Farm meteorological and soil environment monitoring automatic inspection unmanned vehicle based on AIoT

Abstract

The invention discloses an AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle, which integrates an all-terrain adaptive mobile platform, a multi-mode mechanical arm and an intelligent environment sensing module to construct a whole-process unmanned operation system: autonomous obstacle avoidance and path planning of a complex farmland are realized through laser navigation and a three-dimensional environment sensing technology; a switchable execution mechanism is integrated at the tail end of the mechanical arm, accurate loading, unloading and transferring of a fruit basket are completed by utilizing a self-adaptive clamping module, and soil in-situ sampling is realized by combining a soil probe; a multispectral visual system is carried, crop maturity and pest and disease distribution are synchronously analyzed based on a deep learning algorithm, and canopy microclimate data are dynamically acquired in cooperation with a liftable environment monitoring device. The technical bottleneck that traditional agricultural equipment is single in function and poor in collaboration is broken through, intelligent collaboration of efficient turnover of fruit baskets, lossless sampling of soil and multi-dimensional monitoring of the growth environment is achieved, and an all-weather closed-loop solution is provided for precision agriculture.

Description

Technical Field

[0001] The present invention relates to the technical fields of agricultural automation technology and intelligent robot technology, and particularly relates to an automatic inspection unmanned vehicle for farm meteorological and soil environment monitoring based on AIoT. Background Technique

[0002] AIoT (Artificial Intelligence of Things) = AI (Artificial Intelligence) + IoT (Internet of Things). AIoT integrates AI technology and IoT technology. Through the Internet of Things, a large amount of data from different dimensions is generated and collected, stored in the cloud and edge devices, and then through big data analysis and higher forms of artificial intelligence, everything is digitalized and everything is intelligently connected. With the deep integration of the Internet of Things, artificial intelligence and robot technology, agricultural inspection equipment is evolving towards all-terrain adaptation, multi-task collaboration and all-weather operation. The unmanned inspection system can effectively replace manual labor to complete repetitive tasks such as orchard inspection, fruit basket replacement, and soil sampling by integrating high-precision navigation, intelligent robotic arms and multi-modal sensing technology, significantly improving the standardization and data level of agricultural production, and promoting the leapfrog development of precision agriculture towards the intelligent operation and maintenance stage.

[0003] Existing agricultural inspection robots only use lidar for navigation and do not integrate GPS data. Therefore, the positioning error in open farmland may reach or exceed 30 centimeters. Moreover, the robotic arms of existing picking robots have a single function and only support clamping operations, and cannot meet the dual task requirements of fruit basket replacement and soil detection at the same time. Summary of the Invention

[0004] The present invention provides an automatic inspection unmanned vehicle for farm meteorological and soil environment monitoring based on AIoT, which solves the problems of low efficiency, scattered data, insufficient collaboration ability and energy dependence in traditional agricultural inspections.

[0005] To achieve the above object, the present invention provides the following technical solution: An automatic inspection unmanned vehicle for farm meteorological and soil environment monitoring based on AIoT, comprising: A vehicle body, on the top of which a lidar, a GNSS positioning module, a binocular multi-spectral inspection camera, a six-degree-of-freedom robotic arm and a three-stage telescopic meteorological pole are integrated; A binocular depth camera and a battery are arranged at the front part of the upper layer of the vehicle body, and a solar inverter and an ultrasonic radar are installed at the lower layer; Full-terrain adaptive crawlers are configured on both sides of the vehicle body, and an extension tray is provided at the tail, and the upper end of the tray is used to carry empty and full fruit baskets; The end effector of the robotic arm includes an arc-shaped locking jaw and a soil probe, which are respectively used for the loading and unloading operation of the fruit basket and the collection of soil data; A navigation system that constructs an environmental map and realizes autonomous obstacle avoidance by using the SLAM algorithm through integrating the point cloud data of a lidar, the positioning signal of a GNSS positioning module, the stereo vision data of a binocular depth camera, and the detection data of an ultrasonic radar; Adjust the height through the three-stage telescopic meteorological pole to collect environmental parameters of different vertical layers; An energy system, including a solar panel arranged on one side wall of the three-stage telescopic meteorological pole and a lithium iron phosphate battery on the upper layer of the vehicle body, used to supply power to each module; An Internet of Things transmission system, integrated inside the upper layer of the vehicle body, including a multi-mode communication baseband chip and an encryption coprocessor; A data link system that synchronizes the data collected by the lidar, the binocular multi-spectral inspection camera, the robotic arm, and the three-stage telescopic meteorological pole to the cloud platform in real time to form a full-element farm environment monitoring system.

[0006] Preferably, the end effector of the robotic arm is driven by a motor, and the arc-locking gripper is internally provided with a torque sensor to dynamically adjust the clamping force through closed-loop control, and schedule the loading and unloading tasks of the robotic arm according to the full-load state priority of the fruit basket; the soil probe shares the drive shaft system with the gripper, and switches the clamping state and the soil sampling state through the forward and reverse rotation of the motor; when the meteorological data exceeds the preset threshold, trigger the soil sampling action; complete the obstacle avoidance operation and the traversal of environmental data collection points synchronously during the path planning process.

[0007] Preferably, when the robotic arm is in the clamping state, the soil probe is stored in the gripper base; when switching to the soil sampling state, the soil probe vertically inserts into the soil and collects temperature, humidity, and conductivity data through the soil temperature and humidity sensor provided on the gripper base.

