An agricultural non-point source pollution open space integrated monitoring system and method based on power transmission tower edge cooperation

Abstract

An integrated air-ground monitoring system and method for agricultural non-point source pollution based on transmission tower edge collaboration is disclosed. The system includes tower edge sensing nodes, UAV collaborative terminals, and a cloud management platform. The tower edge sensing nodes are equipped with self-powered power supply, multi-dimensional sensing, edge computing gateways, and heterogeneous communication modules. The UAV collaborative terminals are equipped with automatic drone nests and inspection UAVs with electromagnetic compatibility design. The cloud management platform realizes functions such as digital twins and data fusion. The monitoring method collects data through the tower edge sensing nodes in a normalized mode triggered by rainfall. The data is initially screened for pollution through an improved MASFNet network built into the edge gateway. When pollution is suspected, the UAV is awakened. An electromagnetic environment adaptive flight path is planned through an electromagnetic-environment integrated potential energy field model. After the UAV collects detailed evidence, it transmits the data back to the gateway and then uploads it to the cloud management platform via OPGW optical fiber. The cloud management platform completes the pollution diffusion display and source tracing analysis.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) cruise monitoring technology, specifically relating to an integrated air-ground monitoring system and method for agricultural non-point source pollution based on the edge coordination of power transmission towers. Background Technology

[0002] Agricultural non-point source pollution is characterized by its wide distribution, strong concealment, and short outbreak time. Current monitoring methods are mainly divided into two categories: fixed water quality monitoring stations and mobile drone inspections. Fixed water quality monitoring stations are installed at key sections of the river channel and transmit data back in real time through sensors. Mobile drone inspections use drones equipped with cameras to photograph and analyze large areas of water.

[0003] However, power supply and communication in the field are difficult, and some terrains are complex, such as deep mountains and canyons, which leads to serious problems of power supply difficulties and weak public network signals for fixed monitoring stations. Laying dedicated power supply lines and optical cables is extremely costly.

[0004] Current drone inspections mostly employ a full-coverage blind flight mode, resulting in short flight times and difficulty in achieving 24 / 7 monitoring. Furthermore, drones typically cannot identify pollution in real time, requiring video feeds to be transported back to ground stations for analysis, leading to data lag. In inspections involving the vicinity of power facilities, ordinary drones lack path planning capabilities against strong electromagnetic fields, making them prone to flight control malfunctions or communication interruptions due to electromagnetic interference. Moreover, existing transmission line inspection drones primarily focus on detecting defects in the power facilities themselves, such as tower hardware and insulators, without optimizing their sensor configurations and algorithms for the surrounding ecosystem.

[0005] Currently, simply borrowing existing power line inspection drones cannot meet specific environmental protection needs such as identifying eutrophication in water bodies and tracing non-point source pollution, and there is a lack of a dedicated monitoring logic that focuses on the environment below the tower. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides an integrated air-ground monitoring system and method for agricultural non-point source pollution based on transmission tower edge collaboration. This system fully utilizes the spatial, electrical, and communication resources of high-voltage transmission corridors to construct a three-dimensional monitoring network with tower sentinels and drone patrols.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: An integrated air-ground monitoring system for agricultural non-point source pollution based on transmission tower edge collaboration includes tower edge sensing nodes, UAV collaborative terminals, and a cloud management platform; The tower edge sensing node is deployed on the tower body of the power transmission tower and includes a self-powered power supply module, a multi-dimensional sensing module, an edge computing gateway, and a heterogeneous communication module. The drone collaborative terminal includes an automatic drone nest, an inspection drone, and a mission payload deployed on the flood control platform of the tower base. The drone collaborative terminal is equipped with electromagnetic compatibility design. The cloud management platform is deployed in the data center and its main functions include: digital twin scenario construction, multi-source data fusion analysis, pollution spread trend prediction, and joint command and dispatch.

[0008] The self-powered module includes an inductive power harvesting unit and a power management unit. The inductive energy harvesting unit adopts the principle of open-type current transformer and is installed on the B-phase transmission line. It converts the load current of the primary transmission line into low-voltage AC current on the secondary side through electromagnetic induction. The power management unit includes a rectifier and filter circuit, a DC-DC wide voltage regulator circuit, and an overvoltage protection circuit. The output of the power management unit is connected in parallel with an energy storage battery to ensure the continuous operation of the equipment when the power transmission line is out of power or the load current is insufficient. The multidimensional sensing module includes a binocular multispectral camera and a micro-meteorological sensor suspended below the edge computing gateway. The binocular multispectral camera is used to retrieve the concentration of chlorophyll and suspended matter in the water, and the micro-meteorological sensor monitors rainfall, wind speed and wind direction in real time. The edge computing gateway is installed on the tower body and uses an embedded industrial computer. A lightweight AI inference engine is deployed inside the gateway to perform local real-time analysis of the video stream. The heterogeneous communication module includes a northbound interface and a southbound interface; The northbound interface is directly fused to the optical fiber composite overhead ground wire core at the top of the power transmission line via an optoelectronic converter, and connects to the dedicated power data communication network. The southbound interface integrates a LoRa radio frequency module and a Wi-Fi module, which are used to interact with the tower-based drone collaborative terminal for command interaction and large data transmission.

