Intelligent high-precision dynamic sensing method and equipment for coarse-grain-size dam material based on 5G and vision

Abstract

The invention discloses a high-precision dynamic sensing method for coarse-grain-size dam materials based on 5G and visual intelligence, and the method comprises the steps: setting a 5G base station or 5G access equipment in a dam region, and collecting dam surface image data through employing an unmanned plane; transmitting the acquired image data in real time through a 5G network; a database server and an image processing server are arranged on the ground, and the database server receives and stores collected data through a 5G network; a dam material analysis model constructed based on a YOLOv8 model is arranged in the image processing server; the image data in the database server is called by the cloud server and input into a dam material analysis model, and the dam material analysis model calculates the particle size of stones and aggregates by adopting a minimum bounding rectangle method based on an instance segmentation result of a YOLOv8 model, and identifies oversized stones, oversized aggregates and the position and area of a concentrated region of the oversized stones and the oversized aggregates; and the image processing result is sent to the database server for storage by the image processing server. According to the invention, the labor is reduced and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of earth-rock dam engineering and UAV detection technology, and particularly relates to a method and device for high-precision dynamic perception of coarse-grained dam materials based on 5G and visual intelligence. Background Art

[0002] At present, during the construction process of large earth-rock dams, the complexity of the construction environment and the strict requirements for filling quality make it a key link to timely identify and handle oversize stones and concentrated areas of coarse aggregates to ensure project quality. The existence of oversize stones and coarse aggregates will not only affect the stability of the dam body, but may also lead to potential safety hazards during the subsequent use of the dam body. Therefore, how to efficiently and accurately detect these problems has become an important task in the management of earth-rock dam engineering.

[0003] Traditional detection methods mainly rely on manual inspections and visual recognition by airborne cameras. Although these methods can complete the detection tasks to a certain extent, they have significant limitations. First of all, manual inspections are not only time-consuming and laborious, but also easily affected by human factors, resulting in instability and inaccuracy of the detection results. Secondly, the visual recognition technology of airborne cameras performs poorly in complex construction environments, especially when there is dust, light changes and narrow spaces on the surface of the earth-rock dam, which often leads to a decrease in recognition accuracy. For example, dust coverage may block the surface features of oversize stones, and light changes may result in insufficient image contrast, thus affecting the recognition ability of the algorithm. In addition, traditional methods are inefficient when dealing with large-scale areas and are difficult to meet the dual requirements of the project for real-time and accuracy.

[0004] With the commercial deployment of 5G networks and the rapid development of unmanned aerial vehicle (UAV) technology, combined with a high-bandwidth and low-latency network environment and advanced visual intelligence algorithms, it provides a new solution for realizing large-scale and real-time high-precision detection. The high-bandwidth characteristic of 5G networks makes it possible to transmit large amounts of data in real time, greatly improving the efficiency of data processing. The maturity of UAV technology makes it possible to perform efficient image acquisition in complex terrains and environments. By combining the two, it is possible to conduct a comprehensive and rapid inspection of the dam surface at the construction site.

[0005] In the existing literature, although preliminary research has been conducted on the detection of oversize stones or the identification of coarse aggregates in a small range, most of the research is still limited to traditional detection methods and lacks a systematic solution for low-power algorithms, real-time transmission networks, and comprehensive field applications. Current research mainly focuses on the improvement and optimization of algorithms. However, in practical applications, how to effectively combine these algorithms with UAV technology and 5G networks to form a complete detection system is still an urgent problem to be solved. Summary of the Invention

[0006] The present invention provides a high-precision dynamic perception method and device for coarse-grained dam materials based on 5G and visual intelligence to solve the technical problems existing in the known technology.

[0007] The technical solution adopted by the present invention to solve the technical problems existing in the known technology is as follows:

[0008] A high-precision dynamic perception method for coarse-grained dam materials based on 5G and visual intelligence. A 5G base station or 5G access device is set in the dam area to form a 5G full-coverage network in the dam area; a drone is used to collect dam surface image data; through the 5G network, the image data collected by the drone is transmitted in real time from the collection device end of the drone to an external device; a database server and an image processing server that communicate with each other are set on the ground. The database server is used to receive and store data, and the image processing server is used to process the image data; the database server receives and stores the dam surface image data collected from the drone through the 5G network; an analysis model of dam materials constructed based on the YOLOv8 model is set in the image processing server, which calls the image data in the database server and inputs it into the analysis model of dam materials; based on the instance segmentation result of the YOLOv8 model, the analysis model of dam materials calculates the particle size of stones and aggregates by using the minimum circumscribed rectangle method, identifies oversize stones and oversize aggregates, as well as the positions and areas of the concentrated areas of the two; the image processing result is sent by the image processing server to the database server and stored by the database server.

