A method and device for analyzing the degree of damage of tobacco based on feature change

Abstract

The application provides a tobacco disaster degree analysis method and device based on feature change, and belongs to the field of tobacco health detection. The method extracts specific time sequence features through time sequence remote sensing images and the time of tobacco suffering from waterlogging disaster, then predicts the tobacco features under the assumption that the disaster does not occur according to the time sequence features before the disaster, and further analyzes the influence of the waterlogging disaster on the health condition of the tobacco based on the change detection between the predicted features and the actual features. The method analyzes the features of the tobacco in the growth period when the tobacco suffers from disaster, selects the corresponding vegetation index features for extraction, and further ensures that the remote sensing features used for evaluating the health condition of the tobacco have higher sensitivity to the health condition of the tobacco in the period, and the accurate disaster situation monitoring can be realized through the remote sensing technical means.

Description

Technical Field

[0001] This invention belongs to the field of tobacco health detection, and in particular relates to a method and apparatus for analyzing the degree of tobacco damage based on characteristic changes. Background Technology

[0002] Flue-cured tobacco is a drought-resistant crop. Flooding during the ripening period causes the main chemical components of flue-cured tobacco to become less balanced and the content of aroma substances to decrease, which is not conducive to the formation of high-quality tobacco leaves.

[0003] With global warming and the increasing frequency of extreme weather events, the impact of floods on tobacco production is intensifying. Objectively and rapidly monitoring the extent of waterlogging in flue-cured tobacco, quantitatively assessing its impact on the health of the tobacco plants, is of significant practical importance for estimating later-stage tobacco yields, implementing timely and effective control measures, and minimizing losses for tobacco farmers.

[0004] As the demand for efficient, timely, and precise flood and waterlogging monitoring increases, traditional manual methods of flood and waterlogging disaster investigation are gradually being phased out. Other flood and waterlogging disaster monitoring methods, based on precipitation data from meteorological stations, typically only provide a high-precision result for a single monitoring point and are difficult to extend to the entire spatial plane, thus failing to achieve precise regional monitoring.

[0005] In recent years, with the maturity of satellite remote sensing technology, research on monitoring agricultural disasters using remote sensing image analysis has been increasing. A method for monitoring spring inchworm infestations in Populus euphratica forests based on hyperspectral remote sensing (patent application number: CN201810470030.7) has been published, including the following steps: First, spectral measurements of sample trees are performed using a spectrometer to obtain ground hyperspectral data of all plants within the monitoring area; second, the obtained ground hyperspectral data of healthy and damaged plants are processed separately; third, the spectral characteristics of Populus euphratica are analyzed, and the red-edge effect is used to monitor the severity of spring inchworm infestations in Populus euphratica forests; fourth, a satellite remote sensing monitoring model for spring inchworm infestations in Populus euphratica forests based on red-edge feature parameters is constructed; fifth, Hyperion data of the area to be monitored is incorporated into the satellite remote sensing monitoring model for spring inchworm infestations in Populus euphratica forests based on red-edge feature parameters.

[0006] This publicly available scheme monitors the severity of spring inchworm infestations in poplar forests solely through red-edge feature parameters, failing to consider the sensitivity of features extracted from remote sensing images to the state of crops at different growth stages. Sensitive vegetation index features differ at different growth stages, especially for short-term crops with one or two crops per year. The timing of the disaster corresponds to different growth stages of the crop. If disaster monitoring is based solely on specific vegetation index features, significant errors will occur, resulting in inconsistent monitoring outcomes. The scheme constructs a model using ground hyperspectral data from healthy and damaged plants. Although the model's accuracy was screened to use the most accurate model for monitoring, it still ignores the differences in crop growth before and after the disaster. For poplar forests, with their long growth period and minimal changes, this has little impact. However, for crops like tobacco, with their short growth period and rapid growth changes, ignoring these differences in crop growth leads to significant errors in the monitoring results. Summary of the Invention

[0007] This application provides a method and apparatus for analyzing the degree of tobacco damage based on feature changes, aiming to solve the problems of existing technologies for monitoring the degree of crop damage, which fail to select different sensitive vegetation index features according to different growth stages of crops and ignore the differences brought about by the growth of crops themselves, resulting in significant errors in monitoring results and inconsistent monitoring effects.

