Crop real-time monitoring method and system based on unmanned aerial vehicle image recognition and Beidou short message technology

Abstract

The invention relates to a crop real-time monitoring method and system based on unmanned aerial vehicle image recognition and Beidou short message technology, and the method comprises the steps: collecting image data of a target farmland through an unmanned aerial vehicle, and obtaining an RGB-IR four-channel data set with a geographic tag; inputting an image in the RGB-IR four-channel data set with the geographic tag into a YOLOv7-tiny model, identifying pest and disease damage characteristics in the image, and outputting a pest and disease damage image; positioning the unmanned aerial vehicle to obtain coordinate data of the pest image; and transmitting the pest and disease image and the coordinate data to a user side according to a Beidou short message technology, recombining the pest and disease image by the user side, superposing coordinate information, and sending an instruction to the unmanned aerial vehicle according to the recombined pest and disease image. According to the invention, the target positioning error can be reduced, and the pesticide application precision is obviously improved.

Description

Technical Field

[0001] The present invention relates to the field of satellite positioning technology, and in particular to a method and system for real-time monitoring of crops based on unmanned aerial vehicle image recognition and Beidou short message technology. Background Art

[0002] Crop pests and diseases have always been a threat to food security. Traditional monitoring methods mainly rely on manual field inspections, but the efficiency is extremely low - each person covers less than 5 mu of area per day, the missed detection rate exceeds 30%, and the average delay from discovery to treatment is 48 hours. This lag is particularly fatal during the outbreak of pests and diseases. For example, locust swarms can migrate up to 150 kilometers a day, and traditional methods can hardly deal with it effectively. With technological advances, agricultural monitoring has gradually evolved towards automation and intelligence, but each has significant defects.

[0003] Satellite remote sensing monitoring technology uses satellites to obtain multispectral images and combines them with vegetation indices for large-scale analysis. Although it can cover thousands of square kilometers, it is limited by two major bottlenecks: first, the spatial resolution is insufficient. Most commercial satellite images are 10-30 meters per pixel, and early pests and diseases are almost impossible to identify at this resolution; second, the revisit cycle is long. Low-orbit satellites need at least 5 days to cover the same area again, making it difficult to capture the dynamic process of pest and disease spread. In contrast, ground-based Internet of Things monitoring technology achieves continuous data collection by deploying field sensor networks (such as soil moisture meters and insect monitoring lights), but the cost of deploying a single acre of sensors exceeds 200 yuan, and fixed points cannot fully reflect the overall condition of the field. For example, the degree of pests and diseases in two adjacent areas may be significantly different, and sensors can only provide discrete point data.

[0004] In recent years, drone inspection technology has developed rapidly. Consumer drones are equipped with visible light cameras for aerial photography, and images are transmitted back to the cloud for analysis through 4G / 5G networks. However, this technology faces three challenges: First, real-time performance is limited. The cloud processing process causes end-to-end delays of more than 6 hours, which cannot meet the needs of rapid response to pests and diseases; second, the network dependence is high. About 78% of the world's farmland is located in areas where 4G / 5G signals are not covered. Especially in mountainous or remote areas, the data collected by drones cannot be transmitted back in real time; third, the positioning accuracy is insufficient. The error of a single GPS positioning in hilly areas can reach 5-10 meters, which makes it difficult to accurately mark the occurrence point of pests and diseases, resulting in serious deviations in subsequent pesticide application operations. In addition, the coordinated errors of positioning and identification seriously restrict the accuracy of pesticide application. Existing systems generally adopt a separate architecture, that is, the flight coordinates are first located by GPS, and then the location of pests and diseases is estimated by image analysis. This means that the pesticide application drone may deviate from the target area, resulting in pesticide waste or missed spraying. Summary of the invention

[0005] The object of the present invention is to provide a real-time monitoring method and system for crops based on UAV image recognition and Beidou short message technology, which solves the problems of limited real-time performance in crop monitoring and difficulty in accurately marking the occurrence points of pests and diseases.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A real-time monitoring method for crops based on UAV image recognition and Beidou short message technology, comprising:

[0008] Collecting image data of the target farmland by a UAV to obtain a geotagged RGB-IR four-channel data set;

[0009] Inputting the images in the geotagged RGB-IR four-channel data set into the YOLOv7-tiny model to identify the pest and disease characteristics in the images and output pest and disease images;

[0010] Positioning the UAV to obtain the coordinate data of the pest and disease images;

[0011] Transmitting the pest and disease images and the coordinate data to the user terminal according to the Beidou short message technology, the user terminal recombines the pest and disease images and overlays the coordinate information, and sends instructions to the UAV according to the recombined pest and disease images.

