based on the contribution rate and the feature parameters, a tobacco nicotine diagnostic index is constructed. The tobacco nicotine diagnostic index is used to comprehensively characterize the nicotine content. In this embodiment, the correlation between the spectral and texture features and nicotine content is quantified using correlation analysis. Highly correlated feature parameters are selected based on the correlation, including reflectance and texture indices in specific wavelength bands. The feature parameters are optimized using a redundancy removal algorithm to obtain a simplified feature combination, which includes red band, red edge band, near-infrared band, contrast, and coarseness. The feature parameters in the simplified feature combination are standardized to obtain standardized feature parameters. The contribution rate of the standardized feature parameters to nicotine content is calculated using the entropy weight method, and the standardized feature parameters are weighted and combined according to the contribution rate to construct the tobacco nicotine diagnostic index. The tobacco nicotine diagnostic index is optimized using a nonlinear transformation method.

[0035] For example, the Pearson correlation coefficient is used to initially quantify the correlation between various image features and nicotine content, and features with high correlation are selected. Then, the continuous projection algorithm is used to further remove redundant information between features and optimize the feature combination. Finally, the recursive feature elimination method is used to simplify the feature set, retaining the feature combination that is most sensitive to changes in nicotine content and is most representative, ensuring that the features used for model construction are both concise and informative. After testing with experimental data, five feature parameters were finally selected: red band (R), red edge band (RE), near-infrared band (NIR), contrast (CON), and coarseness (COA).

[0036] The R-band plays a crucial role in characterizing nicotine content. Chlorophyll is extremely sensitive to light absorption in the red light band, and its content is closely related to nitrogen nutrition. Nicotine synthesis is directly linked to nitrogen supply; a high nicotine content in tobacco leaves often indicates ample nitrogen supply, leading to a corresponding increase in chlorophyll content and a significant decrease in red light reflectance. Therefore, R-band reflectance and nicotine content typically exhibit a negative correlation.

[0037] The RE band, located between red and near-infrared light, reflects the transitional characteristics of vegetation spectra and is highly sensitive to changes in chlorophyll concentration, leaf internal structure, and nitrogen compounds. When nicotine content increases, leaf nitrogen levels rise, causing a "red-edge shift"—the reflectance curve moves towards longer wavelengths in the red-edge region. This change reflects the dynamic adjustment of tobacco leaf physiological state and chemical composition, making the RE band an important spectral indicator for characterizing nicotine content.

[0038] The NIR band is primarily influenced by leaf cell structure. High nicotine content is usually accompanied by a denser, more compact leaf cell structure, resulting in a relatively higher near-infrared reflectance. Conversely, when nicotine content is low, the leaf structure is relatively loose, and near-infrared reflectance decreases. Therefore, the NIR band can sensitively capture changes in the tissue structure of tobacco leaves under varying nicotine content and, combined with R and RE, form a more robust characterization feature of nicotine content.

[0039] CON reflects the degree of variation in pixels on the surface of tobacco leaves. It is mainly determined by the orderliness of the arrangement of tobacco leaf cells, the regularity of cell morphology, and the uniformity of the distribution of chemical components within the cells. Nicotine content is positively correlated with tobacco leaf contrast. Specifically, tobacco leaves with high nicotine content have a more uneven distribution of internal chemical components, resulting in more obvious contrast differences in the image, that is, the difference between bright and dark areas will be more pronounced.

[0040] The Coarseness Index (COA) primarily reflects the fineness of the cell and tissue structure of tobacco leaves. It is negatively correlated with nicotine content. For tobacco leaves with high nicotine content, the cell structure is relatively more compact, with smaller intercellular spaces, resulting in a denser overall structure. Such leaves exhibit a finer texture and a lower coarseness index. Conversely, tobacco leaves with low nicotine content have a relatively looser cell structure with larger intercellular spaces, leading to a relatively higher coarseness index.

[0041] Index Construction: The entropy weight method is used to calculate the contribution rate (i.e., weight) of each parameter in the sensitive feature combination to nicotine content. The entropy weight method can objectively and accurately reflect the importance of each feature in characterizing changes in nicotine content. Based on the weight results, the Tobacco Leaf Nicotine Diagnostic Index (TLNDI) is constructed. This index is a comprehensive quantitative representation of tobacco leaf nicotine content, which can integrate multi-dimensional image feature information into a single, intuitive indicator. A quantitative relationship is established between this index and the field-collected tobacco leaf nicotine content data, laying a core foundation for subsequent nicotine content inversion and prediction.

