Early identification method for tobacco field waterlogging disaster based on unmanned aerial vehicle hyperspectral
By using UAV hyperspectral technology and spectral angle matching algorithm to classify and identify tobacco fields, the problem of timely and accurate assessment of the extent of waterlogging damage in tobacco fields has been solved, enabling rapid and accurate identification and assessment of waterlogged areas in tobacco fields.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HENAN AGRICULTURAL UNIVERSITY
- Filing Date
- 2021-10-29
- Publication Date
- 2026-06-30
AI Technical Summary
Existing technologies make it difficult to estimate the extent of damage from flooding in tobacco fields in a timely and accurate manner. Traditional manual surveys are labor-intensive, and satellite remote sensing has a time lag, making it impossible to obtain relevant image data in a timely manner.
Hyperspectral data were collected using a drone hyperspectral instrument, and tobacco fields were classified and identified using a spectral angle matching algorithm. The fields were divided into three types of post-disaster surface: flooded tobacco fields, normal tobacco fields, and soil. The classification accuracy was calculated using ROI and confusion matrix.
It enables accurate and rapid identification and assessment of flood-affected tobacco fields, provides a new disaster assessment scheme, reduces workload, and improves the timeliness of assessment.
Smart Images
Figure CN115935111B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of agricultural natural disaster monitoring technology, specifically to a method for identifying waterlogging in tobacco fields based on UAV hyperspectral imaging. Background Technology
[0002] Floods, caused by heavy rainfall, low-lying terrain, and insufficient drainage, are a common and widespread natural disaster with a significant impact on modern agricultural production. Tobacco is a crucial economic crop in my country, playing a vital role in the national economy. Henan Province, a major tobacco-producing province, experiences frequent floods during July and August, which coincides with the critical growth period for flue-cured tobacco. During this time, the tobacco plants are tall and have large leaves, making them susceptible to damage from heavy rainfall, resulting in substantial losses for tobacco production.
[0003] Accurate and rapid monitoring of the extent and severity of flooding in tobacco fields, and quantitative assessment of the potential losses to flue-cured tobacco yield, are crucial for timely remedial measures and minimizing losses for tobacco farmers. Traditional flood disaster surveys often require manual on-site investigations, categorizing disasters and assessing losses based on experience. This method is labor-intensive, time-consuming, and has a limited scope. In recent years, with the maturity of satellite remote sensing technology, research on agricultural disaster analysis and monitoring using remote sensing image data has increased. However, satellite remote sensing is limited by the temporal and spatial resolution of satellites, often exhibiting lag and being affected by weather factors, making it impossible to acquire relevant regional image data in a timely manner, thus affecting the timeliness of disaster assessment. Compared with satellite remote sensing, unmanned aerial vehicles (UAVs) have advantages such as high mobility, high spatial resolution, and wide applicability. Research on drought stress, frost damage, and flooding has already been conducted on crops such as sugarcane, corn, rapeseed, and rice. Currently, UAV remote sensing technology has been widely used in tobacco production, with numerous reports on its application in flue-cured tobacco nutrient diagnosis, tobacco plant phenotype, and tobacco plant protection. However, to date, there have been no reports of using drones for flood disaster monitoring and research during the tobacco production process.
[0004] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the present invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention
[0005] The purpose of this invention is to provide an early identification method for tobacco field flooding based on UAV hyperspectral imaging, so as to solve the technical problem that existing technologies are unable to estimate the extent of tobacco field flooding losses in a timely and accurate manner.
[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:
[0007] Design a method for identifying waterlogging in tobacco fields based on UAV hyperspectral imaging, including the following steps:
[0008] (1) Use a drone equipped with a hyperspectral instrument to collect hyperspectral data of the area to be observed, and perform preprocessing to obtain the canopy reflectance data of the observation area;
[0009] (2) Select the reflectance data of the blue, red and green bands from the previous canopy reflectance data to synthesize a true color image of the observation area;
[0010] (3) Select pink areas, white and light green pixel sets, and green clusters on the true color image obtained in the previous step to establish three types of ROI corresponding to soil, disaster-stricken tobacco fields, and normal tobacco fields, respectively. Use the average spectral reflectance of each type of ROI as the spectral reflectance of that type of landform to generate the corresponding spectral curve.
