A method and system for detecting surface defects of a pultruded plate for a wind turbine blade

Abstract

The application provides a kind of detection method and system of surface defect of pultruded plate for wind power blade, it is related to image processing technical field, method includes: obtaining the cross stroboscopic image of pultruded plate surface under reflection field and backlight field;Split image to obtain reflection field and backlight field image;Based on reflection field image, construct reflection field data set, based on backlight field image, construct backlight field data set;Construct visual inspection model under reflection field and backlight field;Use reflection field data set to train the model under reflection field, use backlight field data set to train the model under backlight field;Obtain the cross stroboscopic image of pultruded plate surface to be detected under reflection field and backlight field;Split image to obtain to-be-detected reflection field and backlight field image;To-be-detected reflection field image is input into the model under reflection field, and reflection field detection result is obtained, to-be-detected backlight field image is input into the model under backlight field, and backlight field detection result is obtained;According to reflection field and backlight field detection result, obtain final detection result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a kind of wind turbine blade with the detection method and system of pultruded plate surface defect. BACKGROUND

[0002] Pultruded plate is a kind of composite material plate manufactured by impregnating glass reinforced fiber with resin and then using pultrusion process, which has a specific cross-sectional shape and continuous length. Due to its excellent mechanical properties and stable quality, pultruded plate is widely used in the manufacture of wind turbine blades. However, during the production process, pultruded plate may have various defects such as bubbles, raw yarn oxidation, inclusions, cracking and dry yarn. These defects may change during the operation of wind power system, thereby seriously affecting the normal operation of wind power system, resulting in significant economic loss and safety risk. Therefore, it is crucial to effectively detect defects in pultruded plate.

[0003] In existing technical solutions, various techniques for pultruded plate defect detection have been proposed, including image-based detection methods and non-image-based detection methods. These techniques aim to reduce human intervention and improve the detection accuracy of small target defects.

[0004] However, existing image-based defect detection methods do not fully consider the semi-transparent nature of pultruded plate for wind turbine blades, which makes it difficult to effectively extract some internal defects under traditional reflected light field, thereby affecting the accuracy of detection. Meanwhile, non-image-based defect detection methods, although providing more comprehensive defect information, also face problems such as noise interference, high equipment and maintenance costs, high technical requirements for operators and high technical difficulty. SUMMARY

[0005] To solve the technical problems that traditional image-based defect detection methods do not fully consider the semi-transparent nature of pultruded plate for wind turbine blades, which makes it difficult to effectively extract some internal defects under traditional reflected light field, thereby affecting the accuracy of detection, and non-image-based defect detection methods, although providing more comprehensive defect information, also face problems such as noise interference, high equipment and maintenance costs, high technical requirements for operators and high technical difficulty, the present application provides a kind of wind turbine blade with the detection method and system of pultruded plate surface defect.

[0006] The technical solutions provided by the embodiments of the present application are as follows:

[0007] First aspect:

[0008] The detection method for surface defects of pultruded plate for wind turbine blades provided by the embodiments of the present application comprises:

[0009] S1: Obtain cross-stroboscopic images of pultruded plate surface under reflected field and backlight field;

[0010] S2: split the cross stroboscopic image to obtain a reflection field image and a backlight field image;

[0011] S3: construct a reflection field dataset based on the reflection field image, and construct a backlight field dataset based on the backlight field image;

[0012] S4: construct a deep learning-based visual detection model under the reflection field and the backlight field, respectively;

[0013] S5: train the visual detection model under the reflection field using the reflection field dataset, and train the visual detection model under the backlight field using the backlight field dataset;

[0014] S6: obtain cross stroboscopic images of a surface of a pultruded plate to be detected under the reflection field and the backlight field;

[0015] S7: split the cross stroboscopic images of the surface of the pultruded plate to be detected to obtain a to-be-detected reflection field image and a to-be-detected backlight field image;

[0016] S8: input the to-be-detected reflection field image into the trained visual detection model under the reflection field for detection to obtain a reflection field detection result, and input the to-be-detected backlight field image into the trained visual detection model under the backlight field for detection to obtain a backlight field detection result;

[0017] S9: obtain a final detection result according to the reflection field detection result and the backlight field detection result.

