Fusion of farmland environment perception, high-precision positioning and navigation of agricultural machinery and operation planning method

Abstract

The application discloses a kind of fusion farmland environment perception's agricultural machinery high-precision positioning navigation and operation planning method, belong to wisdom agriculture field.For the positioning defects caused by farmland shelter, sludge, electromagnetic interference, construct "Beidou No.3 dual-band+UWB+soil humidity sensor" multi-source fusion system, match positioning accuracy and operation demand with crop leaf area index (LAI) dynamic.The data of Beidou B1C / B2a terminal and TOF principle UWB base station carried by agricultural machinery are collected, and the UWB stake point calibrated by GDOP optimization is corrected with TDR soil humidity sensor drift simultaneously;Improved Kalman filter is used to fuse data, and improved A* algorithm is used to generate optimal path, and ARM architecture edge node realizes low delay processing within 50ms.The positioning error of the application is less than or equal to 3cm, the operation efficiency is improved by 20%, the NLOS scene stability is improved by 40%, and the application is suitable for accurate seeding, variable irrigation, and meets the needs of wisdom agriculture.

Description

Technical Field

[0001] This patent relates to the intersection of smart agriculture and agricultural machinery engineering, specifically to agricultural machinery positioning and navigation technology based on multi-source sensor fusion and operation path planning technology combined with crop growth models. It is applied to scenarios such as automatic agricultural machinery operation and precision farming, and falls under the category of modern agricultural technologies that are key national support. Background Technology

[0002] 1. Describe the existing technology With the development of smart agriculture, automated and precision agricultural machinery operations have become key to improving agricultural production efficiency, and high-precision positioning and navigation are the core supporting technologies for achieving this goal. Currently, agricultural machinery positioning and navigation technologies mainly rely on satellite positioning systems (such as BeiDou and GPS), inertial navigation systems, and single short-range positioning technologies (such as UWB and Bluetooth). Among them, the BeiDou-3 global satellite navigation system has achieved global service, with positioning accuracy of 1-5 meters in open environments, which can be improved to centimeter level through differential technology; UWB (ultra-wideband) technology, due to its anti-multipath interference and high positioning accuracy, has been widely used in indoor and short-range scenarios, with positioning accuracy within 10cm; inertial navigation can maintain positioning for a short period when satellite signals are interrupted, but it suffers from drift accumulation problems.

[0003] Existing agricultural machinery positioning solutions include some that combine BeiDou navigation with inertial navigation to improve adaptability to complex scenarios. For example, Beijing UniStrong's "Huinong" BeiDou navigation-based agricultural machinery autopilot system achieves precise control through the fusion of BeiDou / GNSS high-precision positioning and MEMS sensors. Other solutions attempt to introduce short-range positioning technology to optimize accuracy. For instance, Wuhan Yixun Company integrates multi-source sensing technology with BeiDou positioning to improve the stability of agricultural machinery operations. These solutions have already demonstrated value in open plains or simple operational scenarios. Meanwhile, research on the correlation between crop growth cycles and operational needs is gradually underway, such as optimizing fertilization timing using crop growth period data, but this has not yet been combined with dynamic matching of positioning accuracy.

[0004] 2. Shortcomings of existing technologies Existing technologies have significant limitations in complex farmland environments: First, positioning stability is insufficient. Tall crops in farmland can cause BeiDou signal interruptions, leading to a sharp increase in single positioning errors exceeding 10 meters. Slippage of agricultural machinery in muddy conditions exacerbates inertial navigation drift, and existing solutions lack a grounding status judgment mechanism based on soil parameters. Electromagnetic interference in farmland affects UWB positioning accuracy, and the layout of monitoring points is not optimized for farmland scenarios. Second, the accuracy does not match the requirements. Existing solutions use a fixed accuracy mode and do not establish a dynamic correlation between crop growth parameters (such as LAI) and positioning accuracy, resulting in insufficient accuracy during sowing or wasted computing power during irrigation. Third, data processing latency is high. Reliance on cloud processing leads to delayed navigation commands, and edge devices lack low-power heterogeneous architecture design and secure encryption mechanisms. Fourth, UWB positioning does not incorporate GDOP to optimize monitoring point spacing, resulting in insufficient stability of the positioning reference network, and it lacks NLOS error identification and suppression methods.