[0008] Preferably, the lidar in the navigation system is used to generate an environmental point cloud map with centimeter-level accuracy; the GNSS positioning module realizes the spatio-temporal alignment of agricultural situation data and the GIS map, and provides absolute positioning data in the global coordinate system through the GNSS positioning module; the provided binocular depth camera constructs a three-dimensional contour of obstacles through stereo vision algorithms, and cooperates with the ultrasonic radar to detect obstacles in the near-field blind area of the vehicle body, and through an improved SLAM algorithm, dynamically removes dynamic obstacles in real time and generates an optimized path.

[0009] Preferably, the three-stage telescopic meteorological pole includes a modular lifting mechanism with a pulley block and a redundant safety locking device built therein; high-strength fiber cables for controlling the gradual extension of the meteorological pole; and a micro meteorological station module integrated at the upper end of the three-stage telescopic meteorological pole body, including a gas analysis unit, an air temperature and humidity sensor, and an ultrasonic wind speed and direction sensor, for collecting environmental parameters of different vertical layers, including carbon dioxide concentration, temperature, humidity, light intensity, wind speed, and wind direction. The data collected by the meteorological pole at different height layers is bound with time stamps and geographical coordinates for constructing a spatio-temporal evolution model of the farm microclimate.

[0010] Preferably, the energy system includes: Solar panels that achieve maximum power point tracking through the MPPT algorithm; Lithium iron phosphate batteries equipped with a low-temperature self-heating circuit, with an operating temperature range of -20°C to 50°C; The energy system can continuously operate for no less than 72 hours under non-sunshine conditions.

[0011] Preferably, the data link system includes: A LoRa wireless communication module for transmitting preprocessed environmental monitoring data; An edge computing unit for locally compressing and encrypting multi-spectral images, meteorological data, and robotic arm operation data; A cloud platform for real-time display of visual data on soil moisture, crop growth, and pest distribution.

[0012] Preferably, the Internet of Things transmission system includes: A 5G micro base station module that supports dual-band carrier aggregation to achieve the transmission of 4K multi-spectral video streams within a radius of 2 kilometers; LoRa / NB-IoT multi-hop self-organizing network, which relays data transmission through adjacent nodes in signal blind areas, with a maximum number of hops not exceeding 5 and a packet loss rate lower than 3%; A Beidou short message module that sends soil moisture and pest level data every 30 seconds; An SM9 encryption communication unit that establishes an end-to-end secure channel and supports dynamic MTU adjustment and a low-latency transmission mode at the 50ms level.

[0013] Preferably, the Internet of Things transmission system further includes a multi-spectral image background filtering module deployed in the edge computing unit, with the YOLO-Agri algorithm built therein. Through transfer learning, the accuracy of agricultural scene background segmentation is optimized, the false alarm rate of pest identification is reduced, and real-time background frame filtering of multi-spectral images is performed through the YOLO-Agri algorithm.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. The present invention is equipped with a multi-source fusion navigation system, which integrates the laser radar SLAM point cloud data with the global positioning of the GPS module, combines the binocular depth camera stereo vision and the ultrasonic radar near-field obstacle avoidance, and adopts the improved LOAM+ SLAM algorithm to achieve centimeter-level mapping accuracy, and the dynamic path planning response delay is ≤200ms.

[0015] 2. The present invention realizes the collaborative operation of intelligent robotic arms. The end of the six-degree-of-freedom robotic arm integrates a dual module of an arc-shaped locking clamp and a soil probe. The operation mode switching is realized based on motor control. The arc-shaped locking clamp has a built-in torque sensor and a PID force control algorithm to dynamically adjust the clamping force to avoid damage to the fruit basket due to slippage.

[0016] 3. The present invention realizes layered meteorological monitoring. The three-stage telescopic meteorological pole uses high-strength fiber cables to extend step by step to the crop canopy, integrates a gas analysis unit, a temperature and humidity sensor, and an ultrasonic wind speed and direction meter, and uses the Kalman filter algorithm to fuse layered data to predict microclimate trends.

[0017] 4. The present invention optimizes environmental perception. The binocular multispectral inspection camera is equipped with an improved YOLOv11 model, and SIoU Loss and Inner-CIoU Loss are used for collaborative optimization, which reduces the missed detection rate of pests and diseases identification by 60% and achieves a maturity grading accuracy of 96.7%.

[0018] 5. The present invention sets up an Internet of Things transmission architecture, integrates a 5G micro base station and a Beidou / LoRa three-mode communication module, and realizes real-time backhaul of 4K video streams in farmland with a radius of 2 kilometers based on 3.5GHz+700MHz carrier aggregation technology. The signal blind area automatically switches to the LoRa multi-hop self-organizing network. In extreme environments, the core parameters of soil moisture conditions are sent at a frequency of 30 seconds / time through the Beidou short message module. Combined with dynamic MTU adjustment and SM9 national secret algorithm encryption, the transmission efficiency in open farmland multipath environment is improved by 23%.

