A method and system for monitoring rice bacterial blight in the field based on multi-source data

By using multi-source data fusion and convolutional neural networks, cross-stage monitoring of rice bacterial blight was achieved, solving the problems of time-consuming and labor-intensive monitoring and chemical control in existing technologies, and improving the reliability and accuracy of monitoring.

CN114923908BActive Publication Date: 2024-11-12ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210608339.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-31
Publication Date
2024-11-12
Estimated Expiration
2042-05-31

AI Technical Summary

Technical Problem

Existing technologies lack methods for monitoring rice bacterial blight across different growth stages, making field disease monitoring time-consuming, labor-intensive, and susceptible to human factors. Chemical control leads to pathogen resistance and environmental pollution.

Method used

A multi-source data fusion method was adopted to acquire multispectral images, accumulated temperature data, and rice variety resistance information through UAVs. A disease severity regression model was constructed using a convolutional neural network and an attention mechanism module was used to determine the optimal band combination, thereby enabling rapid monitoring of the severity of rice bacterial blight.

Benefits of technology

This has improved the reliability and accuracy of monitoring rice bacterial blight across different growth stages, increased the efficiency and accuracy of disease monitoring, and reduced human intervention and environmental pollution.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114923908B_ABST
    Figure CN114923908B_ABST
Patent Text Reader

Abstract

The present invention relates to a method and system for monitoring rice bacterial blight in the field based on multi-source data, and relates to the field of crop disease monitoring. The method comprises: obtaining multispectral images, rice bacterial blight disease extent and accumulated temperature data of rice fields at different growth stages; obtaining resistance of rice varieties planted in rice fields to bacterial blight disease; extracting the average canopy spectral reflectance of each plot in the rice field; based on the average canopy spectral reflectance and the rice bacterial blight disease extent, using a convolutional neural network to regress the rice bacterial blight disease extent and output deep spectral features; using the accumulated temperature data, deep spectral features and resistance to bacterial blight disease corresponding to each plot as input, and using the corresponding rice bacterial blight disease extent as output to train a disease extent regression model; using the disease extent regression model to monitor the rice bacterial blight disease extent of the monitored rice field. The present invention realizes rapid monitoring of rice resistance to bacterial blight in the field at different growth stages.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of crop disease monitoring, and in particular to a method and system for monitoring rice bacterial blight in a field based on multi-source data. Background Art

[0002] Rice is attacked by pests and diseases during its growth, resulting in irreversible yield reduction and economic losses. Among them, bacterial leaf blight is one of the typical diseases that affect rice growth. It mainly causes leaf dryness, affects rice photosynthesis and nutrient transport, and thus causes yield reduction. In the field, once bacterial leaf blight occurs, rice yield can be reduced by 10-20%, and in severe cases, by more than 50%, or even a total crop failure, seriously threatening the safe production of rice.

[0003] At present, the main method for preventing and controlling rice bacterial blight is chemical control. Long-term use of chemicals will not only make pathogens resistant, but also cause environmental pollution. In addition, under field conditions, the incidence of rice often requires on-site investigation by plant protection personnel, which is time-consuming and labor-intensive, and will be affected by human subjective judgment and the environment. Timely monitoring of disease conditions and understanding of the severity of diseases are of great significance to rice field management and disease control.

[0004] In recent years, UAV remote sensing technology has developed rapidly. It has the advantages of strong flexibility, real-time data collection, and low cost. Among them, multispectral cameras carried by UAVs provide support for the detection and quantification of field crop diseases. Su et al. used five multispectral cameras carried by UAVs to develop a wheat yellow rust monitoring system suitable for farmland scale, and the average accuracy rate of identifying healthy and susceptible wheat reached 89.2%; Chivasa et al. conducted phenotypic analysis of the response of corn varieties to corn stripe virus disease based on multispectral data obtained by UAVs. The correlation between the phenotypic-related data analyzed from the spectral data of different growth periods and the scores of artificial corn stripe virus disease was significant (r = 0.74-0.84).