[0009] The automated cell has automatic recharge, waterproof door sealing, and constant temperature and dehumidification functions. The automated cell has a built-in LoRa receiver terminal. The flight control compass, GPS module, and data transmission link of the inspection drone are all equipped with permalloy shielding covers to prevent interference from high-voltage electromagnetic fields.

[0010] A method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination includes the following steps: S1: Data is collected through the pole edge sensing nodes based on the normalized monitoring mode; S2: The edge gateway invokes the built-in improved multi-level attention and spatial sampling fusion network MASFNet to perform inference on the acquired images; S3: Based on the improved MASFNet for initial contamination screening, the edge gateway sends a message containing [data / information] to the tower-based data center via the LoRa module. Upon receiving the wake-up command, the inspection drone automatically exits its cabin and takes off. S4: The onboard computing module of the inspection drone constructs an electromagnetic-environment integrated potential energy field model and plans the flight path in real time based on the model. The electromagnetic-environment integrated potential energy field model calculates the flight path of the inspection drone at any position. The virtual combined force received Perform path generation; S5: Refined evidence collection. After the inspection drone arrives at the target point, it uses airborne visual SLAM and RTK fusion positioning technology to hover in weak GPS signal environment and obtain high-definition evidence. S6: The inspection drone returns to its home base and transmits video and water quality data to the tower edge gateway via a high-speed Wi-Fi link. The gateway compresses and encrypts the data and then uploads it to the cloud management platform via the OPGW fiber optic communication link. S7: The cloud management platform is based on a GIS geographic information system. It overlays the identified pollution spread range as a dynamic heat map onto the three-dimensional digital twin model of the power transmission channel and automatically associates it with possible upstream agricultural emission sources to generate a source tracing analysis report.

[0011] In S1, the routine monitoring mode is specifically as follows: The binocular multispectral camera first conducts fixed-point patrol photography of the target water area below the tower according to a preset strategy. At the same time, the sensing nodes at the edge of the tower collect rainfall data in real time through micro-meteorological sensors, and use this as a trigger condition to dynamically adjust the monitoring strategy of the multispectral camera. When the system detects that the cumulative rainfall in the past hour is less than the preset threshold, it maintains a low-frequency fixed-point cruise shooting mode every 30 minutes to save energy. When the system detects that the cumulative rainfall in the past hour is greater than or equal to the preset threshold, it determines that it is a high-incidence period of non-point source pollution. The edge computing gateway immediately and automatically switches to high-frequency alert mode, increases the image acquisition frequency to once every 5 minutes, and simultaneously activates the built-in AI recognition module to perform real-time analysis of water features.

[0012] In S2, the improved multi-level attention and spatial sampling fusion network MASFNet is as follows: S2.1: Input the water body image acquired by the binocular multispectral camera in the multidimensional sensing module of the tower edge sensing node into MASFNet; S2.2: The input water body image is subjected to convolutional feature extraction through the lightweight backbone network of MASFNet to obtain the initial water body feature map; S2.3: The initial water feature map is subjected to multi-scale fusion and anti-halo processing using the spatial sampling fusion processing module in MASFNet; S2.4: Decision Output: The model outputs the pollution confidence level of the current water body. (0~1); like If this occurs, a suspected pollution alarm event is generated, and the center latitude and longitude coordinates of the polluted area are extracted. .

[0013] In S2.3, the spatial sampling fusion processing module performs multi-scale fusion on the initial water feature map, as follows: The shallow local detail feature maps and deep global semantic feature maps output by the backbone network are aligned in resolution using bilinear interpolation and then concatenated along the channel dimension to generate a multi-scale fused feature map. ; The spatial sampling fusion processing module performs anti-halo processing on the initial water feature map, as follows: An anti-halo penalty mask is generated by determining a brightness threshold, and the multi-scale fused feature map is weighted element-wise to suppress strong water surface reflectivity noise. Specifically, this includes: The brightness response values ​​of the multi-scale fused feature map at each spatial location are extracted. Regions with brightness response values ​​greater than a preset specular threshold are assigned a value of 0, while other regions are assigned a value of 1, thus generating a binarized anti-halo penalty mask matrix. The multi-scale fused feature map With the anti-halo penalty mask matrix Element-wise multiplication is performed to suppress high-frequency light and shadow noise, and a pure water body feature map is output. Its computational logic is expressed as follows: ; In the formula: This represents the Hadamard product of matrices.

[0014] Current water pollution confidence level The formula for calculating (0~1) is: ; In the formula: This is a global average pooling operation used to reduce the dimensionality of feature maps; and The weight matrix and bias terms of the fully connected classification layer; It is a normalized exponential function; This formula outputs the probability value that a water sample belongs to the suspected non-point source pollution category, i.e., the pollution confidence level. .

