Intelligent agricultural irrigation visual control method and system based on AI voice control

Abstract

The invention discloses a smart agricultural irrigation visual control method and system based on AI voice control, and relates to the technical field of smart agricultural irrigation control, and the method comprises a monitoring system construction stage and a smart irrigation control stage based on AI voice. A digital twinborn model of the farmland to be irrigated is constructed, and the digital twinborn model is provided with a digital twinborn layer; a ground monitoring point, a soil monitoring point, a crop growth monitoring point and an irrigation point are arranged in a farmland to be irrigated, an environment data monitoring device is arranged at the ground monitoring point, a soil monitoring device is arranged at the soil monitoring point, a crop growth state monitoring device is arranged at the crop growth monitoring point, and an irrigation device is arranged at the irrigation point. Through deep fusion of digital twinborn and multi-modal monitoring and AI voice control, precision, convenience and visualization of intelligent agricultural irrigation are realized, multi-modal data acquisition covers full dimensions of farmland environment soil crops, decision deviation caused by single data is avoided, and comprehensive data support is provided for irrigation decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent agricultural irrigation control technology, and in particular to an intelligent agricultural irrigation visual control method and system based on AI voice control. BACKGROUND

[0002] Current agricultural irrigation control technology is developing in a diversified manner. Commonly seen are timing irrigation systems that operate according to preset times; there are also timely irrigation systems based on sensors that can start or stop irrigation according to sensor data such as soil moisture. In some modern farms, Internet of Things technology has been introduced to realize the functions of remote monitoring and control of irrigation equipment. Some advanced systems combine meteorological data to try to more accurately adjust irrigation strategies. Timing irrigation systems operate entirely according to preset times and cannot adjust irrigation volume and irrigation time in real time according to actual weather changes, crop growth stages, and soil moisture conditions. In dry weather, irrigation may be insufficient, while after rainfall, irrigation may be excessive, affecting crop growth and wasting water resources. For crops with large differences in water requirements at different growth stages, it is difficult to provide suitable irrigation solutions. Existing sensor-based irrigation systems are subject to various factors that interfere with the accuracy of the data collected by the sensors. For example, soil moisture sensors are easily affected by soil type, texture, and salt content, resulting in large deviations between collected data and actual soil moisture conditions. Moreover, some sensors have poor stability and their accuracy decreases over time, affecting the reliability of irrigation decisions. In addition, when the sensors are not properly laid out, the data collected cannot represent the true conditions of the entire field. Traditional irrigation systems are mostly manually operated or remotely controlled, and farmers need to manually set irrigation parameters through controllers or mobile phone APPs, which is cumbersome. For farmers with lower education levels, the learning cost is high. Moreover, when working outdoors, it is inconvenient to use a mobile phone APP to operate, making it difficult to quickly respond to the needs of field irrigation. Most irrigation systems can collect data, but they cannot intuitively and comprehensively display field irrigation-related information. Farmers have difficulty quickly obtaining key information such as soil moisture distribution and crop water demand trends from a large amount of data. In terms of decision-making, they mainly rely on manual experience or simple threshold judgments and cannot use big data and artificial intelligence technology for intelligent analysis to develop optimal irrigation strategies.

[0003] Therefore, it is urgent to design an intelligent agricultural irrigation visual control method and related system hardware based on AI voice control to improve the effect of intelligent agricultural irrigation. SUMMARY

[0004] To solve the above technical problems, the present application provides an intelligent agricultural irrigation visual control method and system based on AI voice control. The following technical solutions are adopted: The application discloses an AI voice control-based intelligent agricultural irrigation visual control method, which comprises a monitoring system construction stage and an AI voice-based intelligent irrigation control stage. The monitoring system construction stage comprises the following steps. Step 1: constructing a digital twin model of a farmland to be irrigated, wherein the digital twin model is provided with a digital twin layer. Step 2: setting ground monitoring points, soil monitoring points, crop growth monitoring points and irrigation points in the farmland to be irrigated, wherein an environmental data monitoring device is arranged at the ground monitoring points, a soil monitoring device is arranged at the soil monitoring points, a crop growth state monitoring device is arranged at the crop growth monitoring points, and an irrigation device is arranged at the irrigation points. Step 3: an intelligent agricultural irrigation server collects multi-modal data, analyzes the multi-modal data to generate digital twin layer data and irrigation action suggestion data, and maps the data to the digital twin layer of the digital twin model. The AI voice-based intelligent irrigation control stage comprises the following steps. Step 4: a manager analyzes the digital twin layer data, makes an irrigation action decision, and describes an irrigation action instruction by using a voice instruction. Step 5: the intelligent agricultural irrigation server generates irrigation instruction data based on AI model recognition of the voice instruction, and sends the irrigation instruction data to a corresponding irrigation device. Step 6: the irrigation device executes the irrigation instruction data to complete irrigation.

