Power transmission tower corrosion collaborative diagnosis system and method based on low-altitude Internet of Things

Abstract

The invention provides a power transmission tower corrosion collaborative diagnosis system and method based on a low-altitude internet of things, and belongs to the technical field of electric power internet of things. Comprising a mobile edge node and a cloud diagnosis platform, the mobile edge node is used for acquiring a power transmission tower image and performing corrosion analysis based on the initial corrosion detection model; a decision-making module and a low-power-consumption communication module are integrated on the mobile edge node; wherein the decision-making module is used for making a decision according to a false positive filtering rule and a risk weighting rule and outputting a salient region image; the low-power-consumption communication module is used for compressing and packaging the salient region image into a data packet and transmitting the data packet to a far-end gateway; and the cloud diagnosis platform is used for analyzing the image blocks and the metadata, and performing corrosion segmentation and quantitative analysis by using a high-precision corrosion analysis model. According to the invention, the balance between the diagnosis precision and the diagnosis efficiency can be realized on the basis of reasonably distributing the computing resources of the corrosion collaborative diagnosis system.

Description

Technical Field

[0001] This invention belongs to the field of power Internet of Things (IoT) technology, and particularly relates to a collaborative diagnostic system and method for transmission tower corrosion based on low-altitude IoT. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Traditional corrosion detection of power transmission towers mainly relies on manual inspections, which is not only inefficient and costly but also poses safety risks. In recent years, with the development of computer vision and drone technology, methods based on drone photography and deep learning for automatic diagnosis have become a research hotspot. These methods typically deploy object detection models (such as YOLO and SSD) or image segmentation models (such as U-Net) on edge computing devices on drones (such as NVIDIA Jetson series modules) in an attempt to achieve real-time diagnosis.

[0004] However, existing technical solutions of this type have obvious limitations: (1) Difficulty in balancing accuracy and efficiency in real time: To ensure that the model can run in real time on edge devices with limited computing power, existing methods generally use pre-compressed fixed lightweight models (such as MobileNet, SqueezeNet, etc.). Although these models have low computational cost, their feature extraction capabilities are limited, and their detection accuracy will decrease significantly when faced with complex backgrounds, changes in lighting, or subtle corrosion. Conversely, if a high-precision large model is deployed, it cannot meet the real-time requirements, resulting in diagnostic delays.

[0005] (2) Rigid allocation of computational resources: Existing solutions mostly adopt a "one-size-fits-all" processing mode, that is, the same amount of computation is invested in every frame of image or every region. For regions with simple backgrounds and no or obvious corrosion, lightweight models are sufficient, and investing too much computation is a waste of resources; while for regions with complex backgrounds and suspected corrosion, the capabilities of lightweight models are inadequate. This static resource allocation mode cannot achieve the optimization of computational efficiency.

[0006] (3) Lack of adaptive perception capability: Most methods fail to utilize the prior knowledge of "salience of rusted areas". The core objective of a diagnostic system is to identify "problematic" areas, rather than averaging the entire image. Therefore, current technology cannot automatically identify "high-value" suspicious areas in an image. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, this invention provides a collaborative diagnostic system and method for transmission tower corrosion based on low-altitude Internet of Things, which can achieve a balance between diagnostic accuracy and diagnostic efficiency by rationally allocating the computing resources of the collaborative diagnostic system for corrosion.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a collaborative diagnostic system for corrosion of transmission towers based on low-altitude Internet of Things.

[0009] A collaborative diagnostic system for corrosion of transmission towers based on low-altitude IoT includes: mobile edge nodes and a cloud diagnostic platform; The mobile edge node consists of a drone equipped with an image acquisition module and an edge computing module, used to acquire images of the transmission tower and perform corrosion analysis based on an initial corrosion detection model; The mobile edge node integrates a decision-making module and a low-power communication module, specifically: The decision module is used to make decisions on the corrosion analysis results based on the false positive filtering rules based on geometric dimensions and the risk weighting rules based on regional importance, so as to output a salient area image; The low-power communication module is used to compress the obtained salient region image and encapsulate it into a data packet for transmission to the remote gateway. The cloud diagnostic platform is used to receive data packets from a remote gateway and parse out image blocks and metadata, and to perform corrosion segmentation and quantitative analysis using a high-precision corrosion analysis model; wherein, the high-precision corrosion analysis model is a semantic segmentation network with an integrated quantitative analysis module.

[0010] Furthermore, the false positive filtering rule based on geometry size includes: setting a minimum effective size threshold for the rusted area, and filtering out all detection boxes smaller than the minimum effective size threshold.

[0011] Furthermore, the risk weighting rule based on regional importance includes: calculating the confidence score of the detection frame by weighting it according to the preset risk weight coefficients for different structural parts of the transmission tower, based on the key area mapping map of the transmission tower.