[0018] The second aspect:

[0019] The embodiment of the present application provides a kind of pultruded plate surface defect detection system for wind power blade, comprising: memory and one or more processors;

[0020] The memory stores one or more applications, and the one or more applications are adapted to be executed by the one or more processors to implement the above-mentioned pultruded plate surface defect detection method for wind power blade.

[0021] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects:

[0022] In the present application, by acquiring the cross stroboscopic images of the pultruded plate surface under the reflection field and the backlight field, and splitting the cross stroboscopic images to obtain the reflection field image and the backlight field image, the translucent characteristics of the pultruded plate for wind power blades are fully considered, which helps to effectively extract internal defects under the traditional reflection light field. By respectively constructing the visual detection model based on deep learning under the reflection field and the backlight field, the pultruded plate defect detection no longer faces the problems of noise interference, high equipment and maintenance cost, high technical requirements for operators and high technical difficulty. By constructing the reflection field dataset based on the reflection field image, constructing the backlight field dataset based on the backlight field image, training the visual detection model under the reflection field using the reflection field dataset, and training the visual detection model under the backlight field using the backlight field dataset, the accuracy of the pultruded plate surface defect detection is improved. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0024] Figure 1 A flowchart of a pultruded plate surface defect detection method for wind power blades provided by the present application is shown in the figure.

[0025] Figure 2 A structural diagram of a multi-view field imaging device for the pultruded plate surface of the wind power blades provided by the present application is shown in the figure.

[0026] Figure 3 A structural diagram of a pultruded plate surface defect detection system for wind power blades provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0027] The technical solutions in the present application will be described below with reference to the drawings.

[0028] In the embodiments of the present application, the words such as "example", "for example" and the like are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0029] In order to make the technical problems, technical solutions and advantages of the present application clearer, the following will be described in detail with reference to the drawings and specific embodiments.

[0030] Referring to the accompanying drawings Figure 1 , a flowchart of a method for detecting surface defects of a pultruded plate for a wind turbine blade is shown.

[0031] The present application provides a method for detecting surface defects of a pultruded plate for a wind turbine blade, which can be implemented by a pultruded plate surface defect detection device for a wind turbine blade, which can be a terminal or a server. The processing flow of the method for detecting surface defects of the pultruded plate for the wind turbine blade can include the following steps:

[0032] S1: Obtain cross stroboscopic images of the pultruded plate surface under reflection field and backlight field.

[0033] Referring to the accompanying drawings Figure 2 , a structure diagram of a multi-view imaging device for the surface of the pultruded plate for the wind turbine blade is shown.

[0034] In one possible implementation, S1 is specifically:

[0035] A time-sharing stroboscopic strategy based on encoder trigger signal is used to alternately collect, and cross stroboscopic images of the pultruded plate surface under reflection field and backlight field are obtained.

[0036] Specifically, two light sources of different angles and one linear array camera are installed on the upper and lower surfaces of the pultruded plate, a total of four light sources and two linear array cameras. The installation angle and brightness of the corresponding one linear array camera and two light sources of different angles on the upper and lower surfaces of the pultruded plate are adjusted respectively, and the reflection field and backlight field of the upper surface of the pultruded plate and the reflection field and backlight field of the lower surface of the pultruded plate are formed respectively. A time-sharing stroboscopic strategy based on encoder trigger signal is used, and the linear array camera is triggered by the one-way motion of the pultruded plate, so that the linear array camera and the light source on the same surface cooperate to perform the rapid alternating collection action of the backlight field image and the reflection field image, thereby obtaining the cross stroboscopic images of the upper and lower surfaces of the pultruded plate under the reflection field and the backlight field.

[0037] It should be noted that the light source on the same surface of the pultruded plate is a reflection light source and a backlight light source, and the distance between the reflection light source and the backlight light source and the surface of the pultruded plate is 12 cm and 3 cm respectively, and the included angle of the acute angle type formed between them and the surface of the pultruded plate is 70° and 80° respectively.

[0038] It should be noted that the time-sharing stroboscopic strategy based on the encoder trigger signal is:

[0039] The installation roller-encoder structure is used to trigger the light source stroboscopic signal and the acquisition signal of the camera when the pultruded plate moves unidirectionally.