[0012] Optionally, obtaining the geotagged RGB-IR four-channel data set includes: generating three-dimensional point cloud data of the target farmland by lidar, aligning the lidar coordinates with the image data in terms of timestamp, and obtaining the geotagged RGB-IR four-channel data set.

[0013] Optionally, positioning the UAV to obtain the coordinate data of the pest and disease images includes:

[0014] Performing real-time positioning on the UAV by using PPP-AR through Beidou BDS-3 satellites to obtain the coordinate data of the UAV;

[0015] Converting the coordinate data of the UAV into the plane coordinates of the pest and disease image area;

[0016] Obtaining the minimum circumscribed rectangle of the pest and disease image area and calculating the center point coordinates of the minimum circumscribed rectangle;

[0017] Converting the center point coordinates into Beidou positioning coordinates to obtain the coordinate data of the pest and disease images.

[0018] Optionally, performing real-time positioning on the UAV by using PPP-AR through Beidou BDS-3 satellites includes: decomposing the floating-point ambiguity of the ionosphere-free combination observation into an integer wide-lane ambiguity and a floating-point narrow-lane ambiguity.

[0019] Optionally, the floating-point ambiguity of the ionosphere-free combined observation value is:

[0020]

[0021] where represents the floating-point ambiguity of the undifferenced ionosphere-free combined observation value of receiver r at satellite s; and represent the integer wide-lane ambiguity and the floating-point narrow-lane ambiguity respectively; are the frequencies at B1 and B2 frequency points respectively.

[0022] Optionally, transmitting the pest and disease image and the coordinate data to the user terminal according to the Beidou short message technology includes: compressing the pest and disease image into JPEG format, adopting a dynamic packet splitting strategy for the compressed image data, performing descending-order block transmission according to the threat level, and encapsulating the coordinate data into a target priority message.

[0023] Optionally, obtaining a pest and disease heat map from the recombined pest and disease image and sending an instruction to the UAV includes:

[0024] Generating a digital elevation model from the three-dimensional point cloud data through a surface reconstruction algorithm;

[0025] Generating a red-yellow-green three-color graded pest and disease heat map through kernel density analysis;

[0026] Overlaying the digital elevation model and the red-yellow-green three-color graded pest and disease heat map on the recombined pest and disease image;

[0027] Sending an instruction to the UAV according to the overlaid pest and disease image.

[0028] The present invention also provides a real-time crop monitoring system based on UAV image recognition and Beidou short message technology for implementing the monitoring method, including: a multi-spectral image acquisition module, an edge computing processing module, a dual-frequency Beidou positioning module, a Beidou short message transmission module, and a user terminal interaction module;

[0029] The multi-spectral image acquisition module is used to acquire image data of a target farmland by means of a UAV and obtain an RGB-IR four-channel data set with geographical tags;

[0030] The edge computing processing module is used to input the images in the RGB-IR four-channel data set with geographical tags into the YOLOv7-tiny model, identify the pest and disease features in the images, and output pest and disease images;

[0031] The dual-frequency Beidou positioning module is used to position the UAV and obtain the coordinate data of the pest and disease image;

[0032] The Beidou short message transmission module is used to transmit the pest and disease image and the coordinate data to the user terminal according to the Beidou short message technology;

[0033] The user terminal interaction module is used for the user terminal to recombine the pest and disease image and overlay the coordinate information, and send instructions to the UAV according to the recombined pest and disease image.

[0034] The beneficial effects of the present invention are as follows: Firstly, a lightweight YOLOv7-tiny model is deployed on the UAV side to realize real-time recognition and coordinate binding on the UAV, compress the end-to-end delay, and solve the "storage-processing" delay problem; Secondly, the Beidou BDS-3 satellite short message service is used to realize network-free data transmission and solve the geographical limitation of traditional communication. Finally, combined with the Beidou BDS-3 satellite PPP-AR real-time positioning and lidar three-dimensional modeling, the target positioning error is reduced, and the spraying accuracy is significantly improved. Description of the Drawings

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0036] Figure 1 It is a flowchart of a method for real-time monitoring of crops based on UAV image recognition and Beidou short message technology in an embodiment of the present invention;

[0037] Figure 2 It is a framework diagram of a system for real-time monitoring of crops based on UAV image recognition and Beidou short message technology in an embodiment of the present invention. Detailed Embodiments

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0039] In order 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 drawings and specific embodiments.