[0005] 3. The technical problem to be solved by the present invention In summary, existing agricultural machinery positioning and navigation technologies cannot adapt to the complex environments of farmland, such as obstruction, mud, and electromagnetic interference. They suffer from problems such as low positioning accuracy, poor stability, mismatch between accuracy and operational requirements, high data processing latency, and unreasonable UWB reference network design. This invention aims to provide a high-precision positioning, navigation, and operation planning method for agricultural machinery that integrates farmland environmental perception. By employing BeiDou dual-band selection, UWB-GDOP stakepoint optimization, improved Kalman filter fusion, LAI dynamic accuracy matching, and low-power edge computing technology, this method addresses the aforementioned technical deficiencies, achieving high-precision, highly stable positioning and efficient operation planning in complex farmland environments. Summary of the Invention

[0006] 1. Detailed description of the technical solution of the present invention The technical solution of this invention comprises four parts: a positioning and sensing system deployment, a multi-source data fusion algorithm, a dynamic job planning model, and an edge data processing module. The technical details and parameters of each part are as follows: Deployment of the Positioning and Sensing System: This system uses "satellite wide-area positioning + short-range high-precision calibration + environmental state perception" as its core architecture, overcoming the performance bottleneck of single sensors in complex farmland environments. ① The BeiDou positioning unit selects a BeiDou-3 B1C / B2a dual-band receiver module. The core principle of prioritizing the dual-band design is that the B1C band has anti-narrowband interference capability, and the B2a band can reduce the impact of ionospheric delay. Multipath suppression (MPC) technology weakens the interference of crop canopy reflection signals. The positioning update frequency is set to 10Hz, and the initial positioning accuracy is ≤1m using ephemeris differential data. ② The UWB positioning unit uses the Time-of-Flight (TOF) method to achieve short-range high-precision positioning. Fixed calibration stakes are deployed at the farmland boundary and inside to form a positioning reference network. The stakes are made of C30 concrete with a burial depth ≥0.8m to ensure stability under complex working conditions. The stake spacing is optimized based on positioning accuracy requirements and geometric dilution factor (GDOP) and set to 50-80m. The boundary stakes are ≥2m away from the field ridge to avoid collisions with agricultural machinery. The time synchronization error between the UWB base station and the pile tag is controlled within ≤1ns. Anti-electromagnetic interference capability is improved through bandwidth extension technology (500MHz-1GHz), achieving a bare positioning accuracy within 10cm. ③ The soil moisture sensing unit is based on the time domain reflectance (TDR) principle. A probe is inserted 3cm into the soil to collect the dielectric constant signal. The moisture content is inverted through the mapping relationship between the dielectric constant and soil moisture. The acquisition frequency is 5Hz, and the measurement accuracy is ±1%vol. Its core function is to determine the grounding status of agricultural machinery through sudden changes in soil moisture, providing environmental basis for positioning drift correction.

[0007] Multi-source data fusion algorithm: An adaptive fusion mechanism of "error complementarity + state feedback" is constructed. The core adopts an improved Kalman filter algorithm to realize the dynamic fusion of Beidou and UWB data. At the same time, soil moisture signal is introduced as a drift correction trigger condition. The specific principle and formula are as follows: (1) Construction of state equations: using the plane coordinates (x, y) and velocity (v) of the agricultural machinery. x The state vector X consists of the state vector X and the positioning error correction (Δx, Δy). k Establish the discrete-time state equation:

[0008] Where A is the state transition matrix, representing the continuity of the agricultural machinery's motion; B is the control matrix, u k For agricultural machinery control parameters such as steering angle and speed; w k The noise is Gaussian white noise, and its variance is determined by the variance of the agricultural machinery's acceleration. When the agricultural machinery is moving at a constant velocity in a straight line, the state transition matrix can be simplified to:

[0009] (T is the location update period, with a value of 0.1s) (2) Construction of observation equations: The BeiDou positioning results (x_GNSS, y_GNSS) and UWB positioning results (x_UWB, y_UWB) are used as the observation values ​​Z. k Establish the observation equation:

[0010] Where H is the observation matrix, v k To observe noise, the BeiDou noise variance is dynamically adjusted by the signal-to-noise ratio (SNR), while the UWB noise variance is calculated using the signal flight time variance.

