Crop variable rate fertilization control device based on multispectral sensor and installation method of crop variable rate fertilization control device

Abstract

The invention discloses a crop variable rate fertilization control device based on a multispectral sensor and an installation method thereof, and belongs to the technical field of agricultural fertilization. The system comprises a vehicle-mounted control terminal, a multi-source data acquisition module, an intelligent decision-making system and a variable execution mechanism. All the function modules carry out data interaction through a CAN bus. According to the technical scheme, a closed-loop control system integrating data acquisition, intelligent analysis and accurate execution is constructed, and the scientificity of fertilization decision and the high precision of execution are ensured through a multi-level cooperative working mechanism.

Description

Technical Field

[0001] This invention belongs to the field of agricultural fertilization technology, and relates to a crop variable fertilization control device based on a multispectral sensor and its installation method. Background Technology

[0002] With the rapid increase in my country's population and the sharp decrease in arable land, the rational and scientific application of fertilizers has become a necessary measure and a key issue that urgently needs to be addressed for the sustainable development of modern agriculture. As an important component of modern precision agriculture, variable-rate fertilization technology has become a key research direction in the global agricultural machinery equipment field. In agricultural production, scientific fertilization is a core element in improving crop yield and quality. However, traditional fertilization methods generally suffer from problems such as inaccurate fertilization, low fertilizer utilization, and environmental pollution, which seriously restrict the sustainable development of agriculture.

[0003] In existing technologies, most mechanical fertilizer applicators employ fixed fertilizer application rates or manual adjustment of fertilizer application rates, such as the soil quantitative fertilization devices described in patents CN 222621609 U and CN 222564385 U. While these devices can achieve basic quantitative fertilization functions, their biggest technical drawback is their inability to dynamically adjust based on the actual growth status of crops. Due to the significant spatial heterogeneity of soil nutrient distribution and obvious regional differences in crop growth, adopting a uniform fertilization plan inevitably leads to insufficient fertilization in some areas and excessive fertilization in others. This extensive fertilization method not only results in low fertilizer utilization (currently, the utilization rate of chemical fertilizers in my country is less than 40%), but also causes a series of environmental problems such as soil acidification and eutrophication of water bodies. Furthermore, these devices lack real-time monitoring capabilities and cannot adapt to changes in nutrient requirements at different growth stages of crops, leading to severely delayed fertilization decisions that directly affect crop yield and quality.

[0004] To address the shortcomings of traditional fertilizer applicators, some intelligent fertilization devices have begun to incorporate detection technologies in recent years. For example, patent CN 221768727 U describes a "precision fertilization device based on real-time detection of soil nutrient requirements and fertilizer nutrients." This technology analyzes soil nutrient content using a near-infrared spectroscopy acquisition module and combines a flow meter and variable mechanism to adjust the fertilization rate. However, a significant drawback of this device is its reliance on single soil nutrient data, neglecting the physiological state indicators of the crop itself. In actual agricultural production, there is often a significant difference between soil nutrient content and crop absorption status. For instance, under drought stress, even with sufficient soil nitrogen, crop root absorption capacity can be severely inhibited. Furthermore, this device does not consider key physiological indicators such as the photosynthetic efficiency and chlorophyll content of the crop canopy, resulting in a lack of dynamic feedback on crop growth in fertilization decisions, thus failing to achieve truly precise fertilization.

[0005] Another representative technology is the variable fertilization device based on crop seedling information proposed in patent CN 212876676 U. This device collects normalized difference vegetation index (NDVI) information from the crop canopy and combines it with a positioning system to achieve variable fertilization. Although this technology considers crop growth information, its technical solution has obvious limitations: First, relying solely on a single vegetation index is insufficient to comprehensively reflect the true nutritional status of the crop; second, the device completely ignores the detection of soil nutrient status, resulting in a lack of soil-based data to support fertilization decisions; finally, its control strategy is too simplistic, adjusting only according to the principle of "more fertilization for weak seedlings and less fertilization for vigorous seedlings," failing to achieve truly scientific and precise regulation.

[0006] Through in-depth analysis of existing technologies, the main problems of current variable fertilization technology can be summarized as follows: (1) Single data collection: Existing equipment either relies solely on soil data or focuses solely on crop information, lacking fusion analysis of multi-source data; (2) Simplified decision-making model: Most technologies use static thresholds or simple rules for decision-making, which cannot adapt to complex farmland environments; (3) System response lag: Some systems that rely on cloud computing have significant response delays, making it difficult to meet real-time control requirements; (4) Poor environmental adaptability: Existing technologies lack the ability to dynamically respond to environmental factors such as weather changes, soil moisture, and pests and diseases. With the development of new technologies such as artificial intelligence, modern agriculture has put forward higher requirements for intelligent fertilization equipment. An ideal variable fertilization system should have the following characteristics: (1) It can simultaneously collect multi-dimensional data such as soil nutrients, crop physiology, and environmental parameters; (2) It establishes a dynamic fertilizer requirement algorithm based on crop growth models to achieve scientific and precise regulation; (3) It adopts an edge computing architecture to ensure rapid system response; (4) It is equipped with a high-precision variable fertilization mechanism to achieve centimeter-level precise regulation.

[0007] In summary, existing variable fertilization technologies still have significant shortcomings, and there is an urgent need for an innovative solution that can integrate multi-source data and possess intelligent decision-making capabilities. The crop variable fertilization control device based on multispectral sensors proposed in this invention is a systematic solution addressing the aforementioned technical bottlenecks. Summary of the Invention

[0008] The purpose of this invention is to solve the aforementioned technical problems by providing a crop variable fertilization control device based on a multispectral sensor and its installation method. This device integrates a multispectral sensor, a soil nutrient detection module, and an environmental monitoring unit to achieve real-time acquisition and fusion analysis of multidimensional soil-crop-environment data. It employs an edge computing architecture to construct a dynamic fertilization demand decision model, comprehensively considering crop growth stages, physiological states, and environmental factors to achieve scientific and precise fertilization regulation. Simultaneously, it is equipped with a high-precision variable actuator to ensure that the fertilization amount is dynamically adjusted according to crop needs. This invention solves the problems of single data sources, simple decision models, and delayed response in existing technologies, significantly improving fertilizer utilization and fertilization accuracy, and providing an innovative solution for the sustainable development of modern agriculture.