[0015] In S4, virtual combined force Including target gravity Electromagnetic repulsion ; Virtual resultant force on inspection drones in space Defined as: ; Target gravity The drone is guided towards the pollution source center identified by the edge node, and the calculation formula is as follows: ; In the formula: This is the gravitational gain coefficient. Let the target position vector be... The current position vector of the inspection drone (402); Electromagnetic repulsion Pointing away from high-voltage transmission lines, to achieve dynamic obstacle avoidance based on the power grid's operating status, and based on Ampere's circuital law and the principle of potential field gradient descent, real-time sensing of the transmission line load current is introduced. A dynamic electromagnetic repulsion model is constructed, and the calculation formula is as follows: ; In the formula: The real-time vertical Euclidean distance between the inspection drone (402) and the high-voltage power line; The set electromagnetic safety threshold distance; The transmission line load current value is sensed in real time by the edge computing gateway (202) through the sensing energy harvesting unit (201); To adjust the proportional coefficient of the electromagnetic repulsion force intensity.

[0016] Inspection drones according to Adjust the course in real time according to the direction of the resultant force; When the inspection drone is outside the electromagnetic safety distance, it is mainly controlled by gravity, and its path smoothly crosses the airspace below the power transmission corridor. When the inspection drone approaches the high-voltage line, At that time, the repulsive force increases dramatically, and the inspection drone will fly and hover along the equipotential lines to avoid the risk of communication interruption caused by forced crossing.

[0017] The main beneficial effects of this invention are as follows: 1) Overcame energy and communication bottlenecks in complex field environments, enabling cross-domain reuse of infrastructure: Regarding power supply: Addressing the challenge of a stable mains power supply in remote, wilderness environments, this invention innovatively utilizes magnetic field induction energy harvesting technology from high-voltage transmission lines. Unlike solar power, which is susceptible to weather conditions, the CT (Cyclic Detector) obtains power stably from the load current of the transmission lines. Combined with an energy storage battery, this creates a dual-redundant power supply system of online energy harvesting and offline energy storage, ensuring uninterrupted 24 / 7 operation for the monitoring equipment even during continuous rainy days or at night.

[0018] In terms of communications: Addressing the pain point of numerous blind spots in public network (4G / 5G) signal coverage in deep mountain and canyon areas, this invention reuses the unique optical fiber composite overhead ground wire (OPGW) resources of the power system. By connecting the edge gateway to the dedicated power fiber optic communication network, a high-bandwidth, low-latency, and highly secure data transmission highway is constructed, completely solving the industry problem of the inability to transmit massive amounts of high-definition video streams and monitoring data in real time.

[0019] 2.) A highly efficient end-edge-cloud collaboration mechanism has been established, significantly improving monitoring efficiency and reducing operation and maintenance costs: This invention changes the inefficient traditional model of blind, full-coverage inspections using drones. By deploying a lightweight AI algorithm, namely an improved MASFNet, at the edge of the pole, millisecond-level local early warning of water anomalies is achieved.

[0020] The drone will only take off and upload high-resolution evidence to the cloud when an edge node confirms a suspected contamination event. This event-driven operating mode reduces redundant data transmission and significantly lowers bandwidth consumption compared to the traditional timed data transmission mode. At the same time, the frequency of invalid drone flights is reduced, greatly extending the drone's battery life and the maintenance cycle of mechanical components.

[0021] 3) An adaptive path planning algorithm based on electromagnetic environment perception was proposed to ensure flight safety under power corridors: To address the risk of flight control failure or crashes caused by strong electromagnetic interference near high-voltage power lines for ordinary drones, this invention introduces an electromagnetic-environment integrated potential energy field model.

[0022] This algorithm quantifies the invisible and intangible electromagnetic field strength into a virtual repulsive field. The drone can perceive the strength of the surrounding electromagnetic environment in real time and autonomously decide its flight strategy: when the electromagnetic environment is safe, it chooses a path directly to the pollution source to improve efficiency; when electromagnetic interference is extremely strong, such as during fault discharges or under extremely high loads, it automatically hovers at a safe boundary to avoid the source. This combination of rigidity and flexibility in control strategy achieves a perfect balance between the safety of drone operations and the accuracy of source tracing in complex electromagnetic environments.

[0023] 4) The multi-source heterogeneous data fusion and verification mechanism significantly reduces the false alarm rate of non-point source pollution monitoring: Visual recognition alone is easily affected by environmental factors such as changes in lighting, tree reflections, and water ripples, leading to false alarms. This invention innovatively introduces micrometeorological data as a verification factor.

