Unmanned live working control system based on autonomous decision-making algorithm

Abstract

An unmanned live working control system based on an autonomous decision-making algorithm. It uses drones to conduct flight inspections on transmission lines along a predetermined path and utilizes image processing technology to analyze the status of transmission lines in real time. Then, a multi-stage feature extraction method is adopted, that is, first extract shallow features from the original image, then further mine deep semantic features, and finally integrate this information through a feature joint perception module to generate the final decision result. In addition, a feedback mechanism is designed. When a suspected anomaly is detected, the drone can automatically adjust its position and hover over the target position to collect more detailed high-definition images and transmit this data back to the ground server for further analysis. In this way, not only is the speed of fault diagnosis improved, but also the possibility of false alarms is reduced, thus enhancing the reliability and practicality of the entire system.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology, and more specifically, to an unmanned live working control system based on an autonomous decision-making algorithm. Background Art

[0002] In the power transmission system, the maintenance and repair of high-voltage transmission lines are key links to ensure stable power supply. Traditional live working usually requires manual participation, which is not only inefficient but also poses high safety risks. Especially when dealing with high-voltage lines, workers may face dangers such as electric shock. This method is difficult to meet the requirements of rapid response of modern power grids. With the development of technology, unmanned devices such as unmanned aerial vehicles (UAVs) and crawling robots have been introduced into the power industry to achieve more efficient and safer live working.

[0003] Unmanned Aerial Vehicles (UAVs) and Crawling Robots can carry various sensors, such as high-definition cameras, infrared thermal imagers, Light Detection and Ranging (LiDAR), etc., and can replace humans to perform various tasks of live working on high-voltage lines without direct human intervention, such as inspection, repair, fault repair, or maintenance. These unmanned devices can work in complex terrains and adverse weather conditions, greatly improving work efficiency and safety.

[0004] However, traditional unmanned live working control schemes usually rely on pre-programmed paths or remote operator instructions, lacking real-time adaptability and autonomous decision-making capabilities. This means that they cannot effectively respond or adjust strategies when encountering unforeseen situations. In addition, traditional image processing techniques and sensor data analysis methods may not provide sufficient resolution and accuracy to identify subtle line damages or other problems, which may lead to missed detections or false alarms of faults.

[0005] Therefore, an optimized unmanned live working control system is desired. Summary of the Invention

[0006] This Summary of the Invention section is provided to introduce concepts in a brief form, which will be described in detail in the following Detailed Description section. This Summary of the Invention section is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0007] In a first aspect, there is provided an unmanned live working control system based on an autonomous decision-making algorithm, which includes:

[0008] A UAV flight control module for controlling the UAV to fly along the power transmission line to be detected;

[0009] The transmission line segment image acquisition module is used to collect the surface state images of a predetermined line segment of the transmission line to be detected through the camera of the unmanned aerial vehicle, so as to obtain a time queue of the surface state images of the transmission line segment;

[0010] The image-level difference coefficient calculation module is used to calculate the image-level difference coefficient between every two adjacent surface state images of the transmission line segment in the time queue of the surface state images of the transmission line segment, so as to obtain a time queue of the image-level difference coefficient;

[0011] The suspected abnormal point determination module is used to determine the suspected abnormal points based on the comparison between the time queue of the image-level difference coefficient and a preset threshold;

[0012] The autonomous decision-making module is used to input the surface state image of the transmission line segment corresponding to the suspected abnormal point into an autonomous decision-making model based on deep learning to obtain a decision result, and the decision result is whether the transmission line segment is damaged;

[0013] The unmanned aerial vehicle hovering acquisition module is used to control the unmanned aerial vehicle to hover and collect the high-definition image of the transmission line segment and transmit it back to the ground server in response to the decision result that the transmission line segment is damaged.

[0014] Optionally, the autonomous decision-making module includes: an image shallow feature extraction unit, which is used to extract the shallow features of the transmission line segment image from the surface state image of the transmission line segment to obtain a shallow feature map of the surface state of the transmission line segment; an image deep semantic feature extraction unit, which is used to extract the deep semantic features of the transmission line segment image from the surface state image of the transmission line segment to obtain a deep semantic feature map of the surface state of the transmission line segment; a feature joint perception processing unit, which is used to perform feature joint perception processing guided by the semantic information field on the shallow feature map of the surface state of the transmission line segment and the deep semantic feature map of the surface state of the transmission line segment to obtain a deep and shallow joint perception coding feature of the surface state of the transmission line segment; a decision result determination unit, which is used to determine the decision result based on the deep and shallow joint perception coding feature of the surface state of the transmission line segment.

[0015] Optionally, the image shallow feature extraction unit is used to: extract the shallow features of the transmission line segment image from the surface state image of the transmission line segment through a multi-scale feature extractor for the surface state of the transmission line segment based on the dilated pyramid network to obtain the shallow feature map of the surface state of the transmission line segment.

[0016] Optionally, the image deep semantic feature extraction unit is used to: extract the deep semantic features of the transmission line segment image from the surface state image of the transmission line segment through the multi-scale feature extractor for the surface state of the transmission line segment based on the dilated pyramid network to obtain the deep semantic feature map of the surface state of the transmission line segment.

[0017] Optionally, the feature joint perception processing unit includes: an upsampling subunit, configured to upsample the deep semantic feature map of the surface state of the transmission line segment to obtain an upsampled deep semantic feature map of the surface state of the transmission line segment, wherein the upsampled deep semantic feature map of the surface state of the transmission line segment has the same size as the shallow feature map of the surface state of the transmission line segment; a semantic information field calculation subunit, configured to calculate the semantic information field between the upsampled deep semantic feature map of the surface state of the transmission line segment and the shallow feature map of the surface state of the transmission line segment to obtain the semantic information field of the surface state of the transmission line segment, and map the upsampled deep semantic feature map of the surface state of the transmission line segment to the semantic information field of the surface state of the transmission line segment to obtain a field-modulated deep semantic feature map of the surface state of the transmission line segment; a feature joint perception subunit, configured to input the shallow feature map of the surface state of the transmission line segment and the field-modulated deep semantic feature map of the surface state of the transmission line segment into a feature joint perception module based on a multi-attention structure to obtain a shallow and deep joint perception encoded feature map of the surface state of the transmission line segment as the shallow and deep joint perception encoded feature of the surface state of the transmission line segment.

[0018] Optionally, the semantic information field calculation subunit includes: a semantic information field prediction secondary subunit, configured to input the upsampled deep semantic feature map of the surface state of the transmission line segment and the shallow feature map of the surface state of the transmission line segment after feature connection into a semantic information field predictor based on gated convolution to obtain the semantic information field of the surface state of the transmission line segment; a feature map mapping secondary subunit, configured to map the upsampled deep semantic feature map of the surface state of the transmission line segment to the semantic information field of the surface state of the transmission line segment to obtain the field-modulated deep semantic feature map of the surface state of the transmission line segment.