[0009] To achieve the above objectives, the present invention mainly provides the following technical solutions:

[0010] First, this invention provides a crop variable fertilization control device based on a multispectral sensor, including an on-board control terminal, a multi-source data acquisition module, an intelligent decision-making system, and a variable execution mechanism. The on-board control terminal, as the core hub, is installed in a waterproof electrical box behind the agricultural machinery cab. It is connected to the multi-source data acquisition module, the intelligent decision-making system, and the variable execution mechanism via a CAN bus master node and is directly connected to the agricultural machinery's 24V power supply. In the multi-source data acquisition module, a multispectral imager is fixed at the front and directly connected to the control terminal via gigabit Ethernet. Near-infrared soil probes are distributed at three points on the chassis and connected via a CAN FD subnet. An environmental sensor is integrated on the top of the cab and communicates via RS-485. The intelligent decision-making system is embedded in the computing unit of the control terminal, acquires data through the PCIe channel, and outputs decision commands. The control valve group of the variable execution mechanism is installed on the suspension frame, and the PWM nozzle array is deployed on the fertilization crossbeam. It receives commands via the CANopen protocol and simultaneously provides feedback on the execution status through dual channels. Each module is connected to the control terminal via shielded cables to achieve anti-interference, forming a closed-loop control link of "data acquisition - edge computing - precise execution," with the overall latency controlled within 150ms.

[0011] The vehicle-mounted control terminal, as the core processing unit of the system, adopts an industrial-grade embedded processor to process multi-channel sensor data in real time, execute control algorithms, and perform real-time task scheduling.

[0012] The multi-source data acquisition module is used for crop information acquisition, soil testing, environmental monitoring, and data synchronization and quality control.

[0013] The intelligent decision-making system is used to preprocess data, normalize and extract features from multi-source data, comprehensively evaluate crop nutrition status through fuzzy logic algorithm, output nutrition requirement level, and generate optimal fertilization plan based on deep reinforcement learning algorithm, taking into account crop growth stage, current nutrition status, soil nutrient level and environmental conditions.

[0014] The variable execution mechanism is used to achieve precise fertilization control.

[0015] Furthermore, the device also includes a communication system: the communication system adopts a CAN 2.0B bus (500kbps, 29-bit extended frame format) and supports the ISO 11783 (ISOBUS) protocol to ensure compatibility with agricultural machinery. As the infrastructure layer running through all modules and serving as the data interaction hub of the entire system, it achieves efficient and reliable communication between modules via the CAN 2.0B bus (500kbps, 29-bit extended frame format) and the ISO 11783 (ISOBUS) protocol. Specifically, the system uses the vehicle-mounted control terminal as the main node, responsible for scheduling the real-time data upload of multi-source data acquisition modules (such as RTK-GPS, IMU, and multispectral sensors) (e.g., IMU 100Hz high-frequency data is transmitted via priority ID 0x18FFA001), and issuing fertilization commands from the intelligent decision-making system to the variable actuators (response delay <50ms), while simultaneously receiving feedback signals from the actuators (e.g., hydraulic valve opening) to form a closed-loop control. Its technical features include: plug-and-play compatibility with agricultural implements based on the ISOBUS protocol; a hybrid mechanism of event triggering and periodic polling to ensure the real-time performance of critical data (emergency message delay <10ms); and bus redundancy detection and fault isolation capabilities (such as automatic switching to local cache mode to handle sensor communication timeouts). In practical applications, the system not only effectively suppresses electromagnetic interference in farmland through a twisted-pair physical layer design (120Ω terminating resistor), but also supports the expansion of third-party agricultural implements (such as seeders and sprayers), achieving standardized control of "one terminal for multiple implements." It also complies with the ISO 7637-2 impulse immunity standard, providing high-reliability communication assurance for the system.

[0016] Furthermore, the vehicle-mounted control terminal includes a data fusion processing layer. This layer employs Extended Kalman Filter (EKF) for spatiotemporal registration of multi-source data, fusing GPS (RTK positioning, ±2cm accuracy), IMU (Inertial Measurement Unit, 100Hz sampling rate), and multispectral data to ensure data consistency. The data fusion processing layer is the core data processing hub of the intelligent variable fertilization control system. It is not independent of the four main modules but is a key software function deeply integrated into the vehicle-mounted control terminal. Its interaction with the four main modules is as follows: receiving real-time heterogeneous data from multi-source data acquisition modules (RTK-GPS, IMU, multispectral sensors, etc.) and performing spatiotemporal registration and fusion through Extended Kalman Filter (EKF); providing high-precision, spatiotemporally consistent standardized data packets for the intelligent decision-making system; and simultaneously receiving feedback information from the variable actuators for dynamic calibration. Its core function is to achieve closed-loop processing from raw perception to precise control (delay <200ms) through multi-source data fusion (positioning accuracy improved to ±1.5cm), heterogeneous data spatiotemporal alignment (synchronization error <1ms) and dynamic weight adjustment. It is the key link connecting "perception-decision-execution" and directly determines the system's operational accuracy and reliability in complex farmland environments.

[0017] Furthermore, the device also includes a wireless communication module, which uses a Quectel RM500Q 5G module, supports SA / NSA dual-mode, has a maximum downlink rate of 2.5Gbps, and has built-in GNSS (Global Navigation Satellite System) to ensure remote monitoring and real-time data upload. As a key extension component of the intelligent variable fertilization control system, independent of the four core modules (vehicle control terminal, multi-source data acquisition module, intelligent decision-making system, and variable actuator), the wireless communication module uses a high-performance Quectel RM500Q 5G module (supporting SA / NSA dual-mode, with a maximum downlink rate of 2.5Gbps) to build a collaborative "end-edge-cloud" communication architecture for the system through a 5G wide area network. This module not only integrates a multi-frequency GNSS receiver (compatible with BeiDou / GPS / Galileo) to provide auxiliary positioning for the vehicle terminal, but also achieves remote interconnection of four modules through a low-latency (<50ms) and highly reliable (99.99% connection availability) wireless link: real-time uploading of soil moisture, crop growth, and other data collected by multi-source sensors to the cloud decision-making platform; dynamic reception of fertilizer prescription maps optimized based on meteorological big data; and synchronous monitoring of the operating status of the actuators and support for remote fault diagnosis. Its unique "5G+CAN" dual-channel redundancy mechanism (automatically switching to 5G transmission of control commands within 300ms when bus signal loss is detected) significantly improves system robustness, while its ISO 14990-compliant electromagnetic compatibility design ensures a signal reception sensitivity of -110dBm even in highly interfering farmland environments. Furthermore, the module's integrated FOTA function supports seamless upgrades of all subsystems, and combined with AES-256 encrypted transmission, provides all-weather, highly secure data services for precision agriculture, ultimately achieving a closed-loop cloud-based collaborative optimization of "perception-decision-execution."

