A device for acquiring canopy temperature in real time and an irrigation scheme generation system and method

Abstract

This invention discloses a device for real-time acquisition of canopy temperature and an irrigation scheme generation system and method. Utilizing a device capable of real-time canopy temperature acquisition, based on RGB and IR images of farmland, the system performs image conversion and registration, followed by image segmentation using the multimodal lightweight multilayer semantic segmentation model Multi-MambaSeg to accurately extract the canopy temperature. Based on this device, a system and method for real-time generation of irrigation schemes are constructed. This system can generate irrigation schemes for target farmland in real time based on canopy temperature, meteorological data, and other information, and directly distribute them to terminals or irrigation execution ends. This ensures the timeliness of the irrigation schemes, meets crop needs, and achieves precision irrigation. Compared to existing offline processing methods, this invention achieves near real-time and automation throughout the entire process, reducing labor costs and improving the intelligence and efficiency of farmland management.

Description

Technical Field

[0001] This invention relates to the field of agricultural production management technology, and in particular to a method and system for real-time generation of canopy temperature acquisition and farmland irrigation decisions based on 5G, edge devices and image processing technologies. Background Technology

[0002] Irrigation methods in agricultural planting are often based on experience or fixed patterns, making it difficult to accurately match crop needs and leading to water waste. With technological advancements, advanced imaging equipment, such as drones, agricultural machinery, or high-resolution smart cameras deployed in fields, can acquire comprehensive and multi-angle image information of farmland and crops. Using professional image processing software and powerful algorithm models, these images can be analyzed and processed in depth to determine the current water requirements of crops and the soil moisture content, thereby planning the timing, volume, and method of irrigation and generating an irrigation plan. CN113919615A discloses a smart irrigation decision-making system for farmland based on drone remote sensing data inversion. However, in practical applications, due to the large volume of image data collected, offline data processing is usually required, necessitating manual image labeling and other time-consuming tasks. Furthermore, existing equipment is often independent, and data may not be directly readable or transmitted between devices, requiring intermediate conversion, which also leads to inefficiency.

[0003] In summary, although the above technologies utilize image processing and other methods to obtain canopy temperature, these solutions still have the following shortcomings:

[0004] 1. The decision-making process is lengthy and slow. Traditional methods typically require specialized software for offline image processing after acquiring farmland image data to obtain crop or farmland feature data, resulting in long data collection cycles. Data update frequencies are measured in hours, days, or even weeks, leading to poor real-time performance and delayed response.

[0005] 2. Image processing methods rely excessively on RGB spectra and lack fusion with other spectral data, which limits the deep extraction of crop information; secondly, image processing model architectures are complex and computationally intensive, making it difficult to achieve lightweight deployment.

[0006] 3. Lack of standard interfaces for data transmission and high communication costs. The absence of standardized interfaces during data transmission hinders direct communication between units, resulting in low data transmission efficiency and difficulty in system integration. Furthermore, the continuous transmission of high-frequency, multi-dimensional data collected by farmland sensors to the cloud places extremely high demands on network bandwidth, generating significant data transmission and cloud storage costs.

[0007] In summary, current agricultural irrigation decision-making processes still suffer from shortcomings such as complex models, poor timeliness, lack of standard data transmission interfaces, high communication costs, and the inability to make real-time irrigation decisions. If too much time passes and factors such as climate and temperature change, the irrigation plan may become unsuitable for the current crop needs, impacting irrigation effectiveness. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this application proposes a device for real-time acquisition of canopy temperature and a system for real-time generation of irrigation plans. It effectively integrates RGB and IR to achieve deep extraction of crop information. Furthermore, by constructing a multimodal, lightweight, multi-layer semantic segmentation model, it accurately extracts canopy temperature information. Based on the canopy temperature information, combined with the planting environment and crop status, it generates irrigation plans in real time, ensuring the timeliness of the irrigation plans, meeting crop needs, and achieving precision irrigation.

[0009] A device for real-time acquisition of canopy temperature includes an image acquisition unit and a canopy temperature extraction unit;

[0010] The image acquisition unit acquires RGB and IR images of the farmland in real time;

[0011] The canopy temperature extraction unit includes an image conversion module, an image registration module, an image segmentation module, and a canopy temperature extraction module. The image conversion module converts the IR image into a TIFF image. The image registration module registers the RGB image with the IR image and the TIFF image. The image segmentation module uses the multimodal lightweight multilayer semantic segmentation model Multi-MambaSeg to segment the background in the registered RGB image to obtain a canopy mask image. The canopy temperature extraction module applies the canopy mask image to the TIFF image to extract the canopy temperature.