[0005] In reality, the phenotypic evaluation models applicable to different growth periods are different, and there is currently a lack of a universal disease monitoring method across different growth stages. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for monitoring rice bacterial blight in the field based on multi-source data, so as to realize rapid monitoring of rice resistance to bacterial blight in the field at different growth stages.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for monitoring rice bacterial blight in the field based on multi-source data, comprising:

[0009] Obtain multispectral images of rice fields at different growth stages;

[0010] Obtaining data on the degree of rice bacterial blight disease and accumulated temperature in the rice fields at different growth stages;

[0011] Obtaining the resistance of the rice variety planted in the rice field to bacterial blight;

[0012] Based on each of the multispectral images, extracting the average canopy spectral reflectance of each plot in the rice field; the rice field is divided into a plurality of plots;

[0013] Based on the average canopy spectral reflectance and rice bacterial blight disease severity of each plot, a convolutional neural network was used to regress the rice bacterial blight disease severity and output deep spectral features.

[0014] The accumulated temperature data, deep spectral characteristics and resistance to rice bacterial blight corresponding to each plot are used as input, and the corresponding rice bacterial blight disease severity is used as output to train the disease severity regression model, thereby obtaining a trained disease severity regression model;

[0015] The trained disease severity regression model is used to monitor the degree of rice bacterial blight disease in the monitored rice fields.

[0016] Optionally, it also includes:

[0017] Utilize the attention mechanism module in the convolutional neural network to output the best band combination, where the best band combination is a combination of wavelengths corresponding to the first N items of the wavelength weight values ​​output by the attention mechanism module arranged from large to small;

[0018] Constructing a spectral index for monitoring rice bacterial blight disease according to the optimal band combination;

[0019] The rice bacterial blight disease monitoring spectral index of the rice field to be monitored is obtained, and the rice bacterial blight disease monitoring spectral index of the rice field to be monitored is mapped to the multispectral image to obtain a visualized image of the rice bacterial blight disease degree of the rice field to be monitored.

[0020] Optionally, the acquiring of multispectral images of rice fields at different growth stages specifically includes:

[0021] Remote sensing images of rice fields at different growth stages were obtained by using drones equipped with multispectral cameras;

[0022] A multispectral image of the entire rice field is obtained by stitching the multiple remote sensing images.

[0023] Optionally, the obtaining of the rice bacterial blight disease severity and accumulated temperature data in the rice fields at different growth stages specifically includes:

[0024] The extent of rice bacterial blight in each plot was determined based on the proportion of rice bacterial blight-affected area in each plot to the total area;

[0025] The highest and lowest temperatures of each plot on the day when each multispectral image was acquired were recorded, and the accumulated temperature data of each plot was calculated using the accumulated temperature formula.

[0026] Optionally, extracting the average canopy spectral reflectance of each plot in the rice field based on each of the multispectral images specifically includes:

[0027] Using a threshold segmentation method to remove the background of each multispectral image of the rice field;

[0028] The average canopy spectral reflectance of each plot was extracted from the multispectral image after background removal.

[0029] Optionally, the disease severity regression model is a partial least squares regression model.

[0030] The present invention also discloses a field rice bacterial blight monitoring system based on multi-source data, comprising:

[0031] A multispectral image acquisition module, used to acquire multispectral images of rice fields at different growth stages;

[0032] A module for acquiring data on the degree of rice bacterial blight disease and accumulated temperature, for acquiring data on the degree of rice bacterial blight disease and accumulated temperature in the rice fields at different growth stages;

[0033] A module for obtaining resistance of rice varieties to bacterial blight disease, used for obtaining resistance of rice varieties planted in the rice field to bacterial blight disease;

[0034] An average canopy spectral reflectance extraction module is used to extract the average canopy spectral reflectance of each cell in the rice field based on each of the multispectral images; the rice field is divided into a plurality of cells;

[0035] The deep spectral feature output module is used to regress the degree of rice bacterial blight disease using a convolutional neural network based on the average canopy spectral reflectance and the degree of rice bacterial blight disease corresponding to each plot, and output the deep spectral features;

[0036] The disease severity regression model training module is used to take the accumulated temperature data, deep spectral characteristics and resistance to rice bacterial blight disease corresponding to each plot as input, and take the corresponding rice bacterial blight disease degree as output to train the disease severity regression model, so as to obtain a trained disease severity regression model;

[0037] The disease severity monitoring module is used to monitor the degree of rice bacterial blight disease in the monitored rice fields using a trained disease severity regression model.

