Intelligent diagnosis system and method for rice field drought

Abstract

The present invention relates to the field of agricultural drought prevention and mitigation technology, and in particular to an intelligent rice field drought diagnosis system and method. The system first obtains paddy field images corresponding to each inspection point, obtains spatiotemporal data and the rice growing period corresponding to each inspection point, and then inputs the paddy field images, spatiotemporal data, and rice growing period corresponding to each inspection point into a patrol point drought grade diagnosis model for identification, thereby obtaining the drought grade corresponding to each inspection point. Based on the drought grade corresponding to each inspection point, the area ratio corresponding to each drought grade within the inspection area is determined. Finally, based on the area ratio corresponding to each drought grade within the inspection area, the drought grade corresponding to the inspection area is calculated. The system comprehensively considers the rice growing period and the spatiotemporal data, thereby accurately identifying the drought grade corresponding to each inspection point and the drought grade corresponding to the inspection area.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural drought prevention and resistance, and in particular to an intelligent diagnosis system and method for drought conditions in rice fields. Background Art

[0002] As a typical water-demanding crop, rice has significant differences in its water requirements during different growth stages. These complex water requirements make the diagnosis of drought in rice fields more complicated than in general farmland, and differentiated assessments need to be conducted based on the specific growth stage.

[0003] At present, the diagnosis and identification of rice field drought in my country mainly have the following problems: the traditional manual reporting method has problems such as inaccurate, untimely and highly subjective reporting; the image quality of remote sensing monitoring technology is seriously affected by the cloudy and rainy climate conditions in the south, so the accuracy of drought identification in different growth stages of rice is not high enough; in addition, the application of joint intelligent analysis of massive data of multiple factors such as rice images, time and space is still insufficient; most existing methods use a unified drought judgment standard, ignoring the differences in water requirements of rice in different growth stages. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent diagnosis system and method for rice field drought conditions to improve recognition accuracy.

[0005] To achieve the above object, the present invention provides an intelligent diagnosis method for drought in rice fields, the method comprising:

[0006] Navigate to each inspection point within the inspection area to obtain the paddy field image corresponding to each inspection point;

[0007] Obtaining the spatiotemporal data and rice growth period corresponding to each inspection point; the spatiotemporal data includes location information, image acquisition time, and meteorological data;

[0008] Inputting the paddy field image, the spatiotemporal data and the rice growth period corresponding to each inspection point into the inspection point drought level diagnosis model for identification, thereby obtaining the drought level corresponding to each inspection point;

[0009] Based on the drought levels corresponding to each inspection point, determine the area proportion corresponding to each drought level within the inspection area;

[0010] The drought level corresponding to the patrol area is calculated based on the area ratio corresponding to each drought level within the patrol area.

[0011] Optionally, the method further includes:

[0012] The paddy field images, spatiotemporal data and rice growing period corresponding to each historical inspection point are used as the data input set, and the drought level corresponding to each historical inspection point is used as the data output set. The multimodal agricultural scene perception network (MASPN) is used for training to obtain a drought level diagnosis model for the inspection point.

[0013] Optionally, the paddy field images, spatiotemporal data, and rice growth period corresponding to each historical inspection point are used as the data input set, and the drought level corresponding to each historical inspection point is used as the data output set. The multimodal agricultural scene perception network MASPN is used for training to obtain a drought level diagnosis model for the inspection point, specifically including:

[0014] Based on the ConvNeXt-V2 deep convolutional neural network architecture, feature extraction is performed on the paddy field images corresponding to each historical inspection point to obtain the image visual feature vector;

[0015] Based on the spatiotemporal data collected at each historical inspection point, feature extraction is performed to obtain spatiotemporal feature vectors;

[0016] Based on the water demand characteristics of the rice growth period corresponding to each historical inspection point, feature extraction is performed to obtain the growth period feature vector;

[0017] Adopting an adaptive cross-modal attention mechanism, training is performed based on the image visual feature vector, the spatiotemporal feature vector, the growth period feature vector, and the drought levels corresponding to each historical inspection point, and a total loss value is calculated;

[0018] Determine whether the total loss value meets the set requirements; if so, construct a drought level diagnosis model for the inspection point based on the weights corresponding to the three eigenvectors; if not, adjust the weights corresponding to the three eigenvectors through the back propagation algorithm.

[0019] Optionally, the method further includes:

[0020] A spatial cluster analysis is performed on the drought data corresponding to each inspection point in the inspection area to generate a spatial distribution heat map and hot spots corresponding to the inspection area; the drought data includes longitude and latitude, the drought level corresponding to each inspection point, and the coding of the growth period.

[0021] Optionally, the method further includes:

[0022] A method combining inverse distance weighted interpolation and Kriging interpolation was used to perform spatial interpolation on the drought level corresponding to each inspection point and generate a continuous drought distribution surface.