[0012] Furthermore, the structure of the multimodal lightweight multilayer semantic segmentation model Multi-MambaSeg includes two encoder layers and one decoder layer. The two encoder layers take RGB images and TIFF images as input, respectively; the specific structure is as follows:

[0013] For an encoder that takes RGB images as input, it consists of one PatchEmbed module, three MobileMamba Block modules, and two DownSample modules, with the two DownSample modules arranged alternately among the three MobileMamba Block modules.

[0014] The encoder for TIFF images as input consists of one Normalization module, one PatchEmbed module, and one MobileMamba Block module connected in sequence.

[0015] The decoder includes: an MLP Layer module connected to the outputs of two encoders respectively, and the two MLP Layer modules are then connected to an MLP and an upsampling layer in sequence to obtain a canopy mask.

[0016] Furthermore, the PatchEmbed module is composed of... The input image is processed by a ConvBNReLU operation with a step size of 2 and a ConvBN operation with a step size of 2. Downsampling.

[0017] Furthermore, the canopy mask image is denoted as Where F represents the output feature after passing through the MLP layer, H represents the height of the input RGB image, and W represents the width of the input image.

[0018] Furthermore, the canopy temperature extraction module extracts canopy temperature as follows: a mask is applied to the TIFF image, and the mask is used to filter out temperature data belonging only to the crop canopy, thus obtaining the average canopy temperature within the region. , denoted as:

[0019] ;

[0020] Where T is the TIFF temperature matrix and M is the mask matrix, both having a size of [missing information]. The numerator is the sum of all elements after multiplying the temperature matrix T by the mask M element by element, and the denominator is the sum of all elements in the mask M.

[0021] Furthermore, the canopy temperature extraction unit is deployed and operates on a cloud server, edge device, or other device capable of performing these functions.

[0022] A system capable of generating irrigation plans in real time includes:

[0023] The aforementioned device for real-time acquisition of canopy temperature extracts canopy temperature information based on real-time acquired RGB and IR images.

[0024] The motion mechanism equipped with the above-mentioned equipment drives the equipment to acquire canopy temperature information of farmland piece by piece;

[0025] A cloud data center that interacts with the device includes a meteorological data receiving and storage module, an evapotranspiration calculation module, an irrigation scheme generation module, a data storage module, and a task distribution module; wherein,

[0026] The meteorological data receiving and storage module is used to acquire local real-time meteorological data;

[0027] The evapotranspiration calculation module calculates the reference evapotranspiration ET0 and crop evapotranspiration ETc of the crop based on meteorological data.

[0028] The irrigation scheme generation module calculates the irrigation weight coefficient for each zone based on the canopy temperature information of each zone. Then, combine the crop evapotranspiration (ETC) to form an irrigation plan;

[0029] The data storage module pre-stores farmland zoning coordinate data, and obtains zoning image data block by block based on the zoning coordinate data.

[0030] The task distribution module distributes irrigation plans to terminals or irrigation execution terminals.

[0031] Furthermore, the motion mechanism is an aircraft or a mobile gantry used to carry the device that can acquire canopy temperature in real time and move relative to the farmland according to a preset trajectory.

[0032] Furthermore, the cloud data center communicates with each unit using 5G communication.

[0033] Furthermore, an UAV equipped with an RTK module was used to perform orthophoto surveys on the target farmland to obtain the zone coordinates.

[0034] Furthermore, handheld GIS data collectors were used to collect coordinate information for each target zone.

[0035] Furthermore, based on farmland meteorological information, the reference evapotranspiration ET0 and crop evapotranspiration ETc are calculated and denoted as: ;

[0036] ;

[0037] Where Kc represents the crop coefficient; other parameters are collected from meteorological stations, ET0 represents the reference crop evapotranspiration; Δ represents the slope of the saturated vapor pressure-temperature curve; R n Represents net surface radiation; G represents soil heat flux; γ represents hygrometer constant; T represents daily average temperature; u2 represents wind speed at 2 meters; e s Indicates saturated vapor pressure; e a This indicates the actual water vapor pressure.

