Wind turbine blade orientation angle recognition method and system based on rotating box detection

Abstract

The application discloses a wind turbine blade orientation angle identification method and system based on a rotating frame detection, and the method comprises the following steps: collecting a wind turbine blade image; detecting a cabin and a blade in the image by using a rotating frame detection algorithm to obtain a rotating frame of the cabin and the blade and an angle of the rotating frame; the rotating frame detection algorithm adopts an LSKnet model fused with a PSC technology; determining a cabin center and a blade center according to a positional relationship of the rotating frame of the cabin and the blade; taking the cabin center as an origin and the blade center as a terminal point to obtain an orientation angle and determine a quadrant in which the orientation angle is located; and calculating the orientation angle of the blade according to the quadrant in which the orientation angle is located and the angle of the rotating frame. The application can realize accurate and efficient identification of the orientation angle of the wind turbine blade.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing and deep learning, in particular to a wind turbine blade orientation angle recognition method and system based on rotated box detection. BACKGROUND

[0002] With the continuous development of renewable energy, wind turbine blades as a clean and efficient energy form have received widespread attention. In the operation and maintenance process of wind turbine blades, it is of great significance to accurately identify and monitor the orientation angle of wind turbine blades. Traditional identification methods are usually based on manual feature extraction, but this method has limitations in dealing with complex backgrounds and rotating targets.

[0003] In recent years, the rapid development of deep learning technology has provided a new solution for rotating target detection. Compared with traditional methods, deep learning migrates the detection method of general targets to rotating target detection, solves the limitations of horizontal frame labeling targets, introduces angle information into the neural network, increases the branch angle prediction, rotates the original horizontal frame according to the predicted angle to fit the actual target, and improves the detection ability and speed of rotating targets. However, when deep learning is applied to wind turbine blade orientation angle recognition, there are still some challenges, such as the discontinuity of the loss function caused by the periodicity of the angle, the 180-degree interval of the rectangular symmetry of the detection frame, the inability to identify the 360-degree orientation of the wind turbine blade target, and the influence of image quality on the detection result.

[0004] Therefore, how to combine deep learning and rotating target detection technology to achieve accurate and efficient identification of wind turbine blade orientation angle has become a problem to be solved. SUMMARY

[0005] In order to achieve accurate and efficient identification of wind turbine blade orientation angle, the present application provides a wind turbine blade orientation angle recognition method and system based on rotated box detection.

[0006] In a first aspect, the present application provides a wind turbine blade orientation angle recognition method based on rotated box detection, comprising:

[0007] Collecting wind turbine blade images;

[0008] Detecting the nacelle and the blade in the image using a rotated box detection algorithm to obtain the rotated box of the nacelle and the blade and the angle of the rotated box; the rotated box detection algorithm uses a LSKnet model fused with PSC technology;

[0009] According to the positional relationship between the nacelle and the blade rotation frame, the center of the nacelle and the center of the blade are determined; the heading angle is obtained with the center of the nacelle as the origin and the center of the blade as the end point, and the quadrant in which the heading angle is located is judged; according to the quadrant in which the judged heading angle is located, the heading angle of the blade is calculated and obtained in combination with the rotation frame angle.

[0010] By adopting the above scheme, the deep learning model is adopted, preferably the LSKnet model, to realize the rapid and accurate identification of the heading angle of the wind power blade; at the same time, the phase-shifting coder (PSC) is adopted to encode and decode the feature information in the image data, to solve the problem of discontinuous loss function caused by the periodicity of the angle, and to improve the identification effect of the heading angle of the wind power blade; the 360-degree heading identification of the wind power blade target is realized by relying on the center distance relationship judgment of the nacelle and the blade rotation frame, and the accuracy of the angle identification is further improved.

[0011] Preferably, it further comprises:

[0012] The heading angle of the wind power blade is cyclically changed from 0 degrees to 360 degrees at a certain rate under the control of the wind power blade control system, and the time required for the heading angle of the wind power blade to change from 0 degrees to 360 degrees at a certain rate is counted as the standard time;

[0013] During the process of the heading angle of the wind power blade being cyclically changed from 0 degrees to 360 degrees at a certain rate under the control of the wind power blade control system, the image of the wind power blade at any time is collected, and the heading angle of the blade is detected and calculated by using the rotation frame detection algorithm, which is recorded as the first blade heading angle; the image of the wind power blade after a certain time interval from the certain time is collected, and the heading angle of the blade is detected and calculated by using the rotation frame detection algorithm, which is recorded as the second blade heading angle;

[0014] The error between the first blade heading angle and the second blade heading angle is compared to determine whether it is greater than a preset error, and if it is greater, it is determined that the result of the heading angle of the blade is deviated, and the LSKnet model fused with the PSC technology is optimized and trained until the error between the first blade heading angle and the second blade heading angle detected and calculated by the LSKnet model fused with the PSC technology after the optimization and training is not greater than the preset error, and the optimization and training is stopped.