[0023] The present invention also discloses an intelligent diagnosis system for drought conditions in rice fields, the system comprising:

[0024] A mobile terminal, a server and a service end of a drought inspection system; the mobile terminal is connected to the service end through the server;

[0025] The mobile terminal includes: a camera and a GPS positioning module; the camera is used to obtain paddy field images corresponding to each inspection point; the GPS positioning module is used to locate each inspection point and obtain location information;

[0026] The drought inspection system includes: a data acquisition module and an intelligent recognition module;

[0027] The data acquisition module is used to obtain paddy field images, spatiotemporal acquisition data and rice growth period corresponding to each inspection point; the spatiotemporal acquisition data includes location information, image acquisition time and meteorological data;

[0028] The intelligent recognition module is used to input the paddy field image, the spatiotemporal data and the rice growth period corresponding to each inspection point into the inspection point drought level diagnosis model for identification, so as to obtain the drought level corresponding to each inspection point;

[0029] The server includes: an intelligent analysis module;

[0030] The intelligent analysis module is used to receive the drought level corresponding to each patrol point sent by the mobile terminal; and determine the area ratio corresponding to each drought level in the patrol area based on the drought level corresponding to each patrol point; and calculate the drought level corresponding to the patrol area based on the area ratio corresponding to each drought level in the patrol area;

[0031] The server includes: a data display module;

[0032] The data display module is used to display the drought level corresponding to each inspection point and / or the drought level corresponding to the inspection area sent by the server.

[0033] Optionally, the drought inspection system further includes:

[0034] User management module, used to integrate patrol personnel type selection, personal information management, and patrol task management functions;

[0035] The map navigation module is used to realize paddy field inspection location positioning, inspection point marking, inspection range query, inspection map zooming and inspection route planning.

[0036] Optionally, the server further includes:

[0037] a data processing module for receiving paddy field images uploaded by the mobile terminal, performing format standardization and quality inspection processing, and serving as a data input set for subsequent optimization of the drought grade diagnosis model of the inspection point;

[0038] The storage module is used to adopt a distributed storage architecture and use MongoDB to store processed data.

[0039] Optionally, the intelligent analysis module is also used to perform spatial cluster analysis on the drought data corresponding to each patrol point in the patrol area, generate a spatial distribution heat map and hot spot areas corresponding to the patrol area, and send them to the server for display; the drought data includes longitude and latitude, the drought level corresponding to each patrol point, and the coding of the growth period.

[0040] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0041] The present invention discloses an intelligent diagnosis system and method for rice field drought conditions. The system first obtains paddy field images corresponding to each inspection point; then obtains spatiotemporal data and the rice growing period corresponding to each inspection point; then, the paddy field images, spatiotemporal data, and rice growing period corresponding to each inspection point are input into a patrol point drought grade diagnosis model for identification, thereby obtaining the drought grade corresponding to each inspection point; then, based on the drought grade corresponding to each inspection point, the area ratio corresponding to each drought grade within the inspection area is determined; finally, based on the area ratio corresponding to each drought grade within the inspection area, the drought grade corresponding to the inspection area is calculated. The present invention comprehensively considers the rice growing period and spatiotemporal data, and can accurately identify the drought grade corresponding to each inspection point and the drought grade corresponding to the inspection area. The present invention can provide accurate, timely, and objective decision-making support for drought prevention and relief, significantly improving the intelligent level of rice field drought monitoring and relief work. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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 any creative work.

[0043] Figure 1 This is a structural diagram of the intelligent diagnosis system for rice field drought conditions according to an embodiment of the present invention;

[0044] Figure 2 This is a specific flow chart of the multimodal agricultural scene perception network MASPN training inspection point drought level diagnosis model according to an embodiment of the present invention;

[0045] Figure 3 This is a flow chart of the intelligent diagnosis method for rice field drought according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0047] The purpose of the present invention is to provide an intelligent diagnosis system and method for rice field drought conditions to improve recognition accuracy.

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1

[0050] Figure 1 This is a structural diagram of the intelligent diagnosis system for rice field drought according to an embodiment of the present invention. Figure 1 As shown, the present invention discloses an intelligent diagnosis system for rice field drought, which includes: a mobile terminal 1, a server 2 and a server 3; the mobile terminal 1 is connected to the server 3 through the server 2. The mobile terminal 1 of the present invention includes a drought inspection system, a camera, a GPS positioning module and a network module; the drought inspection system is set on the mobile terminal 1 in the form of an APP; the camera is used to obtain the paddy field image corresponding to each inspection point; the camera supports automatic focus, image stabilization and light compensation functions; the GPS positioning module is used to locate each inspection point to obtain location information; the network module is used to realize data transmission between the mobile terminal 1 and the server 2. The paddy field image mentioned in the present invention is a JPEG format image with a resolution of 1024×768 pixels. A firewall is set between the network module and the server 2. The network module is constructed through a 4G / 5G mobile network or a WiFi network, and has a network disconnection and transmission resume function. When the network is interrupted, it can automatically reconnect and retransmit data. The server 3 is connected to the server 2 through a dedicated network or VPN to ensure the security and stability of data transmission.

[0051] The server 3 of the present invention is used to query and calculate drought distribution statistics. The server 3 adopts a B / S architecture, supports multi-user concurrent access, and has a complete authority management and data security protection mechanism.

[0052] The mobile terminal 1 of the present invention can be a smartphone, tablet, or dedicated device equipped with GPS positioning and camera functions. Its camera can only be activated within a designated paddy field inspection area. GPS positioning verifies the legitimacy of the inspection location. Beyond the designated range, the camera function is automatically disabled, allowing the drought inspection system to be activated. The drought inspection system utilizes a separate front-end and back-end architecture, with the front-end communicating with the back-end (i.e., the mobile terminal 1) via a 4G / 5G wireless network, enabling two-way data exchange with the back-end (i.e., the mobile terminal 1).