[0024] The system utilizes a fuzzy logic algorithm to fuse visual and meteorological features, effectively distinguishing between false anomalies caused by light and shadow and genuine non-point source pollution caused by heavy rain. Experiments show that compared to single visual monitoring, this system reduces the overall false alarm rate, providing environmental protection departments with more accurate and reliable decision-making support. Attached Figure Description

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is a flowchart illustrating the control logic of the coordinated linkage between pole edge sensing and UAV in this invention. Figure 3 This is a simulation comparison diagram of path planning based on the electromagnetic-environmental integrated potential energy field under weak interference environment in this invention; Figure 4 This is a simulation comparison diagram of path planning based on the electromagnetic-environmental integrated potential energy field under strong interference environment in this invention; In the diagram: Cloud management platform 100; Inductive energy harvesting unit 201; Edge computing gateway 202; Binocular multispectral camera 203; Micro-meteorological sensor 204; OPGW fiber optic communication link 300; Automated machine nest 401; Inspection drone 402; Non-point source pollution monitoring area 500. Detailed Implementation

[0026] Example 1: The intelligent monitoring system constructed in this example relies on the 110kV / 220kV high-voltage transmission tower infrastructure that traverses the Danjiangkou Reservoir area and surrounding farmland; like Figure 1 As shown, an integrated air-ground monitoring system for agricultural non-point source pollution based on transmission tower edge collaboration includes tower edge sensing nodes, UAV collaborative terminals, and a cloud management platform 100. The tower edge sensing node is the core sensing and control unit of this system. It is deployed under the crossarm of the tower body 15 to 25 meters above the ground. It utilizes the height advantage of the tower to achieve wide-area coverage. It includes a self-powered module, a multi-dimensional sensing module, an edge computing gateway 202, and a heterogeneous communication module. The drone collaborative terminal includes an automatic drone nest 401 deployed on the flood control platform of the tower base, an inspection drone 402, and a mission payload. The drone collaborative terminal is equipped with electromagnetic compatibility design. The automatic cell 401 features automatic recharging, waterproof door sealing, and constant temperature and dehumidification. The cell integrates a LoRa / ZigBee radio frequency receiver terminal, which is in a low-power listening mode, waiting for a wake-up command from the edge computing gateway 202. The inspection drone 402 is a quadcopter industrial-grade inspection drone. In response to the strong electromagnetic environment under high-voltage lines, the electronic compass and GPS module inside the fuselage have been treated with permalloy electromagnetic shielding. The inspection drone 402 is equipped with a portable water sampler containing a pH test strip color development module and a high-definition zoom camera.

[0027] The cloud management platform 100 is deployed on the data center server of the power company or environmental protection department. The platform receives data transmitted from the front end through the OPGW fiber optic link 300, constructs a digital twin model of the power transmission channel and the watershed environment based on WebGIS technology, and displays a real-time heat map of pollution diffusion. Its main functions include: digital twin scenario construction, multi-source data fusion analysis, pollution diffusion trend prediction, and coordinated command and dispatch. The OPGW fiber optic communication link 300 is the system's data backhaul channel. The uplink communication port of the edge computing gateway 202 is fused to the OPGW fiber optic composite overhead ground wire core at the top of the tower through an optoelectronic converter, and then connected to the dedicated power data communication network to achieve low-latency transmission of video streams.

[0028] The self-powered module includes an inductive power harvesting unit 201 and a power management unit. The inductive energy harvesting unit 201 adopts the principle of an open-type current transformer and is installed on the B-phase transmission line. It converts the load current of the primary transmission line into low-voltage AC current on the secondary side through electromagnetic induction. In this embodiment, the inductive energy harvesting unit is a CT inductive energy harvesting device. The device includes an open-type current transformer (CT) coil, which is snapped onto the B-phase conductor of the transmission line. It uses the electromagnetic induction principle of the line load current to obtain electrical energy, and the rear end is connected to a power management module and a 24V / 50Ah lithium iron phosphate energy storage battery pack.

[0029] Operating mode: When the current in the conductor... When the CT device is in use, it directly powers the equipment and charges the battery; when the power line fails or the load is extremely low, it seamlessly switches to battery power to ensure the system is online 24 / 7.

[0030] The power management unit includes a rectifier and filter circuit, a DC-DC wide voltage regulator circuit, and an overvoltage protection circuit. The output of the power management unit is connected in parallel with an energy storage battery to ensure the continuous operation of the equipment when the power transmission line is out of power or the load current is insufficient. The multi-dimensional sensing module includes a binocular multispectral camera 203 and a micro-meteorological sensor 204 suspended below the edge computing gateway 202. The binocular multispectral camera 203 is used to retrieve the concentration of chlorophyll and suspended matter in water, and the micro-meteorological sensor 204 monitors rainfall, wind speed and wind direction in real time. The binocular multispectral camera 203 is suspended below the edge computing gateway 202 via a gigabit Ethernet port (GigE). The lens is vertically downward or tilted towards the non-point source pollution monitoring area 500 below the tower base. The camera includes visible light and near-infrared bands to capture the spectral characteristics of water eutrophication and suspended sediment. The micro-weather sensor is connected to the edge computing gateway 202 via RS485 bus to monitor local rainfall in real time with an accuracy of 0.1mm, as well as wind speed and direction, providing environmental data support for the system to determine whether it has entered the high-incidence period of post-rain erosion.