[0040] Embodiment 1:

[0041] This embodiment provides a real-time monitoring method for crops based on UAV image recognition and Beidou short message technology, including:

[0042] Collect image data of the target farmland by UAV to obtain a geotagged RGB-IR four-channel dataset;

[0043] Input the images in the geotagged RGB-IR four-channel dataset into the YOLOv7-tiny model to identify the pest and disease characteristics in the images and output pest and disease images;

[0044] Locate the UAV to obtain the coordinate data of the pest and disease images;

[0045] Transmit the pest and disease images and coordinate data to the user terminal according to the Beidou short message technology. The user terminal recombines the pest and disease images and overlays the coordinate information, and sends instructions to the UAV according to the recombined pest and disease images.

[0046] Specifically, obtaining the geotagged RGB-IR four-channel dataset includes: generating three-dimensional point cloud data of the target farmland by lidar, aligning the lidar coordinates with the image data in terms of timestamp, and obtaining the geotagged RGB-IR four-channel dataset.

[0047] Specifically, the UAV flies according to a preset route, and collects images at 2 frames per second through a multispectral camera; the lidar synchronously scans to generate three-dimensional point clouds, which are stored in association with the Beidou PPP-AR positioning coordinates (longitude, latitude, altitude).

[0048] Furthermore, locating the UAV to obtain the coordinate data of the pest and disease images includes:

[0049] Perform real-time positioning of the UAV by using PPP-AR through Beidou BDS-3 satellites to obtain the coordinate data of the UAV;

[0050] Convert the coordinate data of the UAV into the plane coordinates of the pest and disease image area;

[0051] Obtain the minimum bounding rectangle of the pest and disease image area, and calculate the center point coordinates of the minimum bounding rectangle;

[0052] Convert the center point coordinates into Beidou positioning coordinates to obtain the coordinate data of the pest and disease images.

[0053] Even further, performing real-time positioning of the UAV by using PPP-AR through Beidou BDS-3 satellites includes: decomposing the float ambiguity of the ionosphere-free combination observations into an integer wide-lane ambiguity and a float narrow-lane ambiguity.

[0054] Furthermore, the floating-point ambiguity of the ionosphere-free combined observation value is:

[0055]

[0056] where represents the floating-point ambiguity of the undifferenced ionosphere-free combined observation value of receiver r for satellite s; and represent the integer wide-lane ambiguity and the floating-point narrow-lane ambiguity respectively; are the frequencies at B1 and B2 frequency points respectively.

[0057] Specifically, the floating-point wide-lane ambiguity parameter is obtained through the Melbourne-Wubbena (MW) formula, and the floating-point narrow-lane ambiguity parameter is obtained using the floating-point ionosphere-free combined observation value ambiguity and the integer wide-lane ambiguity.

[0058] Furthermore, transmitting the pest image and coordinate data to the user terminal according to the Beidou short message technology includes: compressing the pest image into the JPEG format, adopting a dynamic packetization strategy for the compressed image data, performing descending-order block transmission according to the threat level, and encapsulating the coordinate data into a high-priority message.

[0059] Specifically, the single-packet data volume of Beidou short message transmission ≤ 1000 bytes.

[0060] Furthermore, obtaining the pest heat map based on the reorganized pest image and sending instructions to the drone includes:

[0061] generating a digital elevation model from the three-dimensional point cloud data through a surface reconstruction algorithm;

[0062] generating a red-yellow-green three-color graded pest heat map through kernel density analysis;

[0063] overlaying the digital elevation model and the red-yellow-green three-color graded pest heat map on the reorganized pest image;

[0064] sending instructions to the drone according to the overlaid pest image.