[0038] Furthermore, the irrigation weight coefficient for each zone is calculated. , denoted as:

[0039] ;

[0040] in, Indicates the partition number. Indicates the first Average canopy temperature of each zone This represents the maximum average canopy temperature across all zones. This represents the minimum average canopy temperature across all zones. and All of these are hyperparameters.

[0041] Furthermore, the amount of water injected into each zone. Recorded as: .

[0042] A method for generating irrigation plans in real time, based on the aforementioned system capable of generating irrigation plans in real time, includes the following steps:

[0043] Step 1: Obtain target farmland zoning data;

[0044] Step 2: Based on the motion trajectory of the motion mechanism, acquire RGB and IR images of farmland area by area;

[0045] Step 3: Convert the IR image to TIFF image format;

[0046] Step 4: Register the RGB image and the IR image, aligning the farmland sub-regions in the RGB image and the IR image to the same pixel coordinate system;

[0047] Step 5: The registered RGB image is segmented using the multimodal lightweight multilayer semantic segmentation model Multi-MambaSeg to filter out the soil background in the RGB image and obtain the canopy mask.

[0048] Step 6: Apply the canopy mask to the TIFF image to obtain the average canopy temperature within the region. ;

[0049] Step 7: Repeat the above steps to obtain the average crop canopy temperature for all zones in the entire farmland. The irrigation weight coefficients for each zone were calculated. The irrigation volume in each zone is equal to the crop evapotranspiration ET between two irrigation intervals. c With the district The product of these factors forms the variable irrigation scheme.

[0050] The beneficial effects of this invention are:

[0051] 1. This invention mounts edge devices, image acquisition equipment, and other devices on an aircraft and interacts with a cloud data center in real time. This enables online processing of image and other data, real-time generation of irrigation plans, and timely distribution. Compared with the offline processing methods in the prior art, this invention can achieve near real-time and automation of the entire process, reduce labor costs, and improve the intelligence and efficiency of farmland management.

[0052] 2. Based on the characteristics of RGB and TIFF images, this invention creatively proposes a dual-channel processing multimodal lightweight multilayer semantic segmentation model, which can extract features from the visible light features of RGB images and the temperature features of TIFF images, thereby achieving accurate segmentation of different crop canopies and backgrounds.

[0053] 3. This invention achieves real-time processing and decision-making of field data by deploying a lightweight AI model on edge devices, reducing the irrigation plan generation cycle from days to seconds. It offers high timeliness, and the irrigation plan meets the current crop needs, enabling precise irrigation control.

[0054] 4. This invention enables real-time and efficient communication between cloud data centers, edge devices, and terminals, integrating them into a real-time irrigation decision generation system.

[0055] 5. This invention offloads core computing tasks to edge nodes, significantly reducing bandwidth pressure and cloud service costs for data uploads, while improving work efficiency and ensuring the timeliness of irrigation solutions. Attached Figure Description

[0056] Figure 1 This is a flowchart of the irrigation scheme generation method of the present invention.

[0057] Figure 2 This is an example of an RGB image of farmland.

[0058] Figure 3 This is an example of an IR image of farmland.

[0059] Figure 4 This is an example of a TIFF format image and its parsing.

[0060] Figure 5 This is an example of a registered RGB image.

[0061] Figure 6 This is a sample mask image after segmenting an RGB image of farmland.

[0062] Figure 7 This is an architecture diagram of a lightweight, multi-layer semantic segmentation model (Multi-MambaSeg).

[0063] Figure 8 This is a schematic diagram of the irrigation scheme generation system framework of the present invention.

[0064] Figure 9 This is a flowchart of the partition coordinate acquisition unit.

[0065] Figure 10 This is a schematic diagram of the experimental site's zoning. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1-10 The present invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0067] Example 1

[0068] This invention proposes a device for real-time acquisition of canopy temperature. The device mainly includes an image acquisition unit and a canopy temperature extraction unit, as detailed below:

[0069] The image acquisition unit specifically includes an RGB image acquisition device and an IR image acquisition device, used to acquire RGB and IR images of farmland in real time, respectively. The camera used in this embodiment is a DJI Zenmuse H20T.