[0011] (3) Adaptive filter update: Introduce a fuzzy inference system to dynamically adjust the filter gain K k When the BeiDou SNR is ≥ 35dB, its confidence weight is increased; when the SNR is < 25dB, the UWB data weight is automatically increased. The filter update formula is:

[0012] (P is the covariance matrix, and R is the observation noise matrix) (4) Drift correction trigger: The soil moisture change rate λ is defined as the ratio of the moisture difference between adjacent moments to time. When λ > 15%, it is determined that the agricultural machinery is at risk of slippage, triggering UWB pile point calibration, and the cumulative drift is corrected by the three-point positioning method.

[0013] (x_{UWB,i} represents the UWB positioning results of the three surrounding stake points, and σ_i represents the corresponding positioning error) The fusion algorithm is further optimized through the NLOS (non-line-of-sight) error identification module. It uses a signal power attenuation model to determine the propagation path of the UWB signal and applies mean drift suppression processing to the NLOS signal to ensure fusion accuracy in complex occlusion environments.

[0014] Dynamic operation planning model: Based on the collaborative logic of "agronomic needs-precision matching-path optimization", a dynamic planning system integrating crop growth model and terrain features is constructed to overcome the resource waste problem of traditional fixed-precision planning.

[0015] (1) Growth cycle-accuracy correlation model: Crop leaf area index (LAI) is introduced as a key parameter to establish a quantitative relationship between positioning accuracy requirement P and LAI. The correlation formula is obtained by training through the crop growth database of the Ministry of Agriculture and Rural Affairs:

[0016] The LAI (Local Area Imagery) was acquired through inversion from UAV multispectral imagery and calibrated using field sampling data, achieving an inversion accuracy of R² ≥ 0.92. Centimeter-level accuracy was used during the sowing period to ensure uniform plant spacing (coefficient of variation ≤ 5%), while decimeter-level accuracy was used during the irrigation period to reduce computational power consumption, achieving a dynamic balance between accuracy and energy consumption.

[0017] (2) Improved A* path optimization algorithm: For obstacles such as irrigation ditches and field ridges in farmland, the heuristic function of the traditional A* algorithm is optimized by introducing a terrain cost coefficient C (C=10 when elevation change is >10cm, otherwise C=1), and a new evaluation function is constructed:

[0018] Where g(n) is the actual cost from the starting point to node n, h(n) is the estimated cost from node n to the ending point (calculated using Manhattan distance), ω is the heuristic weight (dynamically adjusted from 1.2 to 1.8), and C(n) is the terrain cost of node n. The algorithm automatically identifies obstacle areas and generates optimal paths by rasterizing the farmland digital elevation model (DEM, 0.1m resolution). The path spacing is adaptively adjusted according to the crop row spacing (0.3-1.2m), and the path generation time is ≤10s, representing a 30% improvement in efficiency compared to traditional algorithms.

[0019] (3) Dynamic allocation of tasks: Combining the agricultural machinery operation radius and energy consumption model, a greedy algorithm is used to realize the path allocation of multi-agricultural machinery collaborative operation, avoiding task overlap and omission, and the operation coverage rate reaches 100%.

[0020] - Edge data processing module: It adopts a hybrid architecture of "edge computing + cloud calibration" to solve the technical pain points of high network latency and unstable signal in farmland, which is in line with the cutting-edge trend of smart agriculture data processing.

[0021] (1) Edge node hardware architecture: The low-power computing unit is built based on the ARM architecture, integrating CPU and GPU heterogeneous computing modules, with computing power ≥200GFLOPS and power consumption ≤10W, and is compatible with the 12V power supply system of agricultural machinery. The hardware adopts a vibration-resistant design to meet the needs of bumpy farmland conditions and supports multi-interface data access (CAN, Ethernet, 4G / 5G).