[0065] The method of this embodiment will be further described below with reference to the accompanying drawings:

[0066] As Figure 1 shown, this embodiment discloses a real-time crop monitoring method based on drone image recognition and Beidou short message technology, including the following steps:

[0067] Step 1: The drone flies according to a preset route and collects images at 2 frames per second through a multispectral camera;

[0068] Step 2: The laser radar synchronously scans to generate a 3D point cloud, which is bound and stored with the Beidou PPP-AR positioning coordinates (longitude, latitude, and elevation);

[0069] Step 3: Run the lightweight YOLOv7-tiny model to identify the characteristics of pests and diseases in the image;

[0070] Step 4: Extract the minimum circumscribed rectangle of the target area and calculate the Beidou positioning coordinates of its center point;

[0071] Step 5: Compress the pest image into JPEG format and transmit it to the user end through Beidou short message packetization;

[0072] Step 6: The user reorganizes the image and overlays the coordinate information of the center point of the pest area calculated by the drone coordinates, and detects the data integrity through the sliding window algorithm;

[0073] Step 7: The user sends a reverse command, and the drone responds and updates the task queue.

[0074] In this embodiment, in step 1, the drone is equipped with a multispectral camera and a laser radar. The multispectral camera includes four bands: green, red, red edge, and near infrared, with a resolution of 120 million pixels, and can capture surface details at 5 cm / pixel.

[0075] In this embodiment, in step 2, the laser radar generates a three-dimensional point cloud model of the farmland at a scanning frequency of 240kHz with an accuracy of ±3cm. During the flight, the system adjusts the exposure time through real-time feedback of the NDVI vegetation index to eliminate backlight or shadow interference, and aligns the laser radar coordinates with the multispectral image for timestamps to generate a RGB-IR four-channel data set with geo-tags, where geo-tags refer to the use of specific symbols or text to identify the type of crops in each area, without coordinate information.

[0076] In the process of positioning the UAV using PPP-AR via the BeiDou BDS-3 satellite, the floating-point ambiguity of the ionospheric-free combined observation value is decomposed into integer wide-lane (WL) ambiguity and floating-point narrow-lane (NL) ambiguity:

[0077]

[0078] In the formula, is the floating point ambiguity of the undifferenced ionospheric-free composite observation of receiver r at satellite s; and They represent integer wide lane ambiguity and floating point narrow lane ambiguity respectively; They are the frequencies at B1 and B2 frequency points respectively.

[0079] The floating-point wide-lane ambiguity parameter is obtained through the Melbourne-Wubbena (MW) formula, and the HMW (Hatch-MW) combined observation value of the BDS-3 satellite is expressed as:

[0080]

[0081] In the formula, is the floating-point wide-lane ambiguity; is its integer part; d r,WL and are the fractional parts of the hardware delays at the receiver end and the satellite end; and respectively represent the pseudorange and phase observation values of the receiver r at the i-th frequency (i = B1, B2) of the satellite s; are the wavelengths of the B1 and B2 frequencies respectively; λ WL is the wide-lane wavelength.

[0082] The floating-point narrow-lane ambiguity parameter is obtained by using the floating-point ionosphere-free combined observation value ambiguity and the integer wide-lane ambiguity:

[0083]

[0084] In the formula, is the integer part of the narrow-lane ambiguity; d r,NL and are the fractional parts of the hardware delays at the receiver end and the satellite end respectively.

[0085] In this embodiment, in step three, an embedded AI chip is used as the processing core, and a solid-state storage unit is integrated. The lightweight YOLOv7-tiny model framework is deployed at the algorithm level to improve the recognition accuracy of targets such as rice blast and aphids. In the processing flow, the NDVI and OSAVI vegetation indices are used to implement multi-scale pest and disease area segmentation through the YOLOv7-tiny model, and finally a report containing coordinates (longitude, latitude, altitude), disease types, and confidence levels is output, and a pixel-level mask map is generated synchronously.

[0086] The loss function of the YOLOv7-tiny model consists of localization loss, classification loss, and confidence loss:

[0087] Among them, the localization loss is used to measure the accuracy of the model's prediction of the target position. The YOLOv7-tiny model uses the Mean Square Error loss function to calculate the localization loss. Its formula is as follows:

[0088]

[0089] In the formula, L loc represents the localization loss, λcoord is the weight parameter for balancing the localization loss and the classification loss, S is the size of the feature map, B is the number of predicted bounding boxes per grid cell, i represents the index of the feature map, and j represents the index of the bounding box in each grid cell. indicates whether the j-th bounding box on the i-th feature map contains an object (1 means contains, 0 means does not contain), tx i and ty i respectively represent the position coordinates of the i-th bounding box predicted by the model, tx i and ty i respectively represent the position coordinates of the i-th bounding box in the ground truth label.