[0040] Specifically, the pultruded plate enters the device from the entrance and extends out of the exit of the device, starts to move unidirectionally inside the device, rotates the roller through the movement, and then triggers the encoder to send a trigger signal, so that the four light sources occur cross stroboscopic, and the two linear array cameras acquire cross stroboscopic images of the upper and lower surfaces of the pultruded plate under the reflection field and the backlight field, respectively.

[0041] In the present application, by installing light sources of different angles on the upper and lower surfaces of the pultruded plate, images under different lighting conditions can be obtained, which helps to reveal surface defects more comprehensively. This setting can capture different defect characteristics under the reflection field and the backlight field, thereby improving the accuracy and comprehensiveness of defect detection. The time-sharing stroboscopic strategy allows the reflection field and the backlight field images to be quickly alternately acquired on the same surface, improving the efficiency of image acquisition. This method can obtain the required images under multiple lighting conditions in a short time, which is suitable for large-scale production and detection.

[0042] S2: split the cross stroboscopic image to obtain the reflection field image and the backlight field image.

[0043] In a possible implementation, S2 is specifically:

[0044] The cross stroboscopic image is split using a gray-scale threshold-based image splitting method to obtain the reflection field image and the backlight field image.

[0045] In a possible implementation, the cross stroboscopic image is split using a gray-scale threshold-based image splitting method to obtain the reflection field image and the backlight field image, specifically including:

[0046] The cross stroboscopic image is gray-scaled to obtain a gray-scale image.

[0047] The global pixel gray-scale value information of the gray-scale image is obtained to determine the gray-scale threshold.

[0048] It should be noted that the size of the gray-scale threshold can be set by a person skilled in the art according to actual needs, which is not limited in the present application.

[0049] The gray-scale image is split into multiple image lines with image lines as the splitting unit.

[0050] The row pixel gray-scale mean value of each image line is calculated.

[0051] The image lines with the row pixel gray-scale mean value greater than the gray-scale threshold are selected as the reflection field image lines, and the image lines with the row pixel gray-scale mean value less than or equal to the gray-scale threshold are selected as the backlight field image lines.

[0052] The screened reflection field image rows and backlight field image rows are spliced according to the arrangement order of the image rows of the cross stroboscopic image, and the reflection field image and the backlight field image are obtained respectively.

[0053] In the present application, by setting the gray threshold, the image regions under the reflection field and the backlight field can be accurately distinguished. The flexible adjustment of the gray threshold makes the method adaptable to different lighting conditions and material surface characteristics, so as to more accurately extract the required image information. By effectively separating the images of the reflection field and the backlight field, the defect features under different lighting conditions can be extracted respectively. This method helps to discover and identify defects that may be ignored under single lighting conditions, thereby improving the overall detection accuracy. By calculating the gray mean value of each image row and screening, the image noise caused by environmental light or surface reflection can be effectively reduced. This helps to generate clearer reflection field and backlight field images, enhancing the reliability of subsequent defect detection.

[0054] In a possible implementation, after S2 and before S3, further comprising:

[0055] The reflection field image and the backlight field image are cut using an image-based region extraction method to obtain a reflection field detection region image and a backlight field detection region image.

[0056] In a possible implementation, the reflection field image and the backlight field image are cut using an image-based region extraction method to obtain a reflection field detection region image and a backlight field detection region image, specifically comprising:

[0057] The reflection field image is binarized to obtain a binary image.

[0058] The binary image is subjected to an opening operation and a dilation operation to extract black and white pixel regions in the binary image.

[0059] According to the coordinates of the boundary lines of the black and white pixel regions, the cutting boundary coordinates are determined.

[0060] According to the cutting boundary coordinates, the reflection field image and the backlight field image are cut respectively to obtain a reflection field detection region image and a backlight field detection region image.

[0061] In the present application, by cutting the image, the truly valuable detection area is retained and extracted, which can effectively reduce the irrelevant information or interference factors in the image. In this way, the subsequent defect detection can be more focused on the key parts, and the detection accuracy can be improved. Through the binarization processing and the opening operation and the inflation operation, the noise or irregular area in the image can be effectively removed. This process can make the image clearer, which is helpful for accurately positioning the defect area and improving the overall detection efficiency.