[0038] Optionally, it also includes:

[0039] An optimal band combination determination module is used to output an optimal band combination using the attention mechanism module in the convolutional neural network, wherein the optimal band combination is a combination of wavelengths corresponding to the first N items after the wavelength weight values ​​output by the attention mechanism module are arranged from large to small;

[0040] A rice bacterial blight disease monitoring spectral index construction module, used to construct a rice bacterial blight disease monitoring spectral index according to the optimal band combination;

[0041] The module for determining the degree of rice bacterial blight disease by visualizing an image is used to obtain the spectral index for monitoring rice bacterial blight disease of the rice field to be monitored, map the spectral index for monitoring rice bacterial blight disease of the rice field to be monitored into a multispectral image, and obtain a visualizing image of the degree of rice bacterial blight disease of the rice field to be monitored.

[0042] Optionally, the multispectral image acquisition module specifically includes:

[0043] A remote sensing image acquisition unit, used to acquire remote sensing images of rice fields at different growth stages through a drone equipped with a multispectral camera;

[0044] The image stitching unit is used to stitch the multiple remote sensing images to obtain a multispectral image of the entire rice field.

[0045] Optionally, the bacterial blight disease extent and accumulated temperature data acquisition module specifically includes:

[0046] A bacterial blight disease degree determination unit is used to determine the bacterial blight disease degree in each plot according to the proportion of the bacterial blight disease area in each plot to the total area;

[0047] The accumulated temperature data determination unit is used to record the highest temperature and the lowest temperature of each plot on the day when each multispectral image is obtained, and calculate the accumulated temperature data of each plot through the accumulated temperature formula.

[0048] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0049] The present invention discloses a field rice bacterial blight monitoring method and system based on multi-source data, which realizes the monitoring of rice bacterial blight across different growth stages according to the fusion of multi-spectral data, field accumulated temperature data and rice variety resistance information to bacterial blight, and is conducive to improving the reliability of disease monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A method for monitoring rice bacterial blight in the field based on multi-source data is shown in the following figure. Figure 1 ;

[0052] Figure 2 A method for monitoring rice bacterial blight in the field based on multi-source data is shown in the following figure. Figure 2 ;

[0053] Figure 3 It is a schematic diagram of the convolutional neural network and multi-source data fusion method of the present invention;

[0054] Figure 4 This is a schematic diagram of fitting the true value and predicted value of the prediction set when only the deep spectral feature is used to train the disease severity regression model to predict bacterial blight in an embodiment of the present invention;

[0055] Figure 5 This is a schematic diagram of fitting the true value and predicted value of the prediction set when the disease severity regression model is trained using multi-source data to predict bacterial blight in an embodiment of the present invention;

[0056] Figure 6 A schematic diagram of the optimal band combination and its corresponding weights selected for an embodiment of the present invention;

[0057] Figure 7 This is a visualization diagram of the incidence of rice bacterial blight in the field according to an embodiment of the present invention;

[0058] Figure 8 The present invention is a schematic diagram of the structure of a field rice bacterial blight monitoring system based on multi-source data. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] The purpose of the present invention is to provide a method and system for monitoring rice bacterial blight in the field based on multi-source data, so as to realize rapid monitoring of rice resistance to bacterial blight in the field at different growth stages.

[0061] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0062] Example 1

[0063] Figure 1 Schematic diagram of the process of a method for monitoring bacterial blight of rice in the field based on multi-source data of the present invention Figure 1 , Figure 2 Schematic diagram of the process of a method for monitoring bacterial blight of rice in the field based on multi-source data of the present invention Figure 2 , such as Figure 1-2 shown, a method for monitoring bacterial blight of rice in the field based on multi-source data includes the following steps:

[0064] Step 101: Obtain multi-spectral images of paddy fields at different growth stages.