[0070] The canopy temperature extraction unit receives RGB and IR images acquired by the image acquisition device, and extracts canopy temperature information after processing the RGB and IR images. The canopy temperature extraction unit specifically includes an image conversion module, an image registration module, an image segmentation module, and a canopy temperature extraction module. The functions of each module are as follows:

[0071] (1) The image conversion module converts the IR image into a TIFF image; for example, the size of the acquired RGB image is 4056*3040*3 (e.g., Figure 2 As shown), the size of the IR image is 640*512*3 (as shown). Figure 3 (As shown); converting IR images to TIFF image format digitally presents the information carried in the image. Converting IR images to TIFF format not only preserves the key metadata carried by the original data to the greatest extent, including but not limited to radiometric information, shooting parameters, GPS data, timestamps, etc., but also facilitates subsequent data processing. For example... Figure 4 As shown, a TIFF format image is parsed into a two-dimensional matrix of the same size, where each element represents the temperature of that pixel.

[0072] (2) The image registration module registers the RGB image with the IR image and the TIFF image. Because the RGB and IR images have different resolutions, the position of the same object is different in the two images; that is, the farmland sub-regions in the RGB image and the IR image cannot be directly matched. Therefore, registration is necessary to align the farmland sub-regions in the RGB and IR images to the same pixel coordinate system. For example, registering the RGB image with the IR image results in a registered RGB image size of 640*512*3 (e.g., ...). Figure 5 (As shown).

[0073] (3) The image segmentation module segments the background in the registered RGB image to obtain the canopy mask image; it then uses a lightweight multi-modal semantic segmentation model (Multi-MambaSeg) to segment the registered RGB image, filtering out background elements such as soil to obtain the canopy mask. Figure 6 As shown. The architecture of the Multi-MambaSeg model is as follows. Figure 7 As shown, it consists of two parts: an encoder and a decoder.

[0074] The Multi-MambaSeg encoder consists of two layers, with RGB and TIFF images as inputs, respectively. In terms of storage structure, RGB images are three-dimensional arrays, while TIFF images are actually two-dimensional arrays representing temperature matrices. Furthermore, in terms of feature representation, RGB images reflect the visual features of an object, while TIFF images reflect its temperature characteristics. Therefore, this invention designs corresponding encoders specifically for the characteristics of RGB and TIFF images.

[0075] Using RGB images as input ( The upper-layer encoder is an improvement on the MobileMamba model (the MobileMamba model can adopt the model structure disclosed in "MobileMamba: Lightweight Multi-Receptive Visual Mamba Network"), consisting of one PatchEmbed module, three MobileMambaBlock modules, and two DownSample modules. The structures of the MobileMamba Block and DownSample modules are the same as in the MobileMamba model. The difference is that in this invention, the PatchEmbed module is composed of... The process consists of one ConvBNReLU operation with a step size of 2 and one ConvBN operation with a step size of 2, performing the following steps on the input image: Downsampling. The ConvBNReLU operation and the ConvBN operation are shown in equations (1) and (2), respectively:

[0076] (1)

[0077] (2)

[0078] Where Conv is the convolution operation, BN (Batch Normalization) is the batch normalization operation, and ReLU is the activation function. As input features, For output features.

[0079] In addition, the output feature size of each MobileMamba Block module in this invention is: Where H represents the height of the input RGB image; W represents the width of the input image; These correspond to three MobileMamba Block modules; C i Indicates the first The number of channels for the output features of each MobileMamba Block module; As an adjustable parameter The adjustment model focuses on the degree of coarse or fine granularity; when the crop canopy is large, A value of 2 can be used, such as for mid-to-late stage corn; when the crop canopy is small, the value can be appropriately increased. The value, Option 3 is acceptable, such as early wheat.

[0080] Using TIFF images as input ( The lower-level encoder of the TIFF image consists of a Normalization module, a PatchEmbed module, and a MobileMamba Block module. First, the TIFF image is processed by a Normalization module to perform max-min normalization and then multiplied by 255 to convert the pixel value of the TIFF image to the order of the pixel value of the RGB image, as shown in formula (3).

[0081] (3)

[0082] in, These are the pixel values ​​of the TIFF image. The minimum pixel value of the TIFF image. This represents the maximum pixel value of the TIFF image. These are the pixel values ​​after processing by the Normalization module.

[0083] Then it goes through one PatchEmbed module and one MobileMamba Block module, which are structurally the same as the PatchEmbed module and the MobileMamba Block module with i=1 in the upper layer encoder.

[0084] In Multi-MambaSeg, the output of each encoder passes through an MLP Layer module. The first step in this module involves an MLP operation (i.e., a linear convolutional operation) to unify the number of channels in the input features to D. The second step upsamples the multi-level features of the RGB and TIFF images. .