[0005] Optionally, in step 1, the digital twin layer comprises an environmental data layer, a soil monitoring data layer, a crop growth visual data layer and an irrigation action suggestion data layer; in step 4, the digital twin layer data comprises environmental data, soil monitoring data, crop growth visual data and irrigation action suggestion data, the environmental data is mapped to the environmental data layer, the soil monitoring data is mapped to the soil monitoring data layer, the crop growth visual data is mapped to the crop growth visual data layer, and the irrigation action suggestion data is mapped to the irrigation action suggestion data layer.

[0006] Optionally, in step 3, the multi-modal data comprises environmental data, soil physical index data, soil chemical index data, soil biological index data, crop growth visual data and irrigation action data.

[0007] Optionally, step 3 comprises the following substeps. Step 31: environmental data analysis: calculating a reference crop evapotranspiration amount and predicting future weather in a set time interval. Step 32: calculating soil available water content based on soil data. Step 33, analyzing crop growth visual data using a deep learning model to output growth data analysis results, including plant height estimation, leaf area index inversion, canopy coverage calculation, phenological period identification, disease and pest symptom automatic identification and classification; Step 34, training an irrigation action decision model based on historical irrigation data, inputting environmental data analysis results, soil data analysis results, and crop growth visual data analysis results into the irrigation action decision model, and outputting current irrigation action recommendation data; Step 35, mapping environmental data analysis results to the environmental data layer, mapping soil data analysis results to the soil monitoring data layer, mapping growth data analysis results to the crop growth visual data layer, and mapping current irrigation action recommendation data to the irrigation action recommendation data layer.

[0008] Optionally, in step 34, collect irrigation data records within a set time in the past, including irrigation time data, duration data, water quantity data, corresponding irrigation points, and associate corresponding irrigation points with environmental analysis data, soil data analysis results, and crop growth data analysis results at the same period to form a training sample set, input the training sample set into a time convolution network TCN, and obtain the irrigation action decision model after training, the input of the irrigation action decision model is the effective soil moisture content, the reference crop evapotranspiration in the future set time period, the future set time interval weather, the crop water stress index and the growth data analysis results, and the output of the irrigation action decision model is the irrigation area, the irrigation water quantity and the irrigation time.

[0009] Optionally, step 5 includes the following specific steps: Step 51, the management personnel issues a voice instruction to the intelligent agricultural irrigation server through the intelligent handheld terminal; Step 52, the intelligent agricultural irrigation server receives the audio stream and performs preprocessing; Step 53, using a Conformer end-to-end ASR model for speech-to-text operation; Step 54, using a BERTCRF model to extract voice instruction key elements; Step 55, judging whether the voice instruction key elements are complete, if so, analyzing the irrigation target, irrigation water quantity and irrigation time, and generating irrigation instruction data.

[0010] Optionally, in step 55, if the voice instruction key elements are incomplete, extract the incomplete instruction key elements, use the irrigation action recommendation data to complete the incomplete instruction key elements, and send the completed irrigation action recommendation data to the intelligent handheld terminal again to request the management personnel to confirm, if confirmed, execute step 6.

[0011] Optionally, if the irrigation action instruction does not appear after the set time threshold is executed by the management personnel, the intelligent agricultural irrigation server automatically executes the irrigation instruction data.

[0012] The AI voice control-based intelligent agricultural irrigation visual control system comprises an intelligent agricultural irrigation server, an intelligent handheld terminal, an environment data monitoring device, a soil monitoring device, a growth state monitoring device and an irrigation device.

[0013] Optionally, the AI voice control-based intelligent agricultural irrigation visual control system further comprises an audible and light alarm device.

[0014] In summary, the AI voice control-based intelligent agricultural irrigation visual control method and system have at least one of the following beneficial technical effects: The AI voice control-based intelligent agricultural irrigation visual control method and system can realize precise, convenient and visual intelligent agricultural irrigation through deep integration of digital twin, multi-modal monitoring and AI voice control, multi-modal data acquisition covering the full dimension of farmland environment, soil and crops, avoiding decision bias caused by single data, and providing comprehensive data support for irrigation decision-making.

[0015] The visual mapping of the digital twin model enables the management personnel to intuitively grasp the real-time state of each region of the farmland, and to make precise irrigation decisions based on data, combined with the irrigation recommendations generated by the system, and to eliminate the blindness of traditional experience-based watering.

[0016] The AI voice control simplifies the operation process, and the management personnel do not need to operate through complex keys or screens, but can quickly issue instructions through the voice of the intelligent handheld terminal when patrolling in the field, which is especially suitable for farmers who are not familiar with smart devices and reduces the operation threshold.

[0017] Through data-driven decision-making, voice-simplified operation and twin visual management, the AI voice control-based intelligent agricultural irrigation visual control method and system can improve irrigation efficiency, save resources, ensure crop yield, promote the transformation of agricultural production from experience-based to intelligent, and have significant economic, social and ecological benefits. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of the AI voice control-based intelligent agricultural irrigation visual control method of the present application; Figure 2 is a component connection principle diagram of the AI voice control-based intelligent agricultural irrigation visual control system of the present application.