[0065] Among them, step 101 specifically includes:

[0066] Obtain remote sensing images of paddy fields at different growth stages through an unmanned aerial vehicle equipped with a multi-spectral camera.

[0067] Perform image stitching on multiple remote sensing images to obtain the overall multi-spectral image of the paddy field.

[0068] Step 102: Obtain the degree of bacterial blight of rice and accumulated temperature data in the paddy field at different growth stages.

[0069] Among them, step 102 specifically includes:

[0070] Determine the degree of bacterial blight in each plot according to the proportion of the diseased area of bacterial blight of rice in the total area of each plot. Specifically, conduct on-site investigations in the field on the day when the unmanned aerial vehicle collects remote sensing images to record the true score of the degree of bacterial blight of rice in the field.

[0071] Record the ratio of the diseased area to the total area as R.

[0072] When R is 0, the rice is healthy without bacterial blight; when 0 < R <= 10%, the disease degree is level 1; when 10% < R <= 25%, the disease degree is level 2; when 25% < R <= 50%, the disease degree is level 3; when 50% < R <= 75%, the disease degree is level 4; when 75% < R <= 100%, the disease degree is level 5.

[0073] Record the highest temperature and the lowest temperature of each plot on the day when each multi-spectral image is obtained, and calculate the accumulated temperature data of each plot through the accumulated temperature formula.

[0074] The accumulated temperature formula is: Among them, AT is the accumulated temperature data on the day when the UAV collects remote sensing images, n is the number of days of unmanned flight (the number of days of collecting remote sensing images), T mean It is the average of the daily maximum and minimum temperatures.

[0075] Step 103: Obtain the resistance of the rice variety planted in the rice field to bacterial blight disease.

[0076] The resistance of rice varieties to bacterial leaf blight includes highly resistant, resistant, moderately resistant, highly susceptible, susceptible and moderately susceptible.

[0077] Step 104: extracting the average canopy spectral reflectance of each plot in the rice field based on each multispectral image; the rice field is divided into a plurality of plots.

[0078] Wherein, step 104 specifically includes:

[0079] The threshold segmentation method is used to remove the background of each multispectral image of the rice field.

[0080] The threshold segmentation method is used to remove the background of each multispectral image of the rice field, including:

[0081] The segmentation threshold was determined based on the difference in reflectance between the background and the rice canopy.

[0082] The background of the multispectral image was removed using the segmentation threshold to obtain the multispectral data of each plot in the rice field.

[0083] The average canopy spectral reflectance of each plot was extracted from the multispectral image of each plot after background removal.

[0084] Step 105: Based on the average canopy spectral reflectance and rice bacterial blight disease severity corresponding to each plot, a convolutional neural network is used to regress the rice bacterial blight disease severity and output a deep spectral feature.

[0085] The convolutional neural network contains an attention mechanism module, which is used to output the weight value of each wavelength in the deep spectral feature after averaging the canopy spectrum input.

[0086] The optimal band combination for field rice bacterial blight monitoring was screened based on the band weight values ​​output by the attention mechanism module in the convolutional neural network model.

[0087] The attention mechanism module in the convolutional neural network is used to output the optimal band combination. The optimal band combination is the combination of wavelengths corresponding to the first N items after the wavelength weight values ​​output by the attention mechanism module are arranged from large to small.

[0088] The spectral index for monitoring rice bacterial blight disease was constructed based on the optimal band combination.

[0089] The rice bacterial blight disease monitoring spectral index of the rice field to be monitored is obtained, and the rice bacterial blight disease monitoring spectral index of the rice field to be monitored is mapped to the multispectral image to obtain a visualized image of the rice bacterial blight disease degree of the rice field to be monitored.

[0090] Step 106: using the accumulated temperature data, deep spectrum characteristics and resistance to rice bacterial blight disease corresponding to each plot as input and the corresponding rice bacterial blight disease degree as output to train a disease degree regression model, thereby obtaining a trained disease degree regression model;

[0091] The disease severity regression model is the partial least squares regression model (PLSR model).

[0092] Step 107: Use the trained disease severity regression model to monitor the degree of rice bacterial blight disease in the monitored rice field.

