Digitalized planting management method for rice

Abstract

The invention discloses a rice digital planting management method, which belongs to the technical field of intelligent agriculture, and mainly comprises the following steps: (1) constructing a multi-modal data acquisition layer, and forming a three-dimensional monitoring system through more than or equal to 10 soil moisture content sensor networks per mu, a 5nm-resolution unmanned aerial vehicle multi-spectral imaging device and a high-precision meteorological station; (2) establishing rice growth digital twins based on a 256-unit LSTM neural network, wherein the space-time alignment error is less than 5 minutes; (3) generating an optimization decision instruction, and when the LAI predicted value is lower than 3.2 of hybrid rice or 2.8 of japonica rice, triggering variable topdressing with the precision of + / -1.5%; and (4) performing closed-loop feedback correction, and dynamically updating the model through 20 million-pixel unmanned aerial vehicle image inversion LAI (error + / -0.15). The method is particularly suitable for the cloudy mountain environment in Guizhou, and tests show that compared with traditional planting, the nitrogen fertilizer utilization rate is increased by 46.9%, the early warning lead of rice planthoppers reaches 9.3 days, the communication networking packet loss rate is only 0.7%, and the proportion of high-quality rice is increased by 26% through block chain traceability.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, specifically a digital rice planting and management method. Background Technology

[0002] Currently, rice cultivation relies heavily on experience-based decision-making, resulting in problems such as low fertilizer and water utilization rates (nitrogen fertilizer utilization rate in Chinese paddy fields is less than 40%), delayed early warning of pests and diseases (loss rates reach 20-30% after rice blast outbreaks), and large yield fluctuations. Manual monitoring costs account for more than 35% of total production costs, and data collection accuracy is limited by the density of field sensor deployment (conventional methods require 5 mu / node).

[0003] Existing technologies suffer from the following problems: low fertilizer and water utilization rate: the nitrogen fertilizer utilization rate in paddy fields in my country is generally less than 40%, and excessive or insufficient fertilization leads to resource waste and environmental pollution.

[0004] Delayed early warning of diseases and pests: Relying on manual inspections, the loss rate after the outbreak of diseases such as rice blast can be as high as 20%-30%, and the control cost is high and the effect is limited.

[0005] Insufficient data acquisition accuracy: The existing field sensor deployment density is low (e.g., 5 mu / node), making it difficult to achieve accurate monitoring. In addition, the satellite remote sensing cycle in cloudy areas is long (7-10 days / time), resulting in poor data timeliness.

[0006] This method solves the core problems of agricultural big data being "unable to be stored, processed, or used effectively" by establishing a digital twin of rice growth and integrating edge computing (field gateways processing image data in real time within 200ms) with a deep learning LSTM model. Summary of the Invention

[0007] The purpose of this invention is to overcome the aforementioned technical difficulties and provide a digital rice planting and management method.

[0008] To achieve the above objectives, the technical solution adopted is: a digital rice planting and management method, comprising the following steps: S1. Construct a multimodal data acquisition layer: Deploy a soil moisture sensor network with a node density of ≥10 per acre and a sampling frequency of 1 time per 10 minutes; configure a UAV multispectral imaging system with a spectral range of 400-2500nm and a resolution of ≤5nm; install a field weather station to monitor the canopy microclimate with an accuracy of ±0.5℃ for temperature and ±3%RH for humidity. S2. Establish a digital twin of rice growth: Construct an organ-level growth model based on LSTM neural network with 256 hidden layers and Dropout=0.2; develop a multi-source data fusion algorithm with a spatiotemporal alignment error of <5 minutes. S3. Generate optimized decision instructions: When the predicted leaf area index (LAI) is lower than the threshold (3.2 for hybrid rice and 2.8 for japonica rice): trigger the topdressing machine, servo motor control, flow error ±1.5%, and adjust the irrigation amount based on the probability of precipitation in the next 72 hours (CMA model of China Meteorological Administration). S4. Closed-loop feedback correction: Using RGB images from drones, the actual LAI value is retrieved at 20 megapixels with an error of ±0.15. The model parameters are dynamically updated with a learning rate of 0.001 and a batch size of 32.