[0053] The Drought Patrol System is available in two versions, Android and iOS, with identical functionality. Patrolmen can download and install the Drought Patrol System on their mobile device (based on the operating system of their mobile device), register, and log in to use it.

[0054] The drought inspection system of the present invention includes: a data acquisition module, an intelligent recognition module, a user management module and a map navigation module. Each module is discussed in detail below:

[0055] The data acquisition module is used to obtain paddy field images, spatiotemporal acquisition data, and rice growth stages corresponding to each inspection point; the spatiotemporal acquisition data includes location information, image acquisition time, and meteorological data, including temperature, humidity, and rainfall; the location information includes longitude and latitude; meteorological data can be obtained by accessing relevant meteorological network data; the above-mentioned location information and image acquisition time can be directly obtained through the mobile terminal 1. The rice growth stage disclosed in the present invention is one of the tillering stage, jointing stage, heading stage, grain filling stage, and maturity stage.

[0056] The user management module integrates functions such as inspector type selection, personal information management, and inspection task management. The drought inspection system offers two types of inspectors: township agricultural technicians and reservoir inspectors. Different inspectors are assigned different inspection areas based on their identities, ensuring that each area does not conflict or overlap.

[0057] When a user first enters the drought inspection system, a loading screen appears, followed by an inspector type selection screen. Inspectors select the township or reservoir to inspect based on their status. Once a patrol point is selected, the system automatically locks that area, preventing other inspectors from selecting it again. Releasing the patrol point requires database processing by server-side administrators.

[0058] The personal information management function includes adding and modifying personal information. Mobile phone number and name are required, while position and company are optional. Inspectors can modify their personal information by clicking the Personal Center in the upper right corner of the homepage.

[0059] Inspection task management provides inspection assistance, modification of inspection types, inspection process management, and other functions. The "Start Inspection" or "Continue Inspection" status prompt is displayed at the bottom of the homepage interface, and the two are dynamically switched. Clicking "Start Inspection" will change to the "Continue Inspection" status.

[0060] The map navigation module integrates functions such as paddy field inspection location positioning, inspection point marking, inspection range query, inspection map zooming, and inspection route planning. It also features offline map functionality and supports switching between satellite and electronic map displays, ensuring normal operation even in remote areas with poor network signals.

[0061] Inspectors use the map navigation module to view real-time map distribution and accurately locate themselves using GPS. Spreading and pinching two fingers together allows zooming in and out of the map, with multi-level zoom supported. Once the inspector has determined the scope of their inspection area, they can mark and save inspection points on the inspection map. The system records the inspector's trajectory data in real time and transmits it to Server 2, where trajectory verification further ensures the authenticity and effectiveness of the inspection.

[0062] Before training, the four-level drought standards were accurately marked:

[0063] Normal water layer (level 0): There is an obvious water layer on the field surface, with a depth of 2-5 cm.

[0064] Soil exposure (Level 1): The water layer on the field surface disappears, and the soil is exposed but without cracks.

[0065] Loss of water layer (Level 2): The field surface is dry and the soil shrinks and small cracks appear.

[0066] Field surface cracks (Level 3): The field surface is severely cracked, with crack width >1 cm.

[0067] The intelligent recognition module is used to input the paddy field images, spatiotemporal data, and rice growth period corresponding to each inspection point into the inspection point drought grade diagnosis model for identification, thereby obtaining the drought grade corresponding to each inspection point. The intelligent recognition module is also used to train the Multimodal Agricultural Scene Perception Network (MASPN) using the paddy field images, spatiotemporal data, and rice growth period corresponding to each historical inspection point as the data input set and the drought grade corresponding to each historical inspection point as the data output set to obtain the inspection point drought grade diagnosis model. The drought grades corresponding to each historical inspection point mentioned in the present invention are normal water layer, exposed soil, missing water layer, and field surface cracks. The MASPN mentioned in the present invention utilizes three feature extraction branches and a cross-modal fusion method. The three feature extraction branches are an image feature branch, a spatiotemporal feature branch, and a growth period feature branch. The image feature branch corresponds to image features, the spatiotemporal feature branch corresponds to spatiotemporal features, and the growth period feature branch corresponds to growth period features.

[0068] like Figure 2 As shown, the present invention uses paddy field images, spatiotemporal data, and rice growth periods corresponding to historical inspection points as data input sets, and drought levels corresponding to historical inspection points as data output sets. The multimodal agricultural scene perception network (MASPN) is used for training to obtain a drought level diagnosis model for inspection points, which specifically includes:

[0069] Image feature branch: Based on the ConvNeXt-V2 deep convolutional neural network architecture, feature extraction is performed on the paddy field images corresponding to each historical inspection point to obtain the image visual feature vector, which specifically includes:

[0070] The paddy field images corresponding to each historical inspection point are size-standardized and pixel-value-normalized through the data processing layer to obtain the preprocessed paddy field images; the preprocessed paddy field images are 1024×768×3 RGB images.

[0071] The preprocessed paddy field image is extracted step by step through the four stages of the ConvNeXt-V2 deep convolutional neural network to obtain a spatial feature map; each stage contains multiple ConvNeXt blocks.

[0072] The spatial feature map is processed sequentially through the global average pooling layer and the feature mapping layer to obtain a 128-dimensional image visual feature vector.