[0031] The edge computing gateway 202 is installed in an industrial-grade protective box on the tower. As the brain of the node, it uses an embedded industrial control computer equipped with an NVIDIA Jetson Orin NX module. It has an AI computing power of no less than 70 TOPS and is responsible for the fusion processing of multi-source data and edge inference. A lightweight AI inference engine is deployed in the gateway to perform local real-time analysis of video streams. The heterogeneous communication module includes a northbound interface and a southbound interface; The northbound interface is directly fused to the optical fiber composite overhead ground wire core at the top of the power transmission line via an optoelectronic converter, and connects to the dedicated power data communication network. The southbound interface integrates a LoRa radio frequency module and a Wi-Fi module, which are used to interact with the tower-based drone collaborative terminal for command interaction and large data transmission.

[0032] The automatic nest 401 has automatic recharge, door sealing and waterproofing, and constant temperature and dehumidification functions. The automatic nest 401 has a built-in LoRa receiver terminal. The flight control compass, GPS module, and data transmission link of the inspection drone 402 are all equipped with permalloy shielding covers to prevent interference from high-voltage electromagnetic fields.

[0033] like Figure 2 As shown, a method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge collaboration includes the following steps: S1: Data is collected through the pole edge sensing nodes based on the normalized monitoring mode; S2: Edge computing gateway 202 calls the built-in improved multi-level attention and spatial sampling fusion network MASFNet to perform inference on the acquired images; S3: Based on the improved MASFNet for initial contamination screening, the edge gateway sends a message containing [data / information] to the tower-based data center via the LoRa module. Upon receiving the wake-up command, the inspection drone 402 automatically exited its cabin and took off. S4: The onboard computing module of the inspection drone 402 constructs an electromagnetic-environment integrated potential energy field model and plans the flight path in real time based on the model. The electromagnetic-environment integrated potential energy field model calculates the flight path of the inspection drone at any position. The virtual combined force received Perform path generation; S5: Refined evidence collection. After the inspection drone 402 flies to the target point, it uses airborne visual SLAM and RTK fusion positioning technology to hover in a weak GPS signal environment and obtain high-definition evidence. S6: Inspection drone 402 returns to its home base and transmits video and water quality data to the tower edge gateway via a high-speed Wi-Fi link. The gateway compresses and encrypts the data and then uploads it to the cloud management platform via the OPGW fiber optic communication link 300. S7: The cloud management platform 100, based on the GIS geographic information system, overlays the identified pollution spread range as a dynamic heat map onto the three-dimensional digital twin model of the power transmission channel, and automatically associates it with possible upstream agricultural emission sources to generate a source tracing analysis report.

[0034] In S1, the routine monitoring mode is specifically as follows: The binocular multispectral camera 203 first conducts fixed-point patrol photography of the target water area below the tower according to a preset strategy of once every 30 minutes. At the same time, the tower edge sensing node collects rainfall data in real time through the micro-meteorological sensor 204, and uses this as a trigger condition to dynamically adjust the monitoring strategy of the binocular multispectral camera 203. Edge gateway 202 reads the cumulative rainfall over the past hour. .

[0035] like If the current condition is not within the scouring period, then low-frequency monitoring should be maintained.

[0036] like If the system detects a high incidence of non-point source pollution, it will automatically switch to high-frequency alert mode, increasing the image acquisition frequency to once every 5 minutes and activating the edge AI inference module. It should be noted that the 10mm mentioned above is only a preferred preset threshold in this embodiment. In practical applications, those skilled in the art can reasonably adjust the rainfall threshold according to the specific terrain and historical meteorological data of the monitoring area. Such adjustments are all within the scope of protection of this invention.

[0037] In S2, the improved multi-level attention and spatial sampling fusion network MASFNet is as follows: S2.1: Input the water body image acquired by the binocular multispectral camera 203 in the multidimensional sensing module of the tower edge sensing node into MASFNet; S2.2: The input water body image is subjected to convolutional feature extraction through the lightweight backbone network of MASFNet to obtain the initial water body feature map; S2.3: The initial water feature map is subjected to multi-scale fusion and anti-halo processing using the spatial sampling fusion processing module in MASFNet; S2.4: Decision Output: The model outputs the pollution confidence level of the current water body. (0~1); like If this occurs, a suspected pollution alarm event is generated, and the center latitude and longitude coordinates of the polluted area are extracted. .