[0019] Optionally, the semantic information field prediction secondary subunit is configured to: input the upsampled deep semantic feature map of the surface state of the transmission line segment and the shallow feature map of the surface state of the transmission line segment after feature connection into a convolutional layer with a convolutional kernel of for processing to obtain a multi-scale semantic fusion feature map of the surface state of the transmission line segment; perform point convolution processing on the multi-scale semantic fusion feature map of the surface state of the transmission line segment to obtain a multi-dimensional semantic modulation feature map of the surface state of the transmission line segment as the semantic information field of the surface state of the transmission line segment.

[0020] Optionally, the feature joint perception subunit is configured to: perform point-by-point convolution processing on the deep semantic feature map of the surface state of the field modulation transmission line segment and then input it into the batch normalization layer, and then perform activation processing using the Sigmoid function to obtain the semantic attention optimized expression feature map of the surface state of the transmission line segment; perform multiple attention optimizations on the shallow feature map of the surface state of the transmission line segment to obtain the semantic attention optimized expression feature map of the surface state of the transmission line segment; add the semantic attention optimized expression feature map of the surface state of the transmission line segment and the semantic attention optimized expression feature map of the surface state of the transmission line segment by position points to obtain the deep and shallow joint perception coding feature map of the surface state of the transmission line segment.

[0021] Optionally, the decision result determination unit is configured to: input the deep and shallow joint perception coding feature map of the surface state of the transmission line segment into a decision module based on a classifier to obtain the decision result.

[0022] With the above technical solutions, it conducts flight inspections on transmission lines by drones according to a predetermined path, and uses image processing technology to analyze the status of transmission lines in real time; then, a multi-stage feature extraction method is adopted, that is, first extract shallow features from the original image, then further mine deep semantic features, and finally integrate this information through a feature joint perception module to generate the final decision result. In addition, a feedback mechanism is also designed. When a suspected anomaly is detected, the drone can automatically adjust its position and hover over the target position to collect more detailed high-definition images, and transmit this data back to the ground server for further analysis. In this way, not only the speed of fault diagnosis is improved, but also the possibility of false alarms is reduced, thereby enhancing the reliability and practicality of the entire system.

[0023] Other features and advantages of the present application will be described in detail in the subsequent specific implementation section. Description of the Drawings

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0025] Figure 1 It is a block diagram of an unmanned live working control system based on an autonomous decision-making algorithm according to an embodiment of the present application.

[0026] Figure 2 It is a block diagram of the autonomous decision-making module in the unmanned live working control system based on the autonomous decision-making algorithm according to the embodiment of the present application.

[0027] Figure 3 It is a block diagram of the feature joint perception processing unit in the unmanned live working control system based on the autonomous decision-making algorithm according to an embodiment of the present application.

[0028] Figure 4 It is a block diagram of the semantic information field calculation sub-unit in the unmanned live working control system based on the autonomous decision-making algorithm according to an embodiment of the present application.

[0029] Figure 5 It is a flowchart of the unmanned live working control method based on the autonomous decision-making algorithm according to an embodiment of the present application. Detailed implementation manners

[0030] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not used to limit the protection scope of the present application.

[0031] It should be understood that the various steps recited in the method embodiments of the present application can be executed in a different order and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this regard.

[0032] The term "including" and its variations used herein are open-ended, that is, "including but not limited to". The term "based on" is "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0033] It should be noted that the concepts such as "first" and "second" mentioned in the present application are only used to distinguish different devices, modules or units, and are not used to limit the order of the functions executed by these devices, modules or units or their interdependent relationships.

[0034] It should be noted that the modifications of "one" and "multiple" mentioned in the present application are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly stated in the context, it should be understood as "one or more".

[0035] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0036] The following will elaborate on the specific implementation manners of the present application in conjunction with the accompanying drawings.

[0037] In response to the above technical problems, in the technical solution of the present application, a control system for unmanned live working based on an autonomous decision-making algorithm is proposed. It combines advanced drone technology and artificial intelligence algorithms, especially deep learning, aiming to improve the intelligent level of automatic inspection of transmission lines. The system conducts flight inspections on transmission lines by drones along a predetermined path and uses image processing technology to analyze the status of transmission lines in real time. To improve the accuracy of fault detection, the system adopts a multi-stage feature extraction method, that is, first extracts shallow features from the original image, then further mines deep semantic features, and finally integrates this information through a feature joint perception module to generate the final decision result. In addition, the system also designs a feedback mechanism. When a suspected anomaly is detected, the drone can automatically adjust its position and hover over the target position to collect more detailed high-definition images and transmit this data back to the ground server for further analysis. This not only improves the speed of fault diagnosis but also reduces the possibility of false alarms, thereby enhancing the reliability and practicality of the entire system.

[0038] Figure 1 As shown in the block diagram of the control system for unmanned live working based on an autonomous decision-making algorithm according to an embodiment of the present application, Figure 1 as shown, the control system 100 for unmanned live working based on an autonomous decision-making algorithm according to an embodiment of the present application includes:

[0039] A drone flight control module 110, configured to control the drone to fly along the transmission line to be detected;

[0040] An image acquisition module 120 for the transmission line section, configured to collect surface state images of a predetermined line section of the transmission line to be detected through the camera of the drone to obtain a time queue of surface state images of the transmission line section;

[0041] An image-level difference coefficient calculation module 130, configured to calculate the image-level difference coefficient between every two adjacent surface state images of the transmission line section in the time queue of surface state images of the transmission line section to obtain a time queue of image-level difference coefficients;

[0042] A suspected anomaly point determination module 140, configured to determine suspected anomaly points based on the comparison between the time queue of image-level difference coefficients and a preset threshold;

[0043] An autonomous decision-making module 150, configured to input the surface state image of the transmission line section corresponding to the suspected anomaly point into an autonomous decision-making model based on deep learning to obtain a decision result, where the decision result is whether there is damage to the transmission line section;

[0044] The drone hovering acquisition module 160 is configured to, in response to the decision result indicating that there is damage in the transmission line segment, control the drone to hover, acquire high-definition images of the transmission line segment, and transmit them back to the ground server.

[0045] In the above-mentioned unmanned live working control system 100 based on the autonomous decision-making algorithm, the drone flight control module 110 is used to control the drone to fly along the transmission line to be detected. Specifically, the drone flight control module 110 first needs to use Geographic Information System (GIS) data, combined with high-resolution maps and terrain information, to set the flight route of the drone, which includes determining the starting point, ending point, and key inspection points along the way. On this basis, considering factors such as weather conditions, air traffic control requirements, or unexpected situations such as bird activities, the flight control system can adjust the flight path in real time during flight to ensure safety and efficiency. This dynamic path adjustment ability is crucial for ensuring the smooth completion of the drone's mission.