[0009] Furthermore, the analysis model of dam materials includes a partial convolution module, a depthwise separable convolution module, a cross-scale feature fusion module, and a YOLOv8 instance segmentation module;

[0010] The convolution layer in the YOLOv8 model is replaced by a partial convolution module and a depthwise separable convolution module; multi-scale feature extraction of stones and aggregates is performed on the collected dam surface image data;

[0011] A cross-scale feature fusion module is introduced into the neck structure of the YOLOv8 model, and the cross-scale feature fusion module fuses feature maps of different scales;

[0012] Pixel-level segmentation of relatively coarse-grained stones and aggregates is performed through the YOLOv8 instance segmentation module. Based on the contour of the instance segmentation, the length and width of the target stones and aggregates are estimated by using the minimum circumscribed rectangle method, and the ground sampling distance is calculated in combination with the flight parameters of the drone, and finally the actual particle size is obtained.

[0013] Furthermore, image data containing different lithologies and particle size distributions are collected from different construction sites for compiling the training data of the analysis model of dam materials. When compiling the training data, data annotation is combined with the Segment Anything Model.

[0014] Further, set the threshold ranges of the following parameters: the particle size of stone materials and aggregates, and the area of concentrated regions of oversize stones and oversize aggregates; the image processing server compares the image processing results with the corresponding threshold ranges. When it determines that the threshold ranges are exceeded, it issues a command signal according to the percentage of the exceeded threshold range and transmits it to the drone through the 5G network. The control system of the drone adjusts the flight altitude, flight speed, and flight attitude according to the command signal.

[0015] Further, the drone is equipped with an auto-zoom camera with a resolution higher than 3840×2160; the focal length adjustment range is 21 - 75 mm.

[0016] Further, the drone flies within the set flight path and altitude range of 20 - 100 meters.

[0017] Further, a display and an alarm are provided in the ground workstation. Both the display and the alarm communicate with the image processing server. The image processing server sends the image data it calls and the image data processing results to the display for display; when the image processing server determines that the image data processing results exceed the set threshold range, it sends an alarm signal to the alarm for early warning.

[0018] Further, adopt a return flight path to cover the entire dam surface, and detect the dam surface section by section. The image acquisition period for each section is 1 - 1.5 hours; when planning the flight path, consider the wind speed and air flow, and set up a standby flight path to cope with emergencies.

[0019] Further, divide the image acquisition area of the drone into grids. For any grid, the image data collected by the drone is located with the center point of the corresponding grid as the reference coordinate origin. The collected image data includes grid numbers, reference coordinates of grid center points, reference coordinates of image data, and resolution information.

[0020] The present invention also provides a device based on the 5G and visual intelligence high-precision dynamic perception method for coarse-grained dam materials, including a memory and a processor. The memory is used to store computer programs; the processor is used to execute the computer programs and implement the steps of the above-mentioned 5G and visual intelligence high-precision dynamic perception method for coarse-grained dam materials when executing the computer programs.

[0021] The advantages and positive effects of the present invention are:

[0022] Set up 5G base stations or miniaturized 5G access devices in the dam area to provide high-speed and low-latency data transmission channels. Through the 5G network, real-time transmission of image data collected by drones is ensured, enabling quick response in complex construction environments and avoiding delays in detection data caused by network latency. In addition, the high-bandwidth feature of the 5G network makes real-time transmission of large-resolution images possible, guaranteeing the accuracy and timeliness of subsequent data processing. By combining the 5G network with drone aerial survey technology, large-area and multi-batch inspections of dam materials are achieved, reducing manual participation and environmental impact. This integration improves the detection efficiency.