[0015] By adopting the above scheme, the time required for the heading angle of the blade to change from 0 degrees to 360 degrees is used to verify whether the heading angle of the blade detected in the same time period is correct, and then the LSKnet model fused with the PSC technology is optimized to obtain more accurate angle identification results.

[0016] Preferably, it further comprises:

[0017] The historical wind power blade images under the combination scenes of different wind speeds, different wind directions and different light conditions are adopted, and the LSKnet model of the fusion PSC technology is trained by using the historical wind power blade images collected under different combination scenes.

[0018] By adopting the above scheme, considering the influence of different environmental conditions on wind power blade image collection, training data under different natural conditions is collected, and the performance of the rotated box detection algorithm in various actual scenes is effectively improved.

[0019] Preferably, it further comprises:

[0020] The orientation angle of the wind power blade is automatically adjusted by the wind power blade control system according to real-time wind direction data, wind speed data and light data, the orientation angle change of the wind power blade in the time sequence is calculated, and the first wind power blade orientation angle change sequence is recorded.

[0021] The wind power blade image is collected in real time, the rotated box detection algorithm is used to detect and calculate the real-time orientation angle of the wind power blade, the orientation angle change of the wind power blade in the time sequence is obtained, and the second orientation angle change sequence is recorded.

[0022] The first orientation angle change sequence and the second orientation angle change sequence are compared, if the difference between the first orientation angle change value and the second orientation angle change value in the continuous preset time is greater than the preset orientation angle change value, the wind power blade images collected under the same conditions of the wind direction data, the wind speed data and the light data in the continuous preset time are added to optimize the training of the LSKnet model of the fusion PSC technology.

[0023] By adopting the above scheme, the control of the control system on the orientation angle of the wind power blade in real-time environment is obtained, the dynamic change of the orientation angle of the wind power blade in the time sequence is determined, and the orientation angle of the wind power blade is compared with the actual detection and acquisition, the continuous tracking and identification of the orientation angle of the wind power blade are completed, and the detection accuracy of the LSKnet model of the fusion PSC technology in part of the environment is determined to be low, and the training of the related environment is increased.

[0024] Preferably, it further comprises:

[0025] The environmental data is collected, and the collected environmental data is compared with the corresponding set environmental data preset threshold; the environmental data includes: light intensity, wind speed and wind power;

[0026] If the type number of the collected environmental data greater than the environmental data preset threshold of the corresponding type is greater than the preset type number, it is judged that the current environmental data is in a complex environment, a two-stage rotating target detection method is used to detect the nacelle and the blade in the image based on the rotating box detection algorithm, otherwise, it is judged that the current environment is not a complex environment, a single-stage rotating target detection method is used to detect the nacelle and the blade in the image based on the rotating box detection algorithm.

[0027] By adopting the above scheme, considering that the difficulty of detecting the orientation angle image of the wind power blade increases under complex environmental conditions, a two-stage rotating target detection method with higher accuracy is used for detection to realize more efficient and accurate recognition.

[0028] Preferably, it further comprises:

[0029] The image processing technology is used to detect whether the blade in the wind power blade image is complete, if not, the wind power blade image collected at different angles is replaced until the blade in the wind power blade image is complete.

[0030] By adopting the above scheme, the wind power blade images at multiple angles are collected in real time to realize the recognition of the orientation angle of the wind power blade with complete blades.

[0031] Preferably, it further comprises:

[0032] The collected image is preprocessed, and the preprocessing includes adjusting the brightness and contrast of the image.

[0033] By adopting the above scheme, the image acquisition is optimized to improve the image quality, thereby improving the recognition effect of the orientation angle of the wind power blade.

[0034] In a second aspect, the application provides a wind power blade orientation angle recognition system based on rotating box detection, comprising:

[0035] An image acquisition module is configured to acquire a wind power blade image.