[0085] More specifically, the MLP Layer module is shown in formula (4), denoted as:

[0086] (4)

[0087] in For the output characteristics of the i-th MobileMamba Block module, This represents the corresponding i-th output feature. This represents the linear convolution operation of an MLP, which converts the number of channels in the input feature layer... Transform into D.

[0088] Next, the outputs of all MLP Layer modules are concatenated to obtain the feature dimension as follows: After fusing all features through one more MLP layer, a final result is calculated. The output is shown in formula (5), denoted as:

[0089] (5)

[0090] in The function represents the feature join operation; For the number of categories (such as separating the background, such as canopy and soil), then =2); This represents the linear convolution operation of an MLP, which converts the number of channels in the input feature layer... Transform into F represents the output feature after passing through the MLP layer.

[0091] Finally, F is upsampled to obtain a dimension of Segmentation mask As shown in formula (6), it is denoted as:

[0092] (6)

[0093] (4) The canopy temperature extraction module applies the canopy mask image to the TIFF image to extract the canopy temperature. Applying the canopy mask to the TIFF image accurately extracts the canopy temperature data. The acquired IR images may contain background information such as crop canopy, bare soil, and roads. The temperature characteristics of these different objects vary greatly (for example, soil temperature at noon may be much higher than plant leaf temperature). To accurately extract the temperature information of the crop canopy and eliminate interference from other backgrounds, a mask must be used to "filter" out the temperature data belonging only to the crop canopy, obtaining the average canopy temperature within the region. Assume the TIFF temperature matrix is ​​T and the mask matrix is ​​M, both of which are of similar size. Mean canopy temperature ( The calculation method for ) is shown in formula (7), and is denoted as:

[0094] (7)

[0095] The numerator is the sum of all elements after multiplying the temperature matrix T by the mask M element by element, and the denominator is the sum of all elements in the mask M (i.e., the number of 1s).

[0096] In this embodiment, the canopy temperature extraction unit can be deployed and run on a cloud server, edge device, or other device capable of performing these functions. Edge devices can reduce data transmission latency, improve real-time performance, reduce bandwidth consumption, enhance privacy protection, optimize energy consumption and computing resource allocation, and significantly improve task execution efficiency and reliability.

[0097] Example 2

[0098] Based on Example 1, this embodiment also proposes a system capable of generating irrigation plans in real time. The system includes the aforementioned device for real-time acquisition of canopy temperature, and further includes:

[0099] The motion mechanism equipped with the above-mentioned equipment drives the equipment to acquire canopy temperature information of farmland piece by piece;

[0100] A cloud data center that interacts with the device includes a meteorological data receiving and storage module, an evapotranspiration calculation module, an irrigation scheme generation module, a data storage module, and a task distribution module.

[0101] The meteorological data receiving and storage module is used to acquire local real-time meteorological data.

[0102] The evapotranspiration calculation module calculates the reference evapotranspiration ET0 and crop evapotranspiration ETc of the crop based on meteorological data.

[0103] The irrigation scheme generation module calculates the irrigation weight coefficient for each zone based on the canopy temperature information of each zone. Then, combine the crop evapotranspiration (ETC) to form an irrigation plan;

[0104] The data storage module pre-stores farmland zoning coordinate data and obtains zoning image data block by block based on the zoning coordinate data; if the zoning is small, the center point coordinates of the zoning can be obtained; if the zoning is large, the coordinate data of multiple points of the zoning can be obtained.

[0105] The task distribution module distributes irrigation plans to terminals or irrigation execution terminals.

[0106] In this embodiment, a motion mechanism, such as an aircraft or a mobile truss, is selected according to the actual situation. It can carry a device that can acquire canopy temperature in real time and move relative to the farmland according to a preset trajectory, acquiring image data of the farmland from top to bottom zone by zone.

[0107] In this embodiment, the target farmland needs to be divided into sections in advance to facilitate the subsequent acquisition of images and other information section by section. Specifically, the target farmland can be divided into several farmland sub-sections of uniform size and shape according to a certain length and width, and the position coordinates in each section can be obtained.

[0108] In this embodiment, the reference evapotranspiration ET0 and crop evapotranspiration ETc of the crop are calculated by collecting meteorological information, as shown in formulas (8) and (9).