[0073] Compared to traditional convolutional neural networks, our ConvNeXt-V2 boasts stronger feature representation and better generalization performance, making it particularly suitable for agricultural image analysis. The network utilizes layer normalization (LayerNorm) and Gaussian Error Linear Unit (GELU) activation functions, resulting in improved training stability.

[0074] Spatiotemporal feature branch: Extract features based on the spatiotemporal data collected at each historical inspection point to obtain spatiotemporal feature vectors, including:

[0075] The location information corresponding to each historical patrol point is standardized through the data processing layer, the image acquisition time corresponding to each historical patrol point is converted, and the meteorological data corresponding to each historical patrol point is normalized.

[0076] The preprocessed data is input into a two-layer fully connected neural network for feature encoding to obtain a 128-dimensional spatiotemporal feature vector; the first layer: the input dimension is 7 (longitude, latitude, timestamp, temperature, humidity, rainfall, collection date), the output dimension is 256, and the rectified linear unit (ReLU) is used for activation; the second layer: the input dimension is 256, the output dimension is 128, and the rectified linear unit (ReLU) is used for activation.

[0077] These spatiotemporal characteristics provide important environmental context information for drought assessment.

[0078] Growth period feature branch: Based on the water demand characteristics of the rice growth period corresponding to each historical inspection point, feature extraction is performed to obtain the growth period feature vector, specifically including:

[0079] The rice growth period is encoded as a digital identifier based on the digital coding layer.

[0080] An embedding layer is used to map the numeric identity into a 64-dimensional continuous vector.

[0081] A fully connected layer is used to expand the 64-dimensional continuous vector into a 128-dimensional growth period feature vector.

[0082] Cross-modal fusion training: Adopting an adaptive cross-modal attention mechanism, training is performed based on the image visual feature vector, the spatiotemporal feature vector, the growth period feature vector and the drought levels corresponding to each historical inspection point, and the total loss value is calculated.

[0083] Determine whether the total loss value meets the set requirements; if so, construct a drought level diagnosis model for the inspection point based on the weights corresponding to the three eigenvectors; if not, adjust the weights corresponding to the three eigenvectors through the back propagation algorithm.

[0084] The adaptive cross-modal attention mechanism disclosed in the present invention can dynamically adjust the weights of the three feature extraction branches according to the water demand characteristics of different growth periods, thereby achieving more accurate drought identification.

[0085] The weights corresponding to the three eigenvectors are:

[0086]

[0087] in, 、 and Represent the attention weights of the image feature branch, spatiotemporal feature branch, and growth period feature branch respectively, 、 and Represent the query weight matrix, 、 and Represent the eigenvectors of the three branches respectively, and represent the key weight matrix, 、 and They represent learnable bias terms, and softmax is a normalized exponential function.

[0088] The specific formula for calculating the total loss value is:

[0089] L = L main + λ 1 ×L stage + λ 2 ×L climate + λ 3 ×L contrastive

[0090] in, L Indicates the total loss value; L main The loss for the main task, i.e., the loss for the four-category drought level identification; L stage The loss for the auxiliary task, i.e., the loss for reproductive period identification; L climate The loss for the auxiliary task, i.e., the loss for regional climate type identification; L contrastive Contrastive learning loss is used to enhance the feature discrimination between different drought levels, so that the model can better distinguish different degrees of drought; λ 1 , λ 2 , λ 3 Both are adjustable hyperparameters used to balance the weights of different task losses.

[0091] The adaptive cross-modal attention mechanism prioritizes image and spatiotemporal features during sensitive periods of rice water demand (such as heading), when water conditions significantly impact rice growth. Meanwhile, it reduces sensitivity to minor drought changes during the ripening period, when water demand is relatively low. Through this dynamic weighted fusion, MASPN can automatically identify four types of drought conditions in paddy field photos: surface cracks, missing water layers, exposed soil, and normal water layers, based on the water requirements of rice at different growth stages.

[0092] Server 2 serves as the core data processing center, supporting load balancing and high-availability deployment. Additionally, Server 2 includes a data processing module, a storage module, an intelligent analysis module, and an information publishing module. The specific functions of each module are as follows:

[0093] The data processing module is used to receive the paddy field images uploaded by the mobile terminal 1, perform format standardization and quality inspection processing, and use them as the data input set for optimizing the drought level diagnosis models of each inspection point in the follow-up. The image data is uniformly converted into the JPEG format, and the resolution is standardized to 1024×768 pixels. The quality inspection includes blur detection, light evaluation, occlusion detection, etc. Unqualified images will be marked and required to be recollected.

[0094] The storage module is used to adopt a distributed storage architecture and use the MongoDB database management system to store the processed data, support massive data storage and fast retrieval, and support automatic backup and disaster recovery.

[0095] The intelligent analysis module is used to receive the drought levels corresponding to each inspection point sent by the mobile terminal 1; and determine the area ratio corresponding to each drought level in the inspection area based on the drought levels corresponding to each inspection point; calculate the drought level corresponding to the inspection area based on the area ratio corresponding to each drought level in the inspection area, specifically including:

[0096] Calculate the drought degree index of the inspection area based on the area ratio corresponding to each drought level in the inspection area. The specific formula is:

[0097] DI = Σ(W i × P i × S i ),

[0098] where DI is the drought degree index of the inspection area, i ranges from 1 to 4, W i is the drought level weight coefficient (normal water layer W1 = 0, soil exposure W2 = 1, water layer absence W3 = 2, field surface cracks W4 = 3), P i is the area ratio of the i-th drought level in the area, and S i is the growth stage adjustment coefficient (tillering stage S = 1.0, jointing stage S = 1.2, heading stage S = 1.5, filling stage S = 1.3, maturity stage S = 0.8).