[0038] In S2.3, the spatial sampling fusion processing module performs multi-scale fusion on the initial water feature map, as follows: The shallow local detail feature maps and deep global semantic feature maps output by the backbone network are aligned in resolution using bilinear interpolation and then concatenated along the channel dimension to generate a multi-scale fused feature map. ; The spatial sampling fusion processing module performs anti-halo processing on the initial water feature map, as follows: An anti-halo penalty mask is generated by determining a brightness threshold, and the multi-scale fused feature map is weighted element-wise to suppress strong water surface reflectivity noise. Specifically, this includes: The brightness response values ​​of the multi-scale fused feature map at each spatial location are extracted. Regions with brightness response values ​​greater than a preset specular threshold are assigned a value of 0, while other regions are assigned a value of 1, thus generating a binarized anti-halo penalty mask matrix. The multi-scale fused feature map With the anti-halo penalty mask matrix Element-wise multiplication is performed to suppress high-frequency light and shadow noise, and a pure water body feature map is output. Its computational logic is expressed as follows: ; In the formula: This represents the Hadamard product of matrices.

[0039] Current water pollution confidence level The formula for calculating (0~1) is: ; In the formula: This is a global average pooling operation used to reduce the dimensionality of feature maps; and The weight matrix and bias terms of the fully connected classification layer; It is a normalized exponential function; This formula outputs the probability value that a water sample belongs to the suspected non-point source pollution category, i.e., the pollution confidence level. .

[0040] like It was determined to be a suspected pollution incident; The edge node generates a command packet containing the estimated coordinates of the pollution source and the current electromagnetic environment level, and sends it to the automated nest 401 at the tower base via a LoRa wireless link. After receiving the command, the nest opens the hatch and wakes up the inspection drone 402. The inspection drone 402 takes off and executes a path planning algorithm for the electromagnetic potential field based on the electromagnetic-environment integrated potential energy field model.

[0041] In S4, virtual combined force Including target gravity Electromagnetic repulsion ; Virtual resultant force on inspection drone 402 in space Defined as: ; Target gravity The system points to the center of the pollution source identified by the edge node, guiding the inspection drone 402 towards the pollution source. The calculation formula is as follows: ; In the formula: This is the gravitational gain coefficient. Let the target position vector be... The current position vector of the inspection drone 402; Electromagnetic repulsion Pointing away from high-voltage transmission lines, to achieve dynamic obstacle avoidance based on the power grid's operating status, and based on Ampere's circuital law and the principle of potential field gradient descent, real-time sensing of the transmission line load current is introduced. A dynamic electromagnetic repulsion model is constructed, and the calculation formula is as follows: ; In the formula: The real-time vertical Euclidean distance between the inspection drone (402) and the high-voltage power line; The set electromagnetic safety threshold distance; The transmission line load current value is sensed in real time by the edge computing gateway (202) through the sensing energy harvesting unit (201); To adjust the proportional coefficient of the electromagnetic repulsion force intensity.

[0042] When the line is operating at low load, that is When the force field is small, the repulsive field weakens, allowing the algorithm to allow the drone to fly closer to the line for a better view; when the line is fully loaded or short-circuited, i.e. When the force increases dramatically, the repulsive field strengthens instantly, forcing the drone to move away quickly, thus achieving dynamic obstacle avoidance based on the power grid's operating status.

[0043] Inspection drone 402 Adjust the course in real time according to the direction of the resultant force; When the inspection drone 402 is outside the electromagnetic safety distance, it is mainly controlled by gravity, and its path smoothly crosses the airspace below the power transmission corridor. When the inspection drone 402 approaches the high-voltage line, that is At that time, the repulsive force increases dramatically, and the inspection drone 402 will fly and hover along the equipotential lines to avoid the risk of communication interruption caused by forced crossing.

[0044] After arriving at the target location, it hovers using SLAM-RTK fusion positioning technology to perform detailed photography or water quality sampling.

[0045] After the inspection drone 402 returns to its home base, it transmits the collected high-definition evidence back to the edge computing gateway 202 via a high-speed Wi-Fi local area network. The gateway then uploads the data to the cloud management platform 100 via the OPGW fiber optic communication link 300, completing the mission loop.

[0046] In this embodiment, the environmental adaptive optimization result is as follows: Efficient crossing in weak interference environments: When the line load is low or the UAV has strong anti-interference capabilities, the proportional coefficient k for adjusting the electromagnetic repulsion force is small, and the algorithm can calculate the balance point of attraction and repulsion in real time, such as... Figure 3 As shown by the solid blue line, when the drone approached the area of ​​the red high-voltage line, it did not blindly cross in a straight line like the traditional fixed route, i.e., the dashed black line. Instead, it automatically planned a smooth S-shaped curve. This path cleverly used inertia and gravity to pass through the weak interference zone of the power transmission corridor and finally reached the source of pollution, thus maximizing efficiency.