[0036] An image detection module is configured to detect a nacelle and a blade in the image by using a rotating box detection algorithm to obtain a rotating box and an angle of the rotating box of the nacelle and the blade; the rotating box detection algorithm uses an LSKnet model fused with PSC technology.

[0037] An orientation angle acquisition module is configured to determine a nacelle center and a blade center according to a positional relationship of the nacelle and the blade rotating boxes, take the nacelle center as an origin and the blade center as a terminal to obtain an orientation angle and determine a quadrant where the orientation angle is located, and calculate the orientation angle of the blade according to the determined quadrant where the orientation angle is located and the angle of the rotating box.

[0038] By adopting the above scheme, the method based on rotation box detection is adopted in combination with deep learning technology to automatically and accurately identify the orientation angle of the wind turbine blade, and the identification precision and efficiency are improved.

[0039] In a third aspect, the present application provides a computer readable storage medium, comprising a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the method as described above when the computer program is running.

[0040] In a fourth aspect, the present application provides a computer device, comprising a memory, a processor and a program stored on the memory and executable by the processor, wherein the program is executed by the processor to implement the steps of the method as described above.

[0041] In summary, the present application has the following beneficial effects:

[0042] 1. The method based on rotation box detection is adopted in combination with deep learning technology, preferably LSKnet model, to realize fast and accurate identification of the orientation angle of the wind turbine blade. Meanwhile, the phase-shifting coder (PSC) is adopted to solve the problem of discontinuous loss function caused by angle periodicity, and the accuracy of the identification of the orientation angle of the wind turbine blade is further improved.

[0043] 2. The wind turbine blade images under different environmental conditions are collected for training to improve the performance of the model detection, and the two-stage or single-stage rotation target detection method is selected for image detection according to the actual environmental conditions, the detection speed and accuracy are balanced, the image quality is optimized, and thus the identification effect of the orientation angle of the wind turbine blade is improved.

[0044] 3. The orientation angle of the wind turbine blade is verified and tracked in real time by combining the real-time adjusted orientation angle of the wind turbine blade or using the rotation angle sensor to collect the orientation angle of the wind turbine blade. The orientation angle of the wind turbine blade obtained by the model detection and calculation is verified and tracked in real time. According to the verification result and the real-time tracking result, the model is further optimized and trained to further improve the detection accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The flowchart of the method for identifying the orientation angle of the wind turbine blade based on the rotation box detection in the embodiment is shown.

[0046] Figure 2 The result diagram of the method for identifying the orientation angle of the wind turbine blade based on the rotation box detection in the embodiment is shown. Figure 2 (a), Figure 2 (b) and Figure 2 (c) are the rotation box detection diagrams of different wind turbine cabin blades.Figure 2 (d) the same wind turbine blade as the model detected by the optimized model Figure 2 (a) a rotating frame detection diagram of the same wind turbine blade;

[0047] Figure 3 A structural schematic diagram of a wind turbine blade orientation angle recognition system based on rotating frame detection according to an embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0049] As shown in Figure 1 The present application discloses a wind turbine blade orientation angle recognition method based on rotating frame detection, and the specific steps are as follows:

[0050] S1, collect wind turbine blade images and perform preprocessing.

[0051] Specifically, a plurality of wind turbine blade images are collected by using a UAV.

[0052] In order to ensure that the collected wind turbine blade images have complete blades as much as possible and ensure the accurate identification of the blade orientation angle, a plurality of UAVs are selected to collect wind turbine blade images at multiple angles. Image processing technology is used to detect whether the blade in the wind turbine blade image is complete, if not, replace the wind turbine blade image collected at a different angle until the blade in the wind turbine blade image is complete and stop replacing.

[0053] In this embodiment, a traditional image processing algorithm (such as Houqh transformation) is selected to detect the blade edge, and the edge contour is used to determine whether the blade in the image is complete.

[0054] The preprocessing of the wind turbine blade image includes image cleaning, image enhancement, etc.; the image enhancement includes adjusting the brightness, contrast, etc. of the image to improve the quality and clarity of the image.

[0055] S2, use a rotating frame detection algorithm to detect the nacelle and blade in the preprocessed image.