[0099] Divide the drought level according to the value of the drought degree index DI of the inspection area: DI≤0.5 is mild drought, 0.5 < DI≤1.5 is moderate drought, 1.5 < DI≤2.5 is severe drought, and DI>2.5 is extreme drought.

[0100] The drought degree index of the inspection area of the present invention adopts a weighted calculation method, comprehensively considers the drought level weight coefficient, the area ratio of each level, and the rice growth stage adjustment coefficient to quantitatively evaluate and obtain the drought degree index of the inspection area, and uses the regional drought degree index to conduct real-time diagnosis and evaluation of the drought degree of the current area. The drought degree of the current area is one of the four drought levels of mild, moderate, severe, and extreme.

[0101] The intelligent analysis module also performs spatial cluster analysis on drought data corresponding to each inspection point within the inspection area, generating a spatial distribution heat map and hotspots for the inspection area, which are then sent to server 3 for display. Drought data includes latitude and longitude, the drought level corresponding to each inspection point, and a code for the growing season. The number of clusters is adaptively determined based on the size of the region. Specifically, machine learning methods such as the K-means clustering algorithm and statistical analysis are used to comprehensively analyze the uploaded drought data. The distribution of drought levels for each growing season in each region is calculated based on the different rice growing seasons. The regional drought severity index is automatically calculated, and a regional rice field current drought severity assessment and spatial distribution heat map are generated.

[0102] The intelligent analysis module also uses a combination of inverse distance weighted interpolation and kriging interpolation to spatially interpolate the drought level corresponding to each inspection point, generating a continuous drought distribution surface and sending it to server 3 for display. Interpolation accuracy is optimized through cross-validation, with a mean absolute error of less than 0.15.

[0103] The information release module is used to send at least one of the drought level corresponding to the inspection point, the drought level corresponding to the inspection area, the drought statistical report, the spatial distribution heat map and hot spot area corresponding to the inspection area, and the continuous drought distribution surface of the region to multiple server ends 3 for display; the spatial distribution heat map corresponding to the inspection area uses different colors to represent different drought levels, and supports multi-level zooming and layer overlay; the drought statistical report contains statistical information such as the area and proportion of each drought level.

[0104] The server 3 of the present invention includes: an inspection management module, a data display module and an information query module. The specific functions of each module are as follows:

[0105] The data display module is used to display the drought level corresponding to each inspection point and / or the drought level corresponding to the inspection area sent by the server 2, and is also used to display drought statistical reports, spatial distribution heat maps and hot spots corresponding to the inspection area, and continuous drought distribution surfaces in the region.

[0106] Inspection management module, specifically including:

[0107] (1) Inspection task management: Administrators can create, assign, and modify inspection tasks, set inspection areas, inspection frequencies, responsible persons, etc. Task division is supported for water systems (watersheds, irrigation areas, reservoirs). The system automatically generates an inspection schedule and pushes it to the corresponding inspectors through the APP.

[0108] (2) Inspection progress monitoring: Real-time monitoring of the inspection progress of each area, showing completed, ongoing, and overdue inspection tasks. The inspection coverage is intuitively displayed on a map, and uncovered areas are marked with different colors. Inspection track playback is supported to verify the authenticity of the inspection.

[0109] (3) Inspection result review: Conduct quality review of drought data uploaded by inspectors, including image quality check, GPS location verification, logical consistency check, etc. Automatic review rules and manual review processes can be set to ensure data quality.

[0110] (4) Permission management: Establish a hierarchical permission management system and assign different operation permissions according to user roles (system administrator, regional administrator, ordinary user). Support data access permission control to ensure data security.

[0111] Information query module, specifically including:

[0112] (1) Multi-dimensional query: Supports combined query based on multiple dimensions such as time, region, drought level, and rice growing period. The time dimension supports different granularities such as day, week, month, season, and year; the region dimension supports different spatial ranges such as point, line, and surface; and the query conditions and sorting rules can be customized.

[0113] (2) Statistical analysis: Provides statistics on drought area, grade distribution, time series, and spatial distribution, etc. Supports year-on-year and quarter-on-quarter analysis, and can generate various statistical charts (bar charts, pie charts, line charts, scatter plots, etc.).

[0114] (3) Data export: Supports exporting query results to various formats such as Excel, PDF, and CSV to meet the data usage needs of different users. Export fields and format templates can be customized.

[0115] The present invention constructs a multimodal agricultural scene perception network (MASPN), which integrates image features, spatiotemporal environmental information, and rice growth period characteristics, and realizes dynamic weight adjustment through an adaptive cross-modal attention mechanism. Compared with traditional single image recognition methods, the recognition accuracy is improved by 10-15%.

[0116] The present invention establishes a special growth period characteristic coding branch and the corresponding adjustment coefficient of the branch based on the differences in water requirements of rice in different growth periods, realizes differentiated drought assessment, and effectively solves the technical defect of the existing technology that ignores the differences in growth periods.