[0012] Furthermore, the key areas of the transmission tower include the main material nodes, bolted connections, and load-bearing components of the transmission tower.

[0013] Furthermore, the low-power communication module uses a lightweight image encoding algorithm to compress the salient region image, and encapsulates the compressed salient region image, metadata related to the salient region image, and device identifier into a data packet conforming to the LoRaWAN protocol specification.

[0014] Furthermore, the quantitative analysis results output by the cloud diagnostic platform include: a corrosion segmentation map, and quantitative data predicting the corrosion area, contour, and depth based on the corrosion segmentation map.

[0015] The second aspect of this invention provides a collaborative diagnostic method for corrosion of transmission towers based on low-altitude Internet of Things.

[0016] A collaborative diagnostic method for corrosion of transmission towers based on low-altitude Internet of Things includes: Real-time images of the transmission towers are acquired based on the mobile edge nodes; On the mobile edge node, the initial corrosion detection model is used to perform corrosion analysis on the obtained real-time image; a false positive filtering rule based on geometric size and a risk weighting rule based on regional importance are used to make decisions on the corrosion analysis results to output a salient region image; the obtained salient region image is compressed and encapsulated into a data packet and transmitted to the remote gateway. On the cloud diagnostic platform, data packets are received from the remote gateway and image blocks and metadata are parsed out. A high-precision corrosion analysis model is used for corrosion segmentation and quantitative analysis. The high-precision corrosion analysis model is a semantic segmentation network with a built-in quantitative analysis module. Furthermore, the false positive filtering rule based on geometry size is used to remove small false positive boxes caused by noise and texture interference; the risk weighting rule based on regional importance is used to weight the confidence of the retained rusted areas and generate a risk score to guide maintenance priorities.

[0017] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the collaborative diagnosis method for corrosion of transmission towers based on low-altitude Internet of Things as described in the second aspect of the present invention.

[0018] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the collaborative diagnosis method for corrosion of transmission towers based on low-altitude Internet of Things as described in the second aspect of the present invention.

[0019] The above one or more technical solutions have the following beneficial effects: (1) The corrosion collaborative diagnosis system of the present invention includes a mobile edge node and a cloud diagnosis platform. The mobile edge node performs image acquisition and corrosion analysis, and integrates a decision module for decision-making and a low-power communication module for communication. The cloud diagnosis platform then performs corrosion segmentation and quantification analysis. Thus, on the edge side, the initial corrosion detection model carried by the UAV can perform rapid inference on the real-time video stream, ensuring low latency and high efficiency. On the cloud side, a high-precision corrosion analysis model is deployed to perform in-depth segmentation and quantification analysis on the salient area images transmitted from the edge, which can ensure the accuracy of the final diagnosis. This division of labor and cooperation overcomes the problems of insufficient accuracy of lightweight models or high latency of large models in existing methods, and simultaneously meets the requirements of real-time performance and high accuracy under limited edge computing power.

[0020] (2) This invention achieves dynamic optimization of computing resource allocation by introducing a false positive filtering rule based on geometric dimensions and a risk weighting rule based on regional importance. The system automatically filters out small false positive boxes caused by noise or texture interference, which can reduce unnecessary computing overhead; at the same time, by weighting the suspicious areas according to the key area mapping map of the transmission tower, computing resources can be preferentially allocated to high-value areas. This adaptive resource allocation mechanism avoids the waste and inadequacy of resources caused by the "one-size-fits-all" processing mode and significantly improves computing efficiency.

[0021] (3) This invention endows the system with adaptive perception capabilities by integrating prior knowledge of "salience of rusted areas". The decision module uses risk weighting rules based on regional importance to automatically identify high-value suspicious areas in the image (such as main material nodes, bolt connections, etc.) and generate weighted risk scores for these areas. This not only improves the targeting of the detection, but also provides maintenance personnel with a basis for priority decision-making, enhancing the engineering practicality and guidance of the system.

[0022] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a structured flowchart of the collaborative diagnosis system for transmission tower corrosion based on low-altitude Internet of Things in Embodiment 1 of the present invention. Detailed Implementation

[0025] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0027] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0028] Example 1 This embodiment discloses a collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things.

[0029] like Figure 1 As shown, the collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things includes: mobile edge nodes and a cloud diagnostic platform; The mobile edge node consists of a drone equipped with an image acquisition module and an edge computing module, used to acquire images of the transmission tower and perform corrosion analysis based on an initial corrosion detection model; The mobile edge node integrates a decision-making module and a low-power communication module, specifically: The decision module is used to make decisions on the corrosion analysis results based on the false positive filtering rules based on geometric dimensions and the risk weighting rules based on regional importance, so as to output a salient area image; The low-power communication module is used to compress the obtained salient region image and encapsulate it into a data packet for transmission to a remote LoRaWAN gateway. The cloud diagnostic platform is used to receive data packets from the LoRaWAN gateway and parse out image blocks and metadata, and to perform corrosion segmentation and quantitative analysis using a high-precision corrosion analysis model; wherein, the high-precision corrosion analysis model is a semantic segmentation network with an integrated quantitative analysis module.