[0023] Drones equipped with zoom cameras can fly at low altitudes along set routes to collect real-time images of the dam surface. The flexibility and efficiency of drones enable them to cover large construction areas, avoiding the limitations of traditional manual inspections. Through the 5G link, the collected images with a resolution of 3840×2160 are transmitted to the ground workstation in real time, ensuring data timeliness and providing support for subsequent image processing.

[0024] Run the improved YOLOv8 model on the ground workstation, adopting partial convolution modules (PConv), depthwise separable convolution modules (DWConv), and cross-scale feature fusion modules (CCFM) to reduce the computing power requirements and improve the detection speed. This improvement in the model structure enables efficient object detection even under limited resources and adapts to the complexity of the earth-rock dam construction environment.

[0025] Based on the instance segmentation results of YOLOv8, calculate the particle size of stones and aggregates through efficient methods such as the minimum bounding rectangle, and identify the positions and areas of oversize stones and coarse aggregate concentration areas. This process not only improves the detection accuracy but also provides data support for subsequent construction decisions, ensuring the control of project quality.

[0026] Store the detection results in the cloud and visualize them in real time; if significant oversize or concentration areas are found, issue warnings in a timely manner to guide the construction party to take necessary measures. This dynamic perception mechanism not only improves the intelligent level of the construction process but also provides real-time data support for construction management, ensuring timely response to emergencies.

[0027] The flight altitude, camera focal length, and particle size threshold parameters can be adjusted according to different dam surface environments to ensure the accuracy and generality of the detection results. The flexibility of this method enables it to adapt to different construction environments and requirements, providing wide possibilities for future engineering applications.

[0028] By combining the 5G network with drone aerial survey technology, large-area and multi-batch inspections of dam materials are achieved, reducing manual participation and environmental impact. This integration improves the detection efficiency.

[0029] The present invention provides flexible expansion and transplantation capabilities, adapting to engineering sites of different scales and types. The modular design enables the system to be adjusted according to specific requirements, ensuring stable operation in different environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the working principle of a high-precision dynamic perception method for coarse-grained dam materials based on 5G and visual intelligence of the present invention.

[0031] Figure 2 It is a schematic diagram of the flight path of an unmanned aerial vehicle in a high-precision dynamic perception method for coarse-grained dam materials based on 5G and visual intelligence of the present invention.

[0032] In the figure:

[0033] Ncls represents: the number of target categories.

[0034] 4*reg_max represents: 4 times the maximum value of the regularization term.

[0035] H represents: Height, the height of the target.

[0036] W represents: width, the width of the target. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] The present invention will be described in detail below with reference to the drawings and in conjunction with embodiments. It should be understood that the preferred embodiments described herein are only for illustrating and explaining the present invention, and are not used to limit the present invention.

[0038] The terms "connected" and "coupled" used in the present invention should be understood in a broad sense. For example, it can be a fixed connection or a detachable connection; it can be directly connected or indirectly connected through an intermediate component; it can also be an electrical connection or a signal transmission; for those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0039] The Chinese interpretations of the following English words, phrases and English abbreviations are as follows:

[0040] YOLOv8: A target detection model.

[0041] Pconv: Partial convolution. The basic principle is to utilize the redundancy of the feature map to reduce calculations and memory access.

[0042] DWConv: Depthwise separable convolution. It is extended from separable convolution and is a lightweight convolution operation.

[0043] CCFM: Cross-Scale Feature Fusion Module, a lightweight cross-scale feature fusion module.

[0044] Segment Anything Model (SAM): An image segmentation model released in 2023.

[0045] YOLOv8-seg: An enhanced version of the YOLO series for segmentation tasks, which is an object detection model.

[0046] GSD: Ground Sampling Distance: The physical distance corresponding to each pixel in the image on the actual ground.

[0047] Conv2d: Two-dimensional Convolution: An operation used in deep learning and computer vision for feature extraction from two-dimensional data.

[0048] batchNorm2d: A function that normalizes the input data for each mini-batch.

[0049] SiLU: Sigmoid Linear Unit: An activation function.

[0050] Feature map: Feature map.

[0051] Upsample: Upsampling, which refers to the process of increasing the sampling rate of a signal in signal processing. The basic principle is to insert new sample points between the original signal's sampling points.

[0052] MaxPool2d: A pooling operation used in convolutional neural networks (CNNs), mainly for reducing the dimensionality of data and extracting image features.