[0117] The present invention has built a complete intelligent system from data collection to decision support. Through machine learning algorithms such as K-means clustering and statistical analysis, it can automatically generate regional drought severity index and spatial distribution heat map, providing quantitative support for drought prevention and relief decision-making.

[0118] Example 2

[0119] like Figure 3 As shown, the present invention also discloses an intelligent diagnosis method for drought in rice fields, the method comprising:

[0120] Step S1: Navigate to each inspection point within the inspection area to obtain the paddy field image corresponding to each inspection point.

[0121] Step S2: Obtaining the spatiotemporal data and rice growth period corresponding to each inspection point; the spatiotemporal data includes location information, image acquisition time and meteorological data.

[0122] Step S3: inputting the paddy field image, the spatiotemporal data and the rice growth period corresponding to each inspection point into the inspection point drought level diagnosis model for identification, and obtaining the drought level corresponding to each inspection point.

[0123] Step S4: Determine the area ratio corresponding to each drought level within the inspection area based on the drought level corresponding to each inspection point.

[0124] Step S5: Calculate the drought level corresponding to the inspection area based on the area ratio corresponding to each drought level within the inspection area.

[0125] As an optional implementation, the present invention uses paddy field images, spatiotemporal data, and rice growth periods corresponding to historical inspection points as the data input set, and the drought levels corresponding to historical inspection points as the data output set. The Multimodal Agricultural Scene Perception Network (MASPN) is trained to obtain a drought level diagnosis model for inspection points. Specifically, the model includes:

[0126] Based on the ConvNeXt-V2 deep convolutional neural network architecture, feature extraction is performed on the paddy field images corresponding to each historical inspection point to obtain the image visual feature vector.

[0127] Based on the spatiotemporal data collected at each historical inspection point, feature extraction is performed to obtain the spatiotemporal feature vector.

[0128] Based on the water demand characteristics of the rice growing period corresponding to each historical inspection point, feature extraction is performed to obtain the growing period feature vector.

[0129] An adaptive cross-modal attention mechanism is used to train and calculate the total loss value based on the image visual feature vector, the spatiotemporal feature vector, the growth period feature vector and the drought levels corresponding to each historical inspection point.

[0130] Determine whether the total loss value meets the set requirements; if so, construct a drought level diagnosis model for the inspection point based on the weights corresponding to the three eigenvectors; if not, adjust the weights corresponding to the three eigenvectors through the back propagation algorithm.

[0131] As an optional embodiment, the method of the present invention further comprises:

[0132] A spatial cluster analysis is performed on the drought data corresponding to each inspection point in the inspection area to generate a spatial distribution heat map and hot spots corresponding to the inspection area; the drought data includes longitude and latitude, the drought level corresponding to each inspection point, and the coding of the growth period.

[0133] As an optional embodiment, the method of the present invention further comprises:

[0134] A method combining inverse distance weighted interpolation and Kriging interpolation was used to perform spatial interpolation on the drought level corresponding to each inspection point and generate a continuous drought distribution surface.

[0135] The parts that are the same as those in Example 1 are specifically referred to in Example 1 and will not be discussed again here.

[0136] Example 3

[0137] The present invention also discloses a drought inspection method, comprising the following steps:

[0138] S1. User login and navigation: The patrol personnel hold a mobile terminal, log in through the user management module of the drought patrol system, select the patrol type, and use the GPS positioning function of the map navigation module to navigate to each patrol point in the patrol area.

[0139] S2. On-site data collection: Start the camera in the designated inspection area to capture typical paddy field images reflecting the degree of drought in the paddy field. At the same time, record the specific rice growth period and spatiotemporal data of the current rice in the drought inspection system.

[0140] S3. Intelligent identification and confirmation: The intelligent identification module automatically activates the Multimodal Agricultural Scene Perception Network (MASPN), analyzes paddy field images based on the water demand characteristics of rice during its growth period and spatiotemporal data collection, and provides identification results and confidence levels in four drought levels. The inspector confirms or corrects the results and uploads them to the server.

[0141] S4. Intelligent analysis and diagnosis: After the server receives the recognition results, the intelligent analysis module automatically uses machine learning algorithms to perform statistical analysis and processing to calculate the regional drought index and the corresponding drought level of the patrol area.

[0142] The detailed implementation process of the drought inspection method is as follows:

[0143] S1. User login and navigation:

[0144] S1.1 System Login: Patrol personnel activate the drought inspection system on mobile terminal 1. First-time users are required to register an account and enter their name, mobile phone number, ID number, and affiliation. Registered users can log in directly by entering their username and password. Biometric login methods such as fingerprint and facial recognition are supported.

[0145] S1.2 Identity Verification: The system verifies the inspector's identity and authority, confirming their patrol qualifications and responsibility areas. GPS positioning is used to confirm the inspector's current location and verify whether they are within the authorized area.

[0146] S1.3 Task selection: The inspector selects the inspection type (township agricultural technician or reservoir inspector) on the main interface, and the system displays a list of currently executable inspection tasks, including task name, inspection area, deadline and other information.

[0147] S1.4 Navigation and Positioning: After selecting a specific inspection task, the system automatically loads a map of the corresponding area, displaying the inspection route and inspection points. GPS navigation guides inspectors to the designated paddy field area, supporting voice broadcast and route planning optimization.