[0011] (4) Based on the spectral angle matching algorithm, the pixels are classified by processing the similarity difference between the spectral curves of each pixel in the original image of the observation area and the spectral curves of the three types of ROIs. Thus, the observation area is divided into three types of post-disaster surface types: flooded tobacco fields, normal tobacco fields, and soil. After classification, the area, location and other information of each type can be further clarified, so as to carry out subsequent statistics, evaluation, and planning of countermeasures to reduce and remedy post-disaster losses.
[0012] In step (1), the UAV is set to fly at an altitude of 90-110m and a speed of 2.8-3.2m / s when collecting data.
[0013] In step (1), the hyperspectral instrument is a Pika-L spectrometer with a spectral range of 400–1000 nm, a spectral resolution of 2 nm, and a sampling interval of 4 nm.
[0014] In step (1), the hyperspectral data preprocessing method includes:
[0015] First, the data is segmented based on GPS data, and useless data during takeoff, landing, and turns is deleted. Then, the core area data is segmented according to the flight path. Subsequently, the segmented data is geometrically and radiometrically corrected using radiometric calibration files. The corrected data is then georeferenced and image-stitched into a hyperspectral cube using ArcGIS and ENVI. Finally, the canopy reflectance data of the observation area is obtained by inverting the target data of the known reflectance curves pre-placed in the observation field area.
[0016] Compared with the prior art, the main beneficial technical effects of the present invention are as follows:
[0017] 1. This invention utilizes hyperspectral images of flood-affected tobacco fields acquired by a UAV hyperspectral instrument, constructs regions of interest (ROIs) based on the interpretation results, and classifies and identifies the tobacco fields using a spectral angle matching algorithm, classifying the tobacco fields into three post-disaster surface types: flood-affected tobacco fields, normal tobacco fields, and soil.
[0018] 2. The spectral characteristics extracted by the method of this invention show significant differences between the affected tobacco fields, normal tobacco fields, and soil, enabling accurate extraction of the affected areas from tobacco field images and achieving good classification results. Furthermore, the overall classification accuracy was evaluated using a confusion matrix and the Kappa coefficient, further verifying the feasibility of using the method of this invention to classify and identify tobacco field flooding. This provides a new solution for accurate and rapid assessment of tobacco field flooding and also provides technical support for the application of UAV hyperspectral technology in tobacco field disaster assessment and production recovery. Attached Figure Description
[0019] Figure 1 This refers to the geographical location of the test verification area in this embodiment of the invention.
[0020] Figure 2 This describes the data processing flow for experiments in this embodiment of the invention.
[0021] Figure 3 This is a true-color image of a tobacco field affected by flooding, as tested in an embodiment of the present invention.
[0022] Figure 4 The hyperspectral reflectance characteristics of different land types in the affected tobacco fields in this embodiment of the invention.
[0023] Figure 5 This is the classification result of flooding in tobacco fields in an embodiment of the present invention.
[0024] Figure 6 The figure shows the average reflectance curves of the three classification results and the ROI in this embodiment of the invention. Detailed Implementation
[0025] The specific embodiments of the present invention will be described below with reference to the accompanying drawings and examples. However, the following examples are only used to illustrate the present invention in detail and do not limit the scope of the present invention in any way.
[0026] Example: Verification Experiment on Flood Identification in Tobacco Fields Based on UAV Hyperspectral Data
[0027] 1. Overview of the test area
[0028] Experimental Zone ( Figure 1Located in the Modern Tobacco Agriculture Industrial Park of Xiangcheng County, Xuchang City, Henan Province (113.52°E, 33.89°N), it is situated on the North China Plain at an altitude of about 70 m. It has a warm temperate continental monsoon climate with four distinct seasons, an average annual temperature of 14.7℃, 2281 hours of sunshine, a frost-free period of 217 days, 98-110 days with an average daily temperature above 20℃, and an average annual precipitation of 600-700 mm. Rainfall from June to August accounts for about 60% of the total annual precipitation. The soil type is brown soil.