[0056] Specifically, on the one hand, considering the diversity of the background of the wind turbine blade, in order to reduce the additional environmental information as much as possible, the LSKnet model is selected for the rotation box detection; on the other hand, considering the problem of loss function discontinuity caused by the angle periodicity in the rotation target detection, which leads to inaccurate detection results, in order to improve the accuracy of the detection results in this case, the PSC technology is adopted, which is fused into the LSKnet model, the model features are phase-shift coded, and the model output is decoded to obtain the accurate rotation box angle. Therefore, the rotation box detection algorithm selects the LSKnet model fused with the PSC technology.

[0057] The input of the LSKnet model is the preprocessed image of the wind turbine blade, and the output is the rotation box of the nacelle and the blade in the wind turbine blade image and the angle of the rotation box. The LSKnet model is trained and generated by historical wind turbine blade images labeled with rotation box parameters (including the length, width, center coordinates and rotation angle of the rotation box) in the nacelle and the blade, wherein the rotation angle is phase-shift coded using the PSC technology and represented in the form of phase-shift cosine value; the angle of the rotation box output by the LSKnet model is also represented in the form of phase-shift cosine value and encoded using the PSC technology, and finally the angle of the rotation box is obtained.

[0058] The preprocessed image is input into the LSKnet model fused with the PSC technology to obtain the detection result.

[0059] S3, determine whether the nacelle and the blade exist in the preprocessed image.

[0060] Specifically, the detection result of the preprocessed image input into the LSKnet model fused with the PSC technology is obtained, and whether the nacelle and the blade exist in the preprocessed image is determined according to the detection result. If not, continue to input other preprocessed images into the model for detection; if so, the rotation box of the nacelle and the blade and the angle of the rotation box are obtained.

[0061] S4, determine the quadrant where the orientation angle of the vector formed by the position relationship of the nacelle and the blade rotation box is located.

[0062] Considering the rotation box detection, the long side 135-degree definition is relied on, because the rectangular symmetry angle range interval is 180 degrees, a single rotation box detection cannot identify the orientation of the wind turbine blade target in the 360-degree range. Therefore, according to the positional relationship between the nacelle and the blade rotation box, the nacelle center and the blade center are determined, and the nacelle center is taken as the origin and the blade center is taken as the terminal point. The current orientation angle is determined based on the vector formed by the two points, that is, the angle between the two-point line and the horizontal plane, and the quadrant in which the orientation angle is located is determined. Among them, the quadrant can be determined according to the coordinate system in which the positive Y-axis is marked as 0° and the positive X-axis is marked as 0°, and the quadrant in which the positive X-axis is marked as 0° and the positive X-axis is marked as 0°. The coordinate system in which the positive X-axis is marked as 0° and the positive X-axis is marked as 0° is set clockwise 360°.

[0063] S5, according to the positional relationship between the nacelle and the blade rotation box, the orientation angle of the vector formed is determined, and the orientation angle of the blade is calculated by combining the rotation box angle.

[0064] Specifically, the specific orientation angle of the blade corresponding to the 0°-360° range is determined according to the quadrant in which the orientation angle is located, and the specific orientation angle of the blade is determined according to the specific orientation angle of the blade. Figure 2 (c) shows that the quadrant in which the vector is located is the first quadrant in the coordinate system in which the positive X-axis is marked as 0° and the positive X-axis is marked as 0°, and the specific orientation angle of the blade is determined according to the angle of the rotation box. Figure 2 (c) shows that the angle of the rotation box obtained based on the model is 42°, and the final orientation angle of the blade is 42°.

[0065] As shown in Figure 2 , the rotation box detection diagram of different wind turbine nacelles and blades obtained by the above method is detected. As shown in Figure 2 (a), the confidence of the wind turbine nacelle is 99.99%, and the confidence of the wind turbine blade is 99.92%. As shown in Figure 2 (b), the confidence of the wind turbine nacelle is 99.98%, and the confidence of the wind turbine blade is 99.88%. As shown in Figure 2 (c), the confidence of the wind turbine nacelle is 99.99%, and the confidence of the wind turbine blade is 99.92%.

[0066] In a specific embodiment, in order to further ensure the accuracy of the specific orientation angle obtained by using the model to detect and finally calculate, the rotation box detection model is further optimized by verifying the accuracy of the model detection result, and the method further comprises:

[0067] Using the display device of the wind turbine blade control system, the orientation angle of the wind turbine blade changes cyclically from 0 degrees to 360 degrees at a certain rate under the control of the wind turbine blade control system in real time. The time required for the blade orientation angle to change from 0 degrees to 360 degrees at a certain rate is recorded as the standard time; the certain time is a manually preset time.