[0019] 6. The present invention realizes closed-loop data management. Based on the 5G / LoRa heterogeneous communication network and edge computing unit, it synchronizes soil moisture conditions, meteorological data and inspection instructions in real time, covering an average of 20 acres of farmland per day. The efficiency of fruit basket loading and unloading is increased by 6 times compared with manual operation, forming an unmanned operation system with the entire process of "perception-decision-execution-transmission". BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0021] In the attached picture: Figure 1 It is a front view structural diagram of the whole unmanned vehicle of the present invention; Figure 2It is a side view structure diagram of the overall unmanned vehicle of the present invention; Figure 3 It is a schematic structural diagram of the robotic arm of the present invention; Figure 4 It is a schematic structural diagram of the weather detection module of the present invention; Figure 5 It is a schematic diagram of the improved yolov11 framework of the present invention; Figure 6 It is a schematic diagram of the target detection network architecture of the present invention; Figure 7 It is a schematic structural diagram of the PID force control system of the present invention; Figure 8 It is a schematic diagram of the path planning algorithm of the present invention; Figure 9 It is a schematic diagram of the environmental monitoring network model of the present invention; Figure 10 It is a schematic diagram of the working process of the intelligent unmanned vehicle of the present invention; Figure 11 It is a schematic diagram of the transmission process of the farm Internet of Things of the present invention; Reference numerals in the figure: 1, vehicle body; 2, solar panel; 3, binocular depth camera; 4, ultrasonic radar; 5, lidar; 6, robotic arm; 7, weather pole; 8, binocular multispectral inspection camera; 9, GNSS positioning module; 10, crawler; 11, fruit basket; 12, gripper; 13, motor; 14, soil probe; 15, soil temperature and humidity sensor; 16, gas analysis unit; 17, ultrasonic wind speed and direction sensor; 18, air temperature and humidity sensor; 19, battery; 20, inverter; 21, 5G micro base station module. Specific embodiments

[0022] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not intended to limit the present invention.

[0023] As Figures 1 - 4 shown, the embodiment of the present invention discloses an AIoT-based automatic inspection unmanned vehicle for farm weather and soil environment monitoring, including: a vehicle body 1, on the top of which a lidar 5, a GNSS positioning module 9, a binocular multispectral inspection camera 8, a six-degree-of-freedom robotic arm 6 and a three-stage telescopic weather pole 7 are integrated; a binocular depth camera 3 and a battery 19 are arranged at the front part of the upper layer of the vehicle body 1, and a solar inverter 20 and an ultrasonic radar 4 are installed at the lower layer; all-terrain adaptive crawlers 10 are configured on both sides of the vehicle body 1, and an extension tray is arranged at the tail, and the upper end of the tray is used to carry empty and full fruit baskets 11; The end effector of the robotic arm 6 includes an arc-locking gripper 12 and a soil probe 14, which are respectively used for the loading and unloading operations of the fruit basket 11 and soil data collection; the end effector of the robotic arm 6 is driven by a motor 13. The arc-locking gripper 12 is internally provided with a torque sensor, and the clamping force is dynamically adjusted through closed-loop control. The loading and unloading tasks of the robotic arm 6 are scheduled according to the full-load state priority of the fruit basket 11; the soil probe 14 shares the drive shaft system with the gripper 12, and the clamping state and soil sampling state are switched by the forward and reverse rotation of the motor 13; when the meteorological data exceeds the preset threshold, the soil sampling action is triggered; during the path planning process, obstacle avoidance operations and traversal of environmental data collection points are completed synchronously; when the robotic arm 6 is in the clamping state, the soil probe 14 is stored in the base of the gripper 12; when switched to the soil sampling state, the soil probe 14 is vertically inserted into the soil, and temperature, humidity and conductivity data are collected through a soil temperature and humidity sensor 15 provided on the base of the gripper 12.

[0024] The navigation system constructs an environmental map and realizes autonomous obstacle avoidance by fusing the point cloud data of the lidar 5, the positioning signal of the GNSS positioning module 9, the stereo vision data of the binocular depth camera 3 and the detection data of the ultrasonic radar 4, and adopting the SLAM algorithm; the lidar 5 in the navigation system is used to generate an environmental point cloud map with centimeter-level accuracy; the GNSS positioning module 9 realizes the spatio-temporal alignment of agricultural situation data and the GIS map, and provides absolute positioning data in the global coordinate system through the GNSS positioning module 9; the provided binocular depth camera 3 constructs a three-dimensional contour of the obstacle through the stereo vision algorithm, and cooperates with the ultrasonic radar 4 to detect obstacles in the near-field blind area of the vehicle body 1, and an improved SLAM algorithm is used to continuously eliminate dynamic obstacles and generate an optimized path.

[0025] The three-stage telescopic meteorological pole 7 can adjust its height to collect environmental parameters of different vertical layers; the three-stage telescopic meteorological pole 7 includes a modular lifting mechanism, which is internally provided with a pulley group and a redundant safety locking device; a high-strength fiber cable is used to control the gradual extension of the meteorological pole 7; through a micro-meteorological station module integrated at the upper end of the three-stage telescopic meteorological pole 7, including a gas analysis unit 16, an air temperature and humidity sensor 18 and an ultrasonic wind speed and direction sensor 17, environmental parameters of different vertical layers are collected, including carbon dioxide concentration, temperature, humidity, light intensity, wind speed and wind direction; the data collected by the meteorological pole 7 at different height layers are bound with time stamps and geographical coordinates, and are used to construct a spatio-temporal evolution model of the farm microclimate.

[0026] The energy system includes a solar panel 2 installed on one side wall of the three-stage telescopic weather pole 7 and a lithium iron phosphate battery 19 on the upper layer of the vehicle body 1, which is used to supply power to each module. The energy system includes: a solar panel 2 that realizes maximum power point tracking through the MPPT algorithm; a lithium iron phosphate battery 19 equipped with a low-temperature self-heating circuit, and the operating temperature range is -20°C to 50°C; the energy system can continuously operate for no less than 72 hours under the condition of no sunlight.