[0009] Furthermore, the early warning model for rice planthopper infestation using the digital twin in S2 includes: Input features: insect voiceprint energy value in the 20-100kHz frequency band, diurnal temperature range of the canopy, and relative humidity; Model structure: 1D-CNN with 3 convolutional layers + LSTM with 128 units Performance metrics: Early identification accuracy 87.5%, n=1520 sets of data, early warning lead time 9.3±1.2 days.

[0010] Furthermore, the variable fertilization control in S3 is optimized using reinforcement learning: State space: soil N content, chlorophyll SPAD value, accumulated temperature; Operational space: Urea application rate (0-15g / m²); Reward function: Forecast yield - Nitrogen fertilizer consumption × 0.3; Training results: Under the same yield target, nitrogen fertilizer usage was reduced by 18.3±2.1% compared with the traditional model (p<0.01, t test).

[0011] Furthermore, the data acquisition layer adopts LoRaWAN networking: Communication parameters: SF=10, BW=125kHz, CR=4 / 5 Actual performance: Data packet loss rate of 0.7±0.3% within a 1km² range, and average power consumption of nodes of 28μA.

[0012] Furthermore, it also includes a quality traceability module: blockchain-based evidence storage for fertilization records (time, amount, operator) and pesticide use (type, concentration). Data correlation: The correlation coefficient between rice taste value and the average daily temperature difference during the heading period was r=0.82 (p<0.001).

[0013] The beneficial effects of adopting the above scheme are as follows: After the application of this digital rice planting and management method, the following improvements were achieved: 1. A qualitative improvement in the level of precision agriculture: Through a high-density sensor network of 10 sensors per mu + 5nm resolution multispectral imaging, the spatial resolution of soil moisture monitoring was improved by 8 times compared with the traditional method (measured at 0.1 mu / unit vs. conventional 0.8 mu / unit); nitrogen fertilizer utilization rate reached 58.3%, which is 46.9% higher than that of traditional planting in Guizhou (39.7%); and water use efficiency was improved by 35%.

[0014] 2. Revolutionary breakthrough in pest and disease control: Based on the 1D-CNN 3-layer convolution + LSTM 128-unit model, the early warning time for rice planthoppers reaches 9.3±1.2 days, which is 86% higher than the conventional forecast (3-5 days) of agricultural technology stations; pesticide application is reduced by 2-3 times per season, and the control cost is reduced by 42 yuan / mu.

[0015] 3. Systematic optimization of resource consumption: Reinforcement learning-driven variable fertilization, under the same yield target (7850±210kg / ha): nitrogen fertilizer application was reduced by 18.3% (180→147kg / ha), and nitrogen fertilizer partial productivity increased from 43.6 to 53.9kg / kg (+23.6%). Combined with irrigation strategies based on CMA precipitation forecasts, the water saving rate reached 28.7%.

[0016] 4. Significant improvements in mountain adaptability: LoRaWAN networking (SF=10): Achieved in typical hilly terrain in Guizhou: 98.2% communication success rate within 1km², a 72% improvement over the Zigbee solution (57%), node power consumption reduced to 28μA, and battery life extended to 3 years (compared to the traditional 6 months).

[0017] 5. The added value of agricultural products has increased significantly. The blockchain traceability system shows a strong correlation between the taste value of rice and the temperature difference during the heading stage (r=0.82): the proportion of high-quality rice has increased from 52% to 78%, and the brand premium has reached 2.3 yuan / kg.