[0019] Explanation of reference signs: 1, intelligent agricultural irrigation server; 2, intelligent handheld terminal; 3, environment data monitoring device; 4, soil monitoring device; 5, growth state monitoring device; 6, irrigation device; 7, sound and light alarm device. DETAILED DESCRIPTION

[0020] The application will be further described in detail below with reference to the accompanying drawings.

[0021] The embodiment of the application discloses an intelligent agricultural irrigation visual control method and system based on AI voice control.

[0022] Reference Figure 1 and Figure 2 , embodiment 1, the intelligent agricultural irrigation visual control method based on AI voice control includes a monitoring system construction stage and an intelligent irrigation control stage based on AI voice; The monitoring system construction stage includes the following steps: Step 1, a digital twin model of the farmland to be irrigated is constructed, and the digital twin model sets a digital twin layer; Step 2, ground monitoring points, soil monitoring points, crop growth monitoring points and irrigation points are set in the farmland to be irrigated, environment data monitoring devices 3 are set at the ground monitoring points, soil monitoring devices 4 are set at the soil monitoring points, crop growth state monitoring devices 5 are set at the crop growth monitoring points, and irrigation devices 6 are set at the irrigation points; Step 3, the intelligent agricultural irrigation server 1 collects multi-modal data, analyzes the multi-modal data to generate digital twin layer data and irrigation action suggestion data, and maps the data to the digital twin layer of the digital twin model; The intelligent irrigation control stage based on AI voice includes the following steps: Step 4, the management personnel analyze the digital twin layer data, make irrigation action decisions, and describe the irrigation action instructions by voice instructions; Step 5, the intelligent agricultural irrigation server 1 generates irrigation instruction data based on AI model recognition of the voice instructions, and sends the irrigation instruction data to the corresponding irrigation device 6; Step 6, the irrigation device 6 executes the irrigation instruction data to complete irrigation.

[0023] By adopting the above technical scheme, a virtual mapping model of the farmland is first constructed by digital twin technology, and multiple types of monitoring points and irrigation points are deployed in the entity farmland: the ground monitoring points collect environment data (such as illumination, temperature and humidity, wind speed), the soil monitoring points collect soil moisture (humidity, fertility, pH value), the crop growth monitoring points collect crop state (plant height, leaf humidity, growth stage), and the irrigation points deploy irrigation equipment that can execute instructions.

[0024] The smart agricultural irrigation server, as the core hub, collects the above-mentioned multi-modal data (environment, soil, crops) in real time, generates two types of key data through data analysis algorithms (such as soil condition evaluation model, crop water requirement prediction model): one is the digital twin layer data for visualization (such as the real-time state parameters of each region of the farmland), and the other is the irrigation action recommendation data based on data deduction (such as the need for 50m 3 ) of a certain region, and maps these data to the digital twin layer of the digital twin model, forming a dynamic visual interface of the farmland state.

[0025] The manager views the visual interface of the digital twin model, combines the irrigation recommendations generated by the system, formulates specific irrigation decisions (such as irrigating A plot for 20 minutes with medium flow), and expresses them in the form of natural language voice instructions.

[0026] The smart agricultural irrigation server 1 analyzes the voice instructions through an AI voice recognition model (such as a voice-to-text model based on deep learning, an instruction intent recognition model), converts them into standardized irrigation instruction data (such as specific irrigation area, duration, water volume, etc.), and accurately sends them to the corresponding irrigation device.

[0027] After receiving the instructions, the irrigation device automatically performs irrigation actions (such as opening / closing valves, adjusting water flow speed), completes the irrigation work, and the execution results can be fed back to the digital twin model for state updating.

[0028] Through the deep integration of digital twin, multi-modal monitoring, and AI voice control, the precision, convenience, and visualization of smart agricultural irrigation are achieved. Multi-modal data collection covers the full dimension of farmland environment, soil, and crops, avoiding decision bias caused by single data, and providing comprehensive data support for irrigation decisions.

[0029] The visual mapping of the digital twin model enables the manager to intuitively grasp the real-time state of each region of the farmland, combined with the irrigation recommendations generated by the system, to eliminate the blindness of traditional experience-based watering, and to make accurate irrigation decisions based on data.

[0030] AI voice control simplifies the operation process, and the manager does not need to operate through complex buttons or screens. When patrolling in the field, the manager can quickly issue instructions (such as pausing irrigation in B plot) through the intelligent handheld terminal 2 voice, which is especially suitable for farmers who are not familiar with smart devices, reducing the operation threshold.

[0031] The voice instructions are analyzed in real time by the AI model into device-recognizable instruction data, reducing intermediate links such as manual entry and parameter setting, and shortening the response time from decision to execution (such as traditional operation requiring 510 minutes, voice control can be compressed to within 1 minute).