[0090] Among them, the classification loss is used to measure the accuracy of the model's prediction of the target category. The YOLOv7-tiny model uses the CrossEntropy Loss function to calculate the classification loss. The formula is as follows:

[0091]

[0092] In the formula, L cls represents the classification loss, λ cls is the weight parameter for balancing the classification loss and the localization loss, C is the number of categories, and respectively represent the probability that the i-th bounding box predicted by the model belongs to the c-th category and the probability that the i-th bounding box in the ground truth label belongs to the c-th category.

[0093] Among them, the confidence loss is used to measure the accuracy of the model's detection of the target. The YOLOv7-tiny model uses the BinaryCross Entropy Loss function to calculate the confidence loss. The formula is as follows:

[0094]

[0095] In the formula, L obj represents the object loss, λ obj is the weight parameter for balancing the confidence loss and the non-confidence loss, and respectively represent the probability that the j-th bounding box on the i-th feature map predicted by the model contains an object and the object probability in the ground truth label.

[0096] In summary, the total loss function of the YOLOv7-tiny model is:

[0097] L = L loc + L cls + L obj

[0098] In this embodiment, in step four, the UAV coordinates are converted into the plane coordinates of the image target area according to the actual ratio; the boundary points of the target area are determined, and the circumscribed rectangle is continuously rotated and reduced until it encloses all the boundary points and has the smallest area, so as to obtain the minimum circumscribed rectangle of the target area; the plane coordinates of the midpoint of the diagonal of the minimum circumscribed rectangle are taken as the plane coordinates of the center point of the image target area; the plane coordinates of the center point of the area are converted back to the Beidou positioning coordinates to obtain the image coordinate data.

[0099] In this embodiment, in step five, the Beidou BDS-3 satellite short message service is used for data transmission, with a single-packet capacity of 1000 bytes and a rate of 500 bps. The system adopts a dynamic packet splitting strategy: high-priority data (pest coordinates and threat levels) is sent in binary encoding; the JPEG-compressed image data is transmitted in blocks in descending order of threat level, with each block ≤ 1000 bytes, and the integrity is ensured through verification. To cope with the electromagnetic interference in the farmland, the UAV is equipped with an anti-interference antenna.

[0100] In this embodiment, in step six, the user-side image recombination includes: the receiving end sorts the data packets according to the sequence number, verifies whether the data packets are damaged or lost through the check code, and finally merges the data packets into a complete JPEG file. The coordinate metadata preferentially transmitted through the Beidou short message is overlaid and displayed on the recombined JPEG image using text or icons (the coordinate is the center point coordinate of the image area).

[0101] Step six also includes building a three-dimensional visualization system on the ArcGIS Pro platform: first, based on the lidar point cloud data, a digital elevation model (DEM) with a resolution of 5 cm / cell is generated through a surface reconstruction algorithm; the ArcGIS Pro platform generates a red-yellow-green three-color graded pest heat map through kernel density analysis, and then overlays the red-yellow-green three-color graded pest heat map, supporting screening and display according to crop type and disease level, and linking with the preset flight path of the spraying UAV. The digital elevation model (DEM) and the heat map can be respectively overlaid on the JPEG image to display the elevation distribution and pest degree of the image area.

[0102] Embodiment Two:

[0103] As Figure 2 shown, this embodiment discloses a real-time crop monitoring system based on UAV image recognition and Beidou short message technology, including: a multispectral image acquisition module, an edge computing and processing module, a dual-frequency Beidou positioning module, a Beidou short message transmission module, and a user-side interaction module;

[0104] The multispectral image acquisition module is used to carry visible light, infrared, and multispectral cameras, support an adaptive exposure algorithm to eliminate light interference, and synchronously generate centimeter-level three-dimensional coordinates of the farmland with a lidar;

[0105] The edge computing processing module is used for the embedded AI chip to run the lightweight YOLOv7-tiny model, real-time identify the characteristics of pests and diseases and generate a mask map;

[0106] The dual-frequency Beidou positioning module is used for real-time positioning (longitude, latitude, altitude) of the UAV through the Beidou BDS-3 satellite using PPP-AR, and obtaining the coordinate data of the pest and disease images;

[0107] The Beidou short message transmission module is used for integrating the RDSS terminal, supporting image compression and dynamic packet transmission, and transmitting back the pest and disease images and coordinate data;

[0108] The user-side interaction module is used for GIS map to overlay and display the pest and disease heat map, and provide prevention and control suggestions and operation path planning.