[0030] Based on the above systematic design, this invention can achieve a balance between diagnostic accuracy and diagnostic efficiency by rationally allocating the computational resources of the corrosion collaborative diagnostic system. To facilitate understanding of the technical solution of this invention, the specific implementation methods of this invention will be further explained and described below.

[0031] A collaborative diagnostic system for transmission tower corrosion based on low-altitude IoT includes: mobile edge nodes and a cloud diagnostic platform.

[0032] The mobile edge node, consisting of a drone equipped with an image acquisition module and an edge computing module, is used to acquire images of power transmission towers and perform corrosion analysis based on an initial corrosion detection model.

[0033] The drone, acting as a mobile edge node, acquires images of the power transmission tower using its onboard image acquisition module. This image acquisition module is a multi-sensor integrated system specifically optimized for diagnosing corrosion on power transmission towers, and its core components include: 1) High-resolution visible light camera: As the main sensor, an industrial-grade global shutter camera with a resolution of over 20 megapixels is selected. Its high resolution ensures that the fine textures on the surface of the transmission tower components can be captured from a safe distance, providing sufficient pixel information for identifying early corrosion. The global shutter avoids the rolling shutter effect caused by the vibration of the drone rotor or the equipment itself, ensuring image clarity.

[0034] 2) Telephoto optical zoom lens: Equipped with a telephoto zoom lens with optical image stabilization (e.g., equivalent focal length coverage from wide-angle to 150mm or more). This allows the drone to take close-up shots of specific, manually preset key parts of the transmission tower (such as tower heads, insulator string hanging points, bolt connections, etc.) from a safe distance permitted by regulations, in order to obtain high-quality, detailed images.

[0035] 3) Position and Attitude Sensors: Integrating a high-precision GNSS (Global Navigation Satellite System) receiver and an IMU (Inertial Measurement Unit). The GNSS receiver records the UAV's geographical location during image acquisition, while the IMU records the camera's pitch, yaw, and roll angles. This metadata is crucial for subsequent image geographic registration, flight path planning, and multi-angle image association.

[0036] 4) Onboard computing unit interface: This module connects to the edge computing module of the UAV via a high-speed interface (such as USB 3.0 or GigE), receives control commands from the latter (such as focus, zoom, shutter trigger), and transmits back image data and metadata in real time.

[0037] 5) Controllable Gimbal: The camera and lens are mounted on a high-precision, three-axis self-stabilized mechanical gimbal. This gimbal is controlled by an edge computing module and can actively compensate for changes in the drone's attitude, ensuring that the camera is always pointed at the target area. Especially when shooting at the telephoto end, it can effectively prevent image blurring.

[0038] The specific process of acquiring images of transmission towers based on the onboard image acquisition module includes: 1) Mission Planning and Flight Path Generation: Before the mission begins, the cloud diagnostic platform or ground control station will plan a preset flight inspection route for the UAV. This route is generated based on the digital 3D model of the transmission tower or historical inspection data to ensure that the UAV can circle the transmission tower and cover all critical components from multiple angles. At the same time, a series of shooting stations will be pre-set in the route, with each shooting station corresponding to a part of the tower material that needs to be inspected.

[0039] 2) Automated data acquisition cycle: After the drone autonomously flies to the preset shooting site, it starts the following data acquisition cycle: A. Coarse positioning and gimbal pointing: The UAV's GNSS and IMU data are combined with the preset geographic coordinates of the tower target to drive the gimbal to roughly point the camera at the target area.

[0040] B. Visual Servo Precision Alignment: The edge computing module calls a lightweight target detection model to analyze the real-time video stream transmitted from the visible light camera and automatically identify target components (such as "angle steel L80×8", "bolt M16", etc.). By calculating the pixel deviation between the target's position in the image frame and the center of the image, a control signal is generated to fine-tune the gimbal angle, thereby achieving precise centering of the target through "visual servoing".

[0041] C. Automatic parameter optimization and capture: The camera automatically optimizes exposure parameters (shutter speed, ISO) based on the current lighting conditions; then, the edge computing module issues a command to control the camera to perform high-speed continuous shooting (e.g., taking 3 to 5 shots in succession), and the algorithm selects the clearest shot as the valid data to deal with possible momentary shaking.

[0042] D. Data encapsulation and local recording: Each high-resolution image acquired is immediately bound to its metadata (including GNSS coordinates, IMU attitude, timestamp, camera parameters, and target component ID) to form a structured data packet, which is temporarily stored in the local storage of the edge computing module.