[0068] Under the control of the wind turbine blade control system, the orientation angle of the wind turbine blade changes cyclically from 0 degrees to 360 degrees at a certain rate. At a certain moment, the wind turbine blade image is randomly acquired, and the orientation angle of the blade is detected and calculated using a rotating frame detection algorithm. This is recorded as the orientation angle of the first blade. Then, the wind turbine blade image is acquired after a standard time interval from the previous moment, and the orientation angle of the blade is detected and calculated using the rotating frame detection algorithm. This is recorded as the orientation angle of the second blade.

[0069] Considering that the orientation angle of the wind turbine blades changes cyclically, the orientation angles of the blades before and after the standard time should be consistent or have only a very small error. Therefore, it is determined whether the error between the orientation angle of the first blade and the orientation angle of the second blade is greater than the preset error. If it is greater, it is determined that there is a deviation in the obtained blade orientation angle result. The LSKnet model with PSC technology is optimized and trained, and the error between the orientation angle of the first blade and the orientation angle of the second blade is obtained again using the optimized LSKnet model with PSC technology. The calculated error is compared with the preset error until the error obtained by the optimized LSKnet model with PSC technology is not greater than the preset error, and the optimization training is stopped.

[0070] Training optimization includes adjusting model parameters or optimizing the model structure. For LSKnet models, improvements can be made by introducing more advanced network structures (such as residual networks or attention mechanisms) to enhance the efficiency and accuracy of feature extraction. For example, in the LSKnet model, a residual network structure can be introduced by replacing the original convolutional layers with residual blocks, or an attention mechanism can be introduced after the convolutional layers in the feature extraction stage.

[0071] like Figure 2 As shown in (d), compared to Figure 3 (a) As far as the model is concerned, the confidence level of identifying the wind turbine nacelle is 99.99% and the confidence level of identifying the wind turbine blade is 99.93%. The accuracy of the detection results can be improved by using the optimized model.

[0072] In one specific embodiment, considering that model detection is affected by environmental factors, in order to avoid the influence of environmental factors, training data collected in different environmental scenarios can be selected for training, so as to improve the detection accuracy of the model in the corresponding environmental scenario. The method further comprises:

[0073] The historical wind turbine blade images in different wind speed, different wind direction and different light condition combination scenarios are adopted, and the LSKnet model of the fusion PSC technology is trained by using the historical wind turbine blade images collected in different combination scenarios, so as to improve the detection performance of the LSKnet model of the fusion PSC technology.

[0074] On the basis of the above training, further considering that different environmental factors have different influences on the detection accuracy of the model, for example, complex environment has a greater influence on the detection accuracy of the model, the model detection accuracy in different environments can be combined to determine which environmental training data is specifically added to optimize the performance of the model. The method further comprises:

[0075] Real-time acquisition of the orientation angle of the wind turbine blade control system automatically adjusting the orientation angle of the blade according to real-time meteorological data, calculation of the orientation angle change of the blade in the time sequence, recorded as the first blade orientation angle change sequence; the meteorological data includes wind direction data, wind speed data, wind power data, light data, wind speed and wind power duration and rainfall data, etc.

[0076] Real-time acquisition of the wind turbine blade image, detection and calculation of the real-time orientation angle of the blade by using the rotated frame detection algorithm, acquisition of the orientation angle change of the blade in the time sequence, recorded as the second orientation angle change sequence;

[0077] Comparison of the first orientation angle change sequence and the second orientation angle change sequence, if the difference between the first orientation angle change value and the second orientation angle change value in the continuous preset time is greater than the preset orientation angle change value, the wind turbine blade image collected under the same condition of the meteorological data corresponding to the continuous preset time is added to optimize the training of the LSKnet model of the fusion PSC technology.

[0078] In one specific embodiment, considering that the more complex the environmental conditions are, the lower the clarity of the collected image is, and thus the detection difficulty of the rotated frame for the collected image is increased, a more accurate detection method needs to be provided. The method further comprises:

[0079] Collection of environmental data, comparison of the collected environmental data with the corresponding set environmental data preset threshold; the environmental data includes: light intensity, wind speed and wind power, and wind speed and wind power duration meteorological data; the corresponding set environmental data preset threshold includes: light intensity preset threshold, wind speed preset threshold, wind power preset threshold, certain wind speed duration preset threshold and certain wind power duration preset threshold.