[0027] The Internet of Things transmission system is integrated inside the upper layer of the vehicle body 1 and includes a multi-mode communication baseband chip and an encryption coprocessor. The Internet of Things transmission system adopts a 5G + LoRa + Beidou three-mode communication architecture, including: a 5G micro base station module 21 integrated inside the upper layer of the vehicle body 1, based on the 3.5GHz and 700MHz dual-band carrier aggregation technology, supporting dual-band carrier aggregation, and realizing the transmission of 4K multi-spectral video streams (≥800Mbps) within a radius of 2 kilometers; LoRa / NB-IoT multi-hop self-organizing network, which performs adjacent node relay transmission through adjacent agricultural machinery or sensor nodes in signal blind areas, with a maximum number of hops not exceeding 5 and a packet loss rate lower than 3%; the Beidou short message module is enabled in extreme environments to send core agronomic parameters (soil moisture, pest and disease levels) at a frequency of 30 seconds / time; the Internet of Things transmission system also includes a multi-spectral image background filtering module deployed in the edge computing unit, with the YOLO-Agri algorithm built-in, optimizing the accuracy of agricultural scene background segmentation through transfer learning, reducing the false alarm rate of pest and disease identification, and the system is built with an agricultural-specific AI chip to perform real-time background frame filtering on multi-spectral images through the YOLO-Agri algorithm; combining GNSS / INS combined positioning to complete the spatio-temporal alignment of agricultural situation data and GIS maps, and intelligently pre-caching high-definition maps of key areas based on the historical bandwidth prediction model; data transmission uses the national secret SM9 algorithm to establish an end-to-end encrypted channel, and improves the effective throughput by 23% in the open space multipath environment through dynamic MTU adjustment (128~1420 bytes), and automatically triggers a 50ms-level low-latency transmission mode during the sowing / harvesting stage to form a hierarchical guarantee for the security pipeline of agricultural data.

[0028] The data link system is composed of a LoRa wireless communication module and an edge computing unit. The data link system pre-processes the data collected by the lidar 5, binocular multi-spectral inspection camera 8, robotic arm 6, and three-stage telescopic weather pole 7 locally, and uploads it to the cloud agricultural big data platform through an encryption protocol to realize real-time visual monitoring of soil moisture, crop growth, pest and disease distribution, and environmental risks, and issues agricultural instructions such as irrigation and fertilization to form a "perception - decision - execution" closed-loop management; forming a full-element farm environment monitoring system.

[0029] As Figures 5 - 11 shown, the specific algorithm description includes: In the object detection task, the accuracy of bounding box regression directly affects the detection performance. Traditional loss functions mainly focus on the overlapping area, center point distance, and aspect ratio between bounding boxes, but ignore the direction information between bounding boxes. This neglect leads to the prediction box "wandering" around the object during the training process, slowing down the convergence speed and ultimately affecting the model performance.

[0030] To solve this problem, this embodiment introduces the SIoU loss function. SIoU redefines the loss function by introducing an angle penalty term, enabling the model to converge faster and more accurately during the training process. The SIoU loss function consists of four parts: angle loss, distance loss, shape loss, and IoU loss. The core idea of the angle loss is to accelerate the convergence process by reducing the degrees of freedom related to distance. Specifically, the model will first move the prediction box to the nearest coordinate axis and then continue to adjust along that axis. Angle loss is defined as follows: (1) Where: (2) (3) (4) In equation (2) represents the difference between the maximum and minimum values of the coordinate values of the center points of the ground truth box and the prediction box in the y axis direction; α is the angle between the line connecting the center points of the prediction box and the ground truth box, x is an intermediate calculation variable; represents the Euclidean distance between the center point of the ground truth box and the center point of the prediction box, which is used to measure the distance between the center points of the two boxes in the spatial position; When α is close to 0, the angle loss is minimized, and the model preferentially adjusts along the nearest coordinate axis; when α is close to the angle loss is maximized, and the model needs to adjust both coordinate axes simultaneously.

[0031] Distance loss is redefined based on the angle loss, and the formula is as follows: (5) Where: (6) The distance loss takes into account the angle information. When α is close to 0, the contribution of the distance loss decreases; when α is close to the contribution of the distance loss increases.

[0032] Shape loss Used to measure the difference in aspect ratio between the predicted bounding box and the ground truth bounding box, the formula is as follows: (7) Where: (8) is a hyperparameter that controls the weight of the shape loss. Optimized by genetic algorithm, is usually set to around 4.

[0033] The IoU loss is the traditional intersection over union loss, and the formula is as follows: (9) The final SIoU loss function is defined as: (10) By combining these four parts, the SIoU loss function performs excellently in object detection tasks, especially showing significant advantages when dealing with occlusions and multi-scale objects.

[0034] The core idea of the LSKA module is to decompose the traditional two-dimensional depth convolution kernel into two one-dimensional convolution kernels, which perform convolution operations along the horizontal and vertical directions respectively. This decomposition method not only reduces the number of parameters but also lowers the computational complexity.

[0035] Among them, the output formula of the LSKA module is as follows:

[0036]

[0037]

[0038]

[0039] Where, and are the one-dimensional convolution kernels in the horizontal and vertical directions respectively, is the dilation rate, is the convolution kernel size, represents element-wise multiplication.