[0093] The present invention discloses the following technical effects:

[0094] (1) The fusion of UAV multispectral data, field accumulated temperature data, and rice variety resistance information to rice bacterial blight has achieved stable monitoring of rice bacterial blight across different growth stages, which is conducive to improving the reliability of disease monitoring; (2) Based on the constructed spectral indicators, a field disease distribution map was drawn, providing technical support for the precise monitoring of rice diseases in the field.

[0095] Embodiment 2

[0096] (1) UAV multispectral data collection

[0097] The implementation site of the embodiment is located in the rice experimental base in Longyou County, Quzhou City, Zhejiang Province. The field is set up as 60 plots, each with an area of ​​10.6m×4.72m. Each adjacent plot is planted with a rice variety, involving a total of 60 rice varieties. Different varieties have different resistance to bacterial blight, and the field resistance phenotype is also different. The climate is subtropical monsoon climate, with abundant rainfall and humid air, which is suitable for rice growth. Importantly, this test base is an endemic area for bacterial blight, which occurs naturally after rice planting without additional treatment. A drone equipped with a multispectral camera was used, 25m above the ground, to collect multispectral images on the 16th, 66th and 92nd days after rice transplanting, respectively, involving the tillering stage, jointing stage, heading stage and filling stage of rice growth. The flight speed is 2.5m / s, and the heading and lateral overlap rates are 60% and 75% respectively. The multispectral camera contains 23 bands with a spectral range of 600-875nm.

[0098] (2) Data processing

[0099] Photoscan software was used to stitch the acquired multispectral images of the field and form an orthophoto of the rice field, and the background was removed based on the threshold segmentation method. According to the threshold segmentation method, the difference between the background and the rice canopy reflectance is the largest at 675nm. After the background is removed, the multispectral reflectance information of the rice canopy in the field can be extracted. The multispectral reflectance of each plot is extracted and the average value is taken to represent the reflectance data of the plot.

[0100] (3) Obtaining the actual disease extent of each plot in the field

[0101] On the same day after the drone flew, a field survey was conducted in the field to record the actual extent of rice bacterial blight disease in the field. Specifically, the actual disease extent of each plot in the field was recorded according to the evaluation criteria for the degree of field bacterial blight disease in the national standard GB / T 17980.19-2000. The disease extent ranged from 1 to 5, and healthy and disease-free plants were recorded as 0.

[0102] (4) Accumulated temperature data acquisition

[0103] The highest and lowest temperatures of each day were recorded after the rice was transplanted into the field, and the accumulated temperature data of each drone flight day was calculated based on the accumulated temperature calculation formula.

[0104] (5) Obtaining information on rice varieties

[0105] The rice samples of this embodiment were provided by Zhejiang Academy of Agricultural Sciences, and the resistance information of each variety of rice to bacterial blight was provided and recorded.

[0106] (6) Deep spectral feature extraction

[0107] The average canopy spectral reflectance of each plot was extracted and matched with the actual field disease severity, accumulated temperature data and rice variety resistance information to bacterial blight in each plot, and then the training set and prediction set were divided into 8:2 ratios. The spectral data in the training set was input into the convolutional neural network model to regress the disease severity.

[0108] The structure of the convolutional neural network model is as follows Figure 3As shown in the figure, it consists of an attention mechanism module, a convolution layer, two Dense layers (fully connected layers) and an output layer. The number of convolution kernels in the convolution layer is 16, the kernel size is 3, the stride is 1, and ELU activation is used as the activation function, followed by a Max Pooling layer; the number of neural nodes in the two Dense layers is 128 and 16 respectively, ReLU is used as the activation function, and BatchNormalization is added after each layer to improve the generalization ability of the model; finally, the output layer is reached, L2loss is used as the loss function, and the evaluation output of the disease degree is obtained. In the training stage, the batch size is 5, and a dynamic learning rate is used. At the beginning, a relatively large learning rate of 0.0005 is set to speed up the training process of the first 300 epochs, and then it is reduced to 0.00005 in the next 300 epochs. The stop-first strategy can be used for training. When R 2 When the correlation coefficient exceeds 0.8, the training output can be stopped.