[0032] Through on-demand irrigation (such as dynamically adjusting water volume according to soil moisture conditions and crop growth stage), water resource waste caused by traditional flooding can be avoided, water can be greatly saved, and problems such as soil salinization and nutrient loss caused by excessive irrigation can be reduced.

[0033] Optimize the crop growth environment: Formulate irrigation strategies based on crop growth status monitoring data to ensure that crops receive appropriate water during critical growth periods, helping to improve crop yield and quality.

[0034] Farmland status data is updated to the virtual model in real time, allowing managers to remotely view the global and local status and grasp the dynamics of the farmland without leaving home. This is especially suitable for intensive management of large farms.

[0035] The execution results of irrigation instructions are fed back to the digital twin model to facilitate tracing the operation history and effects; the multimodal data accumulated over a long period of time can be used to optimize the AI ​​model and promote continuous iteration of the system.

[0036] Voice control lowers the threshold for using technology, allowing elderly farmers and farmers with lower educational levels to quickly get started, solving the problem of difficulty in implementing smart agricultural technology.

[0037] The modular design (monitoring points and irrigation devices can be deployed as needed) is adaptable to farmland of different sizes (from small plots of land owned by individual households to farms covering 10,000 mu), facilitating its promotion and application in different scenarios and accelerating the popularization of smart agriculture.

[0038] Through data-driven decision-making, voice-simplified operations, and twin visual management, it improves irrigation efficiency, saves resources, and ensures crop yields, while promoting the transformation of agricultural production from experience-based to intelligent, with significant economic, social, and ecological benefits.

[0039] In Example 2, in step 1, the digital twin layer includes an environmental data layer, a soil monitoring data layer, a crop growth visual data layer, and an irrigation action suggestion data layer; in step 4, the digital twin layer data includes environmental data, soil monitoring data, crop growth visual data, and irrigation action suggestion data, and the environmental data is mapped to the environmental data layer, the soil monitoring data is mapped to the soil monitoring data layer, the crop growth visual data is mapped to the crop growth visual data layer, and the irrigation action suggestion data is mapped to the irrigation action suggestion data layer.

[0040] In Example 3, in step 3, the multimodal data includes environmental data, soil physical indicator data, soil chemical indicator data, soil biological indicator data, crop growth visual data, and irrigation action data.

[0041] In Example 4, step 3 includes the following sub-steps: Step 31, environmental data analysis: calculate reference crop evapotranspiration and predict future weather at set time intervals; Step 32, calculate the effective soil water content based on the soil data; Step 33, analyze the crop growth visual data using a deep learning model, output growth data analysis results, including plant height estimation, leaf area index inversion, canopy coverage calculation, phenological period identification, automatic identification and classification of disease and pest symptoms; Step 34, train the irrigation action decision model based on historical irrigation data, input the environmental data analysis results, soil data analysis results and crop growth visual data analysis results into the irrigation action decision model, and output the current irrigation action suggestion data; Step 35, map the environmental data analysis results to the environmental data layer, map the soil data analysis results to the soil monitoring data layer, map the growth data analysis results to the crop growth visual data layer, and map the current irrigation action suggestion data to the irrigation action suggestion data layer.

[0042] In step 34 of embodiment 5, collect irrigation data records in the past set time, including irrigation time data, duration data, water quantity data, corresponding irrigation points, and associate the corresponding irrigation points with the same period environmental analysis data, soil data analysis results, and crop growth data analysis results to form a training sample set. Input the training sample set into the time convolution network TCN, and obtain the irrigation action decision model after training. The input of the irrigation action decision model is the effective soil water content, the reference crop evapotranspiration in the future set time period, the future set time interval weather, the crop water stress index and the growth data analysis results. The output of the irrigation action decision model is the irrigation area, the irrigation water quantity and the irrigation time.

[0043] By adopting the above technical solution, in the environmental data analysis, the calculation of the reference crop evapotranspiration in the future set time period can be obtained by using the FAO56PenmanMonteith formula; The future set time interval weather can be directly predicted by using the official weather forecast result, or can be predicted based on the environmental data using a deep learning model; The effective soil water content can be calculated based on the soil data, and the SPAW model can be used to input the specific soil real-time monitoring data to obtain the effective soil water content data; The growth state monitoring device 5 can be in the form of a fixed camera plus a multi-spectral camera carried by a unmanned aerial vehicle. The fixed camera automatically takes pictures every hour. The unmanned aerial vehicle carries a multi-spectral camera and flies to take pictures every week. The images are preprocessed by distortion correction, illumination normalization, background segmentation, etc. A multi-task deep learning model is used to analyze the preprocessed images to obtain crop plant height, leaf area index, canopy coverage, phenological period and disease and pest analysis results.

[0044] The irrigation action decision model is trained based on historical irrigation data, including historical irrigation records (time / region / water quantity), contemporaneous environmental data (ET0 / weather forecast), soil moisture data (AWC consumption rate), and crop growth data (water stress index). The environmental data analysis results, soil data analysis results, and crop growth visual data analysis results are input into the irrigation action decision model, and the current irrigation action recommendation data is output.