[0053] C2f: A module that allows YOLOv8 to obtain more abundant gradient flow information while ensuring light weight.

[0054] Split: A data processing calculation that can split a dataset or tensor in a specified manner.

[0055] DarknetBottleneck: A feature extraction module widely used in object detection models.

[0056] SPPF: Spatial Pyramid Pooling - Fast: An upgraded version of the SSP module. Its purpose is to handle the information fusion of local and global features, fuse the features of different receptive fields of the image, and finally obtain a feature block with both local and global features, which is faster than the original SPP.

[0057] conv: A basic network module in the object detection model, mainly used for feature extraction.

[0058] Protonet: A few-shot classification method based on meta-learning, mainly used to solve the few-shot learning problem.

[0059] BCE: Binary Cross Entropy, the binary cross-entropy loss function.

[0060] CIoU: Complete IoU: A loss function used in the object detection task.

[0061] DFL: DeepFaceLab: A deep learning tool used for data synthesis and exchange.

[0062] Mask coeffcient: Mask coefficient.

[0063] Bbox: bounding box: Bounding box. Used to describe and locate objects in the image.

[0064] Cls: Classification: Used for downstream classification tasks. It is a feature vector that can represent the entire text semantics, representing the label of the entire statement and used for direct classification.

[0065] DJIDock: An automated drone hangar designed to simplify the drone operation process, improve operation efficiency and reliability. DJIDock realizes unattended operation through an integrated design and cloud management function, reduces manpower requirements, and can operate stably in harsh environments.

[0066] M30 Series Dock Version: Refers to the drone docking station designed by DJI for the Matrice 30 series of drones.

[0067] Please refer to Figures 1 to 2, A high-precision dynamic perception method for coarse-grained dam materials based on 5G and visual intelligence. Set up 5G base stations or 5G access devices in the dam area to form a 5G full-coverage network in the dam area; use drones to collect dam surface image data; through the 5G network, transmit the image data collected by the drones from the collection device end of the drones to external devices in real time; set up a database server and an image processing server that communicate with each other on the ground. The database server is used to receive and store data, and the image processing server is used to process the image data; the database server receives and stores the dam surface image data collected by the drones through the 5G network; an analysis model of dam materials based on the YOLOv8 model is built inside the image processing server, which calls the image data in the database server and inputs it into the analysis model of dam materials; based on the instance segmentation results of the YOLOv8 model, the analysis model of dam materials calculates the particle size of stones and aggregates using the minimum bounding rectangle method, identifies oversize stones and oversize aggregates, as well as the positions and areas of the concentration areas of the two; the image processing results are sent by the image processing server to the database server and stored by the database server.

[0068] Preferably, the analysis model of dam materials may include a partial convolution module, a depthwise separable convolution module, a cross-scale feature fusion module, and a YOLOv8 instance segmentation module;

[0069] The convolution layer in the YOLOv8 model can be replaced by a partial convolution module and a depthwise separable convolution module; multi-scale feature extraction of stones and aggregates is performed on the collected dam surface image data;

[0070] A cross-scale feature fusion module can be introduced into the neck structure of the YOLOv8 model, and the cross-scale feature fusion module fuses feature maps of different scales;

[0071] Pixel-level segmentation of coarser-grained stones and aggregates is performed through the YOLOv8 instance segmentation module. Based on the contour of the instance segmentation, the length and width of the target stones and aggregates are estimated using the minimum bounding rectangle method, and the ground sampling distance is calculated in combination with the drone flight parameters to finally obtain the actual particle size.

[0072] Preferably, image data containing different lithologies and particle size distributions can be collected from different construction sites for compiling the training data of the analysis model of dam materials. When compiling the training data, data annotation is combined with the Segment Anything Model.

[0073] Preferably, the threshold ranges of the following parameters can be set: the particle size of stone materials and aggregates, and the area of concentrated regions of oversize stones and oversize aggregates. The image processing server compares the image processing results with the corresponding threshold ranges. When it determines that the threshold ranges are exceeded, it can send a command signal according to the percentage exceeding the threshold range and transmit it to the drone through the 5G network. The control system of the drone adjusts the flight altitude, flight speed, and flight attitude according to the command signal.