[0042] The five selected characteristic parameters sensitive to nicotine content in tobacco leaves—R, RE, NIR, CON, and COA—were standardized to obtain standardized characteristic parameters R', RE', NIR', CON', and COA', ensuring they are on the same dimension. Then, the entropy weight method was used to determine their contribution rates (a, b, c, d, e) to nicotine content in tobacco leaves. A linear transformation method was used to improve the sensitivity of the tobacco nicotine diagnostic index to changes.

[0043] The contribution rates (a, b, c, d, e) of red light band, red edge band, near-red band, contrast, and coarseness to the nicotine content of tobacco leaves were determined using the entropy weight method. Then, a diagnostic index of nicotine in tobacco leaves was constructed according to the formula. Among them, TLNDI is the nicotine diagnostic index for tobacco leaves, and R', RE', NIR', CON' and COA' are the standardized red light band, red edge band, near-infrared band, contrast and coarseness, respectively.

[0044] The standardized formula is as follows: Here, x is a specific value in the original data. The minimum value among all values ​​in the dataset. The maximum value among all values ​​in the dataset. The result is the standardized value; then the information entropy is calculated for the standardized feature parameters, using the following formula: in, The proportion of the j-th feature parameter in the i-th sample; The contribution rate of each factor to the nicotine content of tobacco leaves was determined using the entropy weight method, where the formula for calculating the contribution rate is: To further enhance the scientific rigor and applicability of the diagnostic index, this application introduces logarithmic transformation and ratio construction methods based on traditional linear weighting. Specifically, spectral features (R', RE', and NIR') and texture features (CON', COA') are combined through ratios, highlighting the relative relationship between vegetation physiological and structural features and preventing a single feature from overly dominating the diagnostic results. Simultaneously, introducing a logarithmic function outside the ratio layer effectively compresses the interference of extreme values ​​on the model, improving the robustness of the index, and also enhances the sensitivity to subtle changes in the low-value range, making the index more precise in identifying differences in nicotine content. This construction method balances "non-linear expressive power" and "physiological-structural coupling characteristics," offering greater robustness and interpretability compared to simple weighted linear combinations, providing more reasonable and scientific technical support for the accurate monitoring of tobacco nicotine content.

[0045] This step introduces the concept of feature weighting, using the entropy weighting method to calculate the contribution rate (i.e., weight) of each parameter in the sensitive feature combination to the nicotine content. Based on the weighting results, a tobacco nicotine diagnostic index is constructed to achieve comprehensive quantification of multi-dimensional image feature information, highlight the influence of key features, and more accurately reflect the changes in tobacco nicotine content.

[0046] Step S4: Construct an estimation model based on the tobacco nicotine diagnostic index, and use the estimation model to invert the distribution of nicotine content in tobacco leaves across the entire region. In this embodiment, the estimation model is constructed using regression analysis, which maps the relationship between the tobacco nicotine diagnostic index and the actual nicotine content. The estimation model is then evaluated based on preset evaluation indicators, including the model's coefficient of determination, root mean square error, and normalized root mean square error. Based on the performance evaluation results, the parameters of the estimation model are adjusted to obtain an optimized model. Finally, the optimized estimation model is applied to the global image processing to output the global distribution of tobacco nicotine content.

[0047] Specifically, model construction: A nicotine content estimation model based on the tobacco nicotine diagnostic index (TLNDI) is constructed using methods such as multinomial regression. Quadratic multinomial regression is chosen (balancing fitting accuracy and model complexity), with the optimized TLNDI as the independent variable and the actual nicotine content (Y) as the dependent variable. The model expression is as follows: Where β0 is a constant term, β1 and β2 are regression coefficients, and ε is a random error term.