[0109] In this embodiment, the convolutional neural network model architecture is implemented on MXNet1.4.0, and the output of the second Dense layer of the convolutional neural network model is the deep spectral feature.

[0110] (7) Assessment of the severity of rice bacterial blight in the field based on multi-source data

[0111] The deep spectral features and accumulated temperature data as well as the resistance information of rice varieties to bacterial blight were integrated, and a data set was constructed corresponding to the actual disease severity in the field. The data set was divided into a training set and a prediction set in a ratio of 8:2, and a PLSR model was established to predict the disease severity. The PLSR model was trained using the leave-one-out cross-validation method, and the final prediction set R 2 The values ​​are 0.8672, RMSE is 0.63, and RPD is 2.72, indicating that the model prediction results are very good. The fitting diagram of the true value and the predicted value of the prediction set is as follows: Figure 5 (Modeling of deep spectral features fused with accumulated temperature data and rice variety resistance information to bacterial blight) Figure 4 The prediction set results of the disease severity prediction using only the deep spectral features to establish the PLSR model are shown. In comparison, the method of the present invention has better results.

[0112] (8) Determination of the optimal band combination

[0113] In this embodiment, the convolutional neural network model deeply learns the characteristic information in the rice multi-spectral data through convolutional layers and dense layers. The attention mechanism module makes the network pay more attention to the important areas in the input spectral data, and then outputs the weight value corresponding to each band. The weight value corresponding to each band outputted is as follows: Figure 6 As shown, Figure 6 The middle vertical axis is the weight value. The four bands with higher weight values ​​are selected as the best band combination, which are 611nm, 675nm, 713nm and 633nm.

[0114] (9) Construction of spectral index for monitoring rice bacterial blight in the field

[0115] The weight values ​​corresponding to the four wavelengths in the optimal band combination are 1.25, 1.16, 1.12 and 1.09 respectively. The spectral index SI for monitoring rice bacterial blight in the field is constructed, and its calculation formula is:

[0116] SI=1.25R 611 +1.16R 675 +1.12R 713 +1.09R 633

[0117] Where R 611 , R 675 , R 713 and R 633 They represent the spectral reflectance values ​​corresponding to wavelengths of 611nm, 675nm, 713nm and 633nm respectively.

[0118] (10) Visualization of rice bacterial blight severity in the field

[0119] In this embodiment, the constructed field rice bacterial blight disease monitoring spectral index is calculated and mapped to the multispectral image of the entire experimental field, and finally a visualized image of the degree of field rice bacterial blight disease is obtained, such as Figure 7 shown.

[0120] Embodiment 3

[0121] Figure 8 FIG. 1 is a schematic diagram of the structure of a field rice bacterial blight monitoring system based on multi-source data of the present invention. Figure 8 As shown, a field rice bacterial blight monitoring system based on multi-source data includes:

[0122] The multispectral image acquisition module 201 is used to acquire multispectral images of rice fields at different growth stages.

[0123] The bacterial blight disease extent and accumulated temperature data acquisition module 202 is used to acquire the bacterial blight disease extent and accumulated temperature data of rice in rice fields at different growth stages.

[0124] The module 203 for obtaining resistance of rice varieties to bacterial blight disease is used to obtain resistance of rice varieties planted in rice fields to bacterial blight disease.

[0125] The average canopy spectral reflectance extraction module 204 is used to extract the average canopy spectral reflectance of each plot in the rice field based on each multispectral image; the rice field is divided into a plurality of plots.

[0126] The deep spectral feature output module 205 is used to perform rice bacterial blight disease severity regression using a convolutional neural network based on the average canopy spectral reflectance and rice bacterial blight disease severity corresponding to each plot, and output the deep spectral feature.

[0127] The disease severity regression model training module 206 is used to take the accumulated temperature data, deep spectral characteristics and resistance to rice bacterial blight disease corresponding to each plot as input, and take the corresponding rice bacterial blight disease severity as output to train the disease severity regression model, so as to obtain a trained disease severity regression model.

[0128] The disease severity regression model is a partial least squares regression model.