[0040] Through this decomposition design, the LSKA module significantly reduces the computational complexity and memory occupancy while maintaining high performance, and is suitable for object detection tasks in large-scale images and complex backgrounds.

[0041] In object detection tasks, the accuracy of bounding box regression directly affects the detection performance. Although the traditional IoU loss function can effectively measure the overlap between the predicted box and the ground truth box, it has problems such as slow convergence speed and insufficient generalization ability during training. Therefore, in this embodiment, the Inner-IoU loss function is proposed, which accelerates the bounding box regression process by introducing auxiliary bounding boxes and improves the generalization ability of the model in different detection tasks.

[0042] The core idea of Inner-IoU is to dynamically optimize the calculation method of the loss function by adjusting the scale of the auxiliary bounding box. Specifically, for high IoU samples, using a smaller auxiliary bounding box to calculate the loss can accelerate convergence; while for low IoU samples, using a larger auxiliary bounding box to calculate the loss can expand the regression range and improve the regression effect.

[0043] Generation of auxiliary bounding boxes: Given the ground truth box and the predicted box , Inner-IoU controls the scale of the auxiliary bounding box through the scaling factor . The center point of the auxiliary bounding box is the same as that of the original box, but its width and height are scaled proportionally:

[0044]

[0045] where is the center point of the ground truth box, and are the width and height of the ground truth box respectively. The generation method of the auxiliary bounding box for the predicted box is similar.

[0046] Calculation of Inner-IoU: Calculate the IoU value through the auxiliary bounding box, and the formula is as follows:

[0047]

[0048]

[0049] Finally, the Inner-IoU loss function is defined as:

[0050] By introducing auxiliary bounding boxes, Inner-IoU significantly improves the efficiency and accuracy of bounding box regression, and is particularly suitable for dealing with detection tasks of multi-scale objects and complex backgrounds.

[0051] Ablation experiment comparison Network model Accuracy / % Recall / % F1-score / % Average precision / % Model size / MB YOLOv11 82.9 79.4 81.0 82.9 5.19 YOLOv11 + Inner-IoU 84.1 79.2 81.5 87.3 5.19 YOLOv11 + Inner-IoU + SIou 84.4 79.5 81.9 87.8 5.19 YOLOv11 + Inner-IoU + LSKA 83.6 80.8 82.1 88.1 6.87 YOLOv11 + Inner-IoU + SIou + LSKA 87.5 85.7 86.6 92.4 6.87 Comparison of Detection Performance of Different Models Network model Accuracy / % Recall / % F1-score / % Average precision / % Model size / MB YOLOv5 80.7 77.7 79.1 84.7 5.00 YOLOv8 85.2 81.8 83.4 89.2 5.93 YOLOv11 82.9 79.4 81.0 82.9 5.19 YOLOv11 + Inner-IoU + SIou + LSKA 87.5 85.7 86.6 92.4 6.87 The gripper PID force control algorithm synergistically adjusts the clamping force through the proportional term ( ), integral term ( ), and derivative term ( ) to achieve precise grasping by dynamically regulating the clamping force.

[0052] Proportional term ( = 50 N / m): It responds in real-time to the deviation of the clamping force (e). For example, when the fruit basket 11 slides, it instantaneously increases the clamping force to quickly correct the error. Its principle is based on the linear amplification of the current error, similar to the logic of immediately adjusting the water flow to approach the target water level in "filling a water tank".

[0053] Integral term ( = 5 N / (m·s)): By accumulating the historical error ( ), it eliminates the slow slip caused by insufficient long-term static friction. Especially in high-humidity environments, it prevents the fruit basket 11 from slipping out of the hand due to continuous slipping. The integral term is equivalent to "gradually finding the feeling of filling water" and eliminates the steady-state error through small adjustments.

[0054] Derivative term ( = 10 N·s / m): It predicts the inertial motion trend of the fruit basket 11 ( ) and suppresses the clamping force fluctuation in advance. For example, when the fruit basket 11 suddenly accelerates, it anticipates and cancels the inertial influence through the error change rate. Its function is similar to the damping control of "anticipating the trend of water level change", and the formula is as follows:

[0055] The cruise A* path planning algorithm optimizes the global path planning in the farmland environment through a multi-objective cost function. The specific process is as follows: After generating the initial path based on the Dijkstra algorithm, a terrain ruggedness index is introduced to dynamically adjust the path weight. The terrain ruggedness calculates the height variance (Formula 1) of each grid (1m×1m) through the lidar 5-point cloud data. If the variance exceeds the threshold (>0.1 m²), it is marked as an impassable area; the topological map nodes are defined as path inflection points (curvature change >15° / m) or key agricultural nodes (such as irrigation points), and the edge weight is calculated by weighting the path segment length (70% weight) and the maximum height variance (30% weight) (Formula 2); the dynamic replanning trigger conditions include the appearance of new obstacles (point cloud density >200 points / m³ and distance from the path <2m), terrain mutation of adjacent grids (height variance change >0.05 m²), and continuous over-limit of the lateral deviation of path tracking (>0.3m for more than 5 seconds), ensuring the adaptive ability of the unmanned vehicle in complex terrains.

[0056] Formula 1 Terrain Ruggedness Calculation:

[0057] : Variance of grid height (unit: m²), quantifying the degree of terrain undulation. When > 0.1 m², it is determined as impassable.

[0058] : Number of valid laser points in the grid (≥ 50 points). When insufficient, it is regarded as an invalid grid.