[0048] The PolynomialFeatures and LinearRegression modules from Python's Scikit-learn library are used to expand the training set with quadratic features, generating "1 (constant term), TLNDI, The feature matrix of "" is obtained; the regression coefficients are solved by fitting the training set data using the least squares method. Assuming that β0=1.15, β1=0.82, β2=0.36, the model is: The estimation model is optimized by using cross-validation to determine whether the model is overfitting, i.e., overfitting to the training set data and having poor generalization ability. Regularization optimization is then used to recalculate the regression coefficients until the difference between the training set and the validation set is reduced to a preset range. This solves the overfitting problem and yields the optimized estimation model. Model evaluation and application validation: Test set performance evaluation involves substituting the TLNDI of the test set into the estimation model to calculate the predicted nicotine content. Performance is evaluated using three core indicators: coefficient of determination (R²). Calculate the goodness of fit between the predicted and actual values. If R² = 0.83, the model can explain 83% of the variation in nicotine content; Root Mean Square Error (RMSE): If RMSE = 0.12%, which is much smaller than the standard deviation of nicotine content (0.58%), it indicates high prediction accuracy; Normalized root mean square error (NRMSE): If the average actual content is 2.3%, then NRMSE = 5.2%, which is less than 10%, meeting the accuracy requirements of agricultural monitoring models. Let be the actual nicotine content of the i-th sample, max(y) be the maximum actual nicotine content among all samples, and min(y) be the minimum actual nicotine content among all samples. This is the predicted nicotine content of the i-th sample calculated by the estimation model.

[0049] The performance of the models is evaluated by comparing indicators such as the coefficient of determination (R²), root mean square error (RMSE), and normalized root mean square error (NRMSE). The optimal estimation model is selected and further optimized to improve its accuracy and reliability, enabling it to accurately map the complex relationship between the nicotine content diagnostic index and the actual nicotine content.

[0050] Inversion of Nicotine Content Across the Entire Area: The monitoring model is constructed by substituting multispectral image data of the entire area acquired by UAVs. Based on the built-in quantitative relationship, the model can quickly calculate the distribution of nicotine content in tobacco leaves in various plots across the entire area, realizing the efficient conversion from image data to nicotine content information. This allows growers to fully understand the spatial distribution characteristics of nicotine content in the entire planting area without having to sample point by point.

[0051] A nicotine content estimation model based on the nicotine diagnostic index of tobacco leaves was constructed using multinomial regression. By comparing various model evaluation indicators, the optimal estimation model was selected and further optimized and adjusted to improve the accuracy and reliability of the model, enabling it to accurately map the complex relationship between the nicotine content diagnostic index and the actual nicotine content.

[0052] Step S5: Visualize the distribution of nicotine content in the entire tobacco leaf area to generate a distribution map to support planting management decisions; In this embodiment, the distribution of nicotine content in tobacco leaves across the entire region is graded according to a preset grading standard to obtain grading results; the grading results are combined with geographical data of the planting area using geographic information processing technology to generate the distribution map; the distribution map is rendered to highlight the differences between different grade areas; and the distribution map is output to a user terminal, which is used for display and management decision support.

[0053] For example, the results are visualized as follows: Based on the preset nicotine content level standards, the retrieved nicotine content data for the entire region is classified and categorized to divide the area into plots with different nicotine content levels. Then, using professional Geographic Information System (GIS) software, combined with geographic information data of tobacco planting areas, a monitoring map of tobacco nicotine content at the plot scale is drawn, as shown in the attached figure. Figure 3 This monitoring map visually displays the distribution of nicotine content levels in tobacco leaves across different regions, enabling farmers and relevant decision-makers to quickly understand the nicotine content of tobacco-growing areas from a holistic perspective. Decision support: Through the comprehensive plot-scale nicotine content monitoring map, farmers and relevant decision-makers can intuitively identify plots with abnormal nicotine content, thereby developing targeted nicotine content control measures, optimizing planting management plans, improving tobacco quality and yield, and achieving refined management and efficient decision-making in tobacco cultivation. The creation of the comprehensive plot-scale nicotine content monitoring map helps promote the application of intelligent agricultural technologies in tobacco cultivation, improving the accuracy of agricultural management and the efficiency of decision-making.

[0054] Based on the obtained data, a monitoring map of nicotine content in tobacco leaves at the plot scale was drawn, which intuitively shows the nicotine content in different areas and provides a visual basis for agricultural management and decision-making.