[0018] Furthermore, the multi-source data acquisition module employs a high-resolution multispectral imager during crop information acquisition to simultaneously acquire reflectance data of the crop canopy in multiple bands. It extracts multiple vegetation indices such as NDVI, PRI, and NDRE through a patented algorithm to comprehensively characterize the crop's chlorophyll content, photosynthetic activity, and nitrogen status. In the soil testing process, a contact near-infrared soil probe is used, and in the environmental monitoring process, temperature and humidity sensors, light intensity sensors, and soil moisture sensors are employed.

[0019] Specific algorithm steps:

[0020] Step 1: Radiometric Calibration

[0021] To eliminate the effects of sensor dark current, optical system attenuation, and atmospheric scattering, the original digital signal (DN value) is converted into physical reflectivity, as shown in the following formula:

[0022]

[0023] DN λ Original pixel value

[0024] B λ Dark current noise (obtained through black frame calibration)

[0025] G λ Gain coefficient (provided in calibration documentation)

[0026] E λ Solar irradiance in band λ (atmospheric model input)

[0027] θ: Solar zenith angle

[0028] Step 2: Geometric Correction

[0029] By combining GPS / IMU data with camera calibration parameters, image distortion caused by changes in aircraft attitude is eliminated, ensuring spatial alignment across multiple spectral bands.

[0030]

[0031] K: Camera intrinsic parameter matrix

[0032] R|T: Rotation / translation matrix (obtained from calibration)

[0033] Step 3: Band Registration

[0034] The SIFT feature matching algorithm is used to align images of different bands to ensure pixel-level spatial consistency.

[0035] Step 4: Calculation of Vegetation Index

[0036] Crop physiological status can be quantified by specific band combinations (e.g., NDVI reflects chlorophyll content, and PRI indicates photosynthetic efficiency).

[0037] NDVI (Normalized Difference Vegetation Index):

[0038]

[0039] PRI (Photochemical Reflectance Index):

[0040]

[0041] NDRE (Red-edge Normalized Difference Index):

[0042]

[0043] P531, P570, P760, P860, P660, and P710 represent the reflectivity of the 531nm, 570nm, 760nm, 860nm, 660nm, and 710nm wavelength bands, respectively.

[0044] Step 5: Outlier Detection (QC Algorithm)

[0045] Sliding window statistical methods are used to remove outliers caused by cloud shadows, sensor noise, etc., improving data reliability. Outlier removal is based on the statistical 3σ principle.

[0046] Valid data range = [μ-3σ, μ+3σ]

[0047] Where μ and σ are the mean and standard deviation of the vegetation index within the sliding window.

[0048] Step 6: Spatiotemporal data fusion

[0049] The vegetation index is bound to GPS coordinates (longitude and latitude) and timestamps (UTC) to generate GeoJSON format data.

[0050] Furthermore, the variable actuator includes an electromechanical-hydraulic integrated variable actuator system, which includes a fertilizer storage tank, a metering pump, a flow control valve, and a spraying device.

[0051] Furthermore, the electromechanical-hydraulic integrated variable actuator system also includes a pressure compensation device.

[0052] Furthermore, the variable actuator is equipped with dual CAN buses and mechanical backup.

[0053] Second, this embodiment of the invention also provides an installation method for the crop variable fertilization control device based on a multispectral sensor, comprising the following steps:

[0054] (1) Sensor module installation

[0055] Multispectral imager: It is fixed to the front of the agricultural machinery by a fixing device, and the downward angle is adjusted to 30° by an angle control device, covering a working width of 5m.

[0056] Near-infrared soil probe: Installed on the chassis of agricultural machinery, it adopts a hydraulic lifting mechanism. When in operation, the probe is pressed into the soil, and when not in operation, it automatically rises to avoid obstacles.

[0057] Environmental sensors: integrated into the top of the agricultural machinery cab to avoid dust interference;

[0058] (2) Installation of control and execution system

[0059] Vehicle-mounted control terminal: The waterproof housing is fixed to the rear of the agricultural machinery and communicates with the agricultural machinery ECU via CAN bus;

[0060] PWM nozzle array: evenly distributed along the crossbeam of the fertilizer applicator, each nozzle is independently controlled, supporting liquid / granular fertilizer spraying;

[0061] Redundancy design: The critical circuit adopts dual CAN bus, and the main controller and the backup PLC synchronize data in real time.

[0062] Furthermore, in step (1), the spacing of the PWM nozzle array evenly distributed along the crossbeam of the fertilizer applicator is 15cm.

[0063] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0064] This invention, through systematic innovation in multi-source heterogeneous data fusion, edge intelligent decision-making, and high-precision execution control, successfully solves the three major industry problems of "inaccurate measurement, slow calculation, and imprecise control" commonly found in existing variable fertilization technologies. Compared with existing patented technologies, this solution achieves a qualitative leap in data integrity, scientific decision-making, control accuracy, and system reliability, providing a complete, efficient, and reliable solution for precision agriculture with broad market prospects and significant social value. Attached Figure Description

[0065] Figure 1 This invention describes the system hardware composition and static structure of the crop variable fertilization control device based on a multispectral sensor.

[0066] Figure 2 This is a schematic diagram and working diagram of the multispectral sensor in the crop variable fertilization control device based on a multispectral sensor according to the present invention.