[0018] 6. Regionally Adaptive Innovations for Guizhou's Unique Environment: Spatiotemporal alignment error of data fusion algorithm is less than 5 minutes under cloudy weather (compared to more than 15 minutes under normal conditions), LAI threshold for plateau japonica rice is optimized to 2.5, lodging rate is reduced to 2.1% (compared to 11.5% under traditional methods), and packet loss rate of terraced terrain networking is controlled at 1.2% (23.5% under 4G scheme). Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments. The described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Implementation location: Meitan County, Zunyi City, Guizhou Province (107.5°E, 27.8°N) Experimental varieties: Yixiangyou 2115 (hybrid rice), Qianjingyou 1 (japonica rice) Trial period: April-October 2023 (full reproductive period) 1. Construction of Multimodal Data Acquisition Layer Soil moisture monitoring: YG-SM300 sensors (operating frequency band 868MHz, LoRaWAN protocol) manufactured by Guizhou Yaguang Electronic Technology Co., Ltd. were used, deployed at a rate of 12 nodes per acre. Actual measurement data: Sampling accuracy: Volumetric moisture content ±1.8% (compared to drying method) Network performance: With SF=10, the communication success rate in the hilly terrain of Meitan is 98.2% (a 41% improvement compared to traditional Zigbee networking). Multispectral of UAVs: The DJI P4 Multispectral, equipped with a custom-designed 2100nm narrowband filter (4.3nm bandwidth) from Guizhou University, performs as follows at a flight altitude of 50m: NDVI inversion accuracy: R²=0.91 (compared with ASD FieldSpec4 actual measurement) Pest identification: Rice leaf roller detection rate 89% (manual verification) 2. Establishment of a digital twin Growth model training: Using nearly 5 years of data (1-hour time resolution) provided by the Guizhou Meteorological Bureau, the LSTM model was trained in the TensorFlow 2.8 framework: Tillering stage prediction error: 3.2 stems / clump (comparison based on manual counting) Rice planthopper early warning: The actual measured advance warning rate in Fenggang County was 8.7 days (accuracy rate 85.3%, n=326). Data fusion algorithm: The developed spatiotemporal alignment algorithm was tested under cloudy conditions in Guizhou: Image-sensor data alignment error: 3.8 minutes (better than the 7.2 minutes reported in published literature) Variable fertilization comparative experiment: Group Nitrogen fertilizer application rate (kg / ha) Yield (kg / ha) Nitrogen fertilizer has a partial productivity effect Traditional model 180 7850 43.6 This invention 147 7920 53.9 (Data source: Guizhou Academy of Agricultural Sciences 2023 Yield Measurement Report, p<0.05) 4. Application of quality traceability (verification of claim 5) Blockchain-based evidence storage: Data from the Zunyi experimental field was recorded using Guizhou's blockchain "Xianglian" platform. Fertilization record upload delay: 2.3 ± 0.7 seconds Taste score analysis: For every 1℃ increase in temperature difference during the heading stage, the taste score decreases by 0.82 points (verification r=0.79). Example 2 Based on Example 1, the applicant also conducted the following experiments. Experimental design: A three-block randomized controlled trial (RCBD) was adopted, with three replicates per group and an area of ​​3 mu per block.

[0021] Group: Invention Group (T1): Complete technical solutions according to claims 1-5 Partial digitization group (T2): Data acquisition only as claimed in claim 1 Conventional planting group (CK): conventional planting The following comparative experiments were conducted: (1) System performance comparison index Group T1 Group T2 CK group Data collection completeness rate 98.7±0.5% 91.2±3.1% - Pest warning advance 9.1 ± 0.8 days No warning No warning LoRa node power consumption 26μA 30μA - (2) Comparison of agronomic effects parameter Group T1 Group T2 CK group Nitrogen fertilizer utilization rate 58.3%▲ 42.1% 39.7% lodging incidence 2.1%▼ 8.7% 11.5% Brown rice protein content 7.2g / 100g■ 6.8g / 100g 6.5g / 100g (▲47.1% higher than CK, ▼82.3% lower than CK, ■taste score increased by 0.5 points, p<0.05) Example 3 Mountain-specific verification Group T1': Add the SF=12 configuration of claim 4 to verify the optimization of mountain communication.

[0022] Group CK: Local farmers use traditional terraced fields, which demonstrate the advantages of adapting to the terrain.

[0023] Test location: Wanfenglin area, Xingyi City, Qianxinan Prefecture.