[0046] Navigation and positioning are important components of the drone flight control module 110. The Global Positioning System (GPS) provides accurate position information, while the Inertial Measurement Unit (IMU) is used to compensate for offsets caused by wind or other factors and maintain a stable flight attitude. In addition, visual navigation assistance technology is also widely used. The camera installed on the drone captures images and uses computer vision algorithms to identify feature points such as ground markers or power towers as additional navigation references, thereby ensuring that the drone flies along the predetermined path. Through these technical means, even in a complex environment, the drone can accurately find the target and maintain the correct heading.

[0047] To achieve smooth flight, the drone adopts an automatic flight mode. Once the path planning is completed, the built-in flight controller will control the speed, altitude, and direction according to preset parameters. At the same time, the obstacle avoidance function also plays an important role. The application of Light Detection and Ranging (LiDAR), ultrasonic sensors, or stereo vision technology enables the drone to automatically decelerate or change course when approaching an obstacle, avoiding collisions. This feature greatly improves the safety and reliability of drone operations, especially in a complex transmission line environment.

[0048] In a typical power transmission line inspection task, the UAV flight control module 110 loads the map information of this section of the power transmission line from the database and formulates a flight plan based on this. After taking off, the UAV relies on GPS and IMU to maintain the established flight path, and at the same time uses a high-definition camera and an infrared thermal imager to scan the line. When encountering complex terrain or adverse weather conditions, the flight control system will automatically adjust the flight altitude and speed to ensure the safety of the equipment. If an abnormal situation is detected, such as signs of damage to an insulator somewhere, the UAV will suspend its advance, stay above the fault point, and take clearer photos for subsequent analysis. The whole process reflects the high intelligence and flexibility of the UAV flight control module, providing strong technical support for the maintenance of the power system.

[0049] Suppose a report is received stating that an unknown short-circuit accident has occurred on the high-voltage line in a certain area, and it is necessary to quickly dispatch a UAV to investigate. At this time, the flight control module quickly plans a new path that is the shortest and avoids known dangerous areas. After arriving at the scene, the UAV immediately activates the omnidirectional scanning mode, not only recording visible light images but also collecting temperature distribution maps to determine the specific location and severity of the fault. All the acquired data will be synchronously sent to the command center to provide a decision-making basis for the repair team. This emergency response scenario demonstrates the rapid response ability and high-efficiency handling level of the UAV flight control module in emergencies.

[0050] In the above-mentioned unmanned live working control system 100 based on the autonomous decision-making algorithm, the power transmission line segment image acquisition module 120 is used to collect the surface state images of a predetermined line segment of the power transmission line to be detected through the camera of the UAV to obtain a time queue of the surface state images of the power transmission line segment. Among them, the power transmission line segment image acquisition module 120 collects the surface state images of the predetermined power transmission line segment through a high-definition camera or other special sensors (such as an infrared thermal imager) installed on the UAV. This process not only requires precise control of the position and attitude of the UAV to ensure that the image covers the entire inspection area, but also needs to adjust the camera parameters such as focal length and aperture size according to different inspection requirements to obtain high-quality image data. In addition, considering that the power transmission line is usually located in a complex environment and may be affected by factors such as weather conditions and light changes, the image acquisition module also needs to have a certain degree of adaptability and flexibility to ensure the consistency and stability of the image quality.

[0051] When the UAV flies according to the preset path, the transmission line segment image acquisition module 120 will activate the high-definition camera or other dedicated sensors carried on the UAV, and continuously capture the surface state of the transmission line segment at specific time intervals or in a distance-triggered manner. These images are recorded to form a time series, that is, the time queue of the surface state images of the transmission line segment. The image at each time point represents the observation result of the transmission line by the UAV at a certain position, and the entire time queue constitutes a comprehensive state record of this section of the transmission line. To ensure that the collected images can meet the requirements of subsequent analysis, the transmission line segment image acquisition module 120 must consider multiple aspects, including but not limited to image resolution, viewing angle and perspective, light compensation, and environmental adaptability.

[0052] Specifically, in a regular transmission line inspection task, after the UAV takes off from the ground station, it flies along the pre-planned path. At this time, the transmission line segment image acquisition module 120 starts to work. It triggers the camera to take pictures at fixed intervals (such as once per second), and at the same time records the geographical location information corresponding to each photo. As the UAV gradually approaches the target line, the transmission line segment image acquisition module 120 will adjust the camera settings according to the actual situation, such as switching to the wide-angle mode to obtain a larger field of view, or switching to the telephoto lens to focus on certain key parts. During the process of the UAV flying over the entire inspection area, the transmission line segment image acquisition module 120 continuously collects a large number of images and stores them as a time series. Once the inspection is completed, all the images will be transmitted back to the ground station for further processing and analysis. High-resolution images can provide clearer details, which helps to identify subtle fault or damage features; appropriate viewing angles and perspectives are crucial for capturing the complete transmission line structure, and the UAV can optimize the viewing angle by adjusting the flight height and direction; the change of natural light may affect the image quality, so the transmission line segment image acquisition module 120 should include functions of automatic white balance and exposure compensation to maintain the consistency of the images; for image acquisition under different weather conditions, such as foggy days, rainy days, or strong light irradiation, the transmission line segment image acquisition module 120 takes corresponding measures, such as using a polarization filter to reduce the interference of reflected light, or enabling infrared imaging technology to penetrate the smoke, etc.

[0053] In the above-mentioned unmanned live working control system 100 based on the autonomous decision-making algorithm, the image-level difference coefficient calculation module 130 is used to calculate the image-level difference coefficient between every two adjacent surface state images of the transmission line segment in the time queue of the surface state images of the transmission line segment to obtain the time queue of the image-level difference coefficient. Before formally calculating the image-level difference coefficient, it is necessary to preprocess every two adjacent surface state images of the transmission line segment in the time queue of the surface state images of the transmission line segment. This step includes converting the color image to a grayscale image to reduce the computational complexity; ensuring that all images have the same resolution and scale for subsequent feature extraction and matching; using filters (such as Gaussian blur or median filter) to remove random noise in the image and improve the image quality. After the preprocessing is completed, the image-level difference coefficient calculation module 130 enters the core step - feature extraction and matching. Common feature extraction methods include using algorithms such as Harris corner detection to find significant feature points in the image; marking the object contour through the Canny operator or other edge detection techniques; using local binary pattern (LBP) to describe texture features, which helps to distinguish different types of materials or surface conditions; applying scale-invariant feature transform (SIFT) / speeded-up robust features (SURF) to identify stable feature points that are not affected by scale and rotation, which is suitable for matching in complex scenarios.