[0043] 3) Adaptive Acquisition Strategy: The system possesses preliminary on-site adaptive capabilities. During the real-time video stream in the data acquisition loop, when the lightweight detection model identifies an area suspected of severe corrosion, the edge computing module can dynamically adjust the acquisition plan. For example, it can automatically zoom in to take close-up shots of the suspected area; or it can capture multiple images of the same area from different lighting angles by slightly adjusting the drone's position or gimbal angle, aiding in subsequent 3D analysis and shadow interference elimination.

[0044] After acquiring images of the transmission tower, corrosion analysis is performed based on the initial corrosion detection model integrated within the edge computing module to obtain preliminary bounding boxes of the corrosion areas and corresponding confidence scores. This edge computing module is a high-performance, low-power embedded artificial intelligence computing platform specifically designed for UAV onboard environments. The core components and features of the edge computing module include: 1) Core Processor: Utilizes a system-on-a-chip (SoC) with an integrated dedicated neural network processor, such as the NVIDIA Jetson series and Huawei Atlas series. These chips provide powerful AI computing capabilities while meeting the size, weight, and power consumption constraints of airborne equipment.

[0045] 2) Memory and storage: Equipped with ample LPDDR memory (e.g., 8GB) and a high-speed embedded multimedia storage card to carry the operating system, model files, and temporarily store acquired images and analysis results.

[0046] 3) Operating System and Software Stack: It runs a lightweight Linux operating system and comes pre-installed with necessary AI inference frameworks, such as TensorRT or OpenVINO. These frameworks can optimize and convert trained neural network models into engines that can run efficiently on this hardware.

[0047] 4) External Interfaces: Provides a rich set of I / O interfaces, including but not limited to camera interface, communication interface, and flight control interface. The camera interface is used to connect to and control the image acquisition module; the communication interface is used to connect to the low-power communication module and transmit results; the flight control interface communicates with the UAV flight controller to receive flight status information and send commands to adjust the acquisition strategy when necessary.

[0048] 5) Heat dissipation design: Passive cooling or small silent fans are used for active cooling to ensure stable hardware performance under continuous high load computing.

[0049] Considering the computational resource limitations of edge devices, the initial corrosion detection model integrated within the edge computing module is a deeply optimized, lightweight, single-stage target detection model. Its core idea is to achieve a balance between speed and accuracy to quickly locate suspected corrosion areas. The initial corrosion detection model consists of three parts: a backbone network, a feature fusion network, and a detection head.

[0050] 1) Backbone Network. The backbone network acts as a feature extractor, extracting multi-level feature maps from the input RGB image. Shallow feature maps have high resolution and contain rich detail information, while deep feature maps have low resolution and contain high-level semantic information. Specifically, the backbone network uses lightweight networks such as MobileNetV3 or ShuffleNetV2. These networks use operations such as depthwise separable convolution, which can significantly reduce computation and parameter count while maintaining high accuracy. In this embodiment, the input to the backbone network is the normalized image. The size is fixed. Finally, a set of multi-scale feature maps is output; among them, , and" "These represent the height, width, and RGB channels of the input image tensor, respectively."

[0051] 2) Feature Fusion Network. The feature fusion network is used to fuse feature maps of different scales output by the backbone network. This is because rust targets may vary in size; fusing features from different levels allows the model to perceive both details and semantics simultaneously, thereby improving its ability to detect small targets and different morphologies of rust. Specifically, the feature fusion network employs a simplified path aggregation network structure. It combines deep semantic information with shallow detail information through top-down and bottom-up paths to generate a set of enhanced multi-scale feature maps.

[0052] 3) Detection Head. The detection head is the core output of the model, responsible for predicting bounding boxes and class confidence scores in parallel at each preset anchor point location in the fused feature map. Specifically, the detection head consists of two parallel lightweight convolutional sub-networks: a classification sub-network and a regression sub-network. The classification sub-network predicts the confidence score that each anchor point contains rust; the regression sub-network predicts the positional offset of each anchor point, thereby fine-tuning the preset anchor points to more accurately match the actual rusted areas.

[0053] In the specific implementation process, for feature maps ( The first feature map represents the first feature map. Each spatial location on a scale and each preset anchor point box The detection head will output a vector: ; in, and This represents the offset of the bounding box center coordinates output by the regression subnetwork. and These represent the scaling factors for the width and height in the bounding box offsets output by the regression subnetwork, respectively. The confidence score output by the classification subnetwork, after activation by the sigmoid function, represents the probability of corrosion at that location; confidence score The range of values ​​is ; This represents the predicted tensor.

[0054] Final bounding box coordinates It is done by using a preset anchor point frame. Calculated using the offset: ; ; ; ; in, This represents the Sigmoid function. and This is the scaling factor; Indicates the center of the prediction box Axis coordinates Indicates the center of the prediction box Axis coordinates This indicates the width of the prediction box. Indicates the height of the prediction box; Indicates the center of the anchor point frame Axis coordinates Indicates the center of the anchor point frame Axis coordinates; This indicates the width of the anchor point frame. This indicates the height of the anchor point frame.