[0022] (2) Data processing flow: The edge node completes the entire process of "data preprocessing - fusion calculation - path planning - control command output" locally, with a processing delay of ≤50ms. In the data preprocessing stage, a 5-point moving average filter is used to eliminate soil moisture signal noise. In the fusion calculation stage, an improved Kalman filter algorithm is run. In the path planning stage, an improved A* algorithm is called. Finally, steering and throttle control commands are output to the agricultural machinery actuators through the CAN bus.

[0023] (3) Edge-Cloud Collaboration Mechanism: Drawing on the data hub technology of Weichai Lovol Smart Agriculture, when the farmland network is available (4G / 5G signal strength ≥ -90dBm), the edge nodes upload operation data (location trajectory, operation area, energy consumption) to the cloud platform. The cloud uses big data analysis to calibrate algorithm parameters and distribute them to the edge nodes. When the network is interrupted, the edge nodes maintain navigation function for ≥ 30 minutes based on locally cached farmland data (storage capacity ≥ 16GB), and automatically retransmit the data after the network is restored. Data transmission adopts the lightweight MQTT protocol to reduce network bandwidth requirements, and the transmission latency is ≤ 200ms.

[0024] (4) Safety and reliability design: The edge nodes have data encryption function (AES-128 encryption) to prevent the leakage of operation data. At the same time, a watchdog timer is used to realize fault self-recovery, improve system reliability, and the mean time between failures (MTBF) is ≥2000 hours.

[0025] 2. Advantages of the technical solution of the present invention compared with the prior art. ① Significantly improved positioning accuracy and stability: The dual-band design of BeiDou B1C / B2a improves anti-interference capability by 40%. Combined with the UWB-GDOP reference network and NLOS error suppression, the positioning error is ≤3cm in tall corn stalks and rainy weather, which is 70% more accurate than the existing "BeiDou + inertial navigation" solution (error ≥10cm). The UWB calibration mechanism triggered by soil moisture reduces positioning drift by 85% in slippery scenarios. ② Optimized resource utilization: Dynamic accuracy matching based on LAI reduces edge module power consumption by 35%, reduces computing power consumption during irrigation by 50%, and extends agricultural machinery endurance by 2.5 hours. ③ Improved operational efficiency: The improved A* algorithm shortens the path generation time by 12% compared to traditional algorithms, with a path generation time ≤10s (efficiency improvement of 30%). The 50ms low latency of edge computing enables a 95% operational continuity rate, which is 22% higher than the cloud solution, and the overall operational efficiency is improved by 20%. ④ Strong environmental adaptability and security: UWB500MHz-1GHz bandwidth extension technology enhances anti-electromagnetic interference capabilities, making it suitable for plains and hilly terrains; AES-128 encryption and fault self-recovery design ensure system MTBF≥2000 hours, improving data security.

[0026] 3. Key technical points and areas to be protected in this invention Key technical points: ① The fusion mechanism of BeiDou-3 B1C / B2a dual-band and UWB-TOF technology, and the UWB stake point optimization layout method based on GDOP; ② Improved Kalman filter algorithm state equation construction and fuzzy inference weight adjustment strategy, combined with the drift correction trigger logic of soil moisture change rate; ③ LAI-based "growth cycle-accuracy requirement" quantitative model, and improved A* path optimization algorithm with the introduction of terrain cost coefficient; ④ Hardware design of ARM heterogeneous architecture edge nodes, and the collaborative mechanism of "local processing + cloud calibration" and security encryption scheme.

[0027] The points to be protected are: ① the deployment structure of "BeiDou dual-band terminal + UWB base station + TDR soil sensor", including the C30 concrete burial depth of UWB piles, GDOP optimized spacing and other deployment specifications; ② the improved Kalman filter state / observation equation construction, NLOS error suppression and slippage correction strategy; ③ the correlation formula between LAI and positioning accuracy, and the improved terrain cost coefficient and evaluation function design of the A* algorithm; ④ the heterogeneous architecture of the edge computing module, data processing flow, network disconnection caching mechanism and AES-128 encryption implementation. 4. Detailed Implementation Taking precision sowing and irrigation operations using agricultural machinery in the winter wheat planting area of ​​the North China Plain as an example, the technical solution of this invention will be described in detail: (1) Deployment of experimental fields and equipment The experimental field is located in Zhaoxian County, Shijiazhuang City, Hebei Province, covering an area of ​​50 mu (approximately 3.3 hectares). It is a rectangular plot (200 meters long and 166.7 meters wide), with two north-south irrigation ditches (0.8 meters wide and 0.5 meters deep) and one east-west ridge (1 meter wide and 0.3 meters high). The crop is winter wheat, and the experimental period covers the sowing period (mid-October) and the irrigation period (late March of the following year, during the jointing stage).