[0029] The flue-cured tobacco variety was Zhongyan 100, with a row spacing of 120 cm and a plant spacing of 55 cm. Transplanting in 2021 took place from late April to early May. The main growth period is shown in Table 1. On July 15, 2021, at 7:00 AM, Xiangcheng County experienced heavy rainfall due to the combined influence of the Western Pacific subtropical high and Typhoon "Hwahwa". This rainfall lasted for more than three hours, with a cumulative rainfall of approximately 100 mm, a rare occurrence in recent years. The rainfall caused waterlogging in the tobacco fields, resulting in significant damage to the flue-cured tobacco plants, with many plants tilting and lodging, causing substantial losses to tobacco production.
[0030] Table 1. Key growth periods and timings for flue-cured tobacco in Xiangcheng County
[0031] critical reproductive period Seedling stage Rooting period Prosperous Long-Term Maturity Time period Late April to early May Early May to early June Early June to early July Early July to early September
[0032] 2. Data Acquisition and Processing
[0033] (1) Acquisition of hyperspectral data of tobacco fields
[0034] This verification experiment was conducted on July 15, 2021, after heavy rainfall in Xiangcheng County. A near-ground UAV hyperspectral remote sensing flight test was carried out at the Xiangcheng County Modern Tobacco Agricultural Industrial Park. The flight platform was a DJI M600 Pro hexacopter UAV, equipped with a Pika-L spectrometer manufactured by Liga, with a spectral range of 400–1000 nm, a spectral resolution of 2 nm, and a sampling interval of 4 nm. The UAV's flight altitude was set at 100 m and its speed at 3 m / s during data collection.
[0035] (2) Hyperspectral data preprocessing
[0036] The data acquired by the Pika L spectrometer includes GPS inertial navigation data and ground object reflectance energy data recorded by the spectrometer sensors. Preprocessing is required before use. First, the data is segmented based on the GPS data, removing useless data from takeoff, landing, and turns. Then, the core area data is segmented according to the flight path. Subsequently, geometric and radiometric corrections are performed on the segmented data using radiometric calibration files. The corrected data is then georeferenced and image-stitched into a hyperspectral cube using ArcGIS and ENVI. Finally, canopy reflectance data for the observation area is retrieved using target data with known reflectance curves pre-placed at the observation site.
[0037] 3. Hyperspectral data analysis
[0038] (1) Image interpretation of the affected tobacco fields
[0039] A 10 cm resolution true-color image of the study area was synthesized using reflectance data from the processed hyperspectral image in the three visible light bands of 659 nm, 550 nm, and 480 nm. Figure 3 a) The study found that after the flooding, some tobacco plants in the fields were flattened and covered by water, resulting in high reflectivity in the visible light spectrum, appearing as patches of white and light green pixels in the images. Tobacco plants less affected by the flooding had relatively normal reflectivity, appearing as green clusters in the images. Bare ground without tobacco plant cover, affected by soil moisture supersaturation, had high reflectivity in the red light spectrum, appearing pink. Based on the texture characteristics of the images, it was found that normal tobacco plants in the experimental observation area exhibited a regular east-west oriented texture, while the regular texture in the flood-affected area was disrupted, with the tilting and flattening of tobacco plants showing a southeast-to-northwest tilt.
[0040] Based on this, the area is divided into three categories: ① Soil (see Figure 3 ① Pink area in the middle); ② Tobacco fields that have tilted and collapsed due to flooding (e.g., Figure 3 -c1); ③ Normal tobacco fields with less severe damage (such as...) Figure 3 -c2).