[0045] The environmental data layer mapping data types include ET0 values, temperature forecasts, and precipitation probabilities. The visualization methods are as follows: Heat map: ET0 spatial distribution (green to red gradient); Contour: 24-hour temperature change; Animated icon: moving precipitation cloud map; Soil monitoring data layer mapping data types: soil moisture at each depth, AWC consumption rate; The visualization method is to display using a soil profile diagram; Crop growth visual data layer mapping data types: LAI values, canopy coverage, and pest location; The visualization methods include: NDVI heat map (multi-spectral data from a drone); 3D crop model coloring (by phenological stage); Warning icon (pest occurrence location marker); Irrigation recommendation data layer mapping data types: recommended area, water quantity, and time; The visualization method is to highlight the area.

[0046] Embodiment 6, step 5 includes the following specific steps: Step 51, the administrator issues a voice command to the smart agriculture irrigation server 1 through the intelligent handheld terminal 2; Step 52, the smart agriculture irrigation server 1 receives the audio stream and performs preprocessing; Step 53, perform voice-to-text operation using the Conformer end-to-end ASR model; Step 54, use the BERTCRF model to extract voice command key elements; Step 55, determine whether the voice command key elements are complete, and if so, parse the irrigation target, irrigation water quantity, and irrigation time, and generate irrigation instruction data.

[0047] In step 55 of embodiment 7, if it is judged that the key elements of the voice instruction are incomplete, the incomplete instruction key elements are extracted, the incomplete instruction key elements are completed by using the irrigation action suggestion data, and the completed irrigation action suggestion data is sent to the intelligent handheld terminal 2 again to request the manager to confirm. If the manager confirms, step 6 is executed.

[0048] In embodiment 8, if the irrigation action instruction is not executed by the manager after a set time threshold, the intelligent agricultural irrigation server 1 automatically executes the irrigation instruction data.

[0049] By using the above technical solutions, the Conformer end-to-end ASR model is used for voice-to-text conversion. The model has higher voice recognition accuracy in complex environments (such as farmland background noise and dialect accent), and can effectively reduce the error of voice-to-text conversion. The BERTCRF model is combined to extract key elements (irrigation target, water quantity, time, etc.), which can accurately analyze the core intent of the instruction and avoid semantic ambiguity (such as distinguishing between irrigating the No. 3 plot and turning off irrigation for the No. 3 plot), ensuring accurate and reliable conversion from voice to instruction data.

[0050] The key element verification mechanism reduces operation errors: by judging whether the key elements are complete (such as whether the irrigation area, water quantity, and duration are specified), incomplete instructions (such as only saying to water without specifying the plot) can be intercepted in advance, avoiding misoperation of the irrigation device (such as wrong irrigation or missed irrigation) caused by ambiguous instructions, and reducing the execution risk from the source.

[0051] The automatic completion mechanism reduces communication costs: when the key elements of the voice instruction are incomplete, the system automatically fills in the missing information using the generated irrigation action suggestion data (such as recommended water quantity and area based on previous multi-modal data analysis), and feeds back to the manager for confirmation. This process does not require the manager to reorganize complete instructions, reducing the back-and-forth interaction of the system asking the user to supplement incomplete instructions, especially suitable for quick operation in noisy field environments or when the manager is busy.

[0052] Optimize the completion logic by combining historical data: the irrigation action suggestion data used for completion is derived from the system's previous analysis of the state of the farmland, and has scientific basis, which can guide the manager to make more reasonable irrigation decisions (such as avoiding users from making excessive or insufficient irrigation parameters due to experience bias), indirectly improving the rationality of the decision.

[0053] If the irrigation instruction is not executed manually by the manager after a set time threshold, the system automatically triggers the execution process, which can effectively deal with delays caused by negligence, busyness, or remote operation delay of the manager (such as temporarily leaving the farmland without timely confirmation, forgetting to operate). For the irrigation needs of the critical period of crop growth (such as the grain filling period and the seedling period), timely protection can significantly reduce the risk of yield reduction due to water shortage.

[0054] Balancing artificial decision-making and system autonomy: automatic execution is not a complete replacement of artificial, but under the premise that artificial decision-making has been clear (or the system has generated reasonable suggestions and complements), it serves as the last safeguard, both retaining the decision-making dominance of management personnel and making up for the uncertainty of human operation through system initiative.

[0055] Through intelligent completion and fault-tolerant mechanisms, the system can be compatible with non-standardized voice instructions of management personnel (such as colloquial expressions, information omission), without the need for users to strictly follow fixed speech (such as allowing to say more water to the east field rather than increasing the irrigation water volume in area A by 20%), reducing the technical requirements for operators, especially suitable for middle-aged and elderly farmers or groups unfamiliar with digital operation.

[0056] Handheld terminal plus voice interaction enhances flexibility: based on voice instruction interaction of intelligent handheld terminals, the management personnel can issue instructions in real time while patrolling in the field, combining with the automatic completion and confirmation feedback of the system, realizing walk-and-say-and-control, and greatly improving the convenience of field work.