[0055] This implementation achieves efficient monitoring across the entire area, overcoming traditional limitations: Utilizing drones equipped with multispectral sensors, images of the entire planting area are acquired under standard phenological and weather conditions, eliminating the need for manual inspection of each plant, significantly improving monitoring efficiency and coverage. This solves the problems of time-consuming, labor-intensive, and insufficient representativeness associated with traditional methods, enabling large-scale and rapid detection of nicotine content in tobacco leaves. It also enhances feature extraction accuracy, solidifying the monitoring foundation: Through multi-level preprocessing including correction stitching, noise reduction, and color correction, combined with spectral (multi-band reflectance) and texture (contrast, coarseness, etc.) feature extraction, it accurately captures physiological and structural information of tobacco leaves; correlation analysis and redundancy removal optimize feature combinations, providing high-quality data support for nicotine content correlation analysis. Furthermore, it constructs a scientific diagnostic system, enhancing estimation accuracy: The entropy weight method is used to quantify the contribution rate of feature parameters and construct a weighted diagnostic index. Combined with nonlinear transformation to optimize sensitivity, and regression analysis to build an estimation model, along with evaluation using indicators such as the coefficient of determination and parameter adjustments, significantly improves the accuracy of nicotine content retrieval and reduces monitoring errors. Strengthening the value of data application to support precise decision-making: By integrating hierarchical processing with geographic information, the distribution of nicotine is visualized. The differences in grade are highlighted through rendering and output to the terminal, intuitively presenting the spatial distribution characteristics of nicotine across the entire region. This provides data support for growers to make targeted management decisions such as water and fertilizer regulation and harvesting planning, thereby helping to improve the quality of tobacco leaves.

[0056] The following describes the UAV-based nicotine content monitoring system for tobacco leaves provided by this invention. The UAV-based nicotine content monitoring system described below and the UAV-based nicotine content monitoring method described above can be referred to in correspondence.

[0057] This invention also provides a monitoring system for nicotine content in tobacco leaves based on unmanned aerial vehicles (UAVs), such as... Figure 2 As shown, a drone-based tobacco nicotine content monitoring system includes: The multispectral image acquisition module acquires multispectral images of the tobacco planting area using a multispectral sensor mounted on a drone under preset phenological and weather conditions. The multispectral images cover the entire planting area. The feature extraction module preprocesses and extracts features from the multispectral image to obtain the spectral reflectance and texture features of the tobacco leaves; The tobacco nicotine diagnostic index construction module performs feature screening based on the spectral and texture features to select feature parameters that are sensitive to the nicotine content of tobacco leaves. It then uses the entropy weight method to calculate the contribution rate of the feature parameters to the nicotine content and constructs the tobacco nicotine diagnostic index based on the contribution rate and the feature parameters. The tobacco nicotine diagnostic index is used to comprehensively characterize the nicotine content. The inversion module constructs an estimation model based on the tobacco nicotine diagnostic index and uses the estimation model to invert the distribution of nicotine content in tobacco leaves across the entire region. The visualization module visualizes the distribution of nicotine content in tobacco leaves across the entire region, generating a distribution map to support planting management decisions.

[0058] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840. The processor 810, communications interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions from the memory 830 to execute an emotion-recognition-based electronic photo frame interface adjustment method.

[0059] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the emotion recognition-based electronic photo frame interface adjustment method provided by the above methods.

[0061] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles (UAVs), characterized in that, The method includes the following steps: S1. Under preset phenological and weather conditions, multispectral images of the tobacco planting area are acquired using a multispectral sensor mounted on a drone, and the multispectral images cover the entire planting area. S2. Preprocess and extract features from the multispectral image to obtain the spectral and texture features of the tobacco leaves; S3. Based on the spectral and texture features, feature screening is performed to select feature parameters that are sensitive to the nicotine content of tobacco leaves. The contribution rate of the feature parameters to the nicotine content is calculated using the entropy weight method. Based on the contribution rate and the feature parameters, a tobacco nicotine diagnostic index is constructed. The tobacco nicotine diagnostic index is used to comprehensively characterize the nicotine content. S4. Construct an estimation model based on the tobacco nicotine diagnostic index, and use the estimation model to invert the distribution of nicotine content in tobacco leaves across the entire region; S5. Visualize the distribution of nicotine content in the tobacco leaves across the entire region to generate a distribution map to support planting management decisions.