[0054] The next step is to find the corresponding relationship, that is, to pair the feature points of every two adjacent surface state images of the transmission line segment. This usually involves calculating the distance between feature points (such as Euclidean distance), and then screening out the closest pairs. To improve the matching accuracy, geometric constraint conditions can also be introduced, such as verifying whether the pairs are reasonable based on the affine transformation model. After having the matched feature points, the image-level difference coefficient between every two adjacent surface state images of the transmission line segment can be calculated. The specific calculation method of the image-level difference coefficient depends on the selected feature type and the requirements of the application scenario. The following are several common calculation methods: directly comparing the difference in pixel values at the corresponding positions of every two adjacent surface state images of the transmission line segment, and commonly using the mean square error (MSE) or structural similarity index (SSIM) to measure the overall similarity; calculating the length of the displacement vector between the paired feature points, which reflects the position change between every two adjacent surface state images of the transmission line segment; constructing a color or gradient direction histogram, and using statistical indicators such as chi-square distance or cross-entropy to evaluate its similarity; training a convolutional neural network (CNN) to automatically learn the high-level semantic differences between images, and this method can capture more complex visual information. Through any one or a combination of the above methods, the image-level difference coefficient calculation module 130 generates an image-level difference coefficient for every pair of adjacent surface state images of the transmission line segment. The image-level difference coefficients constitute the time queue of the image-level difference coefficient, providing basic data support for subsequent outlier detection.

[0055] In a regular power transmission line inspection task, the unmanned aerial vehicle (UAV) flies along a pre-planned path and triggers the camera to take pictures at regular intervals, forming a time queue of the surface state images of the power transmission line section. After receiving the time queue of the surface state images of the power transmission line section, the image-level difference coefficient calculation module 130 first preprocesses it, including grayscale conversion, size standardization, and noise removal. Then, by applying the corner detection algorithm, a number of stable feature points are found in every two adjacent surface state images of the power transmission line section in the time queue. Subsequently, an attempt is made to establish the correspondence of the feature points between every two adjacent surface state images in the time queue of the surface state images of the power transmission line section, and the length of the displacement vector between each paired feature point is calculated as the image-level difference coefficient. Finally, all the image-level difference coefficients are organized into a time queue of the image-level difference coefficients for further analysis.

[0056] In the above-mentioned unmanned live working control system 100 based on the autonomous decision-making algorithm, the suspected abnormal point determination module 140 is used to determine the suspected abnormal points based on the comparison between the time queue of the image-level difference coefficients and a preset threshold. To effectively distinguish normal minor changes from possible faults, it is first necessary to set a reasonable threshold. This step is usually carried out based on historical data and domain expertise. For example, inspection data from multiple different time periods and different weather conditions can be collected, the distribution of the image difference coefficients under normal conditions can be statistically analyzed, and then an appropriate percentile can be selected as the threshold. In addition, considering the changes in the actual application environment, a dynamic adjustment mechanism can also be set so that the threshold can be adaptively updated according to factors such as seasonal factors and geographical locations.

[0057] Once the threshold is set, it is then necessary to conduct an in-depth analysis of the time series of the image-level difference coefficients. The key here is to not only focus on whether a single difference coefficient exceeds the threshold, but also consider the change trend in the entire time series. Specifically, if a certain difference coefficient suddenly rises or falls significantly and exceeds the set threshold range, then this point is regarded as a preliminary suspect; if several consecutive difference coefficients exceed the threshold, it indicates that there may be a relatively serious state change during this period of time and needs to be focused on; sometimes, due to the influence of the natural environment (such as slight swings caused by wind speed changes), the image difference coefficients may show a certain periodic fluctuation. For this situation, the high-frequency noise can be removed through a filter, and the low-frequency components can be retained to avoid misjudgment caused by normal fluctuations.

[0058] To improve the accuracy of judging suspected abnormal points, it is also necessary to conduct a comprehensive evaluation by combining other relevant information. For example, auxiliary data such as geographical coordinates and meteorological conditions recorded during the flight of the drone can be referred to; multi-sensor fusion technology can also be used to integrate information from different types of sensors (such as infrared thermal imagers, lidar) to form a more comprehensive understanding. Finally, through the auxiliary decision-making of machine learning algorithms or expert systems, determine which points actually have potential problems and mark them as "suspected abnormal points" for subsequent detailed inspection.

[0059] Suppose a certain area is hit by a heavy rainstorm, resulting in damage to some power transmission lines, and it is urgent to dispatch drones for emergency assessment. Since bad weather may lead to a decline in image quality, additional measures must be taken to ensure the effectiveness of the judgment results. For example, the temperature distribution map obtained by an infrared thermal imager can be preferentially selected for feature extraction because such images are less affected by lighting conditions and are more suitable for use at night or in low visibility environments. In addition, considering the interference of reflected light caused by rain, the local binary pattern (LBP) technology can also be combined to focus on analyzing the texture changes on the surface of the utility poles. When analyzing the time series of the image-level difference coefficient, not only pay attention to whether a single difference coefficient exceeds the threshold, but also pay special attention to those data points that rapidly increase within a short period of time, which may be caused by sudden physical damage. At the same time, a dynamic adjustment mechanism is introduced to appropriately relax or tighten the threshold range according to real-time meteorological data to adapt to different environmental conditions. Finally, through the comprehensive application of various methods and technologies, the suspected abnormal point determination module 140 successfully identifies multiple suspected abnormal points, helping the repair team to respond quickly.

[0060] The specific method example for determining suspected abnormal points shows different implementation methods. In a simple implementation, the suspected abnormal point determination module 140 can determine suspected abnormal points by directly comparing the time series of the image-level difference coefficient with a fixed threshold. Specifically, assume there is a difference coefficient time series {Di}, where i represents the time point of the i-th shot. A fixed threshold T is set, then for each Di, if Di>T, then this point is considered a suspected abnormal point. This method is applicable in most cases, especially when historical data shows that the change of the difference coefficient is relatively stable.

[0061] To improve the accuracy of determining suspected abnormal points, a method based on a dynamic threshold can be adopted. This method not only considers the value of the current difference coefficient but also combines the average value and standard deviation within a previous period of time to set the threshold. Specifically, let Di be the difference coefficient of the i-th shot, and mi and si represent the average value and standard deviation of the difference coefficients from the (i - n)-th to the (i - 1)-th shot respectively. Then, the dynamic threshold Ti can be defined as:

[0062] Here, k is an adjustable parameter used to control the sensitivity of the threshold. If Di > Ti, then this point is considered a suspected abnormal point. This method can better adapt to environmental changes and reduce the possibility of false alarms.