[0129] The disease degree monitoring module 207 is used to monitor the degree of rice bacterial blight disease in the monitored rice field using a trained disease degree regression model.

[0130] A field rice bacterial blight monitoring system based on multi-source data also includes:

[0131] The optimal band combination determination module is used to output the optimal band combination using the attention mechanism module in the convolutional neural network. The optimal band combination is a combination of wavelengths corresponding to the first N items after the wavelength weight values ​​output by the attention mechanism module are arranged from large to small.

[0132] The module for constructing the spectral index for monitoring rice bacterial leaf blight disease is used to construct the spectral index for monitoring rice bacterial leaf blight disease based on the optimal band combination.

[0133] The module for determining the degree of rice bacterial blight disease by visualizing an image is used to obtain the spectral index for monitoring rice bacterial blight disease of the rice field to be monitored, map the spectral index for monitoring rice bacterial blight disease of the rice field to be monitored into a multispectral image, and obtain a visualizing image of the degree of rice bacterial blight disease of the rice field to be monitored.

[0134] The multispectral image acquisition module 201 specifically includes:

[0135] The remote sensing image acquisition unit is used to acquire remote sensing images of rice fields at different growth stages through an unmanned aerial vehicle equipped with a multispectral camera.

[0136] The image stitching unit is used to stitch multiple remote sensing images to obtain a multispectral image of the entire rice field.

[0137] The module 202 for acquiring the degree of bacterial blight disease and accumulated temperature data specifically includes:

[0138] The bacterial blight disease degree determination unit is used to determine the bacterial blight disease degree in each plot according to the proportion of the rice bacterial blight disease area in each plot to the total area.

[0139] The accumulated temperature data determination unit is used to record the highest temperature and the lowest temperature of each plot on the day when each multispectral image is obtained, and calculate the accumulated temperature data of each plot through the accumulated temperature formula.

[0140] The average canopy spectral reflectance extraction module 204 specifically includes:

[0141] The threshold segmentation method is used to remove the background of each multispectral image of the rice field.

[0142] The average canopy spectral reflectance of each plot was extracted from the multispectral image after background removal.

[0143] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0144] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for monitoring rice bacterial blight in the field based on multi-source data. It is characterized in that include: Obtain multispectral images of rice fields at different growth stages; Obtaining data on the degree of rice bacterial blight disease and accumulated temperature in the rice fields at different growth stages; Obtaining the resistance of the rice variety planted in the rice field to bacterial blight; Based on each of the multispectral images, extracting the average canopy spectral reflectance of each plot in the rice field; the rice field is divided into a plurality of plots; Based on the average canopy spectral reflectance and rice bacterial blight disease severity of each plot, a convolutional neural network was used to regress the rice bacterial blight disease severity and output deep spectral features. The accumulated temperature data, deep spectral characteristics and resistance to rice bacterial blight corresponding to each plot are used as input, and the corresponding rice bacterial blight disease severity is used as output to train the disease severity regression model, thereby obtaining a trained disease severity regression model; The trained disease severity regression model is used to monitor the degree of rice bacterial blight disease in the monitored rice fields.

2. The method for monitoring rice bacterial blight in the field based on multi-source data according to claim 1, It is characterized in that Also includes: Utilize the attention mechanism module in the convolutional neural network to output the best band combination, where the best band combination is a combination of wavelengths corresponding to the first N items of the wavelength weight values ​​output by the attention mechanism module arranged from large to small; Constructing a spectral index for monitoring rice bacterial blight disease according to the optimal band combination; The rice bacterial blight disease monitoring spectral index of the rice field to be monitored is obtained, and the rice bacterial blight disease monitoring spectral index of the rice field to be monitored is mapped to the multispectral image to obtain a visualized image of the rice bacterial blight disease degree of the rice field to be monitored.

3. The method for monitoring rice bacterial blight in the field based on multi-source data according to claim 1, It is characterized in that The step of obtaining multispectral images of rice fields at different growth stages specifically includes: Remote sensing images of rice fields at different growth stages were obtained by using drones equipped with multispectral cameras; A multispectral image of the entire rice field is obtained by stitching the multiple remote sensing images.