[0008] To achieve the above objectives, the present invention adopts the following technical solution, including:

[0009] The time-series remote sensing data of the tobacco field to be measured is acquired and preprocessed to obtain time-series remote sensing images;

[0010] The growth stage of tobacco in the field to be tested was determined based on the time of the disaster, and vegetation indices were selected according to the growth stage to extract features from the time-series remote sensing images, resulting in the first set of time-series feature images.

[0011] The first time series feature image set is divided according to time to obtain the second time series feature image set of the previous N years when there was no disaster, the third time series feature image set before the disaster in this year, and the first feature image set after the disaster in this year, where N is a natural number;

[0012] The feature change prediction model is trained and tested based on the second time-series feature image set, and an LSTM prediction model is constructed.

[0013] The third time-series feature image set is used as input and substituted into the LSTM prediction model to calculate the second feature image;

[0014] Based on the analysis of feature changes using the first and second feature images, the degree of damage to tobacco in the tested tobacco field is obtained.

[0015] As a preferred option, the disaster is waterlogging disaster; the growth stages of tobacco are divided into seedling stage-rooting stage, vigorous growth stage, and maturity stage.

[0016] As a preferred option, the vegetation index is selected based on the growth stage, specifically:

[0017] When waterlogging occurs during the seedling to rooting stage of tobacco, the soil improvement and vegetation index (MSAVI) should be adjusted. Where NIR is the reflectance in the near-infrared band and RED is the reflectance in the red band;

[0018] When waterlogging occurs during the peak growing season of tobacco, the Red Edge Chlorophyll Vegetation Index (RECI) is selected, RECI = (NIR / RED) - 1;

[0019] When waterlogging occurs during the tobacco ripening period, the Normalized Differential Red Edge Vegetation Index (NDRE) is selected, where NDRE = (NIR - REG) / (NIR + REG), and REG is the reflectance of the red edge band.

[0020] Preferably, the feature change prediction model is trained and tested based on the second time-series feature image set, and an LSTM prediction model is constructed, specifically as follows:

[0021] The second temporal feature image set is divided into a training dataset and a test dataset;

[0022] The training dataset is organized into m sets of input data and standardized, and then used to train the neural network. Here, m is a natural number.

[0023] The test dataset is fed into the trained neural network for testing to obtain the LSTM prediction model.

[0024] As a preferred method, feature changes are analyzed based on the first and second feature images to obtain the degree of damage to tobacco in the tested tobacco field, specifically:

[0025] The feature value of each pixel in the first feature image is subtracted from the feature value of the corresponding pixel in the second feature image in turn. The difference results are then summarized to obtain the first feature value change map.

[0026] Pixels with differences less than 0 in the first feature value change map are extracted, and the absolute value of the difference results is taken according to the preset evaluation criteria and analyzed to obtain the result of the degree of tobacco damage in the tobacco field to be tested.

[0027] As a preferred option, the preprocessing specifically includes:

[0028] First, the time-series remote sensing data is filtered by the tobacco growth stages;

[0029] Then, the vector data of the tobacco field to be tested is used to spatially filter and crop the filtered data;

[0030] Finally, a cloud mask was constructed using the QA60 band, and the filtered and cropped data was processed to remove clouds. Outliers were removed using the SLC band to obtain time-series remote sensing images.

[0031] A device for analyzing the degree of tobacco damage based on characteristic changes, comprising:

[0032] Image data acquisition module: used to acquire time-series remote sensing data of the tobacco field to be measured and perform preprocessing to obtain time-series remote sensing images;

[0033] The temporal feature image set extraction module is used to determine the growth stage of tobacco in the tobacco field to be tested based on the occurrence time of the disaster, and select vegetation index according to the growth stage to extract features from the temporal remote sensing image to obtain the first temporal feature image set.

[0034] The temporal feature image set partitioning module is used to partition the first temporal feature image set according to time, resulting in the second temporal feature image set of the previous N years without disaster, the third temporal feature image set before the disaster in this year, and the first feature image set after the disaster in this year, where N is a natural number.