[0067] Figure 3 This is a schematic diagram of the near-infrared soil probe structure of the crop variable fertilization control device based on a multispectral sensor according to the present invention;

[0068] Figure 4 This is a control flowchart of the intelligent fertilizer applicator device of the crop variable fertilization control device based on multispectral sensors of the present invention;

[0069] Figure 5 This is a schematic diagram of the PWM nozzle array of the crop variable fertilization control device based on a multispectral sensor according to the present invention;

[0070] Figure 6 A flowchart of the algorithm for extracting multiple vegetation indices such as NDVI, PRI, and NDRE;

[0071] Figure 7 This is a schematic diagram of the overall structure of the device of the present invention;

[0072] Among them, 1-angle control device, 2-multispectral imager, 3-optical lens, 4-fixing device, 5-near-infrared soil analyzer, 6-hydraulic lifting mechanism, 7-near-infrared soil probe, 71-first near-infrared soil probe, 72-second near-infrared soil probe, 73-third near-infrared soil probe, 8-fertilizer applicator crossbeam, 9-suspension frame, 10-PWM nozzle array, 11-environmental sensor, 12-vehicle control terminal. Detailed Implementation

[0073] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0074] 1. System Overall Architecture Design

[0075] The intelligent variable fertilization control device of this invention adopts a closed-loop control architecture of "perception-decision-execution". The overall system consists of four main parts: an on-board control terminal 12, a multi-source data acquisition module, an intelligent decision-making system, and a variable execution mechanism. The on-board control terminal 12, as the core processing unit of the system, uses an industrial-grade embedded processor with powerful edge computing capabilities, enabling real-time processing of multi-channel sensor data and execution of control algorithms. The system adopts a modular design, with each functional module interacting via a CAN bus to ensure high reliability and scalability. In particular, this invention innovatively designs a data fusion processing layer, which performs spatiotemporal alignment and feature extraction on heterogeneous data from different sensors, providing a unified data foundation for subsequent intelligent decision-making. The system is also equipped with a 4G / 5G wireless communication module, enabling remote monitoring and data uploading while retaining the real-time advantages of edge computing, effectively solving the latency problem inherent in traditional cloud processing solutions.

[0076] The intelligent variable fertilization control device of the present invention adopts a closed-loop control architecture of "sensing-decision-execution", and consists of the following core modules:

[0077] Vehicle control terminal 12: It adopts NVIDIA Jetson AGX Orin, equipped with an 8-core ARM Cortex-A78AE CPU and a GPU with 2048 CUDA cores, with an AI computing power of 32 TOPS, runs the Ubuntu 20.04 LTS operating system, and is equipped with ROS 2 (Robot Operating System) for real-time task scheduling.

[0078] Communication system: It adopts CAN 2.0B bus (500kbps, 29-bit extended frame format) and supports ISO 11783 (ISOBUS) protocol to ensure compatibility with agricultural machinery.

[0079] Data fusion processing layer: Extended Kalman filter (EKF) is used for spatiotemporal registration of multi-source data, and GPS (RTK positioning, ±2cm accuracy), IMU (inertial measurement unit, 100Hz sampling rate) and multispectral data are fused to ensure data consistency.

[0080] Wireless communication module: adopts Quectel RM500Q 5G module, supports SA / NSA dual mode, maximum downlink speed of 2.5Gbps, and built-in GNSS (Global Navigation Satellite System) to ensure remote monitoring and real-time data upload.

[0081] 2. Multi-source data acquisition module

[0082] Data acquisition is fundamental to precision fertilization. This invention employs a multi-dimensional data acquisition scheme combining multispectral sensing, near-infrared spectral analysis, and environmental monitoring. For crop information acquisition, the system is equipped with a multispectral imager 2, which can simultaneously acquire reflectance data of the crop canopy in multiple bands (including visible light, red-edge, and near-infrared bands). A patented algorithm extracts multiple vegetation indices such as NDVI, PRI, and NDRE to comprehensively characterize the crop's chlorophyll content, photosynthetic activity, and nitrogen status. For soil testing, the system is equipped with a contact near-infrared soil probe 7, which can detect the nitrogen, phosphorus, and potassium content and organic matter levels in the soil in real time during the fertilizer applicator's movement, with a measurement accuracy of ±5%. The environmental monitoring unit integrates temperature and humidity sensors, light intensity sensors, and soil moisture sensors, providing environmental parameter support for fertilization decisions. All sensor data is spatiotemporally synchronized using timestamps and GPS positioning information to ensure data consistency. This invention features a specially designed sensor data quality control algorithm that can automatically identify and remove abnormal data, significantly improving data reliability.

[0083] (1) Crop information collection

[0084] Multispectral imaging: A SENOP MS-12 multispectral camera with 12 interference filters is used, with center wavelengths of 450nm, 510nm, 560nm, 610nm, 660nm, 710nm, 760nm, 810nm, 860nm, 910nm, 960nm, and 1010nm, and a bandwidth of ±5nm. Optical lens 3 is a 25mm fixed-focus lens with a depth of field of 0.5m-∞, achieving a ground resolution of 1.2mm / pixel at a working distance of 1m. The camera transmits data via a GigE Vision interface, employing FPGA for real-time Bayer decoding, with an image transmission latency of <5ms.

[0085] Vegetation index calculation: An improved NDVI (Normalized Difference Vegetation Index) algorithm is used.

[0086]

[0087] P860, P660, and P710 represent the reflectivity of the 860nm, 660nm, and 710nm wavelength bands, respectively.

[0088] (2) Soil testing

[0089] The Veris P4000 near-infrared soil analyzer 5 is equipped with a diamond-coated cutting edge, achieving a penetration resistance of <200N. Sampling depth is calibrated using a laser rangefinder (accuracy ±1mm). The near-infrared light source is a 20W tungsten halogen lamp (2000-hour lifespan), and the detector is cooled to -10℃ to reduce noise. It provides real-time monitoring of soil nitrogen (0-500mg / kg), phosphorus (0-200mg / kg), potassium (0-1000mg / kg), and moisture, with a sampling depth of 0-30cm and a sampling frequency of 10Hz.

[0090] Furthermore, considering the continuous operation requirements of agricultural machinery, this invention employs "traveling scanning + dynamic compensation" technology to ensure that soil profile data acquisition and fertilization control can be performed without stopping. It primarily utilizes multiple probes working alternately. Soil probes are installed on the chassis of the agricultural machinery and are alternately pressed into the soil via a hydraulic system. When probe 1 is pressed into the soil for scanning (0-30cm), probes 2 / 3 are in a raised or ready state. Every 1.5m the agricultural machinery advances, the next set of probes takes over, achieving seamless continuous sampling. Data from each probe is aligned using timestamps and GPS positioning to generate a continuous soil profile. The probe pressing / lifting cycle is ≤0.5 seconds (hydraulic pressure 0.8MPa), and the scanning time for a single profile is ≤0.3 seconds. At a vehicle speed of 10km / h, each set of probes can complete 20 full profile scans per minute, meeting agronomic requirements.

[0091] When data is missing at a certain location due to vibration or other reasons, the system generates an estimated value based on Kriging interpolation combined with historical data, with an error rate of <5%. The interpolation weights consider: actual measurements from previous scan points (60% weight); historical soil data of the site (30% weight); and multispectral canopy data to assist in prediction (10% weight).