[0074] Preferably, the drone can be equipped with an auto-focus camera with a resolution higher than 3840×2160; the focal length adjustment range is 21~75mm.

[0075] Preferably, the drone can fly within a set flight path and altitude range of 20~100 meters.

[0076] Preferably, a display and an alarm can be provided in the ground workstation. Both the display and the alarm can communicate with the image processing server. The image processing server sends the image data it calls and the image data processing results to the display for display. When the image processing server determines that the image data processing results exceed the set threshold range, it sends an alarm signal to the alarm for early warning.

[0077] Preferably, a return flight path can be adopted to cover the entire dam surface, and the dam surface can be divided into sections for detection one by one, so that the image acquisition period for each section is 1~1.5 hours. When planning the flight path, wind speed and air flow can be considered, and a standby flight path can be set to cope with emergencies.

[0078] Preferably, the image acquisition area of the drone can be divided into grids. For any grid, the image data collected by the drone is located with the corresponding grid center point as the reference coordinate origin. The collected image data includes grid numbers, grid center point reference coordinates, image data reference coordinates, and resolution information.

[0079] The present invention also provides a device based on the 5G and visual intelligence high-precision dynamic perception method for coarse-grained dam materials, including a memory and a processor. The memory is used to store computer programs; the processor is used to execute the computer programs and implement the steps of the above-mentioned 5G and visual intelligence high-precision dynamic perception method for coarse-grained dam materials when executing the computer programs.

[0080] The working process and working principle of the present invention will be further described below with a preferred embodiment of the present invention:

[0081] I. 5G Network Deployment and System Initialization

[0082] (1) 5G Base Station Setup: Set up 5G base stations or portable 5G access devices around the dam body to ensure a stable high-speed data transmission environment within the flight area. These base stations should be reasonably arranged according to the terrain and the actual situation of the construction area to cover all key detection areas, ensuring that the drone maintains a good signal connection during flight. At the same time, considering possible environmental interference, the base stations should have the ability to automatically adjust signal strength and frequency to adapt to different construction conditions.

[0083] (2) Ground Workstation Configuration: Equip the ground workstation with real-time data reception and storage devices, and set up a cloud server or edge computing node according to requirements for subsequent model inference and result storage. The workstation should have powerful computing capabilities to support complex image processing and data analysis tasks. In addition, the cloud server should have good scalability to handle a large number of concurrent requests, ensuring the secure storage and efficient access of data. The edge computing node can perform preliminary data processing near the data source to reduce latency and improve response speed.

[0084] II. Drone Flight Plan and Data Collection

[0085] (1) Drone Selection: Prioritize the use of multi-rotor drones equipped with 21 - 75mm zoom lenses. Such drones can adapt to complex construction environments at different flight heights and speeds, ensuring the stability and clarity of image acquisition. At the same time, select drones with long endurance and high payload capacity to support long-duration flight tasks and high-resolution image acquisition.

[0086] (2) Flight Speed and Altitude: Based on the on-site construction progress and image clarity requirements, generally control the flight speed at about 1 m / s. The flight altitude should be appropriately adjusted according to the size of the dam surface area, usually between 20 and 50 meters, to ensure image clarity and coverage. The selection of flight altitude should also consider the safety of the construction site to avoid conflicts with other equipment or personnel.

[0087] (3) Route Design: Plan a route that covers the entire dam surface to avoid missing shots. The dam surface can be divided into sections (such as 76m×37.5m sections) for detection section by section according to a two-day or shorter construction cycle, ensuring that each section inspection is completed within about 1.5 hours. The route design should consider wind speed, air flow, and other environmental factors to ensure the stability of the drone during flight. At the same time, a backup route can be set up to handle emergencies.

[0088] (4) Real-time image upload: The 3840×2160 resolution images captured by the drone are transmitted in real-time to the ground workstation via a 5G link. To ensure the stability and speed of data transmission, it is recommended to use data compression technology to reduce the amount of data and improve transmission efficiency. In addition, during the data transmission process, a redundancy mechanism should be set up to ensure that the data can be effectively recovered in case of network fluctuations or signal loss.

[0089] III. Improved YOLOv8 Model and Instance Segmentation

[0090] (1) Model lightweighting: Replace the original convolutional layers with partial convolution (PConv) and depthwise separable convolution (DWConv) to reduce the number of model parameters and improve the inference speed. This lightweight design not only improves the running efficiency of the model but also reduces the demand for computing resources, enabling the model to perform rapid inference on edge computing devices.