[0035] LSTM prediction model building module: used for training and testing of feature change prediction models based on the second time-series feature image set, and building LSTM prediction models;

[0036] Feature image calculation module: Used to take the third time series feature image set as input, substitute it into the LSTM prediction model to calculate and obtain the second feature image;

[0037] Disaster Severity Analysis Module: This module is used to analyze feature changes based on the first and second feature images to obtain the disaster severity results of tobacco in the tested tobacco field.

[0038] An electronic device includes a memory and a processor, the memory being used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement a method for analyzing the degree of tobacco damage based on feature changes as described above.

[0039] A computer-readable storage medium storing a computer program that, when executed by a computer, implements a method for analyzing the degree of tobacco damage based on feature changes as described above.

[0040] The present invention has the following beneficial effects:

[0041] (1) This scheme analyzes the characteristics of the growth period of tobacco when it is affected by disaster, and selects appropriate vegetation index characteristics for extraction. This ensures that the remote sensing characteristics used to evaluate the health status of tobacco are more sensitive to the health status of tobacco during this period, and can achieve accurate monitoring of the disaster situation through remote sensing technology.

[0042] (2) This scheme uses time-series Sentinel-2 data to predict the characteristic values ​​of tobacco under the assumption that no disaster occurs, and uses this to replace the theoretical characteristics before the disaster occurs (i.e. the characteristics corresponding when no disaster occurs). Then, it analyzes the changes between the predicted characteristic values ​​and the actual characteristic values ​​of tobacco after the disaster occurs. Based on the difference between the predicted characteristic values ​​and the actual characteristic values, it further evaluates the impact of waterlogging disaster on the health status of tobacco. Since the predicted characteristic values ​​are predicted in combination with the crop's own growth, this method weakens the impact of the differences brought about by the crop's own growth during the period from before the disaster to after the disaster to a certain extent, and improves the accuracy of disaster severity judgment.

[0043] (3) Based on time-series Sentinel-2 data, this scheme constructs a complete method for analyzing the impact of waterlogging disasters on tobacco health. It can effectively determine the degree of damage to tobacco affected by waterlogging disasters, guide farmers to carry out post-disaster relief work, and maximize the recovery of production and economic losses. Attached Figure Description

[0044] Figure 1 This is a flowchart of a method for analyzing the degree of tobacco damage based on feature changes in this invention;

[0045] Figure 2 This is a mind map illustrating a specific implementation scheme in Embodiment 1 of the present invention;

[0046] Figure 3 These are effect diagrams of tobacco at different growth stages in Embodiment 1 of the present invention;

[0047] Figure 4 This is a schematic diagram comparing the predicted feature values ​​and the actual feature values ​​in Embodiment 1 of the present invention;

[0048] Figure 5 This is a schematic diagram of the structure of the tobacco damage analysis device in this invention. Detailed Implementation

[0049] Example 1

[0050] like Figure 1 As shown, a method for analyzing the degree of tobacco damage based on characteristic changes includes the following steps:

[0051] S11. Acquire the time-series remote sensing data of the tobacco field to be measured and perform preprocessing to obtain the time-series remote sensing image.

[0052] S12. Determine the growth stage of tobacco in the tobacco field to be tested based on the time of the disaster, and select vegetation index according to the growth stage to extract features from the time-series remote sensing images to obtain the first time-series feature image set.

[0053] S13. Divide the first time series feature image set according to time to obtain the second time series feature image set of the previous N years when there was no disaster, the third time series feature image set before the disaster in this year, and the first feature image set after the disaster in this year, where N is a natural number.

[0054] S14. Train and test the feature change prediction model based on the second time-series feature image set, and construct the LSTM prediction model;

[0055] S15. Take the third time-series feature image set as input and substitute it into the LSTM prediction model to calculate the second feature image;

[0056] S16. Based on the first feature image and the second feature image, perform feature change analysis to obtain the result of the degree of damage to tobacco in the tobacco field to be tested.