[0059] : Z - axis coordinate of a single point (in the vehicle body coordinate system, unit: meter).

[0060] : Average height of the grid (unit: meter).

[0061] Formula 2 Topological Edge Weight Formula:

[0062] : Edge traversal cost weight (dimensionless). The larger the value, the higher the traversal difficulty.

[0063] : Edge length (unit: meter).

[0064] : Maximum height variance of the area covered by the edge (unit: m²).

[0065]

[0066] : Difference in height variance between adjacent grids (unit: m²). When > 0.05 m², a terrain risk warning is triggered.

[0067] Lateral Deviation: Lateral offset between the actual position and the planned path during path tracking (unit: meter).

[0068] Obstacle Density: Number of points per unit volume after point cloud clustering (unit: points / m³).

[0069] The LOAM + SLAM mapping algorithm realizes high - precision 3D mapping and positioning in the farmland environment by fusing lidar 5 and IMU data: First, the motion distortion compensation formula is used to eliminate the point cloud distortion caused by vehicle body jolts, where ∈ SE(3) represents the pose transformation matrix at time k (3D rigid body transformation), and are the angular velocity and angular acceleration measured by the IMU respectively, = 0.05 s is the time interval between 5 frames of the lidar; then, the pose estimation is optimized through the feature matching error function where is the edge point of the current frame (curvature > 0.1), is the adjacent plane point in the map (curvature < 0.05), and T ∈ SE(3) is the pose matrix to be optimized; the objective function of the backend optimization combines the lidar matching residual and the GPS position residual to dynamically adjust the weights (70% for lidar and 30% for GPS) in the RTK signal locked state. The optimization convergence condition is that the pose change ΔT < (translation < 0.1 mm, rotation < 0.001°) or a maximum of 50 iterations (based on Ceres Solver). Finally, centimeter-level mapping accuracy (error ≤ 0.1 m / 100 m) and real-time dynamic obstacle tracking ability (response delay ≤ 200 ms) in the farmland scenario are achieved.

[0070] This algorithm realizes the high-precision real-time monitoring of farmland environmental parameters through the cooperation of dielectric sensors and meteorological sensors: the soil volumetric water content ( ) is calculated based on the linear interpolation method of dielectric constant (Formula 1). When the detected regional humidity deviation exceeds 20% of the historical mean, 1 m × 1 m encrypted sampling is triggered, up to 5 times; the meteorological data fusion adopts Kalman filtering (Formulas 2 and 3). The microclimate trend is predicted through the state vector (temperature, humidity, wind speed, and their change rates), and is dynamically corrected in combination with the measured values of the sensors. Finally, the temperature in the next 5 minutes is predicted based on the historical temperature change rate (α) and the wind speed-humidity coupling factor (β), realizing the real-time analysis and risk warning of farmland environmental data.

[0071] Formula 1 for soil humidity calculation: 100% : volumetric water content (%), ranging from 0% (dry) to 100% (saturated) : soil dielectric constant (measured value of GS3 sensor) 4.0 / 36.0: dry / saturated soil dielectric constant reference value Formula 2 for meteorological Kalman filtering: Prediction: ,

[0072] Update: ,

[0073] : State vector (temperature / humidity / wind speed, unit: °C / % / m / s) Q = 0.01: Process noise covariance (model uncertainty) R = : Observation noise covariance (temperature sensor accuracy ±0.1 °C) : Measured value of the sensor Formula three microclimate prediction:

[0074] : Predicted temperature (°C) after 5 minutes : Current wind speed (m / s) : Current humidity (%), a humidity-temperature coupling correction is triggered when the threshold of 80% is reached. The multi-modal transmission optimization algorithm (MTOA) improves the communication efficiency of the farm through dynamic bandwidth allocation: construct a 5G / LoRa dual-mode channel state matrix (m = 16 frequency bands, n = 5 time slots), adopt the ε-greedy strategy for path selection, and the reward function . When the channel quality degradation (bit error rate > 1E-5) is detected, it automatically switches to the LoRa multi-hop transmission mode, and establishes the optimal relay path through the improved AODV protocol. Data compression adopts the agricultural feature coding framework (AFCE), performs row-by-row DCT transformation on 15-channel multi-spectral images (compression ratio 300:1, PSNR ≥ 40 dB), and reduces the dimension through principal component analysis (PCA) to retain 98.6% of the effective spectral information. In the actual measurement of 30,000 mu of wheat fields, this algorithm increases the 5G coverage radius to 1.8 times that of the conventional environment, and reduces the invalid data transmission volume by 76%.