[0055] Based on this, the initial corrosion detection model can obtain the preliminary corrosion region bounding box and corresponding confidence score through the following forward propagation process: First, preprocessing is performed: the acquired transmission tower images are scaled to the fixed size required by the model and normalized. Then, forward inference is performed: the preprocessed images are input into the initial corrosion detection model deployed on the edge computing module. The model sequentially passes through the backbone network, feature fusion network, and detection head, outputting a series of prediction tensors. Finally, post-processing is performed: all the original outputs of the model are post-processed to obtain the final readable result. The core step of post-processing is non-maximum suppression, which specifically includes: 1) Confidence filtering: Set a low confidence threshold to filter out all [rejections / outcomes]. The predicted bounding box; where, This indicates the set confidence threshold. This step can remove a large number of obviously erroneous background predictions.

[0056] 2) NMS Processing: Perform NMS on the remaining predicted boxes. The NMS algorithm calculates the intersection-union ratio (IUU) among all boxes and suppresses boxes with high overlap with the highest-scoring box. That is, for each category, it calculates the IUU based on confidence score. Sort by confidence from highest to lowest, and select the box with the highest confidence level. Add it to the final result set; calculate the relationship between all other boxes and the box with the highest confidence. The cross-union ratio is calculated, and boxes with high overlap with the highest-scoring box are suppressed; this process is repeated until all boxes have been processed.

[0057] 3) Output results: After NMS, each remaining bounding box and its corresponding confidence score This constitutes the preliminary corrosion analysis results. These results will be passed to subsequent decision-making modules for further processing.

[0058] Through the above process, the edge computing module can utilize a carefully designed lightweight model to quickly and effectively locate and assess the initial corrosion areas of transmission tower images in resource-constrained environments.

[0059] The mobile edge node also integrates a decision module and a low-power communication module. The decision module is used to make decisions on the corrosion analysis results based on the false positive filtering rules based on geometry and the risk weighting rules based on regional importance, so as to output a salient area image. The low-power communication module is used to compress the obtained salient area image and encapsulate it into a data packet for transmission to the remote LoRaWAN gateway.

[0060] The decision module is used to make decisions based on the corrosion analysis results according to the false positive filtering rules based on geometry size and the risk weighting rules based on regional importance, so as to output a salient area image. Specifically: 1) False positive filtering rules based on geometric dimensions: A minimum effective pixel area threshold and / or minimum bounding box size threshold are set for the rusted area. These thresholds can be adjusted according to actual needs; therefore, this embodiment does not impose specific limitations on them. Based on this, by filtering out all detection boxes smaller than the set thresholds, false positive detection results caused by image noise, minor texture interference, etc., can be eliminated, improving the overall accuracy of the system.

[0061] 2) Risk weighting rule based on regional importance: The system pre-stores or associates a key area mapping map of the transmission tower, and calculates the confidence score of the detection frame based on predefined risk weight coefficients for different structural parts of the transmission tower. The core idea of ​​the risk weighting rule based on regional importance is that the final risk score of a defect depends not only on the algorithm's certainty (confidence) in "seeing" the defect, but also on "where it appears" (location risk weight). The entire process can be divided into three stages: The first phase involves system construction. First, a critical area mapping map of the transmission tower is created. This map is a digital, structured "map" that can be linked to the CAD drawings, 3D models, or standard template drawings of the transmission tower. The map divides the two-dimensional projection or three-dimensional surface of the transmission tower into different functional areas, such as critical areas, secondary critical areas, and non-critical areas. Critical areas include main material connection nodes, diagonal material intersections, bolt connections, load-bearing tower legs, wire hanging points, and insulator string hardware; secondary critical areas include auxiliary diagonal materials and non-connected areas of crossarms; and non-critical areas include the flat center of the main tower structure and auxiliary materials without connecting functions. Next, risk weighting coefficients are defined. Each of the different functional areas is assigned a weighting coefficient. This coefficient represents the relative risk of a defect found in that area. In this embodiment: the weight coefficient of critical areas is >1.0 because any defects in critical areas may lead to structural instability, connection failure or major power outage accidents (for the connecting bolts of the main load-bearing legs of the tower foot, the node plates of the main load-bearing members, etc., the maximum weight coefficient can be set between 3.0 and 5.0); the weight coefficient of non-critical areas is =1.0 or <1.0 because the risk of defects in non-critical areas (such as slight corrosion) is relatively low, and may only affect a local area or allow for a longer maintenance cycle.