[0148] S2. Field data collection:

[0149] S2.1 Location Verification: When an inspector arrives at the target inspection point, the system verifies via GPS that their current location is within the designated inspection range. The camera function is activated only if the inspector is within the legal inspection range. If the inspector is outside the inspection range, the system automatically disables the camera function and prompts the inspector to adjust their position.

[0150] S2.2 Environmental Information Recording: Before taking photos, inspectors need to record the specific growth period information of the current rice in the system, including tillering stage, jointing stage, heading stage, grain filling stage, and maturity stage. The system also automatically records GPS coordinates, shooting time, weather conditions, and other environmental information.

[0151] S2.3 Image Acquisition: Activate the camera to capture images of the paddy fields reflecting the extent of drought. The system requires inspectors to take 3-5 photos from different angles to ensure a comprehensive picture of the drought conditions. During capture, the system automatically checks image quality, including clarity, lighting conditions, and composition. Unsatisfactory images require retakes.

[0152] S2.4 Supplementary Information: Inspectors can add text descriptions to record special circumstances or other information that requires clarification. The system supports voice-to-text functionality for easy on-site input.

[0153] S3, Intelligent Identification and Confirmation:

[0154] S3.1 Automatic identification: After receiving the uploaded image data, spatiotemporal information, and growth period characteristics, the intelligent identification module inputs them into the inspection point drought level diagnosis model for identification, and finally outputs the identification results of four drought levels.

[0155] S3.2 Results Display: The system displays the identified drought level (normal water layer, exposed soil, missing water layer, field cracks) and the corresponding confidence score on mobile terminal 1. Key feature areas are also labeled to help inspectors understand the algorithm's judgment basis.

[0156] S3.3 Manual Confirmation: Inspectors review the identification results and can choose to confirm or revise them. If the identification results do not match the actual situation, inspectors can manually correct them to the correct drought level and provide a reason for the correction. These corrections will be used for continuous optimization of the model.

[0157] S3.4 Data upload: After confirmation, the system uploads the image data and recognition results to Server 2 via the network. The upload uses compression transmission and breakpoint resume technology to ensure reliable transmission even in poor network conditions.

[0158] S4. Intelligent analysis and diagnosis:

[0159] S4.1 Data Reception and Processing: After Server 2 receives the drought severity level corresponding to each inspection point, the data processing module first performs data integrity verification and format standardization. The processed image data is stored in the distributed file system and the structured data is written to the database.

[0160] S4.2 Intelligent Analysis: The intelligent analysis module periodically (hourly or daily) performs batch analysis on newly added data. It uses the K-means clustering algorithm to identify the spatial distribution pattern of drought conditions and calculate the area statistics and grade distribution of each region.

[0161] S4.3 Index calculation: Calculate the drought severity index for each region based on the analysis results, taking into account the drought level weight, area ratio and growth period adjustment coefficient.

[0162] S4.4 Visualization: Generate a regional heat map of rice field drought distribution, using different colors and patterns to represent different drought levels. Generate statistical charts to provide intuitive data support for decision makers.

[0163] Example 4

[0164] Taking the drought diagnosis and monitoring of rice heading period in a certain county as an example, the actual application effect of the system is explained in detail:

[0165] Case background: The county has a rice planting area of approximately 150,000 mu. In 2022, during the heading period, there was no effective rainfall for 20 consecutive days. It was urgent to accurately assess the drought situation in order to formulate drought prevention and relief measures.

[0166] Implementation process:

[0167] 1. Inspection Deployment: 25 township agricultural technicians and 8 reservoir inspectors were assigned inspection tasks through Server 3, covering 128 key plots across the county. Each plot was assigned 2-3 inspection points, for a total of 315 inspection points.

[0168] 2. On-site inspection: Inspectors used the drought inspection system on mobile terminal 1 to complete the entire inspection within two days, capturing 945 images of paddy fields and recording detailed information about the growth period and environment. The system verified the authenticity of the inspections using GPS tracking, achieving 100% inspection coverage.

[0169] 3. Intelligent Identification: MASPN automatically identified 945 images, revealing that 30% showed normal water layers, 40% exposed soil, 20% showed missing water layers, and 10% showed cracks in the field surface. Inspectors confirmed the identification results, achieving an accuracy rate of 91.2%.

[0170] 4. Intelligent Analysis: The intelligent analysis module conducted an in-depth analysis of the data and determined that the regional drought severity index (DI) = (0 × 0.3 + 1 × 0.4 + 2 × 0.2 + 3 × 0.1) × 1.5 = 1.1 × 1.5 = 1.65, indicating severe drought in this region. The resulting drought distribution heat map showed that the drought was most severe in the northeastern and central regions, while relatively mild in the southwestern regions near reservoirs. Based on this information, the drought emergency plan was quickly activated, and water resources were promptly allocated, effectively mitigating the impact of the drought on rice production. Compared to traditional manual estimates, the drought information provided by the system is more accurate, timely, and objective, significantly improving the efficiency and effectiveness of drought prevention and relief efforts.