[0075] During specific operations, the all-terrain adaptive tracked 10 chassis is adopted for the AIoT-based automatic inspection unmanned vehicle for farm meteorological and soil environment monitoring. A six-degree-of-freedom robotic arm 6 is integrated on the top of the vehicle body. The end effector realizes seamless switching between fruit basket 11 grasping and soil sampling through an integrated dual-module design: The arc-locking gripper 12 matches the shape of the standard fruit basket 11 based on a bionic curved surface structure, and a torque sensor is built-in to real-time feedback the clamping force signal. When the binocular depth camera identifies the position of the fruit basket 11, the robotic arm 6 drives the arc-locking gripper 12 to accurately position at a constant speed and dynamically adjust the clamping force through inverse kinematics calculation. The initial clamping force is maintained stable through closed-loop control. When the fruit basket 11 is detected to slide, the clamping force is instantaneously increased and the vibration suppression algorithm is triggered to ensure damage-free loading and unloading of the fruit basket 11; after grasping, the robotic arm 6 places the full fruit basket 11 on the rear tray and keeps it fixed until the vehicle transports it back to the warehouse for unloading, and the loading and unloading cycle is controlled within ten seconds. The soil probe 14 module integrates temperature, humidity and conductivity sensors. When the system switches to the sampling mode, the motor 13 in the base of the arc-locking gripper 12 drives the probe to insert into the soil to a set depth with a constant vertical pressure. The encoder real-time feedbacks the insertion depth, and the soil temperature and humidity sensor 15 collects data. After sampling, the probe automatically cleans the residual soil on the surface through high-pressure air flow. The three-stage telescopic meteorological pole 7 adopts a modular truss structure and is gradually extended to the crop canopy height by high-strength fiber cables. A gas analysis unit 16, a temperature and humidity sensor 18 and an ultrasonic anemometer and wind vane 17 are deployed on the top of the pole body. The redundant locking mechanism suppresses the wind swing deviation during the lifting process. The navigation and obstacle avoidance system scans the environment through the lidar 5 and constructs a three-dimensional map, fuses the stereo vision data of the binocular depth camera and the global positioning coordinates, and the dynamic programming algorithm real-time optimizes the travel path. When encountering an obstacle, the speed and steering angle are adjusted based on the multi-sensor fusion data, and the lateral deviation control accuracy reaches the centimeter level. The binocular multi-spectral inspection camera 8 collects crop canopy images, identifies the maturity and pest and disease areas through spectral reflectance analysis, and the detection results are compressed and encrypted by the edge computing unit and then uploaded to the cloud decision-making system to trigger irrigation or fertilization instructions. The meteorological data fuses the hierarchical sensor parameters through Kalman filtering, predicts the microclimate trend and triggers soil encrypted sampling. The energy management system dynamically optimizes the task priority and charge and discharge strategy according to the light intensity and the battery 19 status. Multi-vehicle collaboration improves the overall operation efficiency through the task allocation algorithm, forming a full-closed-loop farm intelligent operation and maintenance system.

[0076] The driverless vehicle realizes multi-source data fusion and intelligent decision-making through the edge computing unit: The navigation system fuses the lidar 5 point cloud data with the RTK-GPS coordinates, constructs a centimeter-level environmental map by improving the SLAM algorithm, and dynamically eliminates moving obstacles based on the stereo vision data of the binocular depth camera. The path planning algorithm generates a global path by integrating the terrain ruggedness and task priorities. Local obstacle avoidance optimizes the trajectory in real time within the 0.1-1.5 m / s speed space through the dynamic window method to ensure that the response delay under sudden obstacles is ≤200 ms. In the environmental perception module, the multi-spectral camera images are input into the improved YOLOv11 model, and the small target detection is strengthened through the cross-stage attention mechanism. Combining the HSV color gamut analysis and LBP texture features, the maturity grading and pest and disease distribution heat maps are output. The detection results are encrypted and transmitted to the cloud through the LoRa wireless module to trigger irrigation or fertilization instructions. The meteorological data fuses the hierarchical parameters of the three-stage telescopic rod through Kalman filtering to predict the microclimate trend in the next 5 minutes. When the temperature drops suddenly by ≥2°C or the humidity deviation >20%, the soil probe 14 is automatically started for encrypted sampling, and the sampling data is interpolated by Kriging to generate the soil moisture distribution map. The energy management system dynamically allocates task execution and charging strategies based on the output power of the solar panel and the remaining battery 19 power through reinforcement learning, and gives priority to ensuring the power supply for core tasks in rainy weather. The entire system synchronizes the clocks of each module through the time-sensitive network protocol. The task scheduler realizes the conflict-free coordination of the robotic arm 6 operation, soil sampling, and environmental monitoring based on the status of the fruit basket 11, meteorological anomaly thresholds, and path planning results, and finally forms a "perception-analysis-decision-execution" closed loop, with a daily inspection area of ≥20 mu, and the efficiency is 6 times higher than that of manual work; this driverless vehicle breaks through the technical bottlenecks of traditional agricultural equipment with single functions and poor collaboration, realizes the intelligent coordination of efficient turnover of the fruit basket 11, non-destructive soil sampling, and multi-dimensional monitoring of the growth environment, and provides an all-weather closed-loop solution for precision agriculture.

[0077] Internet of Things Transmission Module Deployment Solution: Deploy 5G micro base stations (transmitting power 46 dBm, coverage radius 2 km) at farm watchtowers and irrigation hubs. The unmanned vehicle is equipped with a triple-mode communication module (5G NR + LoRa SX1302 + Beidou RDSS). During data transmission, the 4K multi-spectral video stream is directly transmitted to the edge cloud through 5G NSA networking, and the soil sampling data is aggregated to the in-vehicle MEC unit for local caching through LoRa self-organizing network (SF = 7, BW = 125 kHz). When the signal strength is lower than -110 dBm, the Beidou RDSS short message transmission mode is triggered (56 bytes / packet, 30-second interval) to send the core parameters of soil moisture (temperature ±0.5°C, humidity ±2%RH). The communication protocol stack adopts a hierarchical encryption design: the application layer data is signed using the SM9 algorithm, the transport layer establishes a DTLS secure channel, and the physical layer implements frequency hopping anti-interference (frequency hopping rate 50 hops / s). After actual measurement, the average daily effective data transmission volume of this system in an open farmland environment reaches 38 GB, and the communication interruption duration is controlled within 4.3 minutes / 24 h.