[0080] If the number of types of collected environmental data greater than the corresponding type of environmental data preset threshold is greater than the preset number of types, it is determined that the current environmental data is in a complex environment, and a two-stage rotating target detection method is used to detect the nacelle and the blade in the image based on the rotating box detection algorithm. Among them, the first stage uses the feature extraction and target recognition ability of the LSKnet model to generate a series of candidate regions in the image; the second stage identifies the wind turbine nacelle and the blade in these candidate regions and outputs the rotating box position information and the rotating angle. The preset number is considered to be set, such as more than 3 types of environmental data (light, wind, wind speed) greater than the corresponding set environmental data preset threshold.

[0081] Otherwise, based on the rotating box detection algorithm, a single-stage rotating target detection method is used to detect the nacelle and the blade in the image, that is, directly identifying the wind turbine nacelle and the blade and outputting the rotating box position information and the rotating angle.

[0082] As shown in ​ The embodiment of the application discloses a wind turbine blade orientation angle identification system based on rotating box detection, which specifically comprises:

[0083] An image acquisition module 101 is configured to acquire a wind turbine blade image.

[0084] An image detection module 102 is configured to detect the nacelle and the blade in the image by using a rotating box detection algorithm to obtain the rotating box of the nacelle and the blade and the angle of the rotating box. The rotating box detection algorithm uses an LSKnet model fused with PSC technology.

[0085] An orientation angle acquisition module 103 is configured to determine the center of the nacelle and the center of the blade according to the positional relationship of the rotating box of the nacelle and the blade, take the center of the nacelle as the origin and the center of the blade as the terminal point, acquire the orientation angle and determine the quadrant in which the orientation angle is located, and calculate the orientation angle of the blade according to the quadrant in which the orientation angle is located and the angle of the rotating box.

[0086] The system further comprises:

[0087] The image detection optimization module 104 is used to view the situation that the orientation angle of the wind turbine blade changes at a certain rate from 0 degree to 360 degrees in a cycle under the control of the wind turbine blade control system in real time, and to count the time required for the orientation angle of the wind turbine blade to change from 0 degree to 360 degrees at a certain rate, which is recorded as a standard time. During the process that the orientation angle of the wind turbine blade changes from 0 degree to 360 degrees in a cycle under the control of the wind turbine blade control system, an image of the wind turbine blade at a certain moment is collected, and the orientation angle of the wind turbine blade is detected and calculated by using a rotating frame detection algorithm, which is recorded as a first orientation angle of the wind turbine blade. An image of the wind turbine blade after a standard time interval from the certain moment is collected, and the orientation angle of the wind turbine blade is detected and calculated by using a rotating frame detection algorithm, which is recorded as a second orientation angle of the wind turbine blade. It is determined whether the error between the first orientation angle of the wind turbine blade and the second orientation angle of the wind turbine blade is greater than a preset error. If it is greater than the preset error, it is determined that the result of the orientation angle of the wind turbine blade has deviation, and the LSKnet model fused with the PSC technology is optimized and trained until the error between the first orientation angle of the wind turbine blade and the second orientation angle of the wind turbine blade detected and calculated by using the LSKnet model fused with the PSC technology after the optimization and training is not greater than the preset error, and the optimization and training is stopped.

[0088] The image detection optimization module 104 is also used to use historical wind turbine blade images under different wind speed, different wind direction and different light condition combination scenes, and to train the LSKnet model fused with the PSC technology by using the collected historical wind turbine blade images under different combination scenes. It is also used to automatically adjust the orientation angle of the wind turbine blade according to real-time wind direction data, wind speed data and light data by the wind turbine blade control system, to calculate the change of the orientation angle of the wind turbine blade in time sequence, which is recorded as a first orientation angle change sequence. Real-time wind turbine blade images are collected, and the real-time orientation angle of the wind turbine blade is detected and calculated by using a rotating frame detection algorithm, so as to obtain the change of the orientation angle of the wind turbine blade in time sequence, which is recorded as a second orientation angle change sequence. The first orientation angle change sequence and the second orientation angle change sequence are compared. If the difference between the first orientation angle change value and the second orientation angle change value within a continuous preset time is greater than a preset orientation angle change value, the wind turbine blade images collected under the same conditions of the wind direction data, the wind speed data and the light data within the continuous preset time are added to the LSKnet model fused with the PSC technology for optimization and training.