[0057] In this embodiment, the scheme extracts specific temporal features from temporal Sentinel-2 images and the time of tobacco waterlogging disaster. Then, based on the pre-disaster temporal features, it predicts tobacco characteristics under the assumption that the disaster did not occur (the prediction time is consistent with the latest actual features after the disaster). Finally, based on the detection of changes between the predicted and actual features, it further analyzes the impact of waterlogging disaster on the health of tobacco. The mind map of the above scheme is shown below. Figure 2 As shown, the specific process is as follows:

[0058] 1. Acquisition and preprocessing of time-series Sentinel-2 data

[0059] Sentinel-2 is a multispectral imaging satellite consisting of two satellites, 2A and 2B. 2A was launched on June 23, 2015 at 01:52 UTC using a Vega rocket, and 2B was launched on March 7, 2017 at 09:49 UTC using the same Vega rocket. The revisit period for a single satellite is 10 days; the two satellites complement each other, with a revisit period of 5 days. Sentinel-2 carries a multispectral instrument (MSI) covering 13 spectral bands from visible light to shortwave infrared, with ground resolutions of 10m, 20m, and 60m. This study selects its Level-2A product as one of the foundational data for subsequent work; this product is atmospherically corrected low-level atmospheric reflectance data.

[0060] To minimize data processing time and improve work efficiency, this scheme utilizes the GEE (Google Earth Engine) remote sensing cloud computing platform for data acquisition and processing. Regarding data acquisition, firstly, Sentinel-2L2A data for each year are filtered temporally based on specific time ranges (i.e., tobacco growth stages). Then, vector data of the study area (i.e., the area of ​​the tobacco field to be measured) are used to spatially filter and crop the data. Next, cloud masks are constructed using the QA60 band of the imagery for cloud removal. Furthermore, considering that outliers in the imagery can affect crop identification, SLC bands are used for outlier removal.

[0061] 2. Feature Selection and Extraction

[0062] On the one hand, the biochemical components and appearance of tobacco change continuously during its growth process (see pictures of tobacco at different growth stages). Figure 3 Furthermore, considering the influence of environmental factors such as soil conditions, it is difficult to extract a single feature from remote sensing images to describe the entire growth stage of tobacco. On the other hand, it is usually impossible to determine which stage of tobacco growth corresponds to the timing of waterlogging disasters. Therefore, it is essential to select appropriate features to describe the growth status of tobacco based on the specific timing of waterlogging disasters. To address this, this scheme selects features in the following manner:

[0063] a. First, determine the growth stage of the tobacco plant when the waterlogging disaster occurred. If it is in the seedling-rooting stage [within 40 days after transplanting (this is a general value, which can be adjusted according to the actual situation in the study area, the same below)], then follow step b; if it is in the vigorous growth stage, then follow step c [40-70 days after transplanting]; if it is in the maturity stage [70 days after transplanting to harvest], then follow step d.

[0064] b. During the seedling to rooting stage, the main characteristic is that after the tobacco root system recovers, it begins to grow rapidly, while the stem and leaf growth is slow, resulting in weak plants. Combined with soil background interference, its features are not obvious in remote sensing images, making it difficult to distinguish its health status. Therefore, if waterlogging occurs during this stage, this scheme uses the Modified Soil Adjusted Vegetation Index (MSAVI) to describe the pre- and post-disaster characteristics of tobacco. MSAVI is adjusted for soil effects, reducing the impact of soil on crop monitoring results. It is sensitive to early-stage crops in the field and is suitable for situations with a high proportion of bare soil, sparse vegetation, or low chlorophyll content in plants. The MSAVI calculation formula is as follows:

[0065]

[0066] Wherein, NIR and RED represent the reflectance of the near-infrared and red bands of the preprocessed Sentinel-2 image, respectively;

[0067] c. During the vigorous growth period, the growth center of tobacco shifts to the above-ground parts, stems rapidly grow taller and thicker, leaves develop rapidly, leaf area expands rapidly, and metabolism is predominantly nitrogen-based. The Red-edged Chlorophyll Vegetation Index (RECI) significantly responds to the chlorophyll content in nitrogen-nourished leaves, indicating the photosynthetic activity of the canopy. It is often used to describe the characteristics of active vegetation development stages (not applicable to the harvest season). Therefore, if waterlogging occurs during this stage, RECI is chosen here to describe the pre- and post-disaster characteristics of tobacco. The RECI calculation formula is:

[0068] RECI = (NIR / RED) - 1

[0069] d. During the maturity stage, the lower leaves of tobacco gradually age and turn yellow, maturing sequentially from bottom to top, with metabolism shifting to primarily carbon metabolism. If waterlogging occurs during this stage, the Normalized Differential Red Edge Vegetation Index (NDRE) is used to describe the pre- and post-disaster characteristics of tobacco. NDRE is suitable for areas with high vegetation cover and is typically used to monitor crops that have reached maturity. The NDRE calculation formula is:

[0070] NDRE = (NIR - REG) / (NIR + REG),

[0071] REG represents the reflectance of the red-edge band of the preprocessed Sentinel-2 image.

[0072] After selecting appropriate features according to the above feature methods, feature extraction is performed using the preprocessed Sentinel-2 data (i.e., time-series remote sensing images) from Section 1. The extracted time-series feature image set (i.e., the first time-series feature image set) is divided into three groups according to time information: the time-series feature image set DATA1 for the previous N years without disaster (i.e., the second time-series feature image set), the time-series feature image set DATA2 for the current year before the disaster (the third time-series feature image set), and the latest feature image DATA3 for the current year after the disaster (i.e., the first feature image).

[0073] 3. Eigenvalue prediction under the assumption that no disaster will occur

[0074] 3.1 Prediction Model Training

[0075] LSTM is a variant of recurrent neural networks that can effectively capture long-term dependencies when processing time-series data, and is therefore widely used in time-series image prediction tasks. Through its internal gating structure and memory units, LSTM can selectively retain or forget historical information, thus taking more historical information into account when predicting time-series images and improving prediction accuracy.

[0076] The construction of the LSTM prediction model involved in this scheme is mainly carried out in four steps: (1) The DATA1 image set in Section 2 is further organized into an input dataset and divided into a training dataset DATA1_Train and a test dataset DATA1_Test; (2) The LSTM network model is constructed, and the parameters and objectives required for model training and prediction are confirmed; (3) The model is trained using the training data to establish a prediction model for the corresponding feature images; (4) The model is tested using the test dataset.

[0077] This scheme employs a multi-time-point input to multi-time-point output model construction. The DATA1_Train time series data is organized into m sets of input sequences and standardized. A fixed length of data is used as the input to the network for each set, and the data at the time point to be predicted serves as the theoretical output value of the model. A dual-layer LSTM and a dense fully connected layer are combined to form the hidden layer, minimizing the impact of the large-scale dynamics of the data. Since the feature images involved in training are all single-band images, the feature dimension of both the input and output is 1. The learning rate is fine-tuned after determining its magnitude based on the change in the loss value from 0.1 to 0.000001. The number of training iterations is determined based on the convergence of the loss value as the number of training iterations increases. Then, the number of nodes in the hidden layer is determined according to the empirical formula derived from previous research, as shown below.

[0078]

[0079] Where l is the number of hidden layer nodes, α and β are the number of input and output layer nodes respectively, and μ is an integer between 0 and 9. The l value corresponding to the minimum RMSE is found by traversing and optimizing.

[0080] During the training process, the weights of the network model are continuously updated using the predetermined parameters and gradient descent algorithm to obtain the hidden layer network. Finally, the trained network is used to predict the test dataset DATA1_Test, and the data is destandardized. The predicted values ​​at each prediction time point are then output and compared with the actual values ​​to verify the model's performance.

[0081] 3.2 Eigenvalue Prediction under the Assumption that No Disaster Occurs

[0082] Using DATA2 as input data to the LSTM prediction model trained in section 3.1, the model predicts the feature image DATA4 (i.e., the second feature image) at the same time as the feature image DATA3, assuming no disaster occurs. A schematic diagram of the prediction at a specific pixel location is shown below. Figure 4 As shown.