[0062] The second stage involves real-time target detection and location matching. Target detection involves acquiring images using a drone, applying AI algorithms (such as YOLO and Faster R-CNN) to infer meaning from the images, and outputting a series of detection results with category labels (e.g., "loose bolts," "rust," "bird's nest"), confidence scores, and bounding boxes. Location matching involves the system matching the coordinates of the detected bounding boxes with a pre-stored "key region mapping map." This can be achieved in the following ways: A. Based on georeferencing: If the drone is equipped with accurate GPS and RTK, the detection box can be mapped to the precise three-dimensional coordinates of the transmission tower and then matched with the area in the three-dimensional model.

[0063] B. Image registration based: By matching feature points, the real-time image is aligned with a standard transmission tower template image, thereby determining the area where the detection box is located in the template image.

[0064] C. Reasoning based on relative position: For tower-shaped structures with regular patterns, the region to which the detection box belongs can be inferred based on the relative position of the detection box in the image (such as the top, middle, or near the connection).

[0065] The third stage involves risk-weighted calculation and output. For each detected defect, the final risk score is the product of the detected confidence level and the regional risk weight coefficient. In this way, a defect in a critical area, even with a slightly lower confidence level, may have a significantly higher final risk score than a defect in a non-critical area with a high confidence level. This aligns better with the risk assessment logic in actual operation and maintenance.

[0066] 3) Decision Output: The final corrosion risk score is the product of the original confidence level and the risk weight coefficient. Therefore, the weighted score not only reflects the model's judgment on "whether it is corrosion," but also incorporates prior engineering knowledge on "whether the corrosion at this location is dangerous," thus outputting a more engineering-guided priority decision-making strategy to guide maintenance personnel in prioritizing high-risk defects. The following is an end-to-end implementation process from data preparation to system integration, aiming to put the "regional importance risk weighting" rule into practice.

[0067] The first step is to build a knowledge base and model (offline preparation).

[0068] First, the transmission tower structure is digitized and divided into zones. In the implementation process, CAD drawings, 3D models, or standard design drawings of the transmission tower are used as input. Through collaboration with structural engineers, the tower body is decomposed into multiple functional zones, and a digital polygon mask is created for each zone. These masks serve as the "map" for subsequent location matching. Finally, a structured "key zone mapping database" is output, with each record containing a zone ID, zone name (e.g., "Main Material Node K7"), zone polygon coordinate set, and risk weight coefficient.

[0069] Next, risk weighting coefficients are defined. In the specific implementation process, the evaluation dimensions are "the degree of impact of the defect on the structural integrity of this part," "the degree of impact of the failure of this part on the overall stability," and "the level of consequences caused by the failure of this part." Pairwise comparisons or direct scoring are performed on each region to calculate the final weight. In this embodiment, the scaling is as follows: Weight = 1.0: Base area, such as the flat surface of the tower.

[0070] Weight = 1.2 ~ 1.8: Secondary critical areas, such as diagonal members that are not the main load-bearing members.

[0071] Weight = 2.0 ~ 3.0: Critical areas, such as the main diagonal members and crossbeam connection areas.

[0072] Weight = 3.5 ~ 5.0: Extremely critical areas, such as main material nodes, tower feet, and hanging points.

[0073] The second step is to perform real-time detection and data fusion (online operation).

[0074] First, target detection and localization are performed. In practice, drones are used for line inspection, and airborne or ground-based AI models analyze video streams or images to output raw detection results: {defect category, confidence level, bounding box data}.

[0075] Next, image-map registration is performed. Due to the varying angles from which drones are captured, the bounding box data from the real-time images is mapped onto a standard "key area map." In practice, a template matching-based method can be used: pre-render template images of the transmission tower and their corresponding area maps from multiple standard perspectives. Using feature point matching algorithms such as SIFT and ORB, the most similar template to the current image is identified, and a homography transformation matrix is ​​calculated.

[0076] Finally, location matching and weight lookup are performed. Specifically, using the obtained transformation relationship, the center point or the entire bounding box is projected onto the "critical area map," and it is determined which polygon the projected point falls within. Then, the corresponding regional risk weight is retrieved for each detected defect. .

[0077] The third step is risk weighting and decision output. The final risk score can be calculated using the following formula: ; in, Indicates the final risk score. Indicates regional risk weights. The raw risk score represents the visual performance.

[0078] All detected defects are classified according to their final risk. Sort the data in descending order to generate an inspection report with clear priorities, as shown in Table 1.

[0079] Table 1 Inspection Report

[0080] The low-power communication module first processes the image marked with salient regions, and then transmits the processed image data packets to the remote LoRaWAN gateway. Simultaneously, to adapt to the low-power, narrow-bandwidth communication characteristics of LoRaWAN, intelligent front-end processing is performed on the captured image marked with salient regions to reduce the power consumption of the drone terminal device. The low-power communication module can be implemented through the following methods: 1) Data compression: Lightweight image coding algorithms (such as JPEG, JPEG2000, or compression algorithms designed for edge computing) are used to perform lossy or lossless compression on images of salient regions to reduce the image data size.