[0089] The image detection module 102 is further configured to collect environment data, compare the collected environment data with a corresponding set of environment data preset threshold values, and determine whether the current environment is complex based on the comparison result. The environment data includes illumination intensity, wind speed and wind power, and wind speed and wind power duration. If the number of types of the collected environment data that are greater than the corresponding type of environment data preset threshold values is greater than a preset type number, it is determined that the current environment data is in a complex environment. In this case, a two-stage rotating target detection method is used to detect the nacelle and the blade in the image based on the rotating box detection algorithm. Otherwise, it is determined that the current environment is not a complex environment. In this case, a single-stage rotating target detection method is used to detect the nacelle and the blade in the image based on the rotating box detection algorithm.

[0090] The image collection module 101 is further configured to detect whether the blade in the wind power blade image is complete by using an image processing technology. If the blade is not complete, wind power blade images collected at different angles are replaced until the blade in the wind power blade image is complete. The collected image is preprocessed, and the preprocessing includes adjusting the brightness and contrast of the image.

[0091] The embodiment of the present application further discloses a computer readable storage medium.

[0092] Specifically, the computer readable storage medium stores a computer program capable of being loaded and executed by the processor to implement the wind power blade orientation angle recognition method based on the rotating box detection. The computer readable storage medium includes, for example, a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various storage program code media.

[0093] The embodiment of the present application further discloses a computer device.

[0094] Specifically, the computer device includes a memory and a processor. The memory stores a computer program capable of being loaded and executed by the processor to implement the wind power blade orientation angle recognition method based on the rotating box detection.

[0095] The above are preferred embodiments of the present application, and do not limit the protection scope of the present application. Any feature disclosed in the specification (including the abstract and the drawings) can be replaced by other equivalent or similar features unless specifically described. That is, each feature is only an example of a series of equivalent or similar features unless specifically described.

Claims

1. A method for identifying the orientation angle of a wind turbine blade based on rotating box detection, characterized in that, The method comprises the following steps: Collecting a wind turbine blade image; Detecting the nacelle and the blade in the image by using a rotating box detection algorithm to obtain the rotating box and the angle of the rotating box of the nacelle and the blade; The rotating box detection algorithm adopts a LSKnet model fused with PSC technology; According to the positional relationship of the rotating box of the nacelle and the blade, the center of the nacelle and the center of the blade are determined; taking the center of the nacelle as the origin and the center of the blade as the terminal, the heading angle is obtained and the quadrant in which the heading angle is located is determined; According to the quadrant in which the heading angle is located, the heading angle of the blade is calculated in combination with the angle of the rotating box; further comprising: Real-time viewing of the cyclic variation of the heading angle of the blade from 0 degrees to 360 degrees at a certain rate under the control of the wind turbine blade control system, and counting the time required for the heading angle of the blade to change from 0 degrees to 360 degrees at a certain rate, which is recorded as the standard time; During the cyclic variation of the heading angle of the blade from 0 degrees to 360 degrees at a certain rate under the control of the wind turbine blade control system, collecting a wind turbine blade image at an arbitrary time, and detecting and calculating the heading angle of the blade by using the rotating box detection algorithm, which is recorded as the first blade heading angle; collecting a wind turbine blade image after a certain time interval from the standard time, and detecting and calculating the heading angle of the blade by using the rotating box detection algorithm, which is recorded as the second blade heading angle; Determining whether the error between the first blade heading angle and the second blade heading angle is greater than a preset error, and if it is greater, it is determined that the result of the heading angle of the blade is deviated, and the LSKnet model fused with PSC technology is optimized and trained until the error between the first blade heading angle and the second blade heading angle detected and calculated by using the LSKnet model fused with PSC technology after optimization and training is not greater than the preset error, and the optimization and training is stopped.

2. The wind turbine blade orientation angle identification method based on rotating box detection according to claim 1, characterized in that, Further comprising: Using historical wind turbine blade images under different wind speed, different wind direction and different light condition combination scenes, and training the LSKnet model fused with PSC technology by using the collected historical wind turbine blade images under different combination scenes.