4. The method for monitoring rice bacterial blight in the field based on multi-source data according to claim 1, It is characterized in that The obtaining of the rice bacterial blight disease degree and accumulated temperature data in the rice fields at different growth stages specifically includes: The extent of rice bacterial blight in each plot was determined based on the proportion of rice bacterial blight-affected area in each plot to the total area; The highest and lowest temperatures of each plot on the day when each multispectral image was acquired were recorded, and the accumulated temperature data of each plot was calculated using the accumulated temperature formula.

5. The method for monitoring rice bacterial blight in the field based on multi-source data according to claim 1, It is characterized in that The step of extracting the average canopy spectral reflectance of each plot in the rice field based on each of the multispectral images specifically includes: Using a threshold segmentation method to remove the background of each multispectral image of the rice field; The average canopy spectral reflectance of each plot was extracted from the multispectral image after background removal.

6. The method for monitoring rice bacterial blight in the field based on multi-source data according to claim 1, It is characterized in that The disease degree regression model is a partial least squares regression model.

7. A field rice bacterial blight monitoring system based on multi-source data, It is characterized in that include: A multispectral image acquisition module, used to acquire multispectral images of rice fields at different growth stages; A module for acquiring data on the degree of rice bacterial blight disease and accumulated temperature, for acquiring data on the degree of rice bacterial blight disease and accumulated temperature in the rice fields at different growth stages; A module for obtaining resistance of rice varieties to bacterial blight disease, used for obtaining resistance of rice varieties planted in the rice field to bacterial blight disease; An average canopy spectral reflectance extraction module is used to extract the average canopy spectral reflectance of each cell in the rice field based on each of the multispectral images; the rice field is divided into a plurality of cells; The deep spectral feature output module is used to regress the degree of rice bacterial blight disease using a convolutional neural network based on the average canopy spectral reflectance and the degree of rice bacterial blight disease corresponding to each plot, and output the deep spectral features; The disease severity regression model training module is used to take the accumulated temperature data, deep spectral characteristics and resistance to rice bacterial blight disease corresponding to each plot as input, and take the corresponding rice bacterial blight disease degree as output to train the disease severity regression model, so as to obtain a trained disease severity regression model; The disease severity monitoring module is used to monitor the degree of rice bacterial blight disease in the monitored rice fields using a trained disease severity regression model.

8. The field rice bacterial blight monitoring system based on multi-source data according to claim 7, It is characterized in that Also includes: An optimal band combination determination module is used to output an optimal band combination using the attention mechanism module in the convolutional neural network, wherein the optimal band combination is a combination of wavelengths corresponding to the first N items after the wavelength weight values ​​output by the attention mechanism module are arranged from large to small; A rice bacterial blight disease monitoring spectral index construction module, used to construct a rice bacterial blight disease monitoring spectral index according to the optimal band combination; The module for determining the degree of rice bacterial blight disease by visualizing an image is used to obtain the spectral index for monitoring rice bacterial blight disease of the rice field to be monitored, map the spectral index for monitoring rice bacterial blight disease of the rice field to be monitored into a multispectral image, and obtain a visualizing image of the degree of rice bacterial blight disease of the rice field to be monitored.

9. The field rice bacterial blight monitoring system based on multi-source data according to claim 7, It is characterized in that The multispectral image acquisition module specifically includes: A remote sensing image acquisition unit, used to acquire remote sensing images of rice fields at different growth stages through a drone equipped with a multispectral camera; The image stitching unit is used to stitch the multiple remote sensing images to obtain a multispectral image of the entire rice field.

10. The field rice bacterial blight monitoring system based on multi-source data according to claim 7, It is characterized in that The module for acquiring the degree of bacterial blight disease and accumulated temperature data specifically comprises: A bacterial blight disease degree determination unit is used to determine the bacterial blight disease degree in each plot according to the proportion of the bacterial blight disease area in each plot to the total area; The accumulated temperature data determination unit is used to record the highest temperature and the lowest temperature of each plot on the day when each multispectral image is obtained, and calculate the accumulated temperature data of each plot through the accumulated temperature formula.