[0063] In the above-mentioned unmanned live working control system 100 based on the autonomous decision-making algorithm, the autonomous decision-making module 150 is configured to input the surface state image of the transmission line segment corresponding to the suspected abnormal point into the autonomous decision-making model based on deep learning to obtain a decision result, and the decision result is whether there is damage to the transmission line segment. Figure 2 It is a block diagram of the autonomous decision-making module in the unmanned live working control system based on the autonomous decision-making algorithm according to an embodiment of the present application, as Figure 2 shown, the autonomous decision-making module 150 includes: an image shallow feature extraction unit 151, configured to extract the shallow features of the transmission line segment image from the surface state image of the transmission line segment to obtain the shallow feature map of the surface state of the transmission line segment; an image deep semantic feature extraction unit 152, configured to extract the deep semantic features of the transmission line segment image from the surface state image of the transmission line segment to obtain the deep semantic feature map of the surface state of the transmission line segment; a feature joint perception processing unit 153, configured to perform feature joint perception processing guided by the semantic information field on the shallow feature map of the surface state of the transmission line segment and the deep semantic feature map of the surface state of the transmission line segment to obtain the deep and shallow joint perception coding features of the surface state of the transmission line segment; a decision result determination unit 154, configured to determine the decision result based on the deep and shallow joint perception coding features of the surface state of the transmission line segment.

[0064] Specifically, in the above-mentioned unmanned live working control system based on the autonomous decision-making algorithm, it is crucial to input the surface state image of the transmission line segment corresponding to the suspected abnormal point into the autonomous decision-making model based on deep learning for the damage detection process of the transmission line segment. This is because the transmission lines in different regions and at different times may have different aging conditions or damage characteristics. Traditional fault detection methods rely on fixed thresholds or rules, and this method often appears to be insufficiently flexible and accurate when facing complex actual environments. However, the method of using the autonomous decision-making model based on deep learning to detect the damage of the transmission line segment can identify more subtle and complex line defect patterns through a large amount of data training, thereby significantly improving the accuracy of fault detection such as the damage of the transmission line. At the same time, by continuously learning new sample data to adjust its own parameters, the adaptability of the system to various environmental changes can be enhanced.

[0065] Specifically, in the process of autonomous decision-making, the technical concept of this application is to analyze the surface state image of the transmission line segment corresponding to the suspected abnormal point by using image processing and analysis algorithms based on artificial intelligence and deep learning, so as to capture the multi-scale joint perception features of the surface state of the transmission line segment in the image, and thus determine the damage of the transmission line and return the decision result. In this way, when it is detected that there is damage to the transmission line segment, the drone can be controlled to perform further high-definition image acquisition and transmission, ensuring the safe and reliable operation of the power grid, and providing strong technical support for the efficient management and maintenance of the power transmission system.

[0066] In a specific embodiment of this application, in the image shallow feature extraction unit 151 and the image deep semantic feature extraction unit 152, first, the shallow features of the transmission line segment image are extracted from the surface state image of the transmission line segment to obtain the shallow feature map of the surface state of the transmission line segment, and the deep semantic features of the transmission line segment image are extracted from the surface state image of the transmission line segment to obtain the deep semantic feature map of the surface state of the transmission line segment. It should be understood that since the surface state image of the transmission line segment contains multi-faceted feature information about the surface state of the transmission line segment, specifically including the edges, textures, colors, and semantic features of the transmission line segment, these features can not only directly reflect the surface state of the transmission line, such as whether there are obvious cracks or corrosion marks, but also help to understand the state of the entire transmission line segment, rather than just the local performance, providing a basis for subsequent more complex analysis and damage defect detection of the transmission line segment. Specifically, in a specific example of this application, the multi-scale feature extractor of the surface state of the transmission line segment based on the dilated pyramid network is used to perform multi-scale feature extraction on the surface state image of the transmission line segment, and the shallow feature map of the surface state of the transmission line segment and the deep semantic feature map of the surface state of the transmission line segment can be obtained. It is worth mentioning that the combined use of the shallow features and deep semantic features of the surface state of the transmission line segment can complement each other and provide a more rich and comprehensive description of the surface state of the transmission line segment. This can not only improve the accuracy of the system's damage detection of the transmission line segment, but also enhance its adaptability and response speed to complex environmental changes.

[0067] In a specific embodiment of this application, Figure 3 For the block diagram of the feature joint perception processing unit in the unmanned live working control system based on the autonomous decision-making algorithm according to the embodiment of this application, as Figure 3As shown in the figure, the feature joint perception processing unit 153 includes: an upsampling sub-unit 1531, configured to perform upsampling on the deep semantic feature map of the surface state of the transmission line segment to obtain an upsampled deep semantic feature map of the surface state of the transmission line segment, wherein the upsampled deep semantic feature map of the surface state of the transmission line segment has the same size as the shallow feature map of the surface state of the transmission line segment; a semantic information field calculation sub-unit 1532, configured to calculate the semantic information field between the upsampled deep semantic feature map of the surface state of the transmission line segment and the shallow feature map of the surface state of the transmission line segment to obtain a semantic information field of the surface state of the transmission line segment, and map the upsampled deep semantic feature map of the surface state of the transmission line segment to the semantic information field of the surface state of the transmission line segment to obtain a field-modulated deep semantic feature map of the surface state of the transmission line segment; a feature joint perception sub-unit 1533, configured to input the shallow feature map of the surface state of the transmission line segment and the field-modulated deep semantic feature map of the surface state of the transmission line segment into a feature joint perception module based on a multi-attention structure to obtain a deep and shallow joint perception encoded feature map of the surface state of the transmission line segment as the deep and shallow joint perception encoded feature of the surface state of the transmission line segment.

[0068] It should be understood that since the shallow feature map of the surface state of the transmission line segment and the deep semantic feature map of the surface state of the transmission line segment respectively contain shallow features and deep semantic features related to the surface state of the transmission line segment, by combining shallow and deep features, the visual manifestations of these surface defects of the transmission line segment can be captured more comprehensively, thereby improving the accuracy of fault identification and damage detection. Based on this, in order to be able to perform surface state recognition and damage detection of the transmission line segment more fully and comprehensively, in the technical solution of this application, the shallow feature map of the surface state of the transmission line segment and the deep semantic feature map of the surface state of the transmission line segment are further subjected to feature joint perception processing guided by a semantic information field to obtain a deep and shallow joint perception encoded feature of the surface state of the transmission line segment. Through the feature joint perception processing guided by the semantic information field, the shallow and deep features of the surface state of the transmission line segment can be combined, and the semantic information field can be used as a guide to realize the interaction and modulation between these two features, thereby generating a richer and semantically interpretable deep and shallow joint perception encoded feature of the surface state of the transmission line segment.