[0047] Safe hovering in environments with strong interference: When the line is under extremely high load or experiences a fault discharge, causing a sharp increase in the surrounding electromagnetic field strength, the proportional coefficient k for adjusting the electromagnetic repulsion force becomes extremely large. The algorithm will then automatically trigger a safety avoidance mechanism, such as... Figure 4 As shown by the blue solid line, when the drone approaches the red deep water area, the calculated electromagnetic repulsion force rapidly exceeds the target's gravity. The flight control system forces the drone to maintain a safe distance from the guide wire threshold. The drone automatically hovers at the boundary, refusing to execute the crossing command. At this point, the drone will use its onboard high-definition zoom camera to take long-distance photographs of the pollution source from a safe position, and then return along the original route.

[0048] This embodiment demonstrates that the present invention can flexibly adjust strategies according to the strength of the electromagnetic environment while ensuring the electromagnetic compatibility (EMC) safety of unmanned aerial vehicles, thus achieving an organic unity of efficient traceability and intrinsic safety.

Claims

1. A ground-to-air integrated monitoring system for agricultural non-point source pollution based on transmission tower edge collaboration, characterized in that: Including pole edge sensing nodes, drone collaborative terminals, and cloud management platform (100). The pole edge sensing node is deployed on the tower body of the power transmission tower and includes a self-powered module, a multi-dimensional sensing module, an edge computing gateway (202), and a heterogeneous communication module; The drone collaborative terminal includes an automatic drone nest (401) deployed on the flood control platform of the tower base, an inspection drone (402), and a mission payload. The drone collaborative terminal is equipped with electromagnetic compatibility design. The cloud management platform (100) is deployed in the data center and its main functions include: digital twin scenario construction, multi-source data fusion analysis, pollution diffusion trend prediction and joint command and dispatch.

2. The integrated air-ground monitoring system for agricultural non-point source pollution based on transmission tower edge coordination as described in claim 1, characterized in that: The self-powered module includes an inductive power harvesting unit (201) and a power management unit; The inductive energy harvesting unit (201) adopts the principle of open-type current transformer and is installed on the B-phase transmission line. It converts the load current of the primary transmission line into the low-voltage AC current of the secondary side through electromagnetic induction. The power management unit includes a rectifier and filter circuit, a DC-DC wide voltage regulator circuit, and an overvoltage protection circuit. The output of the power management unit is connected in parallel with an energy storage battery to ensure the continuous operation of the equipment when the power transmission line is out of power or the load current is insufficient. The multidimensional sensing module includes a binocular multispectral camera (203) and a micro-meteorological sensor (204) suspended below the edge computing gateway (202). The binocular multispectral camera (203) is used to retrieve the concentration of chlorophyll and suspended matter in the water, and the micro-meteorological sensor (204) monitors the rainfall, wind speed and wind direction in real time. The edge computing gateway (202) is installed on the tower body and uses an embedded industrial computer. A lightweight AI inference engine is deployed in the gateway to perform local real-time analysis of the video stream. The heterogeneous communication module includes a northbound interface and a southbound interface; The northbound interface is directly fused to the optical fiber composite overhead ground wire core at the top of the power transmission line via an optoelectronic converter, and connects to the dedicated power data communication network. The southbound interface integrates a LoRa radio frequency module and a Wi-Fi module, which are used to interact with the tower-based drone collaborative terminal for command interaction and large data transmission.

3. The integrated air-ground monitoring system for agricultural non-point source pollution based on transmission tower edge collaboration as described in claim 1, characterized in that: The automatic nest (401) has automatic recharge, door sealing and waterproofing, and constant temperature and dehumidification functions. The automatic nest (401) has a built-in LoRa receiver terminal. The flight control compass, GPS module and data transmission link of the inspection drone (402) are all equipped with permalloy shielding covers to prevent interference from high voltage electromagnetic fields.

4. A method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination, characterized in that... Includes the following steps: S1: Data is collected through the pole edge sensing nodes based on the normalized monitoring mode; S2: The edge gateway invokes the built-in improved multi-level attention and spatial sampling fusion network MASFNet to perform inference on the acquired images; S3: Based on the improved MASFNet for initial contamination screening, the edge gateway sends a message containing [data / information] to the tower-based data center via the LoRa module. Upon receiving the wake-up command, the inspection drone (402) automatically exits its cabin and takes off; S4: The onboard computing module of the inspection drone (402) constructs an electromagnetic-environment integrated potential energy field model and plans the flight path in real time based on the model. The electromagnetic-environment integrated potential energy field model calculates the flight path of the inspection drone at any position. The virtual combined force received Perform path generation; S5: Refined evidence collection. After the inspection drone (402) arrives at the target point, it uses airborne visual SLAM and RTK fusion positioning technology to hover in a weak GPS signal environment to obtain high-definition evidence. S6: The inspection drone (402) returns to its home and transmits video and water quality data to the tower edge gateway via a high-speed Wi-Fi link. The gateway compresses and encrypts the data and then uploads it to the cloud management platform via the OPGW fiber optic communication link (300). S7: The cloud management platform (100) is based on the GIS geographic information system. It overlays the identified pollution diffusion range on the three-dimensional digital twin model of the power transmission channel in the form of a dynamic heat map, and automatically associates it with possible upstream agricultural emission sources to generate a source tracing analysis report.