[0092] The system monitors vehicle speed in real time and dynamically adjusts the scanning frequency as shown in Table 1:

[0093]

[0094] Flow compensation formula: Q adj = Q ref × [1-(v-v0) / 100]

[0095] v: Real-time vehicle speed (km / h) v0: Base speed (5km / h)

[0096] (3) Environmental monitoring

[0097] Temperature and humidity sensor (SHT85): Accuracy ±0.2℃ (temperature), ±2%RH (humidity).

[0098] Soil moisture sensor (TEROS 12): Measurement range 0-100% VWC (volume water content), accuracy ±3%.

[0099] Light sensor (LI-190R): measures PAR (photosynthetically active radiation), range 0-3000 μmol / m² / s.

[0100] (4) Data synchronization and quality control

[0101] PTPv2 (Precise Time Protocol) is used to ensure time synchronization of sensor data (error <1μs).

[0102] Data anomaly detection: If the data exceeds the range three times consecutively, an alarm will be triggered and the system will switch to a redundant sensor.

[0103] 3. Intelligent Decision-Making System

[0104] The intelligent decision-making system constructs a dynamic nutrient requirement decision-making model based on crop growth models and machine learning algorithms. The system adopts a hierarchical decision architecture: the first layer is a data preprocessing layer, which normalizes and extracts features from multi-source data; the second layer is a state assessment layer, which comprehensively evaluates crop nutrient status using fuzzy logic algorithms and outputs nutrient requirement levels; the third layer is a decision optimization layer, which, based on deep reinforcement learning algorithms, comprehensively considers crop growth stage, current nutrient status, soil nutrient levels, and environmental conditions to generate the optimal fertilization plan. During the training phase, the decision model uses a large amount of field trial data, covering the growth response characteristics of different crop varieties, soil types, and climatic conditions, ensuring the model's generalization ability. The system also innovatively introduces a feedback adjustment mechanism, continuously optimizing decision parameters by comparing crop growth changes before and after fertilization, achieving self-learning and performance improvement. To adapt to different agronomic needs, the system supports flexible configuration of multiple fertilization strategies, allowing users to select appropriate fertilization modes according to crop type and growth goals.

[0105] (1) Neural network architecture

[0106] Input layer: 14 nodes (6 vegetation indices + 4 soil parameters + 4 environmental parameters).

[0107] Hidden layers: 3 layers (32-64-32 nodes), activation function is LeakyReLU (α=0.01).

[0108] Output layer: 3 nodes (nitrogen, phosphorus, and potassium requirements), normalized to the 0-1 range using the Sigmoid function.

[0109] (2) Training and optimization

[0110] Training data: Based on 10,000 sets of field trial data (covering crops such as corn, wheat, and soybeans).

[0111] Optimization algorithm: Adam optimizer is used (learning rate 0.001, batch size 64).

[0112] Online learning mechanism: The model parameters are automatically updated after each hectare of work is completed.

[0113] (3) Decision-making logic

[0114] Fertilization mode selection (currently 5 preset modes) is shown in Table 2:

[0115]

[0116] Dynamic adjustment mechanism: If the soil nitrogen content is <200mg / kg and NDVI <0.6, nitrogen fertilizer will be automatically increased by 10%.

[0117] 4. Variable execution mechanism

[0118] To achieve precise fertilization control, this invention designs a high-precision electromechanical-hydraulic integrated variable execution system. This system consists of a fertilizer storage tank, a metering pump, a flow control valve, and a spraying device. A PID closed-loop control algorithm ensures precise adjustment of the fertilizer application rate. The metering pump is driven by a servo motor and, in conjunction with a high-precision flow sensor, achieves a flow rate adjustment accuracy of 0.1 L / min. The spraying device employs a fan-shaped nozzle array design, using PWM modulation technology to control the opening time and frequency of each nozzle, achieving precise spatial distribution of the fertilizer application rate. The system is also equipped with a pressure compensation device to ensure stable fertilizer application under different travel speeds and terrain conditions. To improve system reliability, the actuators adopt a redundant design; when the main control system fails, the backup system can immediately take over, ensuring continuous operation. This invention specifically optimizes the synergy between the mechanical structure and the control algorithm, eliminating mechanical response delay through a feedforward compensation algorithm, enabling the system to complete the entire process from decision-making to execution within 0.5 seconds, meeting the requirements of real-time precise control. Details are as follows:

[0119] (1) Hydraulic system

[0120] Hydraulic pump: Bosch Rexroth A10VSO axial piston pump, working pressure 20MPa, displacement 28cm³ / rev.

[0121] Flow control: A proportional valve and Coriolis flow meter are used, with an accuracy of ±0.5%.

[0122] (2) Control Algorithm

[0123] Improved fuzzy PID control

[0124]

[0125] Kp, Ki, and Kd are dynamically adjusted based on the error.

[0126] (3) PWM nozzle array

[0127] 25 TeeJet AIXR11003 nozzles, 15cm spacing, PWM control (1kHz frequency, 0.1% duty cycle resolution).

[0128] Response time: <10ms, flow rate adjustment range: 0.1-5L / min.

[0129] (4) Redundancy design

[0130] Dual CAN bus: If the master controller fails, the backup controller will take over within 50ms.

[0131] Mechanical backup: In case of electronic system failure, the manual adjustment valve can maintain basic fertilization function.

[0132] 5. Detailed explanation of the work process

[0133] The intelligent variable-rate fertilization system of this invention employs a control mechanism of multi-sensor collaborative data acquisition, real-time decision-making, and precise execution during operation to ensure high-precision variable-rate fertilization during continuous agricultural machinery movement. The following provides a detailed description of each stage:

[0134] (1) Initialization phase (0-500ms)

[0135] After the system starts up, it first performs initialization operations to ensure that all modules are in normal working order:

[0136] Loading GIS data: Retrieving farmland boundary data (Shapefile format) via 5G network or local storage; loading historical fertilization records (including nitrogen, phosphorus, and potassium application rates and crop response data); analyzing soil type distribution maps (such as sandy soil, clay, loam, etc.) to provide benchmark data for subsequent decision-making.

[0137] Sensor self-test: The multispectral camera performs white balance calibration and checks lens cleanliness (if dirt obscures >5% of the field of view, an alarm is triggered); the near-infrared soil probe 7 performs zero-point calibration (verifying the measurement deviation <±3% in a standard soil sample); the environmental sensor 11 verifies whether the initial readings of temperature and humidity (SHT85) and light intensity (LI-190R) are within a reasonable range; the positioning system mainly confirms that the RTK-GPS (Huace Navigation P7) locks onto a valid satellite signal (HDOP<1.5), and the IMU (TDK ICM-42688-P) completes gyroscope calibration.