[0109] (8)

[0110] (9)

[0111] Where Kc represents the crop coefficient; other parameters are collected from meteorological stations, ET0 represents the reference crop evapotranspiration (mm / day); Δ represents the slope of the saturated vapor pressure-temperature curve (kPa / ℃); R n γ represents net surface radiation (MJ / m² / day); G represents soil heat flux (MJ / m² / day); γ represents hygrometer constant (kPa / ℃); T represents daily average temperature (℃); u² represents wind speed at 2 meters (m / s); e s Indicates saturated vapor pressure (kPa); e a This represents the actual water vapor pressure (kPa).

[0112] In this embodiment, the irrigation scheme generation module calculates the irrigation weight coefficient for each zone based on the average canopy temperature of all zones. The process of generating an irrigation plan is as follows:

[0113] First, based on the average canopy temperature information obtained from all zones, the irrigation weighting coefficient for each zone is calculated. As shown in the following formula (10):

[0114] (10)

[0115] in Indicates the partition number. Indicates the first Average canopy temperature of each zone This represents the maximum average canopy temperature across all zones. This represents the minimum average canopy temperature across all zones. and The hyperparameters were obtained. Indicates the first Irrigation weight coefficients for each zone. The range of values ​​is , The irrigation weighting coefficient for the zone with the lowest average canopy temperature represents the zone with the lowest irrigation volume. The irrigation weighting coefficient represents the zone with the highest average canopy temperature, indicating that the zone receives the most irrigation. The larger the value, the better. Lower limit of the range The smaller; The larger the value, the better. Upper limit of the value range The closer it is to 1.

[0116] Water volume in each zone ET during the two irrigation periods c With the irrigation weighting coefficient of this zone The product of is shown in formula (11).

[0117] (11)

[0118] In this embodiment, 5G communication is used for communication between the cloud data center and various units, modules, and terminals (including but not limited to mobile communication devices, irrigation equipment, etc.). The edge device transmits the canopy temperature information of the zone to the cloud data center in real time via the 5G module.

[0119] The communication interface transmission standard is as follows:

[0120]

[0121] The cloud data center sends decision-making instructions to the irrigation equipment, and the communication interface transmission standard is as follows:

[0122]

[0123] In this embodiment, the edge device used is the Pinli Technology 17F1 UAV-borne computing platform. When the edge device does not support converting IR images to TIFF images due to system reasons (for example, DJI Thermal SDK does not support running on ARM architecture), the IR image can be transmitted to the cloud data center via a 5G communication module. After being converted to a TIFF image in the cloud data center, it is then transmitted back to the edge device.

[0124] In this embodiment, the meteorological data receiving and storage module is used to acquire meteorological information released by local meteorological stations, including but not limited to: reference crop evapotranspiration, saturated water vapor pressure, net surface radiation, soil heat flux, thermometer constant, air temperature, wind speed, water vapor pressure, etc.

[0125] In this embodiment, the methods for obtaining the farmland zoning coordinate data pre-stored in the data storage module include, but are not limited to, the following two: ① Considering high precision requirements, a drone equipped with an RTK module can be used to perform orthophoto surveys on the target farmland. The images acquired during the flight are imported into professional software for processing to generate a high-precision orthophoto map. Finally, the coordinate information of a point in each target zoning can be directly obtained from the generated orthophoto map. The zoning coordinate data is then input into the data storage module for storage and can be directly retrieved during operation. ② Considering economic and convenience requirements, a handheld GIS data collector can be used to collect the coordinates of the center points of each target zoning, and then the zoning coordinate data is input into the data storage module for storage and can be directly retrieved during operation. (Refer to the flowchart for reference.) Figure 9 As shown, when operating on a certain piece of farmland, it is only necessary to call the pre-stored partition coordinate data of the farmland, and the drone will collect the corresponding image data block by block according to the center coordinate of the partition.

[0126] Example 3

[0127] A method for generating irrigation plans in real time, based on the aforementioned system capable of generating irrigation plans in real time, includes the following steps:

[0128] Step 1: Obtain target farmland zoning data;

[0129] Step 2: Based on the motion trajectory of the motion mechanism, acquire RGB and IR images of farmland area by area;

[0130] Step 3: Convert the IR image to TIFF image format;

[0131] Step 4: Register the RGB image and the IR image, aligning the farmland sub-regions in the RGB image and the IR image to the same pixel coordinate system;

[0132] Step 5: The registered RGB image is segmented using the multimodal lightweight multilayer semantic segmentation model Multi-MambaSeg to filter out the soil background in the RGB image and obtain the canopy mask.