[0029] The agricultural machinery used is the Dongfanghong LX904 wheeled tractor, equipped with the positioning and sensing system and edge processing module of this invention: a Beidou terminal (B1C / B2a dual-band) is installed in the center of the top of the cab; UWB base stations (500MHz-1GHz bandwidth) are installed on the left side of the front bumper (1 meter from the front of the vehicle, 1.2 meters from the ground) and the right side of the rear bumper (1 meter from the rear of the vehicle, 1.2 meters from the ground); a soil moisture sensor (TDR principle) is installed on the bracket inside the right rear drive wheel, with the probe facing the ground; eight UWB fixed stakes (C30 concrete, buried at a depth of 0.8m) are deployed in the farmland, including four boundary stakes (at the four corners of the plot) and four internal stakes (diamond distribution, GDOP optimized spacing of 60 meters); the UWB tag on the top of the stake and the frequency of the agricultural machinery base station are both set to 6.5GHz. The edge computing module is installed in the passenger position of the cab and is connected to the agricultural machinery control system via a CAN bus.

[0030] (2) Implementation of sowing operations (accuracy requirement ≤3cm) Step 1: Data import and parameter setting. Import winter wheat sowing period data (LAI=0.8, plant spacing 15cm, row spacing 20cm), experimental field DEM data (resolution 0.1m) and obstacle area coordinates through the edge module. Set the positioning accuracy threshold to 3cm. The UWB calibration trigger condition is soil moisture change rate λ>15%. Improve the A* algorithm heuristic weight ω=1.5.

[0031] Step 2: Initialize the positioning system. After the agricultural machinery is started, the Beidou terminal completes satellite search (number of satellites searched ≥ 8, signal-to-noise ratio ≥ 35dB), the UWB base station establishes communication with surrounding fixed piles (communication success rate ≥ 95%), the soil moisture sensor preheats for 5 minutes, and the initial soil moisture measurement is 22%vol.

[0032] Step 3: Path planning. The edge module runs the improved A* algorithm to generate sowing paths parallel to the field ridges, with a path spacing of 20cm, avoiding irrigation canal areas, generating 10 operation paths with a total length of 1850 meters, and the path generation time is 8.2 seconds.

[0033] Step 4: Operation and Positioning Correction. The agricultural machinery travels along the planned path. The Beidou terminal and UWB base station synchronously collect positioning data. The fusion algorithm dynamically adjusts the weights of the two types of data (Beidou confidence 0.6, UWB confidence 0.4) and outputs the positioning results. When the agricultural machinery travels near the irrigation canal (soil moisture rises to 38% vol, change rate 27%), the system activates UWB calibration, correcting the positioning based on the data from the two most recent stake points. After correction, the positioning error is reduced from 8 cm to 2.3 cm. After the sowing operation is completed, the edge module counts the operation area as 50 mu, with a missed sowing rate of 0.3% and a re-sowing rate of 0.2%.

[0034] (3) Implementation of irrigation operations (accuracy requirement ≤30cm) Step 1: Parameter adjustment. Import winter wheat jointing stage data (LAI=4.2), adjust the positioning accuracy threshold to 30cm, the path planning target is to align the irrigation nozzle with the gap between wheat rows, the path spacing is 60cm, and improve the terrain cost coefficient of the A* algorithm to C=1 (the experimental field has a gentle elevation).