[0138] Establish a work coordinate system: Integrate RTK-GPS (horizontal accuracy ±2cm) and IMU data to generate a high-precision work path (AB line). Calculate the optimal travel trajectory based on the size of the agricultural machinery (e.g., wheel track 2.5m) and the fertilization width (4-6m) to avoid overlapping or missed spraying.

[0139] (2) Data acquisition phase (every 100ms)

[0140] During the movement of agricultural machinery, each sensor collects data synchronously at a fixed interval:

[0141] Multispectral imaging (2ms exposure, 30fps): Exposure time 2ms, acquiring 12 bands (450-1010nm) with a spatial resolution of 1.2mm / pixel (at a height of 1.5m above the ground); real-time calculation of the vegetation index (NDVI), formula as follows:

[0142]

[0143] Soil scanning (10Hz): Three near-infrared soil probes work alternately. The first near-infrared soil probe 71 is pressed into the soil (0-30cm depth) for scanning, taking 0.3s. After the agricultural machinery advances 1.5m, the second near-infrared soil probe 72 and the third near-infrared soil probe 73 take over, ensuring seamless data coverage. The main measurement parameters include: nitrogen content (0-500mg / kg, ±5mg / kg), phosphorus content (0-200mg / kg, ±3mg / kg), and potassium content (0-1000mg / kg, ±10mg / kg).

[0144] Environmental monitoring: Primarily measures air temperature and humidity (range: -20-60℃, 0-100%RH, ±1.5% accuracy), light intensity (0~2000μmol / m² / s, ±5% accuracy), and soil volumetric water content (0-100%VWC, ±3% accuracy).

[0145] (3) Intelligent decision-making stage (<50ms)

[0146] Data preprocessing (5ms): This mainly involves data normalization and spatiotemporal alignment. Normalization involves scaling the data from each sensor to the range of [0,1] (e.g., mapping NDVI to 0.2-0.8). Spatiotemporal alignment is achieved by fusing multi-source data through extended Kalman filtering (EKF) to eliminate spatiotemporal biases caused by agricultural machinery vibration.

[0147] Neural network inference (30ms):

[0148] Input features: 14-dimensional vector (6 vegetation indices + 4 soil parameters + 4 environmental parameters).

[0149] Model structure: 3-layer fully connected network (14-8-3), output N, P, K demand ratio (e.g., 3:1:2).

[0150] Training data: Based on 100,000 sets of field measurement data, the accuracy of the test set is ≥92%.

[0151] Command generation (15ms): Based on vehicle speed (obtained via CAN bus), the fertilizer application rate is dynamically adjusted, generating a PWM control signal (duty cycle 0.1%-100%) and sending it to the corresponding nozzle. Calculation formula:

[0152]

[0153] Where v is the real-time vehicle speed (km / h).

[0154] (4) Precise execution phase (<100ms)

[0155] The variable actuator completes the precise application of fertilizer according to instructions:

[0156] Hydraulic pump speed regulation (50ms): When the target flow rate is 5L / min, the pump speed is stabilized at 1500rpm (PID control, steady-state error <±1%). The pressure sensor (20MPa range) monitors the system pressure in real time and triggers the safety valve when the limit is exceeded.

[0157] Proportional valve positioning: opening adjustment accuracy ±0.5%, response time <20ms, Coriolis flow meter (0.1-20L / min) feedback of actual flow, closed-loop control error <±3%.

[0158] Nozzle modulation (10ms): 25 nozzles are controlled independently. When the duty cycle is 30%, the nozzles are turned on for 3ms / cycle (1kHz PWM) and the flow rate is 1.5L / min.

[0159] (5) Exception handling mechanism

[0160] The system has a built-in multi-layer fault-tolerance mechanism to ensure operational safety:

[0161] Sensor malfunction: If the data exceeds the limit three times consecutively (e.g., soil N content > 600 mg / kg), mark the probe as faulty and switch to the backup probe; if the multispectral camera is obscured by dirt > 10%, trigger the automatic cleaning device (high-pressure gas blowing).

[0162] Execution error: Flow deviation > 5% for 200ms: Pause the current nozzle and switch to the backup pipeline; Hydraulic pressure > 18MPa: Reduce pump speed and trigger an alarm.

[0163] Communication failure: When the CAN bus error rate exceeds 10%, switch to the redundant bus to prevent data loss. When the 5G signal is interrupted, enable LoRa self-organizing network to maintain critical data transmission.

[0164] (6) Data recording and uploading

[0165] The system records all job data and supports remote management.

[0166] Local storage: 1000 hours of data are stored in a loop (ISO 11783 XML format), mainly including timestamps, GPS coordinates, sensor readings, fertilizer application amount, and equipment status.

[0167] Cloud synchronization: Real-time upload to the agricultural cloud platform via 5G (Quectel RM500Q module) (MQTT protocol), key data (such as fertilizer prescription map) supports offline download (USB 3.0 interface).

[0168] 6. Agricultural Machinery Installation Plan

[0169] (1) Sensor module installation

[0170] Multispectral imager 2: It is fixed to the front of the agricultural machinery (1.5m above the ground) by the fixing device 4, and the downward angle is adjusted to 30° by the angle control device 1, covering the working width (5m).

[0171] Near-infrared soil probe: Installed on the chassis of agricultural machinery, it adopts a hydraulic lifting mechanism 6 (stroke 50cm). When working, the probe is pressed into the soil, and when not working, it automatically rises to avoid obstacles.

[0172] Environmental sensor 11: Integrated into the top of the agricultural machinery cab to avoid dust interference.

[0173] (2) Installation of control and execution system

[0174] Vehicle-mounted control terminal 12: The waterproof housing is fixed to the rear of the agricultural machinery and communicates with the agricultural machinery ECU via CAN bus.

[0175] PWM nozzle array 10: evenly distributed along the crossbeam 8 of the fertilizer applicator (15cm spacing), each nozzle is independently controlled, supporting liquid / granular fertilizer spraying.

[0176] Redundancy design: The critical circuit adopts dual CAN bus, and the main controller and the backup PLC synchronize data in real time.

[0177] This invention relates to an intelligent variable fertilization control device and method based on a multispectral sensor, as described below with reference to the accompanying drawings ( Figures 1-5 The specific implementation method is described in detail.