[0133] Step 6: Apply the canopy mask to the TIFF image to obtain the average canopy temperature within the region. ;

[0134] Step 7: Repeat the above steps to obtain the average crop canopy temperature for all zones in the entire farmland. The irrigation weight coefficients for each zone were calculated. The irrigation volume in each zone is equal to the crop evapotranspiration ET between two irrigation intervals. c With the district The product of these factors forms the variable irrigation scheme.

[0135] To illustrate the feasibility and technical effects of this invention, the following uses a piece of farmland as an example to describe the specific implementation process of the technical solution of this invention:

[0136] Target farmland information: such as Figure 10 As shown, the farmland is evenly divided into 20 sections, each measuring 16m x 5m. The crop planted is "Zhenmai 12" winter wheat, planted on November 2, 2024. The red dots in the diagram represent the center of each section.

[0137] Image data acquisition time: The image data was acquired on March 29, 2025.

[0138] The equipment used: the aircraft is a DJI M300 RTK drone, the image acquisition module uses a Zenmuse H20T visible light and thermal infrared dual-view lens, and the edge computing module is a Pinli Technology 17F1.

[0139] The drone, equipped with an image acquisition module and an edge computing module, flies to the corresponding coordinate location to collect image data, and processes it in real time through edge devices to obtain the average canopy temperature of each zone. Then, the average canopy temperature of each zone is sent to the cloud data center via a 5G communication module.

[0140] The cloud data center uses real-time weather station data and the current growth stage (early jointing stage K) to... c =1.1), to obtain the crop evapotranspiration ET between the two irrigation periods. c The thickness is 32.4 mm. The area of ​​each zone is 16m × 5m = 80㎡, resulting in a net irrigation water requirement of 32.4 × 0.08 ≈ 2.592 m³. Finally, multiply by the irrigation weight coefficient for each zone. This yields the actual net irrigation amount for each zone. In this example, we take... , Then the irrigation weight coefficient As shown in formula (12):

[0141] (12)

[0142] The entire process, from drone takeoff to the calculation of the actual net irrigation amount for each zone at the cloud computing center, took approximately 220 seconds. The specific calculation results are shown in the table below:

[0143]

[0144] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A device for real-time acquisition of canopy temperature, characterized in that, Includes an image acquisition unit and a canopy temperature extraction unit; The image acquisition unit acquires RGB and IR images of the farmland in real time; The canopy temperature extraction unit includes an image conversion module, an image registration module, an image segmentation module, and a canopy temperature extraction module. The image conversion module converts the IR image into a TIFF image. The image registration module registers the RGB image with the IR image and the TIFF image. The image segmentation module uses the multimodal lightweight multilayer semantic segmentation model Multi-MambaSeg to segment the background in the registered RGB image to obtain a canopy mask image. The canopy temperature extraction module applies the canopy mask image to the TIFF image to extract the canopy temperature.

2. The device for real-time acquisition of canopy temperature according to claim 1, characterized in that, The lightweight, multi-layer semantic segmentation model Multi-MambaSeg comprises two encoder layers and one decoder layer. The two encoder layers take RGB images and TIFF images as input, respectively. The specific structure is as follows: For an encoder that takes RGB images as input, it consists of one PatchEmbed module, three MobileMamba Block modules, and two DownSample modules, with the two DownSample modules arranged alternately among the three MobileMamba Block modules. The encoder for TIFF images as input consists of one Normalization module, one PatchEmbed module, and one MobileMamba Block module connected in sequence. The decoder includes: an MLP Layer module connected to the outputs of two encoders respectively, and the two MLP Layer modules are then connected to an MLP and an upsampling layer in sequence to obtain a canopy mask.

3. The device for real-time acquisition of canopy temperature according to claim 2, characterized in that, The PatchEmbed module consists of... The input image is processed by a ConvBNReLU operation with a step size of 2 and a ConvBN operation with a step size of 2. Downsampling.

4. The device for real-time acquisition of canopy temperature according to claim 2, characterized in that, Canopy mask image denoted as Where F represents the output feature after passing through the MLP layer, H represents the height of the input RGB image, and W represents the width of the input image.