[0069] Specifically, the upsampling sub-unit 1531 is configured to perform upsampling on the deep semantic feature map of the surface state of the transmission line segment to obtain an upsampled deep semantic feature map of the surface state of the transmission line segment, wherein the upsampled deep semantic feature map of the surface state of the transmission line segment has the same size as the shallow feature map of the surface state of the transmission line segment.

[0070] Specifically, compared with the traditional visual processing architecture, the feature joint perception processing guided by the semantic information field enables the deep semantic feature map of the surface state of the upsampled transmission line segment and the shallow feature map of the surface state of the transmission line segment to interact at the same scale by introducing upsampling and feature connection, solves the problem of multi-scale information fusion, and improves the model's understanding ability of complex scenes.

[0071] This process can be expressed by the formula:

[0072]

[0073] Wherein, is the deep semantic feature map of the surface state of the transmission line segment, represents the upsampling process, is the deep semantic feature map of the surface state of the upsampled transmission line segment.

[0074] Specifically, Figure 4 is the block diagram of the semantic information field calculation sub-unit in the unmanned live working control system based on the autonomous decision-making algorithm according to the embodiment of the present application. As Figure 4 shown, the semantic information field calculation sub-unit 1532 includes: a semantic information field prediction secondary sub-unit 15321, configured to perform feature connection on the deep semantic feature map of the surface state of the upsampled transmission line segment and the shallow feature map of the surface state of the transmission line segment, and then input the result into a semantic information field predictor based on gated convolution to obtain the semantic information field of the surface state of the transmission line segment; a feature map mapping secondary sub-unit 15322, configured to map the deep semantic feature map of the surface state of the upsampled transmission line segment to the semantic information field of the surface state of the transmission line segment to obtain the field-modulated deep semantic feature map of the surface state of the transmission line segment.

[0075] More specifically, the semantic information field prediction secondary sub-unit 15321 is configured to perform feature connection on the deep semantic feature map of the surface state of the upsampled transmission line segment and the shallow feature map of the surface state of the transmission line segment, and then process the result through a convolutional layer with a convolutional kernel of to obtain a multi-scale semantic fusion feature map of the surface state of the transmission line segment; perform point convolution processing on the multi-scale semantic fusion feature map of the surface state of the transmission line segment to obtain a multi-dimensional semantic modulation feature map of the surface state of the transmission line segment as the semantic information field of the surface state of the transmission line segment.

[0076] Moreover, in the feature map mapping secondary subunit 15322, by constructing the semantic information field of the surface state of the transmission line segment and modulating the deep semantic feature map of the upsampled transmission line segment surface state accordingly, it is ensured that the field-modulated deep semantic feature map of the transmission line segment surface state not only contains rich detailed information but also is consistent with the structure of the shallow feature map of the transmission line segment surface state. In particular, using the feature joint method guided by the semantic information field enables the model to better understand complex scenarios and correctly identify the key objects and damage defects of the transmission line segment even in the face of occlusion or deformation. At the same time, this approach can also provide more detailed and interpretable feature representations, helping to detect early small faults or potential risks of transmission line damage and providing more reliable support for autonomous decision-making.

[0077] This process can be expressed by the formula:

[0078]

[0079]

[0080] Where, is the deep semantic feature map of the upsampled transmission line segment surface state, is the shallow feature map of the transmission line segment surface state, represents the feature connection operation, is a convolutional layer with a convolutional kernel of is the point convolution processing, is the semantic information field of the transmission line segment surface state, is the element-wise multiplication by position, is the field-modulated deep semantic feature map of the transmission line segment surface state.

[0081] Specifically, the feature joint perception subunit 1533 is used to perform point convolution processing on the field-modulated deep semantic feature map of the transmission line segment surface state and then input it into the batch normalization layer, and then perform activation processing using the Sigmoid function to obtain the optimized expression feature map of the semantic attention of the transmission line segment surface state; perform multiple attention optimizations on the shallow feature map of the transmission line segment surface state to obtain the optimized expression feature map of the semantic attention of the transmission line segment surface state; perform element-wise addition of the optimized expression feature map of the semantic attention of the transmission line segment surface state and the optimized expression feature map of the semantic attention of the transmission line segment surface state to obtain the deep and shallow joint perception coding feature map of the transmission line segment surface state.

[0082] This process can be expressed by the formula:

[0083] ​

[0084]

[0085]

[0086] Among them, is pointwise convolution processing, is global average pooling operation, is per-channel convolution processing, is batch normalization layer, is Sigmoid function, is the optimized expression feature map of the surface state semantic attention of the transmission line segment, is the optimized expression feature map of the surface state shallow attention of the transmission line segment, is addition by position, is the deep and shallow joint perception coding feature map of the surface state of the transmission line segment, is the deep semantic feature map of the surface state of the transmission line segment modulated by the field, is the shallow feature map of the surface state of the transmission line segment.

[0087] In a specific embodiment of the present application, the decision result determination unit 154 is configured to: input the deep and shallow joint perception coding feature map of the surface state of the transmission line segment into a decision module based on a classifier to obtain the decision result. Then, input the deep and shallow joint perception coding feature map of the surface state of the transmission line segment into a decision module based on a classifier to obtain the decision result. That is to say, the deep and shallow joint perception features of the surface state of the transmission line segment are used for classification processing, so as to make a decision on whether there is damage to the transmission line segment and return the decision result. In this way, when it is detected that there is damage to the transmission line segment, the unmanned aerial vehicle can be controlled to perform further high-definition image acquisition and transmission, ensuring the safe and reliable operation of the power grid and providing strong technical support for the efficient management and maintenance of the power transmission system.

[0088] That is, when the shallow feature map of the surface state of the transmission line segment and the deep semantic feature map of the surface state of the transmission line segment respectively represent the shallow features of the surface state of the transmission line segment and the deep semantic features of the surface state of the transmission line segment determined by shallow feature extraction and deep semantic feature extraction of the surface state image of the transmission line segment, the deep and shallow joint perception coding feature map of the surface state of the transmission line segment will also have the diversity of the joint interaction distribution of heterogeneous surface states of the transmission line segment based on the respective feature fine-grained cores under the cross-feature domain feature distribution. That is to say, due to the information differences between different feature layers (for example, the shallow layer captures local details and the deep layer identifies global anomalies), there is a problem of uneven distribution in the cross-domain fusion of the feature maps. Therefore, when the deep and shallow joint perception coding feature map of the surface state of the transmission line segment is classified by a classifier, it will affect the accuracy of the classification result. In this way, considering that when the weight matrix of the classifier acts on the deep and shallow joint perception coding feature vector of the surface state of the transmission line segment after the deep and shallow joint perception coding feature map of the surface state of the transmission line segment is unfolded, the randomness of the weight response of the weight matrix of the classifier caused by the distribution diversity of the deep and shallow joint perception coding feature map of the surface state of the transmission line segment affects the accuracy of the classification result.