[0057] Embodiment 9, a smart agricultural irrigation visual control system based on AI voice control, comprising a smart agricultural irrigation server 1, an intelligent handheld terminal 2, an environmental data monitoring device 3, a soil monitoring device 4, a growth state monitoring device 5, and an irrigation device 6, the environmental data monitoring device 3, the soil monitoring device 4, the growth state monitoring device 5, and the irrigation device 6 are respectively installed at the ground monitoring points, soil monitoring points, crop growth monitoring points, and irrigation points set in the farmland to be irrigated, and are respectively wirelessly communicated with the smart agricultural irrigation server 1 through a wireless network, and the intelligent handheld terminal 2 is wirelessly communicated with the smart agricultural irrigation server 1 through a wireless network.

[0058] Embodiment 10, further comprising a sound-light alarm device 7, the smart agricultural irrigation server 1 controls the execution action of the sound-light alarm device 7.

[0059] The following uses specific embodiments to illustrate the implementation principle of the present application: Smart agricultural irrigation visual control method based on AI voice control specific embodiment Taking the spring irrigation management of a certain large-scale wheat planting farm (area about 500 mu, divided into 10 planting areas, numbered A1-A10) as an example, the specific implementation process is as follows: I. Monitoring system construction stage; 1. Build a digital twin model: The smart agricultural irrigation server 1 builds a digital twin model of the farm based on the farm GIS map, soil type distribution map, and crop planting plan, and sets up four digital twin layers in the model: environmental data layer, soil monitoring data layer, crop growth visual data layer, and irrigation action suggestion data layer.

[0060] 2. Deployment of monitoring points and irrigation devices: 8 ground monitoring points are evenly set up on the farm, and an environmental data monitoring device 3 (including a temperature and humidity sensor, a light sensor, and an anemometer) is installed at each monitoring point to collect real-time environmental data such as air temperature, humidity, light intensity, and wind speed. 2 soil monitoring points are set up in each planting area (A1-A10), and a soil monitoring device 4 (including a soil moisture sensor, a pH sensor, and a nitrogen, phosphorus, and potassium sensor) is installed to collect data such as soil moisture content, pH value, and effective nitrogen, phosphorus, and potassium content at a depth of 030 cm. 1 crop growth monitoring point is set up in the center of each planting area, and a growth state monitoring device 5 (including a fixed high-definition camera and a multi-spectral camera carried by a drone that flies once a week) is deployed. The fixed camera takes images of the wheat canopy every hour, and the drone collects multi-spectral images. 23 irrigation points are set up in each planting area, and an irrigation device 6 (an intelligent electromagnetic valve and a flow meter) is installed to control the irrigation switch, flow, and duration.

[0061] 3. Multi-modal data collection and digital twin layer mapping The smart agriculture irrigation server 1 collects multi-modal data in real time through a wireless network and analyzes and processes them according to the following process and then maps them to the digital twin model: Environmental data analysis: Calculate the daily reference crop evapotranspiration (ET0), combine official weather forecasts and farm historical data to predict the weather (temperature, precipitation probability) for the next 3 days, and map the ET0 spatial distribution (heat map, green to red gradient indicating evapotranspiration from low to high), 24-hour temperature contour, and precipitation cloud map animation to the environmental data layer. Soil data analysis: Based on soil monitoring data, calculate the effective soil moisture content and consumption rate of each region using the SPAW model, and map it to the soil monitoring data layer in the form of a soil profile graph (030 cm depth on the vertical axis, color depth indicating moisture content). Crop growth data analysis: After preprocessing (distortion correction, background segmentation) of the fixed camera and drone images, use a multi-task deep learning model to analyze: Output the estimated plant height, leaf area index (LAI), and canopy coverage; Identify the current wheat phenological stage (such as the green-up stage and the jointing stage); Automatically identify leaf yellowing, aphid aggregation, and other disease and pest symptoms and classify them; Map the results to the crop growth visual data layer in the form of an NDVI heat map (multi-spectral data, red indicating vigorous growth, and yellow indicating weaker growth), a 3D wheat model with color differentiation (according to the jointing stage / green-up stage), and disease and pest location markers (red warning icons). Generating irrigation action suggestion data: The server calls the irrigation action decision model (time convolution network TCN) trained based on historical irrigation data, inputs the effective soil moisture, future 3-day ET0, weather forecast, wheat water stress index and growth data analysis results, and outputs the irrigation suggestion of each region (such as A3 area needs irrigation, and the recommended water quantity is 8m 3 / acre, and the time length is 20 minutes), and maps to the irrigation action suggestion data layer in the digital twin model by highlighting the A3 area.