3. The wind turbine blade orientation angle identification method based on rotating box detection according to claim 2, characterized in that, Further comprising: Real-time acquisition of the heading angle of the blade automatically adjusted by the wind turbine blade control system according to real-time wind direction data, wind speed data and light data, calculation of the heading angle change of the blade in time sequence, which is recorded as the first blade heading angle change sequence; Real-time collection of a wind turbine blade image, detection and calculation of the real-time heading angle of the blade by using the rotating box detection algorithm, and acquisition of the heading angle change of the blade in time sequence, which is recorded as the second heading angle change sequence; Comparing the first heading angle change sequence with the second heading angle change sequence, if the difference between the first heading angle change value and the second heading angle change value within a continuous preset time is greater than a preset heading angle change value, then the wind turbine blade image collected under the same condition of the wind direction data, the wind speed data and the light data within the continuous preset time is added to the LSKnet model fused with PSC technology for optimization and training.

4. The wind turbine blade orientation angle identification method based on rotating box detection according to claim 1, characterized in that, Further comprising: Collecting environmental data and comparing the collected environmental data with the corresponding set environmental data preset threshold; The environmental data includes: light intensity, wind speed and wind power; If the number of types of the collected environmental data greater than the preset threshold of the corresponding type of environmental data is greater than the preset number of types, it is determined that the current environmental data is in a complex environment. Based on the rotated bounding box detection algorithm, a two-stage rotated object detection method is used to detect the nacelle and the blade in the image. Otherwise, it is determined that the current environment is not a complex environment. Based on the rotated bounding box detection algorithm, a single-stage rotated object detection method is used to detect the nacelle and the blade in the image.

5. The wind turbine blade orientation angle identification method based on rotating box detection according to claim 1, characterized in that, Also includes: The image processing technology is used to detect whether the blade in the wind turbine blade image is complete. If not, the wind turbine blade image collected at different angles is replaced until the blade in the wind turbine blade image is complete.

6. The wind turbine blade orientation angle identification method based on rotating box detection according to claim 1, characterized in that, Also includes: The collected image is preprocessed, and the preprocessing includes adjusting the brightness and contrast of the image.

7. A wind turbine blade orientation angle identification system based on rotating box detection, characterized in that, Includes: An image acquisition module is configured to acquire a wind turbine blade image. An image detection module is configured to detect the nacelle and the blade in the image using a rotated bounding box detection algorithm to obtain the rotated bounding box and the angle of the rotated bounding box of the nacelle and the blade. The rotated bounding box detection algorithm uses a LSKnet model fused with PSC technology. An orientation angle acquisition module is configured to determine the center of the nacelle and the center of the blade according to the positional relationship of the nacelle and the blade rotated bounding box, take the center of the nacelle as the origin and the center of the blade as the terminal point, acquire the orientation angle and determine the quadrant in which the orientation angle is located, and calculate the orientation angle of the blade according to the determined quadrant in which the orientation angle is located and the angle of the rotated bounding box. An image detection optimization model is configured to view in real time the cyclic variation of the orientation angle of the wind turbine blade from 0 degrees to 360 degrees at a certain rate under the control of the wind turbine blade control system, and calculate the time required for the orientation angle of the wind turbine blade to change from 0 degrees to 360 degrees at a certain rate, which is recorded as the standard time. During the cyclic variation of the orientation angle of the wind turbine blade from 0 degrees to 360 degrees at a certain rate under the control of the wind turbine blade control system, an image of the wind turbine blade is collected at an arbitrary time, and the rotated bounding box detection algorithm is used to detect and calculate the orientation angle of the blade, which is recorded as the first blade orientation angle. An image of the wind turbine blade is collected at a time interval of the standard time, and the rotated bounding box detection algorithm is used to detect and calculate the orientation angle of the blade, which is recorded as the second blade orientation angle. It is determined whether the error between the first blade orientation angle and the second blade orientation angle is greater than a preset error. If it is greater, it is determined that the result of the orientation angle of the blade has deviation, and the LSKnet model fused with PSC technology is optimized and trained until the error between the first blade orientation angle and the second blade orientation angle calculated by the LSKnet model fused with PSC technology after optimization and training is not greater than the preset error, and the optimization and training is stopped.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein the computer program controls the device in which the computer-readable storage medium is located to execute the method of any one of claims 1 to 6 when the computer program is running.

9. A computer device, comprising: The computer device includes a memory, a processor, and a program stored on the memory and executable by the processor, and the program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images