5. The device for real-time acquisition of canopy temperature according to claim 4, characterized in that, The canopy temperature extraction module extracts canopy temperature as follows: a mask is applied to the TIFF image, and the mask is used to filter out temperature data that belongs only to the crop canopy, thus obtaining the average canopy temperature within the region. , denoted as: ； Where T is the TIFF temperature matrix and M is the mask matrix, both having a size of [missing information]. The numerator is the sum of all elements after multiplying the temperature matrix T by the mask M element by element, and the denominator is the sum of all elements in the mask M.

6. The device for real-time acquisition of canopy temperature according to claim 1, characterized in that, The canopy temperature extraction unit is deployed and operates on cloud servers, edge devices, or other devices capable of performing these functions.

7. A system capable of generating irrigation plans in real time, characterized in that, include: The device for real-time acquisition of canopy temperature as described in claim 1, wherein the device extracts canopy temperature information based on real-time acquired RGB and IR images; The motion mechanism equipped with the above-mentioned equipment drives the equipment to acquire canopy temperature information of farmland piece by piece; A cloud data center that interacts with the device includes a meteorological data receiving and storage module, an evapotranspiration calculation module, an irrigation scheme generation module, a data storage module, and a task distribution module; wherein, The meteorological data receiving and storage module is used to acquire local real-time meteorological data; The evapotranspiration calculation module calculates the reference evapotranspiration ET0 and crop evapotranspiration ETc of the crop based on meteorological data. The irrigation scheme generation module calculates the irrigation weight coefficient for each zone based on the canopy temperature information of each zone. Then, combine the crop evapotranspiration (ETC) to form an irrigation plan; The data storage module pre-stores farmland zoning coordinate data, and obtains zoning image data block by block based on the zoning coordinate data. The task distribution module distributes irrigation plans to terminals or irrigation execution terminals.

8. A system capable of generating irrigation plans in real time according to claim 7, characterized in that, The motion mechanism uses an aircraft or a mobile frame to move relative to the farmland according to a preset trajectory.

9. A system capable of generating irrigation plans in real time according to claim 7, characterized in that, The cloud data center communicates with each unit using 5G.

10. A system capable of generating irrigation plans in real time according to claim 7, characterized in that, A drone equipped with an RTK module was used to perform orthophoto surveys on the target farmland to obtain the zone coordinates.

11. A system capable of generating irrigation plans in real time according to claim 7, characterized in that, The coordinate information of each target zone was collected using a handheld GIS data collector.

12. The system for generating irrigation plans in real time according to claim 7, characterized in that, The reference evapotranspiration ET0 and crop evapotranspiration ETc of crops are calculated based on farmland meteorological information. Recorded as: ; ； Where Kc represents the crop coefficient; other parameters are collected from meteorological stations, ET0 represents the reference crop evapotranspiration; Δ represents the slope of the saturated vapor pressure-temperature curve; R n Represents net surface radiation; G represents soil heat flux; γ represents hygrometer constant; T represents daily average temperature; u2 represents wind speed at 2 meters; e s Indicates saturated vapor pressure; e a This indicates the actual water vapor pressure.

13. A system capable of generating irrigation plans in real time according to claim 12, characterized in that, Calculate the irrigation weighting coefficient for each zone. , denoted as: ； in, Indicates the partition number. Indicates the first Average canopy temperature of each zone This represents the maximum average canopy temperature across all zones. This represents the minimum average canopy temperature across all zones. and All of these are hyperparameters.

14. A system capable of generating irrigation plans in real time according to claim 13, characterized in that, Water volume for each zone Recorded as: .

15. A method for generating irrigation plans in real time, characterized in that, Based on the system for generating irrigation plans in real time as described in claim 7, the method includes the following: Step 1: Obtain target farmland zoning data; Step 2: Based on the motion trajectory of the motion mechanism, acquire RGB and IR images of farmland area by area; Step 3: Convert the IR image to TIFF image format; Step 4: Register the RGB image and the IR image, aligning the farmland sub-regions in the RGB image and the IR image to the same pixel coordinate system; Step 5: The registered RGB image is segmented using the multimodal lightweight multilayer semantic segmentation model Multi-MambaSeg to filter out the soil background in the RGB image and obtain the canopy mask. Step 6: Apply the canopy mask to the TIFF image to obtain the average canopy temperature within the region. ; Step 7: Repeat the above steps to obtain the average crop canopy temperature for all zones in the entire farmland. The irrigation weight coefficients for each zone were calculated. The irrigation volume in each zone is equal to the crop evapotranspiration ET between two irrigation intervals. c With the district The product of these factors forms the variable irrigation scheme.

Images