[0041] (2) Spectral characteristics of typical areas of affected tobacco fields
[0042] Based on the interpretation results of the research on the images of the affected tobacco fields (see...) Figure 3 ), typical representative areas were selected from the image to construct Regions of Interest (ROIs), and the average spectral reflectance of the ROI was used as the spectral reflectance of that landform type. The extracted spectral reflectance curves for the three landform types are shown in the figure. Figure 4As shown in the figure, the spectral curves of the three landform types differ significantly in shape. Among the three landform types, the spectral curves of normal tobacco fields and damaged tobacco fields are similar in shape, both forming two absorption valleys near the 400-450 nm blue light band and the 660-680 nm red light band, a reflection peak at the 540-550 nm green light band, and a high reflectivity plateau in the 780-1000 nm band. For damaged tobacco leaves, the overall shape of the reflectivity curve is similar to that of normal tobacco leaves, but its reflectivity is generally higher in the visible light band than that of normal tobacco leaves, while it is lower in the near-infrared band.
[0043] 4. Monitoring and classification of flood-prone tobacco fields based on spectral angle
[0044] Classifying affected tobacco fields is crucial for accurately determining the area of damage, which is essential for disaster assessment, disaster relief strategy development, post-disaster recovery, and loss reduction. This experimental study, based on image interpretation of affected tobacco fields, analysis of the affected area, and spectral information from typical affected areas, employs a supervised classification algorithm based on spectral angle matching to classify flooded tobacco fields.
[0045] Spectral angle matching is to have n The algorithm classifies target pixels by using the spectral reflectance curve of each pixel in the hyperspectral data of each band as a multidimensional spatial vector, calculating the generalized angle between it and the spectral curve of a reference class pixel of known category, and then determining the degree of matching. The main steps include:
[0046] ① Construct ROIs for each known category on the image, requiring each ROI to contain only pixels of its own category; ② Extract spectral information of each ROI category; ③ Classify the study area using a classification algorithm and evaluate the accuracy.
[0047] In this experiment, ROIs for three types of land features were first established based on visual image interpretation results, and the reflectance spectral curves of these three types of land features were statistically analyzed (see...). Figure 3 , 4 Based on this, the algorithm is used to classify flooded tobacco fields, and the classification results are as follows: Figure 5 As shown, overlaying the classification results with the original image reveals that the spectral angle matching-based classification algorithm can effectively distinguish between damaged tobacco fields and normal tobacco fields.
[0048] 5. Evaluation of classification result accuracy
[0049] (1) Evaluation of classification performance based on spectral curves
[0050] To evaluate the accuracy of the classification results, the average spectral curves of the three land cover types in the classified tobacco fields were extracted and compared with the spectral curves of the ROI used as a classification reference (the results are shown in the figure). Figure 6 ).Depend on Figure 6 It can be seen that the reflectance curves of the three types of land cover are quite similar in shape to the ROI curve, indicating that the classification algorithm using spectral angle matching can effectively extract the reflectance spectral characteristics of the three types of land cover.
[0051] In addition to graphical comparisons, this experiment also used spectral correlation coefficients and spectral angles, which characterize the morphological similarity between different curves, as indicators to evaluate the classification performance by comparing the similarity between the classification results and the ROI curves. Among these, the spectral similarity coefficient... F The calculation formula is as follows:
[0052] ①;
[0053] In the formula, F Represents the spectral similarity coefficient; , Indicates the spectral curve x and spectral curves y Reflectivity at different wavelengths; , Indicates the spectral curve x and spectral curves y The average reflectance across the entire spectrum; F The larger the value, the higher the similarity in shape between the two spectral curves; conversely, the smaller the value, the greater the difference in shape between the two spectral curves.
[0054] Spectral angle θ The calculation formula is as follows:
[0055] ②;
[0056] In the formula, θ Indicates spectral angle; , Indicates the spectral curve x and spectral curves y Reflectivity at different wavelengths; n Indicates the number of bands; θ The smaller the value, the higher the similarity in shape between the two spectral curves; conversely, the larger the value, the greater the difference in shape between the two spectral curves.
[0057] Table 2. Spectral similarity coefficients and spectral angles between classification results and ROIs
[0058] .
[0059] Table 2 shows that the spectral curves of the affected tobacco fields, normal tobacco fields, and soils are related to the ROI. F The values are all close to 1. θA value close to 0 indicates a high degree of similarity in the spectral curves of the two, and the classification results are in good agreement with the ROI.