[0089] Preferably, inputting the deep and shallow joint perception coding feature map of the surface state of the transmission line segment into a decision module based on a classifier to obtain the decision result includes:

[0090] Calculating the parameter expectation and distribution dispersion of the deep and shallow joint perception coding feature map of the surface state of the transmission line segment, and performing normalization processing on the parameter expectation based on the distribution dispersion to obtain the surface state distribution probability index of the transmission line segment:

[0091] Performing a dot product operation on the deep and shallow joint perception coding feature map of the surface state of the transmission line segment and the reciprocal of the surface state distribution probability index of the transmission line segment to obtain the random robustness map of the surface state of the transmission line segment:

[0092]

[0093]

[0094] Wherein, represents the deep and shallow joint perception coding feature map of the surface state of the transmission line segment, represents the dot product, and respectively represent the parameter expectation and distribution dispersion of the deep and shallow joint perception coding feature map of the surface state of the transmission line segment, that is, the mean and standard deviation of the set composed of all feature values of the deep and shallow joint perception coding feature map of the surface state of the transmission line segment, represents the surface state distribution probability index of the transmission line segment, Represents the random robust graph of the surface state of the transmission line segment;

[0095] Analyze the number of eigenvalues greater than the expected parameter value in the feature value of the combined deep and shallow perception coding feature map of the surface state of the transmission line segment as the probability stack base number;

[0096] After performing point addition on the probability constraint graph of the surface state of the transmission line segment and the probability stack base number, perform binary logarithm conversion to obtain the stable constraint graph of the surface state of the transmission line segment:

[0097]

[0098] Among them, Represents point addition, Represents the probability stack base number, Represents the stable constraint graph of the surface state of the transmission line segment;

[0099] Perform differential operation on the probability constraint graph of the surface state of the transmission line segment with the probability stack base number and then perform inverse element calculation to obtain the suppression graph of the measurement sequence of the surface state of the transmission line segment:

[0100]

[0101] Among them, Represents differential operation, Represents the suppression graph of the measurement sequence of the surface state of the transmission line segment;

[0102] Fuse the features of the stable constraint graph of the surface state of the transmission line segment and the suppression graph of the measurement sequence of the surface state of the transmission line segment to obtain an optimized combined deep and shallow perception coding feature map of the surface state of the transmission line segment;

[0103] Input the optimized combined deep and shallow perception coding feature map of the surface state of the transmission line segment into the decision module based on the classifier to obtain the decision result.

[0104] Correspondingly, in this preferred embodiment, through the statistical distribution characteristics of the combined deep and shallow perception coding feature map of the surface state of the transmission line segment, a dynamic coupling relationship is established between the physical magnitude dimension and the abstract feature dimension. And, use the cross-level short-time sequence data based on statistics to generate a two-channel latent factor topology to achieve two-way knowledge transfer at the intermediate scale. At the same time, integrate the time series correlation constraints to complete the inversion compensation of the latent variable, enhance the optimization stability in the distribution function space, so as to improve the accuracy of the decision result output by the decision module based on the classifier for the combined deep and shallow perception coding feature map of the surface state of the transmission line segment.

[0105] In the above-mentioned unmanned live working control system 100 based on the autonomous decision-making algorithm, the drone hovering and image acquisition module 160 is configured to, in response to the decision result indicating that there is damage in the transmission line section, control the drone to hover, acquire high-definition images of the transmission line section, and transmit them back to the ground server. Once near the target position, the drone hovering and image acquisition module 160 controls the drone to gradually decelerate until it completely stops, and adjusts its attitude to maintain a stable hovering state. This step is crucial for obtaining high-quality images because any unnecessary movement may result in blurred images. To maintain hovering stability, the drone hovering and image acquisition module 160 adopts an advanced flight controller algorithm, combines the information fed back by the IMU, dynamically adjusts the motor output power, and cancels out external interference factors (such as wind force, airflow changes, etc.) to ensure that the drone can remain stationary in the air.

[0106] After completing the image acquisition, the drone hovering and image acquisition module 160 compresses and encodes the obtained high-definition images and transmits them back to the ground server through a wireless communication link. To ensure the reliability and speed of data transmission, the drone hovering and image acquisition module 160 usually adopts high-rate, low-latency communication protocols, such as 5G network or Wi-Fi 6 technology. At the same time, considering the possible signal occlusion problem, the drone hovering and image acquisition module 160 also pre-plans multiple backup transmission paths to ensure that even if one path fails, other available paths can still be found to complete the data upload. In addition, to improve the transmission efficiency, the drone hovering and image acquisition module 160 can also apply an image compression algorithm to reduce the file size without affecting the image quality and accelerate the transmission rate.

[0107] In one embodiment of the present application, Figure 5 is a flowchart of the unmanned live working control method based on the autonomous decision-making algorithm according to the embodiment of the present application, as Figure 5As shown, the unmanned live working control method based on the autonomous decision-making algorithm according to an embodiment of the present application includes: S210, controlling the unmanned aerial vehicle (UAV) to fly along the power transmission line to be detected; S220, collecting the surface state images of a predetermined line segment of the power transmission line to be detected through the camera of the UAV to obtain a time queue of the surface state images of the power transmission line segment; S230, calculating the image-level difference coefficient between every two adjacent surface state images of the power transmission line segment in the time queue of the surface state images of the power transmission line segment to obtain a time queue of the image-level difference coefficients; S240, determining the suspected abnormal points based on the comparison between the time queue of the image-level difference coefficients and a preset threshold; S250, inputting the surface state image of the power transmission line segment corresponding to the suspected abnormal point into an autonomous decision-making model based on deep learning to obtain a decision result, where the decision result is whether there is damage to the power transmission line segment; S260, in response to the decision result indicating that there is damage to the power transmission line segment, controlling the UAV to hover and collect the high-definition image of the power transmission line segment and transmit it back to the ground server.

[0108] Those skilled in the art can understand that the specific operations of each step in the above unmanned live working control method based on the autonomous decision-making algorithm have been described in detail above with reference to Figures 1 to 4 the description of the unmanned live working control system 100 based on the autonomous decision-making algorithm, and therefore, the repeated description thereof will be omitted.

[0109] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosure concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the present application.