[0062] S3: constructing a reflection field dataset based on the reflection field images, and constructing a backlight field dataset based on the backlight field images.

[0063] It should be noted that the number of images in the reflection field dataset is greater than or equal to the first preset number, and the number of images in the backlight field dataset is greater than or equal to the second preset number.

[0064] It should be noted that the first preset number and the second preset number can be set according to actual needs by those skilled in the art, and the present application does not limit them.

[0065] Optionally, the number of images in the reflection field dataset is greater than or equal to 1000, and the number of images in the backlight field dataset is greater than or equal to 1000.

[0066] The reflection field dataset is divided into a reflection field training dataset and a reflection field test dataset according to a first preset ratio, and the backlight field dataset is divided into a backlight field training dataset and a backlight field test dataset according to a second preset ratio.

[0067] It should be noted that the first preset ratio and the second preset ratio can be set according to actual needs by those skilled in the art, and the present application does not limit them.

[0068] Optionally, the reflection field dataset is divided into a reflection field training dataset and a reflection field test dataset according to a ratio of 8:2, and the backlight field dataset is divided into a backlight field training dataset and a backlight field test dataset according to a ratio of 8:2.

[0069] In the present application, by respectively constructing the reflection field dataset and the backlight field dataset, it can be ensured that the features and defects on the surface of the pultruded plate are fully characterized under different lighting conditions. This multi-view data construction can capture more details and provide more comprehensive training data. By dividing the dataset into a training dataset and a test dataset according to a ratio, it is ensured that most of the data is used for training the model, so that the model can obtain sufficient feature information in the learning process. This division method improves the generalization ability of the model, so that it performs more stably in actual application and can effectively deal with unseen data.

[0070] In one possible implementation, after S3 and before S4, the method further includes:

[0071] The reflection field dataset and the backlight field dataset are subjected to data enhancement processing using a random global brightness noise image enhancement method.

[0072] In a possible implementation, the reflection field dataset and the backlight field dataset are subjected to data enhancement processing using a random global brightness noise image enhancement method, specifically including:

[0073] The brightness value distribution of all images in the reflection field dataset and the backlight field dataset is respectively acquired.

[0074] The brightness interval of the reflection field dataset and the backlight field dataset is respectively determined according to the brightness value distribution.

[0075] The brightness of all images in the reflection field dataset and the backlight field dataset is subjected to random-size brightness transformation, so as to perform data enhancement processing on the reflection field dataset and the backlight field dataset.

[0076] In the present application, by performing random-size brightness transformation on images, images under various brightness conditions can be generated to simulate different actual application scenarios. This data diversification enables the model to be exposed to more variations during the training process, thereby improving its robustness when facing uncertain environments. Since the surface of the pultruded plate may encounter various lighting conditions in actual detection, by introducing brightness noise for enhancement processing, the model can learn the feature performance under different lighting conditions, thereby improving the adaptability of the model to lighting changes.

[0077] S4: constructing a deep learning-based visual detection model under the reflection field and the backlight field respectively.

[0078] Optionally, a ternary network-based visual detection model under the reflection field and the backlight field is constructed respectively, and SqueezeNet is selected as the backbone network of the ternary network to perform feature extraction on the input image.

[0079] In the present application, the visual detection model based on the ternary network and using SqueezeNet as the backbone network can realize the lightweight and high efficiency of the model under the premise of ensuring high detection accuracy. This design not only improves the real-time performance and generalization ability of the model, but also provides support for the rapid deployment and efficient operation in actual industrial applications.

[0080] S5: training the visual detection model under the reflection field using the reflection field dataset, and training the visual detection model under the backlight field using the backlight field dataset.

[0081] Specifically, the deep learning-based visual detection models under the reflection field and the backlight field are iteratively trained using the reflection field dataset and the backlight field dataset respectively. During the training process, the network loss values of the deep learning-based visual detection models under the reflection field and the backlight field are calculated using a triplet loss function, and the obtained network loss values are subjected to gradient backpropagation for weight updating of the deep learning-based visual detection models under the reflection field and the backlight field respectively.