[0091] (2) Cross-scale feature fusion (CCFM): Introduce a cross-scale feature fusion module in the neck structure of the network to fully extract the key information of coarse aggregate at various scales and improve the recognition rate of the model for large-size targets. In this way, the model can better handle targets of different sizes and improve the detection accuracy.

[0092] (3) Training data: Combine the Segment Anything Model (SAM) for rapid data annotation and collect images with different lithologies and particle size distributions from different construction sites to enhance the generalization of the model. During the training process, attention should be paid to the diversity of data to ensure that the model can adapt to various environments and conditions.

[0093] (4) Instance segmentation: Use the improved YOLOv8-seg branch to perform precise pixel-level segmentation on coarse aggregate and concentrated coarse aggregate, preparing for subsequent particle size and distribution calculations. The results of instance segmentation will provide reliable data support for subsequent size measurement and dynamic perception.

[0094] IV. Particle Size Measurement and Dynamic Perception

[0095] (1) Size calculation: Based on the contour of instance segmentation, use the minimum bounding rectangle method to quickly estimate the length and width of the target aggregate, and combine the drone flight parameters to calculate the ground sampling distance (GSD), finally obtaining the actual particle size. This process needs to consider the influence of different flight heights on the measurement results to ensure the accuracy of particle size calculation.

[0096] (2) Threshold determination: Set thresholds for oversized stones (such as 100 cm) and concentrated coarse aggregate thresholds (such as 50 cm), etc., and classify according to the particle size of the detected targets. The classification results will be used for subsequent construction decisions to ensure that the construction party can promptly handle oversized stones and concentrated coarse aggregate areas.

[0097] (3) Dynamic update: Since the dam surface environment will change at any time during the construction process, the present invention uses high-frequency flight route repetition and high-speed 5G network transmission to dynamically update and upload the detection results to cloud storage. The dynamic update mechanism ensures that the construction party can obtain the latest detection data in real time and adjust the construction plan in a timely manner.

[0098] (4) Visualization and warning: The detection results are displayed in real time on the ground workstation and mobile terminal, and the detected oversize stones or concentrated areas are highlighted for easy construction personnel to take timely measures (such as crushing or allocation, etc.). The visualization interface should have a friendly user experience to facilitate the construction personnel to quickly understand the detection results.

[0099] V. Multi-scenario adaptation and expansion

[0100] (1) Flexible adjustment: Based on different dam types and construction progress, the flight time, camera focal length, and shooting span can be flexibly adjusted, and thresholds can be set for different particle sizes and distributions. This flexibility ensures that the system can adapt to various complex construction environments and improves the accuracy of detection.

[0101] (2) Model real-time improvement mechanism: Ensure sufficient high frame rate and stability in complex environments, and can cooperate with subsequent 5G upgrades or a larger coverage drone fleet. By optimizing algorithms and hardware configurations, the real-time response ability of the model is improved, enabling it to maintain high-efficiency detection performance in dynamic environments.

[0102] (3) Applicable to other scenarios: This method is not only applicable to the detection of earth-rock dams, but can also be extended to the particle size detection and concentration analysis of port yards, large aggregate storage, etc. This adaptability makes the application scope of the technology more extensive and provides more possibilities for future engineering applications.

[0103] The implementation scheme of the present invention not only improves the intelligent level of the earth-rock dam construction process, but also provides reliable technical support for the real-time monitoring and management of related fields.

[0104] The above-mentioned functions of the dam material analysis model, partial convolution module, depthwise separable convolution module, cross-scale feature fusion module, YOLOv8 model, YOLOv8 instance segmentation module, minimum bounding rectangle method, Segment Anything Model (SAM), drones, auto-focus cameras, monitors, alarms, 5G base stations, 5G access devices, database servers, image processing servers, etc. Modules, devices and algorithms can all adopt the functional modules, devices and algorithms in the prior art, or adopt the functional modules, devices and algorithms in the prior art and construct them by conventional technical means.

[0105] The embodiments described above are only used to illustrate the technical idea and features of the present invention. The purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The patent scope of the present invention cannot be limited only by these embodiments. That is, any equivalent changes or modifications made in accordance with the spirit disclosed by the present invention still fall within the patent scope of the present invention.