[0109] The loss function of the YOLOv7-tiny model in the edge computing processing module consists of localization loss, classification loss and confidence loss:

[0110] Among them, the localization loss is used to measure the accuracy of the model's prediction of the target position. The YOLOv7-tiny model uses the Mean Square Error loss function to calculate the localization loss. The formula is as follows:

[0111]

[0112] In the formula, L loc represents the localization loss, λ coord is the weight parameter used to balance the localization loss and the classification loss, S is the size of the feature map, B is the number of prediction boxes in each grid, i represents the index of the feature map, and j represents the index of the box in each grid. represents whether the j-th box on the i-th feature map contains the target (1 means contains, 0 means does not contain), tx i and ty i respectively represent the position coordinates of the j-th box predicted by the model, tx i and ty i respectively represent the position coordinates of the j-th box in the true label.

[0113] Among them, the classification loss is used to measure the accuracy of the model's prediction of the target category. The YOLOv7-tiny model uses the CrossEntropy Loss function to calculate the classification loss. The formula is as follows:

[0114]

[0115] In the formula, L cls represents the classification loss, λ clsis the weight parameter for balancing the classification loss and the localization loss, C is the number of categories, and represent the probability that the i-th box predicted by the model belongs to the c-th category and the probability that the i-th box in the true label belongs to the c-th category, respectively.

[0116] Among them, the confidence loss is used to measure the detection accuracy of the model for the target. The YOLOv7-tiny model uses the binary cross-entropy loss function to calculate the confidence loss. The formula is as follows:

[0117]

[0118] In the formula, L obj represents the object loss, λ obj is the weight parameter for balancing the confidence loss and the non-confidence loss, and represent the probability that the j-th box on the i-th feature map predicted by the model contains a target and the object probability in the true label, respectively.

[0119] In summary, the total loss function of the YOLOv7-tiny model is:

[0120] L = L loc + L cls + L obj 。

[0121] During the positioning of the drone by the Beidou BDS-3 satellite using PPP-AR, the floating-point ambiguity of the ionosphere-free combination observation is decomposed into an integer wide-lane (WL) ambiguity and a floating-point narrow-lane (NL) ambiguity:

[0122]

[0123] In the formula, represents the floating-point ambiguity of the non-differenced ionosphere-free combination observation of receiver r at satellite s; and represent the integer wide-lane ambiguity and the floating-point narrow-lane ambiguity, respectively; are the frequencies at B1 and B2 frequency points, respectively.

[0124] The floating-point wide-lane ambiguity parameter is obtained through the Melbourne-Wubbena (MW) formula, and the HMW (Hatch-MW) combined observation of the BDS-3 satellite is expressed as:

[0125]

[0126] In the formula, is the floating-point wide-lane ambiguity; is its integer part; d r,WL and are the fractional parts of the hardware delays at the receiver end and the satellite end; and respectively represent the pseudorange and phase observations of receiver r at the i-th frequency (i = B1, B2) of satellite s; are the wavelengths of the B1 and B2 frequencies respectively; λ WL is the wide-lane wavelength.

[0127] The floating-point narrow-lane ambiguity parameter is obtained by using the floating-point ionosphere-free combined observation ambiguity and the integer wide-lane ambiguity:

[0128]

[0129] In the formula, is the integer part of the narrow-lane ambiguity; d r,NL and are the fractional parts of the hardware delays at the receiver end and the satellite end respectively.

[0130] The Beidou short message transmission module implements a dynamic packet splitting strategy, preferentially transmitting image blocks of pest and disease areas with a high threat level, and the single-packet data volume ≤ 1000 bytes; the coordinate data is encapsulated into high-priority messages.

[0131] The user-side interaction module supports the sliding window algorithm to verify data integrity and reconstructs the farmland terrain model through three-dimensional point cloud data; the prevention and control suggestions include pesticide types, spraying amounts, and the flight path planning of unmanned aerial vehicles.