[0178] 1. System hardware composition and static structure ( Figure 1 )

[0179] The intelligent variable fertilization system of the present invention consists of the following core modules:

[0180] (1) Vehicle-mounted control terminal

[0181] Model: NVIDIA Jetson AGX Orin (32 TOPS AI computing power)

[0182] Operating system: Ubuntu 20.04 LTS + ROS 2 (Humble version)

[0183] Functions: Responsible for data fusion, decision calculation, and execution control.

[0184] Interfaces: 2×CAN 2.0B (500kbps, supports ISO 11783 protocol); 1×RS485 (for connecting to near-infrared soil probe); 1×HDMI (for debugging display); 1×5G module (Quectel RM500Q)

[0185] (2) Multispectral sensor module Figure 2 , Figure 3 )

[0186] Model: SENOP MS-12 (12-band, 450-1010nm)

[0187] Installation location: Front of the agricultural machinery (1.5m above the ground, 30° downward angle)

[0188] Data output: Resolution: 1280×960; Frame rate: 30fps; Exposure time: 2ms (automatic adjustment)

[0189] (3) Soil testing module—Near-infrared soil analyzer 5 (Veris P4000)

[0190] Sampling depth: 0-30cm

[0191] Measurement range: N (0-500 mg / kg), P (0-200 mg / kg), K (0-1000 mg / kg)

[0192] Sampling frequency: 10Hz

[0193] Soil moisture sensor (TEROS 12): Measurement range: 0-100% VWC; Accuracy: ±3%

[0194] (4) Implementing agency ( Figure 4 )

[0195] Hydraulic system: Bosch Rexroth A10VSO (28cm³ / rev, 20MPa); Proportional valve: Parker D1FPE50 (±0.5% linearity); Flow meter: Coriolis flow meter (0.1-20L / min, ±0.5% accuracy); PWM nozzle array: 25 TeeJet AIXR11003 fan nozzles; Control mode: PWM (1kHz, duty cycle 0.1%-100%); Flow range: 0.1-5L / min.

[0196] (5) Redundant backup system

[0197] Dual CAN bus: When the master controller fails, the backup controller (STM32H743) takes over within 50ms.

[0198] Mechanical backup valve: Manual adjustment mode ensures fertilization can still be performed in case of electronic failure.

[0199] 2. System Workflow ( Figure 5 )

[0200] (1) Initialization phase (0-500ms)

[0201] (1) Load farmland GIS data

[0202] Import farmland boundaries and historical fertilization records (Shapefile format) via USB or 5G network.

[0203] 2. Sensor self-test

[0204] Multispectral camera: Check lens cleanliness (alarm triggered if dirt > 5%).

[0205] Near-infrared soil probe: zero-point calibration (verified in known standard soil samples).

[0206] 3. Establish the work coordinate system:

[0207] By combining RTK-GPS (Huace Navigation P7, ±2cm accuracy) and IMU (TDK ICM-42688-P) data, a high-precision path is generated.

[0208] (2) Data acquisition phase (loop every 100ms)

[0209] 1. Multispectral imaging:

[0210] Exposure time 2ms, acquire 12 band reflectance data.

[0211] Calculate the vegetation index (e.g., NDVI*, see Formula 1).

[0212] 2. Soil testing

[0213] Near-infrared sensors scan the soil and output the N, P, and K contents (mg / kg).

[0214] 3. Environmental monitoring

[0215] Temperature and humidity (SHT85), light intensity (LI-190R), and soil moisture (TEROS 12) data updated.

[0216] (3) Intelligent decision-making stage (<50ms)

[0217] 1. Data preprocessing:

[0218] Normalization: Scaling all sensor data to the 0-1 range.

[0219] Spatiotemporal alignment: Multi-source data are fused using extended Kalman filtering (EKF).

[0220] 2. Neural Network Inference:

[0221] Input: 14-dimensional features (6 vegetation indices + 4 soil + 4 environment).

[0222] Output: The ratio of N, P, and K demand (2-1-1).

[0223] 3. Control command generation:

[0224] Choose the fertilization pattern according to the crop type (e.g., growth-promoting N:P:K=3:1:2).

[0225] (4) Precise execution phase (<100ms)

[0226] 1. Hydraulic pump speed regulation:

[0227] The target flow rate is 5L / min, and the pump speed is adjusted to 1500rpm (PID control).

[0228] 2. Proportional valve opening adjustment:

[0229] Based on feedback from the Coriolis flow meter, the valve opening is dynamically adjusted (0-100%).

[0230] 3. Nozzle PWM control:

[0231] For example: Nozzle 1 duty cycle = 30%, opening time 3ms / cycle.

[0232] (5) Exception handling

[0233] Sensor failure: Switch to backup sensor when data exceeds the limit 3 times consecutively.

[0234] Communication interruption: Switch to redundant channel when CAN bus error rate > 10%.

[0235] 3. Specific operating steps

[0236] 1. Preliminary preparations:

[0237] The multispectral imager 2 is mounted on the front of the agricultural machinery using the fixing device 4, and its downward angle is set to 30° using the angle control device 1 to ensure that the field of view covers a width of 3m.

[0238] Add ISO VG46 anti-wear hydraulic oil to the hydraulic system and adjust the pressure to 15MPa.

[0239] 2. Field operations:

[0240] The agricultural machinery travels along the preset path (AB line), and the straight-line error guided by RTK-GPS is less than 5cm.

[0241] When an area is detected to have an NDVI*=0.55 (nitrogen deficiency), nitrogen fertilizer will be automatically added by 10%.

[0242] 3. Data Recording:

[0243] Every 10 minutes, the operation data (including fertilizer application amount, location, and sensor status) is uploaded to the agricultural cloud platform via 5G.

[0244] 4. System workflow and collaboration mechanism

[0245] In actual operation, each subsystem operates efficiently through a precise collaborative mechanism. When the fertilizer applicator begins operation, the GPS positioning system first determines the boundaries and location information of the work area. As the machine moves, multispectral sensors and near-infrared soil probes simultaneously collect crop and soil data, while the environmental monitoring unit records environmental parameters in real time. After preprocessing, all data is input into the intelligent decision-making system. The system combines historical data and real-time monitoring data of the current work location to generate a fertilizer application control command for that location. Upon receiving the command, the variable actuator adjusts the metering pump speed and nozzle opening to achieve precise fertilization. Throughout the process, the system collects data and updates decisions at a frequency of 10Hz, ensuring that the fertilizer application rate dynamically adjusts according to changes in crop needs. This invention also incorporates an anomaly handling mechanism. When a sensor malfunction or actuator malfunction is detected, the system automatically switches to a safe mode to avoid erroneous fertilization. After the operation is completed, the system automatically generates a fertilizer prescription map and an operation report, providing data support for agricultural management.