[0024] Key comparative data: Communication performance: T1' group packet loss rate 1.2% (CK group uses 4G transmission, packet loss rate 23.5%), economic benefits: .

[0025] Example 3 Special verification of pest early warning: An artificial insect inoculation control was set up in Rongjiang County (a high-incidence area of ​​rice planthoppers). deal with Early warning methods Prevention and control effect This invention group CNN-LSTM of claim 2 The accuracy rate of timely prevention and control was 89.2%. control group Agricultural technology station routine forecast The accuracy rate of timely prevention and control was 61.4%. Blank control No prevention Production decreased by 47.3%.

[0026] This invention reduces pesticide use by 2 times per season compared to conventional forecasts (data from Guizhou Plant Protection Station). Example 4 Guizhou Mountain Adaptability Verification Location: Dafang County, Bijie City (altitude 1450m) Special improvements: The sensor network adds a terrain correction module, increasing the node density to 15 nodes per acre when the slope is greater than 25°. The digital twin incorporates growth parameters specific to highland japonica rice (e.g., adjusting the LAI threshold to 2.5). Verification results: Yield increased by 17.2% compared to traditional planting (control field yield data). Frost warning accuracy rate: 91% (actual test in September 2023) It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

[0027] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A digital rice planting management method, characterized by: Includes the following steps: S1. Construct a multimodal data acquisition layer: Deploy a soil moisture sensor network with a node density of ≥10 per acre and a sampling frequency of 1 time per 10 minutes; configure a UAV multispectral imaging system with a spectral range of 400-2500nm and a resolution of ≤5nm; install a field weather station to monitor the canopy microclimate with an accuracy of ±0.5℃ for temperature and ±3%RH for humidity. S2. Establish a digital twin of rice growth: Construct an organ-level growth model based on LSTM neural network with 256 hidden layers and Dropout=0.2; develop a multi-source data fusion algorithm with a spatiotemporal alignment error of <5 minutes. S3. Generate optimized decision instructions: When the predicted leaf area index (LAI) is lower than the threshold (3.2 for hybrid rice and 2.8 for japonica rice): trigger the topdressing machine, servo motor control, flow error ±1.5%, and adjust the irrigation amount based on the probability of precipitation in the next 72 hours (CMA model of China Meteorological Administration). S4. Closed-loop feedback correction: Using RGB images from the drone, the actual LAI value is retrieved at 20 megapixels with an error of ±0.15, and the model parameters and learning rate are dynamically updated. 0.001, batch size 32.

2. The digital rice planting management method according to claim 1, wherein the early warning model for rice planthopper pests using a digital twin in step S2 comprises: Input features: insect voiceprint energy value in the 20-100kHz frequency band, diurnal temperature range of the canopy, and relative humidity; Model structure: 1D-CNN with 3 convolutional layers + LSTM with 128 units Performance metrics: Early identification accuracy 87.5%, n=1520 sets of data, early warning lead time 9.3±1.2 days.

3. The digital rice planting management method according to claim 1, characterized in that: The variable fertilization control in S3 is optimized using reinforcement learning: State space: soil N content, chlorophyll SPAD value, accumulated temperature; Operational space: Urea application rate (0-15g / m²); Reward function: Forecast yield - Nitrogen fertilizer consumption × 0.3; Training results: Under the same yield target, nitrogen fertilizer usage was reduced by 18.3±2.1% compared with the traditional model (p<0.01, t test).

4. The digital rice planting management method according to claim 1, characterized in that: The data acquisition layer uses LoRaWAN networking: Communication parameters: SF=10, BW=125kHz, CR=4 / 5 Actual performance: Data packet loss rate of 0.7±0.3% within a 1km² range, and average power consumption of nodes of 28μA.

5. The digital rice planting management method according to claim 1, characterized in that: It also includes a quality traceability module: blockchain-based evidence storage of fertilization records (time, amount, operator) and pesticide use (type, concentration). Data correlation: The correlation coefficient between rice taste value and the average daily temperature difference during the heading period was r=0.82 (p<0.001).