[0035] Step 2: Positioning and Operation. The fusion algorithm reduces the weight of UWB data (BeiDou confidence 0.7, UWB confidence 0.3) to reduce computing power consumption. The edge module generates 5 irrigation paths with a total length of 1780 meters. During the operation, the farmland was hit by rainy weather, and the BeiDou signal-to-noise ratio dropped to 28dB. UWB data became the main positioning source, and the fusion positioning error stabilized at 22cm, meeting the irrigation requirements.

[0036] (4) Comparison of experimental data

[0037] The test results show that the present invention can meet the requirements of operation accuracy during both the sowing and irrigation periods, improve the operation efficiency by 20%-28% compared with the existing technology, reduce the power consumption of the edge module by more than 40%, and significantly reduce the missed reseeding rate, thus verifying the feasibility and superiority of the technical solution. Attached Figure Description

[0038] Figure 1 This is a structural block diagram of the agricultural machinery positioning and navigation system of the present invention. The present invention takes "perception-fusion-planning-control" as its core, emphasizing multi-source perception collaboration, dynamic accuracy matching and edge-cloud closed-loop optimization. The key technical points correspond one-to-one with the invention content.

Claims

1. A method for high-precision positioning, navigation, and operation planning of agricultural machinery that integrates farmland environmental perception, characterized in that, include: The deployment steps for the positioning and sensing module include: installing BeiDou-3 B1C / B2a dual-band positioning terminals, TOF-based UWB base stations, and TDR-type soil moisture sensors at key locations on the agricultural machinery; deploying several UWB fixed calibration points along the farmland boundary and within the farmland according to the GDOP optimization principle; the multi-source data fusion step involves acquiring anti-interference satellite positioning data through the BeiDou terminal, acquiring short-range high-precision positioning data through interaction between the UWB base station and fixed calibration points, collecting soil dielectric constant signals at the grounding point through the soil moisture sensor, fusing the three types of data based on an improved Kalman filter algorithm, and triggering slippage and drift correction; the dynamic operation planning step involves importing target farmland crop LAI data and DEM terrain data, determining positioning requirements based on the LAI-precision correlation model, and generating dynamic operation paths through an improved A* algorithm; and the edge data processing step involves real-time processing of positioning and operation data through ARM architecture heterogeneous computing edge nodes, outputting navigation control commands, and achieving edge-cloud data collaboration.

2. The method according to claim 1, characterized in that, In the multi-source data fusion step, the UWB fixed calibration points are cast with C30 concrete, buried at a depth of ≥0.8m. The spacing between the points is set to 50-80m based on the positioning accuracy requirements and GDOP optimization results, and the boundary points are ≥2m away from the field ridge. The sampling frequency of the UWB tag on the top of the point and the UWB base station at the agricultural machinery end is ≥10Hz, and the time synchronization error is ≤1ns. The soil moisture sensor probe is inserted into the soil 3cm, with a sampling frequency of 5Hz and a measurement accuracy of ±1%vol. When the soil moisture change rate λ>15%, the UWB three-point positioning calibration mechanism is triggered.

3. The method according to claim 1, characterized in that, In the dynamic operation planning step, positioning accuracy requirements are established based on LAI: LAI < 1.2 (sowing / transplanting period) accuracy ≤ 3cm, 1.2 ≤ LAI < 3.5 (tillering period) accuracy 3-10cm, LAI ≥ 3.5 (irrigation / jointing period) accuracy ≤ 30cm; the improved A* algorithm introduces a terrain cost coefficient C, C = 10 when the elevation change is > 10cm, otherwise C = 1, the evaluation function is f(n) = g(n) + ω·h(n) + C(n), the path spacing is adaptively adjusted according to the crop row spacing between 0.3-1.2m, and the path generation time is ≤ 10s.

4. The method according to claim 1, characterized in that, The edge data processing step adopts a CPU+GPU heterogeneous architecture edge module with a computing power of ≥200GFLOPS, power consumption of ≤10W, and compatibility with 12V agricultural machinery power supply; it integrates AES-128 data encryption and watchdog fault self-recovery function, with an MTBF of ≥2000 hours; when the network is available (4G / 5G signal ≥-90dBm), it uploads data via MQTT protocol, and in the event of an interruption, it maintains navigation for ≥30 minutes based on a ≥16GB local cache, with a data processing latency of ≤50ms.

Images