[0132] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A real-time monitoring method for crops based on UAV image recognition and Beidou short message technology, characterized in that, Including: Collecting image data of the target farmland by a drone to obtain a geotagged RGB-IR four-channel data set; Inputting the images in the geotagged RGB-IR four-channel data set into the YOLOv7-tiny model to identify the pest and disease characteristics in the images and output pest and disease images; Locating the drone to obtain the coordinate data of the pest and disease images; Transmitting the pest and disease images and the coordinate data to the user terminal according to the Beidou short message technology. The user terminal recombines the pest and disease images and overlays the coordinate information, and sends instructions to the drone according to the recombined pest and disease images.

2. The real-time monitoring method for crops based on UAV image recognition and Beidou short message technology according to claim 1, characterized in that, Obtaining the geotagged RGB-IR four-channel data set includes: generating three-dimensional point cloud data of the target farmland by lidar, aligning the lidar coordinates with the image data in terms of timestamp, and obtaining the geotagged RGB-IR four-channel data set.

3. The real-time monitoring method for crops based on UAV image recognition and Beidou short message technology according to claim 1, characterized in that, Locating the drone to obtain the coordinate data of the pest and disease images includes: Performing real-time positioning on the drone by using the Beidou BDS-3 satellite with PPP-AR to obtain the coordinate data of the drone; Converting the coordinate data of the drone into the plane coordinates of the pest and disease image area; Obtaining the minimum bounding rectangle of the pest and disease image area and calculating the center point coordinates of the minimum bounding rectangle; Converting the center point coordinates into Beidou positioning coordinates to obtain the coordinate data of the pest and disease images.

4. The real-time monitoring method for crops based on UAV image recognition and Beidou short message technology according to claim 3, characterized in that, Performing real-time positioning on the drone by using the Beidou BDS-3 satellite with PPP-AR includes: decomposing the floating-point ambiguity of the ionosphere-free combined observation values into an integer wide-lane ambiguity and a floating-point narrow-lane ambiguity.

5. The real-time monitoring method of crops based on UAV image recognition and Beidou short message technology according to claim 4, characterized in that, The floating-point ambiguity of the ionosphere-free combined observation values is: Among them, represents the floating-point ambiguity of the undifferenced ionosphere-free combination observation of receiver r on satellite s; and represent the integer wide-lane ambiguity and the floating-point narrow-lane ambiguity respectively; are the frequencies on B1 and B2 frequency points respectively.

6. The real-time monitoring method of crops based on UAV image recognition and Beidou short message technology according to claim 1, characterized in that, Transmitting the pest and disease images and the coordinate data to the user terminal according to the Beidou short message technology includes: compressing the pest and disease images into the JPEG format, adopting a dynamic packet splitting strategy for the compressed image data, performing descending-order block transmission according to the threat level, and encapsulating the coordinate data into a target priority message.

7. The real-time monitoring method of crops based on UAV image recognition and Beidou short message technology according to claim 2, characterized in that, Sending instructions to the drone according to the pest and disease heat map obtained from the recombined pest and disease images includes: Generating a digital elevation model from the three-dimensional point cloud data through a surface reconstruction algorithm; Generating a red-yellow-green three-color graded pest and disease heat map through kernel density analysis; Overlaying the digital elevation model and the red-yellow-green three-color graded pest and disease heat map on the recombined pest and disease images; Sending instructions to the drone according to the overlaid pest and disease images.

8. A real-time crop monitoring system based on UAV image recognition and Beidou short message technology for implementing the method according to any one of claims 1-7, characterized in that, Including: A multispectral image acquisition module, an edge computing processing module, a dual-frequency Beidou positioning module, a Beidou short message transmission module, and a user terminal interaction module; The multispectral image acquisition module is used to collect image data of the target farmland by a drone to obtain a geotagged RGB-IR four-channel data set; The edge computing processing module is used to input the images in the geotagged RGB-IR four-channel data set into the YOLOv7-tiny model to identify the pest and disease characteristics in the images and output pest and disease images; The dual-frequency Beidou positioning module is used to position the UAV and obtain the coordinate data of the pest and disease image; The Beidou short message transmission module is used to transmit the pest and disease image and the coordinate data to the user terminal according to the Beidou short message technology; The user terminal interaction module is used for the user terminal to recombine the pest and disease image and overlay the coordinate information, and send instructions to the UAV according to the recombined pest and disease image.

Images