[0062] II. AI voice-based intelligent irrigation control stage; 4. Management personnel make irrigation decisions: The farm manager logs in to the digital twin model interface through the intelligent handheld terminal 2 and views the data of each digital twin layer: The environmental data layer shows that there is no precipitation in the A3 area in the next 2 days, and the ET0 is high (the thermal map is orange-red); The soil monitoring data layer shows that the effective soil moisture consumption rate in the A3 area is 65% (the color of the 020cm soil layer in the profile is light, and the water content is low); The crop growth visual data layer shows that the wheat in the A3 area is in the jointing stage, and the NDVI value is lower than that in the surrounding area (pale yellow), indicating that the water is slightly lacking; The irrigation action suggestion data layer highlights the A3 area, and suggests that the irrigation water quantity is 8m 3 / acre, and the time length is 20 minutes.

[0063] The manager decides to perform irrigation based on the above information, and issues a voice command through the intelligent handheld terminal 2: Water the A3 area.

[0064] 5. Voice command processing and irrigation instruction generation; The intelligent agricultural irrigation server 1 receives the audio stream sent by the terminal, and first performs preprocessing (noise reduction, normalization); Convert the voice to text using the Conformer end-to-end ASR model to get Water the A3 area; Use the BERTCRF model to extract key elements, find that the irrigation water quantity and time length are missing, and determine that the elements are incomplete; The server calls the irrigation action suggestion data to automatically complete the missing elements (water quantity 8m 3 / acre, time length 20 minutes), and sends confirmation information to the intelligent handheld terminal 2: Confirm the irrigation of the A3 area, water quantity 8m 3 / acre, time length 20 minutes? The manager confirms through voice, and the server generates standardized irrigation instruction data (irrigation area: A3 area; water quantity: 8m 3 / acre; time length: 20 minutes; start time: immediately).

[0065] 6. Irrigation execution and feedback: The smart agriculture irrigation server 1 sends an irrigation instruction to the A3 area irrigation device 6, and the device starts the intelligent electromagnetic valve after receiving the instruction, and starts irrigation according to the set flow rate. The flowmeter records the water quantity in real time; During the irrigation process, the irrigation action suggestion data layer of the digital twin model is updated synchronously A3 area irrigation state (displaying irrigation and remaining time); After 20 minutes, the irrigation device 6 automatically closes, and the irrigation is completed. The actual water quantity is 8.2m 3 / acre is fed back to the server, and the server updates the soil monitoring data layer (profile color deepening, water content rising) and the crop growth visual data layer (NDVI value gradually recovering) of A3 area in the digital twin model.

[0066] Three, special scene processing; Instruction completion and confirmation: If the management personnel's voice instruction is to irrigate A5 area for 10 minutes (lack of water quantity), the server supplements the water quantity 5m 3 / acre based on the suggestion data and requests confirmation, and executes after the management personnel confirms; Automatic execution trigger: If the server generates an irrigation suggestion (such as A7 area needs emergency irrigation, otherwise it will affect the jointing), and the management personnel does not issue an instruction within 30 minutes after checking, the server automatically sends an instruction to the A7 area irrigation device to execute irrigation; Sound and light alarm: When the crop growth visual data layer identifies that there are aphid aggregations in A2 area (pest and disease classification level 3, which needs to be intervened), the server controls the sound and light alarm device 7 near A2 area to start (red light flashing + bee buzzing), and at the same time marks the alarm position in the digital twin model.

[0067] Through the above implementation, the farm realizes the data visualized analysis, voice convenient control, precise execution and closed-loop management of wheat irrigation, and significantly improves the irrigation efficiency and water resource utilization rate.

[0068] The above are preferred embodiments of the present application, not to limit the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. The intelligent agricultural irrigation visualization control method based on AI voice control is characterized by: It includes the monitoring system construction stage and the AI ​​voice-based smart irrigation control stage; The monitoring system construction phase includes the following steps: Step 1: Build a digital twin model of the farmland to be irrigated, and set the digital twin layer in the digital twin model; Step 2: setting up ground monitoring points, soil monitoring points, crop growth monitoring points, and irrigation points on the farmland to be irrigated, setting up an environmental data monitoring device (3) at the ground monitoring point, setting up a soil monitoring device (4) at the soil monitoring point, setting up a crop growth status monitoring device (5) at the crop growth monitoring point, and setting up an irrigation device (6) at the irrigation point; Step 3, the smart agricultural irrigation server (1) collects multimodal data, analyzes the multimodal data to generate digital twin layer data and irrigation action suggestion data, and maps them to the digital twin layer of the digital twin model; The AI ​​voice-based smart irrigation control stage includes the following steps: Step 4: The manager analyzes the digital twin layer data, makes irrigation action decisions, and uses voice commands to describe the irrigation action instructions; Step 5, the smart agricultural irrigation server (1) generates irrigation instruction data based on the AI ​​model recognition voice instruction, and sends the irrigation instruction data to the corresponding irrigation device (6); Step 6: The irrigation device (6) executes the irrigation instruction data to complete the irrigation.