[0246] The control valve group of the variable actuator is installed on the suspension frame 9, and the PWM nozzle array is deployed on the fertilizer beam. It receives commands through the CANopen protocol and simultaneously feeds back the execution status through dual channels. Each module is connected to achieve anti-interference through shielded cables, forming a closed-loop control link of "data acquisition-edge computing-precise execution", with the overall delay controlled within 150ms.

[0247] Through the implementation of the above technical solution, this invention achieves true variable-rate precision fertilization, solving many problems existing in traditional fertilization techniques. The fusion and acquisition of multi-source data ensures comprehensive information, the intelligent decision-making system improves the scientific nature of fertilization, and the high-precision actuator guarantees precise control. These three elements work together to form a complete precision fertilization solution. Practical application shows that this technology can increase fertilizer utilization to over 65% while reducing fertilizer waste by more than 30%, resulting in significant economic and ecological benefits. The modular design of this invention also facilitates functional expansion and maintenance upgrades, providing a new technical path for the development of intelligent agricultural equipment. The modular design of this invention supports future upgrades, such as adding drone-assisted operations or AI-based visual weed recognition functions, providing a complete solution for precision agriculture.

[0248] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Furthermore, although some specific terms are used in this specification, these terms are merely for convenience of explanation and do not constitute any limitation on the present invention.

Claims

1. A crop variable fertilization control device based on a multispectral sensor, characterized in that, The system includes an onboard control terminal, a multi-source data acquisition module, an intelligent decision-making system, and a variable actuator. The onboard control terminal, serving as the core, is installed in a waterproof electrical box behind the agricultural machinery cab. Within the multi-source data acquisition module, a multispectral imager is fixed at the front and directly connected to the control terminal via gigabit Ethernet; near-infrared soil sensors are positioned at three points on the chassis and connected via a CAN FD subnet; and environmental sensors are integrated on the top of the cab and communicate via RS-485. The intelligent decision-making system is embedded in the computing unit of the control terminal, acquiring data and outputting decision commands via a PCIe channel. The control valve assembly of the variable actuator is mounted on the suspension frame, and the PWM nozzle array is deployed on the fertilization beam. It receives commands via the CANopen protocol and simultaneously provides feedback on the execution status through dual channels. The vehicle-mounted control terminal, as the core processing unit of the system, adopts an industrial-grade embedded processor to process multi-channel sensor data in real time, execute control algorithms, and perform real-time task scheduling. The multi-source data acquisition module is used for crop information acquisition, soil testing, environmental monitoring, and data synchronization and quality control. The intelligent decision-making system is used to preprocess data, normalize and extract features from multi-source data, comprehensively evaluate crop nutrition status through fuzzy logic algorithm, output nutrition requirement level, and generate optimal fertilization plan based on deep reinforcement learning algorithm, taking into account crop growth stage, current nutrition status, soil nutrient level and environmental conditions. The variable execution mechanism is used to achieve precise fertilization control.

2. The apparatus according to claim 1, characterized in that, It also includes a communication system that uses the CAN2.0B bus and supports the ISO 11783 protocol to ensure compatibility with agricultural machinery.

3. The apparatus according to claim 1, characterized in that, The vehicle control terminal includes a data fusion processing layer, which uses an extended Kalman filter (EKF) for spatiotemporal registration of multi-source data and fuses GPS, IMU, and multispectral data to ensure data consistency.

4. The apparatus according to claim 1, characterized in that, It also includes a wireless communication module, which uses the Quectel RM500Q 5G module, supports SA / NSA dual-mode, has a maximum downlink rate of 2.5Gbps, and has built-in GNSS to ensure remote monitoring and real-time data upload.

5. The apparatus according to claim 1, characterized in that, The multi-source data acquisition module uses a high-resolution multispectral imager during crop information acquisition to simultaneously acquire reflectance data of the crop canopy in multiple bands, extract NDVI, PRI, and NDRE vegetation indices, and comprehensively characterize the chlorophyll content, photosynthetic activity, and nitrogen status of crops. During soil testing, a contact near-infrared soil probe is used, and during environmental monitoring, temperature and humidity sensors and light intensity sensors are used.

6. The apparatus according to claim 1, characterized in that, The variable actuator includes an electromechanical-hydraulic integrated variable actuator system, which includes a fertilizer storage tank, a metering pump, a flow control valve, and a spraying device.

7. The apparatus according to claim 6, characterized in that, The electromechanical-hydraulic integrated variable actuator system also includes a pressure compensation device.

8. The apparatus according to claim 7, characterized in that, The variable actuator is equipped with dual CAN buses and mechanical backup.

9. A method for installing a crop variable fertilization control device based on a multispectral sensor as described in any one of claims 1-5, characterized in that, Includes the following steps: Step (1) Sensor module installation Multispectral imager: Installed at the front of the agricultural machinery, the downward viewing angle is adjusted to 30° via an angle control device; Near-infrared soil probe: Installed on the chassis of agricultural machinery, it adopts a hydraulic lifting mechanism. When working, the probe is pressed into the soil, and when not working, it automatically rises to avoid obstacles. Environmental sensors: integrated into the top of the agricultural machinery cab to avoid dust interference; Step (2), Installation of Control and Execution System Vehicle-mounted control terminal: It is fixed to the rear of the agricultural machinery through a waterproof enclosure and communicates with the agricultural machinery ECU via a CAN bus; PWM nozzle array: evenly distributed along the crossbeam of the fertilizer applicator, each nozzle is independently controlled, supporting liquid / granular fertilizer spraying; Redundancy design: The critical circuit adopts dual CAN bus, and the main controller and the backup PLC synchronize data in real time.

10. The installation method according to claim 9, characterized in that, In step (1), the multispectral imager is fixed to the front of the agricultural machine by a fixing device, with a height of 1.5m above the ground and a field of view covering an operating width of 5m. The near-infrared soil probe uses a hydraulic lifting mechanism with a stroke of 50cm. In step (2), the PWM nozzle array is evenly distributed along the crossbeam of the fertilizer applicator with a spacing of 15cm.

Images