[0083] 4. Change detection and analysis of the impact of waterlogging disasters on tobacco health status

[0084] By comparing the predicted tobacco feature values ​​(DATA4 feature image) under the assumption that the disaster will not occur with the actual tobacco feature values ​​(DATA3 feature image) after the disaster occurs, the location and extent of the impact of waterlogging disaster on tobacco can be determined. Details are as follows:

[0085] a. First, the difference between the actual tobacco feature values ​​after the disaster and the predicted tobacco feature values ​​under the assumption that the disaster did not occur is calculated. That is, the difference between the two images is calculated by subtracting the feature image DATA4 from the feature image DATA3, and subtracting the pixels of the two images one by one. Finally, the feature value change map DATA5 (i.e. the first feature value change map) is obtained.

[0086] b. Extract the pixels in DATA5 that are less than 0. The corresponding positions of these pixels are the areas where tobacco health is affected. The larger the absolute value of the pixel, the greater the impact on tobacco health and the more severe the damage to tobacco. This judgment method is the preset evaluation standard.

[0087] Example 2

[0088] like Figure 5 As shown, a device for analyzing the degree of tobacco damage based on characteristic changes includes:

[0089] Image data acquisition module 10: used to acquire time-series remote sensing data of the tobacco field to be measured and perform preprocessing to obtain time-series remote sensing images;

[0090] Temporal feature image set extraction module 20: used to determine the growth stage of tobacco in the tobacco field to be tested based on the time of disaster occurrence, and select vegetation index according to the growth stage to extract features from the temporal remote sensing image to obtain the first temporal feature image set;

[0091] The temporal feature image set partitioning module 30 is used to partition the first temporal feature image set according to time to obtain the second temporal feature image set of the previous N years without disaster, the third temporal feature image set before the disaster in this year, and the first feature image set after the disaster in this year, where N is a natural number;

[0092] LSTM prediction model building module 40: used for training and testing of the feature change prediction model based on the second time-series feature image set, and to build the LSTM prediction model;

[0093] Feature image calculation module 50: It is used to take the third time series feature image set as input, substitute it into the LSTM prediction model to calculate and obtain the second feature image;

[0094] Disaster severity analysis module 60: used to analyze feature changes based on the first feature image and the second feature image to obtain the disaster severity results of tobacco in the tobacco field to be tested.

[0095] One embodiment of the above-mentioned device is as follows: In the image data acquisition module 10, time-series remote sensing data of the tobacco field to be tested is acquired and preprocessed to obtain time-series remote sensing images. In the time-series feature image set extraction module 20, the growth stage of tobacco in the tobacco field to be tested is determined according to the time of disaster occurrence, and vegetation indices are selected according to the growth stage to extract features from the time-series remote sensing images to obtain a first time-series feature image set. In the time-series feature image set division module 30, the first time-series feature image set is divided according to time to obtain a second time-series feature image set of the previous N years without disaster, and the current time-series feature image set of the previous N years without disaster. The third time-series feature image set before the annual disaster and the first feature image set after the disaster in the current year, where N is a natural number, are used in the LSTM prediction model construction module 40. Based on the second time-series feature image set, the feature change prediction model is trained and tested to construct the LSTM prediction model. In the feature image calculation module 50, the third time-series feature image set is used as input and substituted into the LSTM prediction model to calculate the second feature image. In the disaster severity analysis module 60, based on the first and second feature images, the feature change is analyzed to obtain the disaster severity result of tobacco in the tobacco field to be tested.

[0096] Example 3

[0097] Based on the above embodiments, this embodiment provides an electronic device.

[0098] Example 4

[0099] Based on the above embodiments, this embodiment provides a storage medium.

[0100] The above description is merely a specific embodiment of the present invention, but the technical features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art within the scope of the present invention are covered by the patent scope of the present invention.