[0081] 2) Information encapsulation: The compressed salient region image, metadata related to the salient region image, and device identifier are encapsulated together into a data packet that conforms to the LoRaWAN protocol specification.

[0082] 3) Uplink transmission: Using LoRa wireless modulation technology, data packets are sent to one or more LoRaWAN gateways in a low-power, long-distance manner to complete data backhaul.

[0083] The cloud diagnostic platform is used to receive data packets from the LoRaWAN gateway and parse out image blocks and metadata, and to perform corrosion segmentation and quantitative analysis using a high-precision corrosion analysis model.

[0084] A high-precision corrosion analysis model deployed in the cloud performs in-depth analysis of the images, and its output results are as follows: 1) High-precision corrosion segmentation map: each corrosion area is identified at the pixel level.

[0085] 2) Quantitative data based on segmentation results: Automatically calculate the area and outline of each rusted area, as well as depth prediction based on prior knowledge (such as rust pit depth estimation).

[0086] The high-precision corrosion analysis model is a typical semantic segmentation network with an encoder-decoder structure, and a quantitative analysis module is added on top of it. Its core objective is to achieve pixel-level classification (i.e., corrosion or non-corrosion) and derive quantitative indicators.

[0087] The encoder portion of the high-precision corrosion analysis model acts as the "eyes" and "brain" of the network, responsible for extracting multi-scale, hierarchical features from the input image. Its structure can employ a pre-trained convolutional neural network (such as ResNet, VGG, or EfficientNet) as the backbone. In practice, the encoder progressively compresses the spatial dimensions (width and height) of the image through a series of convolutional and pooling layers, while simultaneously increasing the number of channels. It captures basic features (such as edges, texture, and color) through shallow networks and high-level semantic features (such as "this is the patchy texture of corrosion" or "this is metallic reflection") through deep networks. Finally, it outputs a low-resolution, high-channel feature map, containing rich information for determining whether each pixel is corroded, but losing precise location details.

[0088] The decoder part of the high-precision corrosion analysis model is used to "translate" the high-level feature maps output by the encoder back to pixel-level classification results and restore their spatial details for accurate localization. Its structure consists of upsampling layers (such as transposed convolution and interpolation) and convolutional layers. The skip connections used are key to achieving high accuracy, enabling channel-wise concatenation of feature maps from the encoder's intermediate (shallow) layers with the upsampling results from the corresponding layers of the decoder. Shallow feature maps contain rich location and detail information (such as "where is the boundary of corrosion"), while deep features contain accurate semantic information (such as "is this corrosion?"). Skip connections fuse these two, allowing the model to accurately identify corrosion and clearly delineate its outline. Finally, a probability map with the same resolution as the input image is output, where the value of each pixel represents the confidence level that the pixel belongs to the "corrosion" category.

[0089] The added quantitative analysis module in the high-precision corrosion analysis model is used for post-processing and data extraction of the segmentation map generated by the decoder, and outputting quantitative indicators usable in engineering. This is not a trained neural network, but an algorithm module based on computer vision and rules. Specifically, the learning process of the model core (encoder-decoder) can be viewed as a function mapping: ; in, This represents the input image tensor, with a size of [value missing]. , , and" "Represents the height, width, and RGB channels of the input image tensor respectively; This represents the entire encoder-decoder network. The output segmentation graph (probabilistic graph) has a dimension of . Each pixel value is between [0, 1].

[0090] The training objective of the model is to minimize the difference between the predicted value and the true label. The differences between (manually labeled, pixel-level binary segmentation images) are achieved using binary cross-entropy loss in this embodiment: ; in, This represents the loss function, used to measure the degree of error in the model's predictions; This represents the total number of pixels in an image; Indicates the location The actual label at the location is either "0" or "1", with "0" representing the background and "1" representing corrosion. Indicates position The probability that a given location is predicted as rust by the model. Parameters are continuously optimized using backpropagation and gradient descent algorithms. This minimizes the loss, allowing the model's predictions to get closer and closer to the manually labeled "standard answer".

[0091] This invention's collaborative reasoning architecture of "preliminary edge screening - detailed cloud analysis" not only ensures system response speed but also greatly improves the accuracy and richness of the final diagnostic results, providing direct data support for predictive maintenance.

[0092] Example 2 This embodiment discloses a collaborative diagnostic method for corrosion of power transmission towers based on low-altitude Internet of Things.