[0078] Finally, it should be noted that the above are only preferred examples of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle, characterized in that: include: The vehicle body, on top of which is integrated with a laser radar, a GNSS positioning module, a binocular multispectral inspection camera, a six-degree-of-freedom robotic arm, and a three-stage telescopic meteorological mast; The upper front part of the vehicle body is provided with a binocular depth camera and a battery, and the lower part is provided with a solar inverter and an ultrasonic radar; All-terrain adaptive crawlers are arranged on both sides of the vehicle body, and an extended tray is arranged at the rear, and the upper end of the tray is used to carry empty and fully loaded fruit baskets; The end effector of the mechanical arm includes an arc-shaped locking gripper and a soil probe, which are used for loading and unloading of the fruit basket and collecting soil data respectively; The navigation system uses the SLAM algorithm to build an environmental map and achieve autonomous obstacle avoidance by integrating the point cloud data of the lidar, the positioning signal of the GNSS positioning module, the stereo vision data of the binocular depth camera, and the detection data of the ultrasonic radar; The height is adjusted by the three-stage telescopic meteorological mast to collect environmental parameters of different vertical layers; The energy system includes a solar panel installed on one side of the three-stage telescopic meteorological mast and a lithium iron phosphate battery on the upper layer of the vehicle body, which is used to power each module; The IoT transmission system is integrated into the upper layer of the vehicle body and includes a multi-mode communication baseband chip and an encryption coprocessor; The data link system synchronizes the data collected by the lidar, binocular multi-spectral inspection camera, robotic arm, and three-stage telescopic meteorological pole to the cloud platform in real time, forming a full-factor farm environment monitoring system.

2. According to claim 1, the AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle is characterized by: The end actuator of the robotic arm is driven by a motor, and the arc-shaped locking clamp is provided with a torque sensor inside. The clamping force is dynamically adjusted through closed-loop control, and the loading and unloading tasks of the robotic arm are prioritized according to the full load status of the fruit basket; the soil probe and the clamp share a driving shaft system, and the clamping state and the soil sampling state are switched by the forward and reverse rotation of the motor; the soil sampling action is triggered when the meteorological data exceeds the preset threshold; the obstacle avoidance operation and the traversal of the environmental data collection points are completed synchronously during the path planning process.

3. According to claim 2, the AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle is characterized by: When the robotic arm is in a clamping state, the soil probe is stored in the clamp base; when switched to a soil sampling state, the soil probe is vertically inserted into the soil and collects temperature, humidity and conductivity data through a soil temperature and humidity sensor provided on the clamp base.

4. According to claim 1, the AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle is characterized by: The laser radar in the navigation system is used to generate an environmental point cloud map with centimeter-level accuracy; the GNSS positioning module realizes the spatiotemporal alignment of agricultural data and GIS maps, and provides absolute positioning data in the global coordinate system through the GNSS positioning module; the binocular depth camera is used to construct the three-dimensional contour of obstacles through the stereo vision algorithm, and cooperates with the ultrasonic radar to detect obstacles in the near-field blind spot of the vehicle body. Through the improved SLAM algorithm, dynamic obstacles are eliminated in real time and an optimized path is generated.

5. According to claim 1, the AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle is characterized by: The three-stage telescopic meteorological pole includes a modular lifting mechanism, which has a built-in pulley block and a redundant safety locking device; a high-strength fiber cable is used to control the step-by-step extension of the meteorological pole; a micro-meteorological station module integrated at the upper end of the three-stage telescopic meteorological pole body, including a gas analysis unit, an air temperature and humidity sensor, and an ultrasonic anemometer, collects environmental parameters of different vertical layers, including carbon dioxide concentration, temperature, humidity, light intensity, wind speed and wind direction; the data collected by the meteorological pole at different altitudes are bound to timestamps and geographic coordinates to construct a spatiotemporal evolution model of farmland microclimate.

6. The AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle according to claim 1 is characterized by: The energy system comprises: Solar panels, achieve maximum power point tracking through MPPT algorithm; Lithium iron phosphate battery, equipped with low-temperature self-heating circuit, operating temperature range is -20℃ to 50℃; The energy system will operate continuously for no less than 72 hours without sunlight.

7. The AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle according to claim 1 is characterized by: The data link system comprises: LoRa wireless communication module, used to transmit pre-processed environmental monitoring data; The edge computing unit locally compresses and encrypts multispectral images, meteorological data, and robotic arm operation data; The cloud platform displays real-time visualization data on soil moisture conditions, crop growth, and distribution of pests and diseases.

8. The AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle according to claim 7 is characterized by: The IoT transmission system includes: 5G micro base station module, supporting dual-band carrier aggregation and enabling multi-spectral video streaming; LoRa / NB-IoT multi-hop ad hoc network, relay transmission through adjacent nodes in signal blind areas, the maximum number of hops does not exceed 5 and the packet loss rate is less than 3%; SM9 encrypted communication unit, establishes end-to-end secure channel and supports dynamic MTU adjustment.

9. The AIoT-based farm weather and soil environment monitoring automatic inspection unmanned vehicle according to claim 8 is characterized by: The IoT transmission system also includes a multispectral image background filtering module deployed in an edge computing unit, with a built-in YOLO-Agri algorithm, which optimizes the background segmentation accuracy of agricultural scenes through transfer learning and reduces the false alarm rate of pest and disease identification.

Images