Claims

1. A method for analyzing the degree of tobacco damage based on characteristic changes, characterized in that, include: The time-series remote sensing data of the tobacco field to be measured is acquired and preprocessed to obtain time-series remote sensing images; The growth stage of tobacco in the field to be tested was determined based on the time of the disaster, and vegetation indices were selected according to the growth stage to extract features from the time-series remote sensing images, resulting in the first set of time-series feature images. The first time series feature image set is divided according to time to obtain the second time series feature image set of the previous N years when there was no disaster, the third time series feature image set before the disaster in this year, and the first feature image set after the disaster in this year, where N is a natural number; The feature change prediction model is trained and tested based on the second time-series feature image set, and an LSTM prediction model is constructed. The third time-series feature image set is used as input and substituted into the LSTM prediction model to calculate the second feature image; Based on the analysis of feature changes in the first and second feature images, the severity of tobacco damage in the tested tobacco field was obtained; the damage was identified as waterlogging; the tobacco growth stages were divided into seedling-rooting stage, vigorous growth stage, and maturity stage; vegetation indices were selected according to the growth stages, specifically: When waterlogging occurs during the seedling to rooting stage of tobacco, the soil improvement and vegetation index (MSAVI) should be adjusted. Where NIR is the reflectance in the near-infrared band and RED is the reflectance in the red band; When waterlogging occurs during the peak growing season of tobacco, the Red-edged Chlorophyll Vegetation Index (RECI) is selected. ; When waterlogging occurs during the tobacco ripening period, the Normalized Differential Red Edge Vegetation Index (NDRE) is selected. Where REG represents the reflectivity of the red-edge band; the feature change prediction model is trained and tested based on the second time-series feature image set, and an LSTM prediction model is constructed, specifically as follows: The second temporal feature image set is divided into a training dataset and a test dataset; The training dataset is organized into m sets of input data and standardized, and then used to train the neural network. Here, m is a natural number. The test dataset is fed into the trained neural network for testing to obtain the LSTM prediction model.

2. The method for analyzing the degree of tobacco damage based on characteristic changes according to claim 1, characterized in that, Based on the analysis of feature changes in the first and second feature images, the degree of damage to tobacco in the tested tobacco field was obtained, specifically as follows: The feature value of each pixel in the first feature image is subtracted from the feature value of the corresponding pixel in the second feature image in turn. The difference results are then summarized to obtain the first feature value change map. Pixels with differences less than 0 in the first feature value change map are extracted, and the absolute value of the difference results is taken according to the preset evaluation criteria and analyzed to obtain the result of the degree of tobacco damage in the tobacco field to be tested.

3. The method for analyzing the degree of tobacco damage based on characteristic changes according to claim 1, characterized in that, The preprocessing is as follows: First, the time-series remote sensing data is filtered by the tobacco growth stages; Then, the vector data of the tobacco field to be tested is used to spatially filter and crop the filtered data; Finally, a cloud mask was constructed using the QA60 band, and the filtered and cropped data was processed to remove clouds. Outliers were removed using the SLC band to obtain time-series remote sensing images.

4. A device for analyzing the degree of tobacco damage based on characteristic changes, characterized in that, To implement the method for analyzing the degree of tobacco damage based on characteristic changes as described in any one of claims 1-3, the method includes: Image data acquisition module: used to acquire time-series remote sensing data of the tobacco field to be measured and perform preprocessing to obtain time-series remote sensing images; The temporal feature image set extraction module is used to determine the growth stage of tobacco in the tobacco field to be tested based on the occurrence time of the disaster, and select vegetation index according to the growth stage to extract features from the temporal remote sensing image to obtain the first temporal feature image set. The temporal feature image set partitioning module is used to partition the first temporal feature image set according to time, resulting in the second temporal feature image set of the previous N years without disaster, the third temporal feature image set before the disaster in this year, and the first feature image set after the disaster in this year, where N is a natural number. LSTM prediction model building module: used for training and testing of feature change prediction models based on the second time-series feature image set, and building LSTM prediction models; Feature image calculation module: Used to take the third time series feature image set as input, substitute it into the LSTM prediction model to calculate and obtain the second feature image; Disaster Severity Analysis Module: This module is used to analyze feature changes based on the first and second feature images to obtain the disaster severity results of tobacco in the tested tobacco field.

5. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement a method for analyzing the degree of tobacco damage based on feature changes as described in any one of claims 1-3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a computer, it implements a method for analyzing the degree of tobacco damage based on feature changes as described in any one of claims 1-3.

Images