Claims

1. A high-precision dynamic perception method for coarse-grained dam materials based on 5G and visual intelligence, characterized in that: A 5G base station or 5G access device is set up in the dam area to form a 5G full coverage network in the dam area; drones are used to collect dam surface image data; the image data collected by the drone is transmitted in real time from the drone's collection device end to external devices through the 5G network; a database server and image processing server that communicate with each other are set up on the ground, the database server is used to receive and store data, and the image processing server is used to process the image data; the database server receives and stores the dam surface image data collected by the drone through the 5G network; the image processing server is equipped with a dam material analysis model built based on the YOLOv8 model, which calls the image data in the database server and inputs it into the dam material analysis model; the dam material analysis model is based on the instance segmentation result of the YOLOv8 model, and the minimum circumscribed rectangle method is used to calculate the particle size of stone and aggregate, and identify oversized stone and oversized aggregate, as well as the location and area of ​​the concentrated area of ​​the two; the image processing result is sent from the image processing server to the database server and stored by the database server.

2. According to the high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1, it is characterized in that: The dam material analysis model includes a partial convolution module, a depthwise separable convolution module, a cross-scale feature fusion module, and a YOLOv8 instance segmentation module; The convolutional layers in the YOLOv8 model are replaced by partial convolutional modules and depthwise separable convolutional modules; multi-scale feature extraction of stones and aggregates is performed on the collected dam surface image data; A cross-scale feature fusion module is introduced into the neck structure of the YOLOv8 model to fuse feature maps of different scales. The YOLOv8 instance segmentation module is used to perform pixel-level segmentation on coarser-grained stones and aggregates. Based on the contour of the instance segmentation, the minimum circumscribed rectangle method is used to estimate the length and width of the target stone and aggregate. The ground sampling distance is calculated in combination with the UAV flight parameters to finally obtain the actual particle size.

3. The high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1 is characterized in that: Image data containing different lithologies and particle size distributions are collected from different construction sites to compile training data for the dam material analysis model. When compiling training data, the Segment Anything Model is used for data annotation.

4. The high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1 is characterized in that: Set the threshold range of the following parameters: particle size of stone and aggregate, concentrated area of ​​oversized stone and oversized aggregate; the image processing server compares the image processing result with the corresponding threshold range. When it is judged that the threshold range is exceeded, it sends a command signal according to the percentage exceeding the threshold range and transmits it to the drone through the 5G network. The control system of the drone adjusts the flight altitude, flight speed and flight attitude according to the command signal.

5. The high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1 is characterized in that: The drone is equipped with an automatic zoom camera with a resolution higher than 3840×2160; the focal length adjustment range is 21 to 75mm.

6. The high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1 is characterized in that: The drone flies within the set route and altitude range of 20 to 100 meters.

7. The high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1 is characterized in that: A display and an alarm are provided in the ground workstation, and both the display and the alarm communicate with the image processing server. The image processing server sends the image data and the image data processing results it calls to the display, which is displayed on the display. When the image processing server determines that the image data processing results exceed the set threshold range, it sends an alarm signal to the alarm, which issues an early warning.

8. The high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1 is characterized in that: A return flight path is adopted to cover the entire dam surface, and the dam surface is divided into sections for inspection. The image acquisition cycle of each section is 1 to 1.5 hours. Wind speed and airflow are taken into consideration when planning the flight path, and backup routes are set up to deal with emergencies.

9. The high-precision dynamic perception method of coarse-grained dam materials based on 5G and visual intelligence according to claim 1 is characterized in that: The image area collected by the drone is divided into grids. For any grid, the image data collected by the drone is positioned with the corresponding grid center point as the reference coordinate origin. The collected image data includes the grid number, grid center point reference coordinates, image data reference coordinates and resolution information.

10. A device for high-precision dynamic perception of coarse-grained dam materials based on 5G and visual intelligence, comprising a memory and a processor, characterized in that: The memory is used to store computer programs; the processor is used to execute the computer program and implement the steps of the high-precision dynamic perception method for coarse-grained dam materials based on 5G and visual intelligence as described in any one of claims 1 to 9 when executing the computer program.

Images