[0171] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0172] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An intelligent diagnosis method for rice field drought, characterized in that: The method comprises: Navigate to each inspection point within the inspection area to obtain the paddy field image corresponding to each inspection point; Obtaining the spatiotemporal data and rice growth period corresponding to each inspection point; the spatiotemporal data includes location information, image acquisition time, and meteorological data; Inputting the paddy field image, the spatiotemporal data and the rice growth period corresponding to each inspection point into the inspection point drought level diagnosis model for identification, thereby obtaining the drought level corresponding to each inspection point; Based on the drought levels corresponding to each inspection point, determine the area proportion corresponding to each drought level within the inspection area; Calculate the drought level corresponding to the patrol area based on the area ratio corresponding to each drought level within the patrol area; The data input sets are paddy field images, spatiotemporal data, and rice growth periods corresponding to each historical inspection point, and the data output sets are drought levels corresponding to each historical inspection point. A multimodal agricultural scene perception network (MASPN) is used for training to obtain a drought level diagnosis model for inspection points. Specifically, the model includes: Based on the ConvNeXt-V2 deep convolutional neural network architecture, feature extraction is performed on the paddy field images corresponding to each historical inspection point to obtain the image visual feature vector; Based on the spatiotemporal data collected at each historical inspection point, feature extraction is performed to obtain spatiotemporal feature vectors; Based on the water demand characteristics of the rice growth period corresponding to each historical inspection point, feature extraction is performed to obtain the growth period feature vector; Adopting an adaptive cross-modal attention mechanism, training is performed based on the image visual feature vector, the spatiotemporal feature vector, the growth period feature vector, and the drought levels corresponding to each historical inspection point, and a total loss value is calculated; Determine whether the total loss value meets the set requirements; if so, build a drought level diagnosis model for the inspection point based on the weights corresponding to the three eigenvectors; if not, adjust the weights corresponding to the three eigenvectors through the back propagation algorithm; The adaptive cross-modal attention mechanism can dynamically adjust the weights of the three feature extraction branches according to the water demand characteristics of different growth stages; The weights corresponding to the three eigenvectors are: ； in, 、 and Represent the attention weights of the image feature branch, spatiotemporal feature branch, and growth period feature branch respectively, 、 and Represent the query weight matrix, 、 and Represent the eigenvectors of the three branches respectively, and represent the key weight matrix, 、 and They represent learnable bias terms, and softmax is a normalized exponential function.

2. The intelligent diagnosis method for rice field drought according to claim 1, characterized in that: The method further comprises: A spatial cluster analysis is performed on the drought data corresponding to each inspection point in the inspection area to generate a spatial distribution heat map and hot spots corresponding to the inspection area; the drought data includes longitude and latitude, the drought level corresponding to each inspection point, and the coding of the growth period.

3. The intelligent diagnosis method for rice field drought according to claim 1, characterized in that: The method further comprises: A method combining inverse distance weighted interpolation and Kriging interpolation was used to perform spatial interpolation on the drought level corresponding to each inspection point and generate a continuous drought distribution surface.

4. An intelligent diagnosis system for rice field drought, characterized in that: The system is applied to the intelligent diagnosis method for rice field drought according to any one of claims 1 to 3, and the system comprises: A mobile terminal, a server and a service end of a drought inspection system; the mobile terminal is connected to the service end through the server; The mobile terminal includes: a camera and a GPS positioning module; the camera is used to obtain paddy field images corresponding to each inspection point; the GPS positioning module is used to locate each inspection point and obtain location information; The drought inspection system includes: a data acquisition module and an intelligent recognition module; The data acquisition module is used to obtain paddy field images, spatiotemporal acquisition data and rice growth period corresponding to each inspection point; the spatiotemporal acquisition data includes location information, image acquisition time and meteorological data; The intelligent recognition module is used to input the paddy field image, the spatiotemporal data and the rice growth period corresponding to each inspection point into the inspection point drought level diagnosis model for identification, so as to obtain the drought level corresponding to each inspection point; The server includes: an intelligent analysis module; The intelligent analysis module is used to receive the drought level corresponding to each patrol point sent by the mobile terminal; and determine the area ratio corresponding to each drought level in the patrol area based on the drought level corresponding to each patrol point; and calculate the drought level corresponding to the patrol area based on the area ratio corresponding to each drought level in the patrol area; The server includes: a data display module; The data display module is used to display the drought level corresponding to each inspection point and / or the drought level corresponding to the inspection area sent by the server.

5. The intelligent diagnosis system for rice field drought according to claim 4, characterized in that: The drought inspection system also includes: User management module, used to integrate patrol personnel type selection, personal information management, and patrol task management functions; The map navigation module is used to realize paddy field inspection location positioning, inspection point marking, inspection range query, inspection map zooming and inspection route planning.

6. The intelligent diagnosis system for rice field drought according to claim 4, characterized in that: The server further includes: a data processing module for receiving paddy field images uploaded by the mobile terminal, performing format standardization and quality inspection processing, and serving as a data input set for subsequent optimization of the drought grade diagnosis model of the inspection point; The storage module is used to adopt a distributed storage architecture and use MongoDB to store processed data.

7. The intelligent diagnosis system for rice field drought according to claim 4, characterized in that: The intelligent analysis module is also used to perform spatial cluster analysis on the drought data corresponding to each patrol point in the patrol area, generate a spatial distribution heat map and hot spot areas corresponding to the patrol area, and send them to the server for display; the drought data includes longitude and latitude, the drought level corresponding to each patrol point, and the code of the growth period.

Images