5. The method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination according to claim 4, characterized in that: In S1, the routine monitoring mode is specifically as follows: The binocular multispectral camera first conducts fixed-point cruise photography of the target water area below the tower according to a preset strategy. At the same time, the tower edge sensing node collects rainfall data in real time through a micro-meteorological sensor (204), and uses this as a trigger condition to dynamically adjust the monitoring strategy of the multispectral camera. When the system detects that the cumulative rainfall in the past hour is less than the preset threshold, it maintains a low-frequency fixed-point cruise shooting mode every 30 minutes to save energy. When the system detects that the cumulative rainfall in the past hour is greater than or equal to the preset threshold, it is determined to be a period of high incidence of non-point source pollution. The edge computing gateway (202) immediately and automatically switches to high-frequency alarm mode, increases the image acquisition frequency to once every 5 minutes, and simultaneously activates the built-in AI recognition module to perform real-time analysis of water features.

6. The method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination according to claim 4, characterized in that: In S2, the improved multi-level attention and spatial sampling fusion network MASFNet is as follows: S2.1: Input the water body image acquired by the binocular multispectral camera (203) in the multidimensional sensing module of the tower edge sensing node into MASFNet; S2.2: The input water body image is subjected to convolutional feature extraction through the lightweight backbone network of MASFNet to obtain the initial water body feature map; S2.3: The initial water feature map is subjected to multi-scale fusion and anti-halo processing using the spatial sampling fusion processing module in MASFNet; S2.4: Decision Output: The model outputs the pollution confidence level of the current water body. (0~1); like If this occurs, a suspected pollution alarm event is generated, and the center latitude and longitude coordinates of the polluted area are extracted. .

7. The method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination according to claim 6, characterized in that: In S2.3, the spatial sampling fusion processing module performs multi-scale fusion on the initial water feature map, as follows: The shallow local detail feature maps output by the backbone network are aligned to the resolution with the deep global semantic feature maps using bilinear interpolation, and then concatenated along the channel dimension to generate a multi-scale fused feature map. ; The spatial sampling fusion processing module performs anti-halo processing on the initial water feature map, as follows: An anti-halo penalty mask is generated by determining a brightness threshold, and the multi-scale fused feature map is weighted element-wise to suppress strong water surface reflectivity noise. Specifically, this includes: The brightness response values ​​of the multi-scale fused feature map at each spatial location are extracted. Regions with brightness response values ​​greater than a preset specular threshold are assigned a value of 0, while other regions are assigned a value of 1, thus generating a binarized anti-halo penalty mask matrix. The multi-scale fused feature map With the anti-halo penalty mask matrix Element-wise multiplication is performed to suppress high-frequency light and shadow noise, and a pure water body feature map is output. Its computational logic is expressed as follows: ； In the formula: This represents the Hadamard product of matrices.

8. The method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination according to claim 6, characterized in that: Current water pollution confidence level The formula for calculating (0~1) is: ； In the formula: This is a global average pooling operation used to reduce the dimensionality of feature maps; and The weight matrix and bias terms of the fully connected classification layer; It is a normalized exponential function; This formula outputs the probability value that a water sample belongs to the suspected non-point source pollution category, i.e., the pollution confidence level. .

9. The method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination according to claim 4, characterized in that: In S4, virtual combined force Including target gravity Electromagnetic repulsion ; Virtual resultant force on the inspection drone (402) in space Defined as: ； Target gravity The inspection drone (402) is guided to approach the pollution source center identified by the edge node. The calculation formula is as follows: ； In the formula: This is the gravitational gain coefficient. Let the target position vector be... The current position vector of the inspection drone (402); Electromagnetic repulsion Pointing away from high-voltage transmission lines, to achieve dynamic obstacle avoidance based on the power grid's operating status, and based on Ampere's circuital law and the principle of potential field gradient descent, real-time sensing of the transmission line load current is introduced. A dynamic electromagnetic repulsion model is constructed, and the calculation formula is as follows: ； In the formula: The real-time vertical Euclidean distance between the inspection drone (402) and the high-voltage power line; The set electromagnetic safety threshold distance; The transmission line load current value is sensed in real time by the edge computing gateway (202) through the sensing energy harvesting unit (201); To adjust the proportional coefficient of the electromagnetic repulsion force intensity.

10. The method for integrated air-ground monitoring of agricultural non-point source pollution based on transmission tower edge coordination according to claim 9, characterized in that: Inspection drone (402) according to Adjust the course in real time according to the direction of the resultant force; When the inspection drone (402) is outside the electromagnetic safety distance, it is mainly controlled by gravity, and its path smoothly crosses the airspace below the power transmission corridor; When the inspection drone (402) approaches the high-voltage line, that is At that time, the repulsive force increases dramatically, and the inspection drone (402) will fly and hover along the equipotential line to avoid the risk of communication interruption caused by forced crossing.

Images