2. The method for visualizing smart agricultural irrigation based on AI voice control according to claim 1 is characterized in that: In step 1, the digital twin layer includes the environmental data layer, the soil monitoring data layer, the crop growth visual data layer and the irrigation action suggestion data layer; in step 4, the digital twin layer data includes environmental data, soil monitoring data, crop growth visual data and irrigation action suggestion data. The environmental data is mapped to the environmental data layer, the soil monitoring data is mapped to the soil monitoring data layer, the crop growth visual data is mapped to the crop growth visual data layer, and the irrigation action suggestion data is mapped to the irrigation action suggestion data layer.

3. The method for visualizing smart agricultural irrigation based on AI voice control according to claim 2 is characterized in that: In step 3, the multimodal data includes environmental data, soil physical indicator data, soil chemical indicator data, soil biological indicator data, crop growth visual data, and irrigation action data.

4. The method for visualizing smart agricultural irrigation based on AI voice control according to claim 3 is characterized in that: Step 3 includes the following sub-steps: Step 31, environmental data analysis: calculate reference crop evapotranspiration and predict future weather at set time intervals; Step 32, calculating the effective soil water content based on the soil data; Step 33: Analyze crop growth visual data using a deep learning model and output growth data analysis results, which include plant height estimation, leaf area index inversion, canopy cover calculation, phenological period identification, and automatic identification and classification of pest and disease symptoms. Step 34: training an irrigation action decision model based on historical irrigation data, inputting environmental data analysis results, soil data analysis results, and crop growth visual data analysis results into the irrigation action decision model, and outputting current irrigation action recommendation data; Step 35, mapping the environmental data analysis results to the environmental data layer, mapping the soil data analysis results to the soil monitoring data layer, mapping the growth data analysis results to the crop growth visual data layer, and mapping the current irrigation action suggestion data to the irrigation action suggestion data layer.

5. The method for visualizing smart agricultural irrigation based on AI voice control according to claim 4 is characterized in that: In step 34, irrigation data records within the past set time are collected. The irrigation data records include irrigation time data, duration data, water volume data, and corresponding irrigation points. The environmental analysis data, soil data analysis results, and crop growth data analysis results of the corresponding irrigation points during the same period are associated to form a training sample set. The training sample set is input into the temporal convolutional network (TCN). After the training is completed, an irrigation action decision model is obtained. The input of the irrigation action decision model is the effective soil water content, the reference crop evapotranspiration in the future set time period, the weather in the future set time interval, the crop water stress index, and the growth data analysis results. The output of the irrigation action decision model is the irrigation area, irrigation water volume, and irrigation time.

6. The method for visualizing smart agricultural irrigation based on AI voice control according to any one of claims 1 to 5, characterized in that: Step 5 includes the following specific steps: Step 51, the manager sends a voice command to the smart agricultural irrigation server (1) via the smart handheld terminal (2); Step 52, the smart agricultural irrigation server (1) receives the audio stream and performs pre-processing; Step 53: Use the Conformer end-to-end ASR model to perform speech-to-text conversion. Step 54: extract key elements of the voice command using the BERTCRF model; Step 55: determine whether the key elements of the voice command are complete. If they are complete, parse the irrigation target, irrigation water volume, and irrigation time to generate irrigation command data.

7. The method for visualizing smart agricultural irrigation based on AI voice control according to claim 6 is characterized in that: In step 55, if it is determined that the key elements of the voice instruction are incomplete, the incomplete key elements of the instruction are extracted, the incomplete key elements of the instruction are completed using the irrigation action suggestion data, and the completed irrigation action suggestion data are sent again to the smart handheld terminal (2) to request the administrator to confirm. If confirmed, step 6 is executed.

8. The method for visualizing smart agricultural irrigation based on AI voice control according to claim 7 is characterized in that: If the irrigation action instruction is not executed by the administrator after the set time threshold, the smart agricultural irrigation server (1) automatically executes the irrigation instruction data.

9. The intelligent agricultural irrigation visualization control system based on AI voice control is characterized by: The method for realizing the AI ​​voice-controlled smart agricultural irrigation visualization control method according to any one of claims 1 to 8, wherein the smart agricultural irrigation visualization control system comprises a smart agricultural irrigation server (1), a smart handheld terminal (2), an environmental data monitoring device (3), a soil monitoring device (4), a growth status monitoring device (5) and an irrigation device (6), wherein the environmental data monitoring device (3), the soil monitoring device (4), the growth status monitoring device (5) and the irrigation device (6) are respectively installed at ground monitoring points, soil monitoring points, crop growth monitoring points and irrigation points set up in the farmland to be irrigated, and are respectively wirelessly connected to the smart agricultural irrigation server (1) via a wireless network, and the smart handheld terminal (2) is wirelessly connected to the smart agricultural irrigation server (1) via a wireless network.

10. The AI ​​voice-controlled smart agricultural irrigation visualization control system according to claim 9 is characterized in that: It also includes an audible and visual alarm device (7), and the smart agricultural irrigation server (1) controls the execution of the audible and visual alarm device (7).

Images