[0110] In addition, although the operations are depicted in a specific order, this should not be construed as requiring the operations to be performed in the specific order shown or in sequential order. In certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the present application. Certain features described in the context of a single embodiment can also be implemented in combination in a single embodiment. Conversely, the various features described in the context of a single embodiment can also be implemented separately or in any suitable sub-combination in multiple embodiments.

[0111] Although the subject matter has been described in language specific to structural features and / or methodological logical acts, it is to be understood that the subject matter defined is not necessarily limited to the specific features or acts described above. On the contrary, the specific features and acts described above are merely example forms of implementation. Regarding the apparatus in the above embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be elaborated here.

Claims

1. An unmanned live working control system based on an autonomous decision-making algorithm, characterized in that, Including: A drone flight control module for controlling the drone to fly along the power transmission line to be detected; A power transmission line segment image acquisition module for acquiring surface state images of a predetermined line segment of the power transmission line to be detected through the camera of the drone to obtain a time queue of power transmission line segment surface state images; An image-level difference coefficient calculation module for calculating the image-level difference coefficient between every two adjacent power transmission line segment surface state images in the time queue of the power transmission line segment surface state images to obtain a time queue of image-level difference coefficients; A suspected abnormal point determination module for determining suspected abnormal points based on the comparison between the time queue of the image-level difference coefficients and a preset threshold; An autonomous decision-making module for inputting the power transmission line segment surface state image corresponding to the suspected abnormal point into an autonomous decision-making model based on deep learning to obtain a decision result, where the decision result is whether the power transmission line segment is damaged; A drone hovering acquisition module for controlling the drone to hover and acquire high-definition images of the power transmission line segment and transmit them back to the ground server in response to the decision result that the power transmission line segment is damaged; Wherein, the image-level difference coefficient calculation module includes: feature extraction and matching and finding corresponding relationships, and the finding corresponding relationships includes: pairing the feature points of every two adjacent power transmission line segment surface state images; Wherein, the autonomous decision-making module includes: An image shallow feature extraction unit for extracting shallow features of the power transmission line segment image from the power transmission line segment surface state image to obtain a shallow feature map of the power transmission line segment surface state; An image deep semantic feature extraction unit for extracting deep semantic features of the power transmission line segment image from the power transmission line segment surface state image to obtain a deep semantic feature map of the power transmission line segment surface state; A feature joint perception processing unit for performing feature joint perception processing guided by a semantic information field on the shallow feature map of the power transmission line segment surface state and the deep semantic feature map of the power transmission line segment surface state to obtain a deep and shallow joint perception coding feature of the power transmission line segment surface state; A decision result determination unit for determining the decision result based on the deep and shallow joint perception coding feature of the power transmission line segment surface state; Wherein, the feature joint perception processing unit includes: An upsampling subunit for upsampling the deep semantic feature map of the power transmission line segment surface state to obtain an upsampled deep semantic feature map of the power transmission line segment surface state, where the upsampled deep semantic feature map of the power transmission line segment surface state has the same size as the shallow feature map of the power transmission line segment surface state; A semantic information field calculation subunit for calculating the semantic information field between the upsampled deep semantic feature map of the power transmission line segment surface state and the shallow feature map of the power transmission line segment surface state to obtain a semantic information field of the power transmission line segment surface state, and mapping the upsampled deep semantic feature map of the power transmission line segment surface state to the semantic information field of the power transmission line segment surface state to obtain a field-modulated deep semantic feature map of the power transmission line segment surface state; The feature joint perception subunit is configured to input the surface state shallow feature map of the transmission line segment and the deep semantic feature map of the field-modulated transmission line segment surface state into a feature joint perception module based on a multi-attention structure to obtain a surface state shallow-deep joint perception encoded feature map of the transmission line segment as the surface state shallow-deep joint perception encoded feature of the transmission line segment.

2. The unmanned live working control system based on the autonomous decision-making algorithm according to claim 1, wherein, The image shallow feature extraction unit is configured to: extract the shallow features of the transmission line segment image through a multi-scale feature extractor for the surface state of the transmission line segment based on an atrous pyramid network from the surface state image of the transmission line segment to obtain the surface state shallow feature map of the transmission line segment.

3. The control system for unmanned live working based on the autonomous decision-making algorithm according to claim 2, wherein The image deep semantic feature extraction unit is configured to: extract the deep semantic features of the transmission line segment image through the multi-scale feature extractor for the surface state of the transmission line segment based on an atrous pyramid network from the surface state image of the transmission line segment to obtain the surface state deep semantic feature map of the transmission line segment.

4. The unmanned live working control system based on the autonomous decision-making algorithm according to claim 3, characterized in that, The semantic information field calculation subunit includes: The semantic information field prediction secondary subunit is configured to input the upsampled deep semantic feature map of the transmission line segment surface state and the shallow feature map of the transmission line segment surface state after feature connection into a semantic information field predictor based on gated convolution to obtain the semantic information field of the transmission line segment surface state; The feature map mapping secondary subunit is configured to map the upsampled deep semantic feature map of the transmission line segment surface state to the semantic information field of the transmission line segment surface state to obtain the deep semantic feature map of the field-modulated transmission line segment surface state.

5. The control system for unmanned live working based on the autonomous decision-making algorithm according to claim 4, characterized in that The semantic information field prediction secondary subunit is configured to: After performing feature connection on the deep semantic feature map of the surface state of the upsampled transmission line segment and the shallow feature map of the surface state of the transmission line segment, it is processed through a convolutional layer with a convolution kernel of to obtain a multi-scale semantic fusion feature map of the surface state of the transmission line segment; Perform point convolution processing on the multi-scale semantic fusion feature map of the transmission line segment surface state to obtain a multi-dimensional semantic modulation feature map of the transmission line segment surface state as the semantic information field of the transmission line segment surface state.

6. The unmanned live working control system based on the autonomous decision-making algorithm according to claim 5, characterized in that The feature joint perception subunit is configured to: Perform pointwise convolution processing on the deep semantic feature map of the field-modulated transmission line segment surface state, then input it into a batch normalization layer, and then perform activation processing using the Sigmoid function to obtain a semantic attention optimized expression feature map of the transmission line segment surface state; Perform multi-attention optimized expression on the shallow feature map of the transmission line segment surface state to obtain a semantic attention optimized expression feature map of the transmission line segment surface state; Perform position-wise point addition on the semantic attention optimized expression feature map of the transmission line segment surface state and the semantic attention optimized expression feature map of the transmission line segment surface state to obtain the surface state shallow-deep joint perception encoded feature map of the transmission line segment.

7. The unmanned live working control system based on the autonomous decision-making algorithm according to claim 6, characterized in that, The decision result determination unit is configured to: input the surface state shallow-deep joint perception encoded feature map of the transmission line segment into a decision module based on a classifier to obtain the decision result.

Images