[0093] A collaborative diagnostic method for corrosion of transmission towers based on low-altitude Internet of Things includes: Real-time images of the transmission towers are acquired based on the mobile edge nodes; On the mobile edge node, the initial corrosion detection model is used to perform corrosion analysis on the obtained real-time image; a false positive filtering rule based on geometric size and a risk weighting rule based on regional importance are used to make decisions on the corrosion analysis results to output a salient region image; the obtained salient region image is compressed and encapsulated into a data packet and transmitted to the remote gateway. On the cloud diagnostic platform, data packets are received from the remote gateway and image blocks and metadata are parsed out. A high-precision corrosion analysis model is used for corrosion segmentation and quantitative analysis. The high-precision corrosion analysis model is a semantic segmentation network with a built-in quantitative analysis module.

[0094] Furthermore, the false positive filtering rule based on geometry size is used to remove small false positive boxes caused by noise and texture interference; the risk weighting rule based on regional importance is used to weight the confidence of the retained rusted areas and generate a risk score to guide maintenance priorities. Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0095] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the collaborative diagnosis method for corrosion of transmission towers based on low-altitude Internet of Things as described in Embodiment 2 of this disclosure.

[0096] Example 4 The purpose of this embodiment is to provide an electronic device.

[0097] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the collaborative diagnosis method for corrosion of transmission towers based on low-altitude Internet of Things as described in Embodiment 2 of this disclosure.

[0098] The steps and methods involved in the apparatuses of Embodiments 1, 3, and 4 above correspond to those in Embodiment 2. For specific implementation details, please refer to the relevant description section of Embodiment 2. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0099] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0100] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things, characterized in that, include: Mobile edge nodes and cloud diagnostic platforms; The mobile edge node consists of a drone equipped with an image acquisition module and an edge computing module, used to acquire images of the transmission tower and perform corrosion analysis based on an initial corrosion detection model; The mobile edge node integrates a decision-making module and a low-power communication module, specifically: The decision module is used to make decisions on the corrosion analysis results based on the false positive filtering rules based on geometric dimensions and the risk weighting rules based on regional importance, so as to output a salient area image; The low-power communication module is used to compress the obtained salient region image and encapsulate it into a data packet for transmission to the remote gateway. The cloud diagnostic platform is used to receive data packets from a remote gateway and parse out image blocks and metadata, and to perform corrosion segmentation and quantitative analysis using a high-precision corrosion analysis model; wherein, the high-precision corrosion analysis model is a semantic segmentation network with an integrated quantitative analysis module.

2. The collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things as described in claim 1, characterized in that, The false positive filtering rule based on geometry includes: setting a minimum effective size threshold for the rusted area, and filtering out all detection boxes smaller than the minimum effective size threshold.

3. The collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things as described in claim 1, characterized in that, The risk weighting rule based on regional importance includes: calculating the confidence score of the detection frame by weighting it according to the risk weight coefficients preset for different structural parts of the transmission tower, based on the key area mapping map of the transmission tower.

4. The collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things as described in claim 3, characterized in that, The key areas of the transmission tower include the main material nodes, bolted connections, and load-bearing components.

5. The collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things as described in claim 1, characterized in that, The low-power communication module uses a lightweight image encoding algorithm to compress the salient region image, and encapsulates the compressed salient region image, metadata related to the salient region image, and device identifier into a data packet that conforms to the LoRaWAN protocol specification.

6. The collaborative diagnostic system for transmission tower corrosion based on low-altitude Internet of Things as described in claim 1, characterized in that, The quantitative analysis results output by the cloud diagnostic platform include: a corrosion segmentation map, and quantitative data predicting the corrosion area, contour, and depth based on the corrosion segmentation map.

7. A collaborative diagnostic method for corrosion of transmission towers based on low-altitude Internet of Things, used in the corrosion collaborative diagnostic system as described in any one of claims 1-6, characterized in that, include: Real-time images of the transmission towers are acquired based on the mobile edge nodes; On the moving edge node, the initial corrosion detection model is used to perform corrosion analysis on the obtained real-time image; False positive filtering rules based on geometric dimensions and risk weighting rules based on regional importance are used to make decisions on the corrosion analysis results to output a salient region image; the obtained salient region image is compressed and encapsulated into a data packet and transmitted to the remote gateway; On the cloud diagnostic platform, data packets are received from the remote gateway and image blocks and metadata are parsed out. A high-precision corrosion analysis model is used for corrosion segmentation and quantitative analysis. The high-precision corrosion analysis model is a semantic segmentation network with a built-in quantitative analysis module.

8. The collaborative diagnosis method for transmission tower corrosion based on low-altitude Internet of Things as described in claim 7, characterized in that, The false positive filtering rule based on geometry size is used to remove small false positive boxes caused by noise and texture interference; the risk weighting rule based on regional importance is used to weight the confidence of the retained rusted areas and generate a risk score to guide maintenance priorities.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the collaborative diagnosis method for corrosion of transmission towers based on low-altitude Internet of Things as described in any one of claims 7-8.

10. An electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the collaborative diagnosis method for corrosion of transmission towers based on low-altitude Internet of Things as described in any one of claims 7-8.

Images