2. The method for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles as described in claim 1, characterized in that, Step S2 includes: S21. Correct and stitch the multispectral images to eliminate image distortion and displacement; S22. The multispectral image after correction and stitching is denoised by a filtering algorithm to obtain the denoised multispectral image; S23. Perform color correction on the denoised multispectral image to obtain the true surface reflectance; S24. Extract the spectral features based on the true surface reflectance, wherein the spectral features include reflectance information of multiple bands; S25. Extract the texture features using an image analysis algorithm. The texture features characterize the differences in the surface structure of tobacco leaves.

3. The method for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles as described in claim 1, characterized in that, Step S3 includes: S31. Quantify the correlation between the spectral and texture features and the nicotine content using correlation analysis; S32. Based on the correlation degree, select highly correlated feature parameters, including reflectance in a specific band and texture index; S33. The feature parameters are optimized by a redundancy removal algorithm to obtain a simplified feature combination. The simplified feature combination includes the feature parameters of red light band, red edge band, near-infrared band, contrast, and coarseness. S34. Standardize the feature parameters in the simplified feature combination to obtain standardized feature parameters; S35. The contribution rate of the standardized feature parameters to the nicotine content is calculated using the entropy weight method, and the standardized feature parameters are weighted and combined according to the contribution rate to construct the tobacco nicotine diagnostic index; S36. The tobacco nicotine diagnostic index is optimized by a nonlinear transformation method, which is used to improve the sensitivity of the tobacco nicotine diagnostic index to changes.

4. The method for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles as described in claim 3, characterized in that, Step S35 includes: The contribution rates (a, b, c, d, e) of red light band, red edge band, near-red band, contrast, and coarseness to the nicotine content of tobacco leaves were determined using the entropy weight method. Then, a diagnostic index of nicotine in tobacco leaves was constructed according to the formula: Among them, TLNDI is the nicotine diagnostic index for tobacco leaves, and R', RE', NIR', CON' and COA' are the standardized red light band, red edge band, near-infrared band, contrast and coarseness, respectively.

5. The method for monitoring nicotine content in tobacco leaves based on UAV imagery as described in claim 1, characterized in that, Step S4 includes: S41. The estimation model is constructed using regression analysis, and the estimation model maps the relationship between the tobacco nicotine diagnostic index and the actual nicotine content; S42. The estimation model is evaluated for performance according to preset evaluation indicators, including the model's coefficient of determination, root mean square error, and normalized root mean square error. S43. Adjust the parameters of the estimation model based on the performance evaluation results to obtain an optimized estimation model; S44. Apply the optimized estimation model to the global image processing to output the global distribution of nicotine content in tobacco leaves.

6. The method for monitoring nicotine content in tobacco leaves based on unmanned aerial vehicles as described in claim 1, characterized in that, Step S5 includes: S51. The distribution of nicotine content in the entire tobacco leaf area is graded according to the preset grading standards to obtain the grading results; S52. The grading results are combined with the geographical data of the planting area using geographic information processing technology to generate the distribution map; S53. Render the distribution map to highlight the differences between different levels of areas; S54. Output the distribution map to the user terminal, which is used to display and manage decision support.

7. A drone-based system for monitoring nicotine content in tobacco leaves, the system comprising: The multispectral image acquisition module acquires multispectral images of the tobacco planting area using a multispectral sensor mounted on a drone under preset phenological and weather conditions. The multispectral images cover the entire planting area. The feature extraction module preprocesses and extracts features from the multispectral image to obtain the spectral reflectance and texture features of the tobacco leaves; The tobacco nicotine diagnostic index construction module performs feature screening based on the spectral and texture features to select feature parameters that are sensitive to the nicotine content of tobacco leaves. It then uses the entropy weight method to calculate the contribution rate of the feature parameters to the nicotine content and constructs the tobacco nicotine diagnostic index based on the contribution rate and the feature parameters. The tobacco nicotine diagnostic index is used to comprehensively characterize the nicotine content. The inversion module constructs an estimation model based on the tobacco nicotine diagnostic index and uses the estimation model to invert the distribution of nicotine content in tobacco leaves across the entire region. The visualization module visualizes the distribution of nicotine content in tobacco leaves across the entire region, generating a distribution map to support planting management decisions.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for monitoring the nicotine content of tobacco leaves based on UAV imagery as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for monitoring the nicotine content of tobacco leaves based on UAV imagery as described in any one of claims 1 to 6.

Images