[0060] (2) Evaluation of classification accuracy based on confusion matrix and Kappa coefficient
[0061] In addition to comparing the spectral curves of the classification results with the ROIs, 30 ROIs (10 per class, totaling 30) were constructed as validation points. A confusion matrix between the validation points and the classification results was then built to obtain the overall classification accuracy and Kappa coefficient for accuracy evaluation. Overall classification accuracy is the ratio of the number of correctly classified pixels to all validation pixels. The Kappa coefficient is an objective indicator used for consistency testing, complementing the overall classification accuracy through a discrete multi-source technique. Its value ranges from 0 to 1; a Kappa value between 0.80 and 1.0 indicates that the two are almost perfectly consistent.
[20] The confusion matrix of this classification result is shown in Table 3. The overall classification accuracy is 91.8% and the Kappa coefficient is 0.85, indicating that the classification result has good consistency with the validation ROI.
[0062] Table 3. Confusion matrix of classification results for flooded tobacco fields
[0063] .
[0064] As the analysis above shows, there are significant differences in spectral characteristics between the affected tobacco fields, normal tobacco fields, and soil. Therefore, the spectral angle matching algorithm can accurately extract the affected areas from the tobacco field images and achieve good classification results (overall classification accuracy reaches 91.8%, Kappa coefficient 0.85).
[0065] The present invention has been described in detail above with reference to the accompanying drawings and embodiments. However, those skilled in the art will understand that, without departing from the concept of the present invention, various specific parameters in the above embodiments can be changed, or related methods and steps can be substituted in an equivalent manner, thereby forming multiple specific embodiments. These are all common variations of the present invention and will not be described in detail here.
Claims
1. A method for early identification of flooding in tobacco fields based on UAV hyperspectral imaging, characterized in that, Includes the following steps: (1) Use a drone equipped with a hyperspectral instrument to collect hyperspectral data of the area to be observed, and perform preprocessing to obtain the canopy reflectance data of the observation area; (2) Select the reflectance data of the blue, red and green bands from the previous canopy reflectance data to synthesize a true color image of the observation area; (3) Select pink areas, white and light green pixel sets, and green clusters on the true color image obtained in the previous step to establish three types of ROI corresponding to soil, disaster-stricken tobacco fields, and normal tobacco fields, respectively. Use the average spectral reflectance of each type of ROI as the spectral reflectance of that type of landform to generate the corresponding spectral curve. (4) Based on the spectral angle matching algorithm, the pixels are classified by processing the similarity difference between the spectral curves of each pixel in the true color image of the observation area and the spectral curves of the three types of ROIs, thereby classifying the observation area into three types of post-disaster surface types: flooded tobacco fields, normal tobacco fields, and soil.
2. The method for identifying flooding in tobacco fields based on UAV hyperspectral imaging according to claim 1, characterized in that, In step (1), the drone is set to fly at an altitude of 90-110 m and a speed of 2.8-3.2 m / s when collecting data. The data collection time is 2-3 hours after the end of the rainfall that caused the flooding.
3. The method for identifying flooding in tobacco fields based on UAV hyperspectral imaging according to claim 1, characterized in that, In step (1), the hyperspectral instrument is a Pika-L spectrometer with a spectral band range of 400–1000 nm, a spectral resolution of 2 nm, and a sampling interval of 4 nm.
4. The method for identifying flooding in tobacco fields based on UAV hyperspectral imaging according to claim 1, characterized in that, In step (1), the hyperspectral data preprocessing method includes: First, the data is segmented based on GPS data, and useless data during takeoff, landing, and turns is deleted. Then, the core area data is segmented according to the flight path. Subsequently, the segmented data is geometrically and radiometrically corrected using radiometric calibration files. The corrected data is then georeferenced and image-stitched into a hyperspectral cube using ArcGIS and ENVI. Finally, the canopy reflectance data of the observation area is obtained by inverting the target data of the known reflectance curves pre-placed in the observation field area.