[0082] In the present application, by training the visual detection models under the reflection field and the backlight field respectively, each model can be adapted to a specific lighting condition, thereby achieving the best detection effect in the corresponding scene. The lighting characteristics of the reflection field and the backlight field are different, so the model needs to extract different features. This separate training method can better capture these features. By using gradient backpropagation to optimize the network loss value, the model can gradually adjust its parameters to better identify defects. This precise optimization process can effectively reduce false positives and false negatives, improving the overall detection accuracy of the model.

[0083] S6: Obtain cross stroboscopic images of the surface of the pultruded plate to be detected under the reflection field and the backlight field.

[0084] S7: Split the cross stroboscopic images of the surface of the pultruded plate to be detected to obtain the reflection field image to be detected and the backlight field image to be detected.

[0085] S8: Input the reflection field image to be detected into the trained visual detection model under the reflection field for detection to obtain the reflection field detection result, and input the backlight field image to be detected into the trained visual detection model under the backlight field for detection to obtain the backlight field detection result.

[0086] Specifically, the reflection field image to be detected is input into the trained visual detection model under the reflection field, and the updated visual detection model weight under the reflection field is used to detect the reflection field image to obtain the reflection field detection result. The backlight field image to be detected is input into the trained visual detection model under the backlight field, and the updated visual detection model weight under the backlight field is used to detect the backlight field image to obtain the backlight field detection result.

[0087] In the present application, using the updated weight after training for detection can ensure that the model can accurately identify and process the image to be detected after sufficient training. The weight optimized during the training process can maximize the detection performance of the model. Inputting the reflection field image and the backlight field image to be detected into the trained special model can optimize the detection effect of each model for a specific lighting condition, thereby improving the detection accuracy.

[0088] S9: obtaining a final detection result according to the reflection field detection result and the backlight field detection result.

[0089] In a possible implementation, S9 specifically includes:

[0090] performing or operating the reflection field detection result and the backlight field detection result to obtain the final detection result:

[0091]

[0092] wherein Out represents the final detection result, OR represents a logical OR operator, Out1 and Out2 respectively represent the reflection field detection result and the backlight field detection result, wherein the values of Out1 and Out2 are 1 or 0, wherein 1 represents detecting a defect, and 0 represents not detecting a defect.

[0093] It should be noted that when the reflection field detection result and / or the backlight field detection result is 1, the final detection result is detecting a defect, and when the reflection field detection result and the backlight field detection result are both 0, the final detection result is not detecting a defect.

[0094] In the present application, the reflection field and the backlight field provide different visual information, and can detect potential defects from different angles and lighting conditions. The logical OR operation can integrate the two kinds of information to ensure that no defects are missed under any condition. Since the reflection field and the backlight field each have their own characteristics and limitations, relying on only one condition may result in missed detection or false detection. By combining the two detection results, the robustness and accuracy of the final detection result are enhanced, and the sensitivity and reliability of the detection system for various defects are improved.

[0095] The technical scheme provided by the embodiments of the present application has at least the following beneficial effects:

[0096] In the present application, by acquiring cross stroboscopic images of the pultruded plate surface under the reflection field and the backlight field, and splitting the cross stroboscopic images to obtain reflection field images and backlight field images, the translucent characteristics of the pultruded plate for wind power blades are fully considered, which helps to effectively extract internal defects under the traditional reflected light field. By respectively constructing a visual detection model based on deep learning under the reflection field and the backlight field, the defect detection of the pultruded plate no longer faces the problems of noise interference, high equipment and maintenance cost, high technical requirements for operators, and high technical difficulty. By constructing a reflection field dataset based on the reflection field images, constructing a backlight field dataset based on the backlight field images, training the visual detection model under the reflection field using the reflection field dataset, and training the visual detection model under the backlight field using the backlight field dataset, the accuracy of the pultruded plate surface defect detection is improved.

[0097] With reference to the accompanying drawings, the present application provides a kind of wind turbine blade with the structure diagram of the detection system of surface defect of pultruded plate shown in the figure. Figure 3 , shows the structure diagram of the detection system of surface defect of pultruded plate for wind turbine blade provided by the present application.

[0098] The present application also provides a kind of wind turbine blade with the detection system 30 of surface defect of pultruded plate, comprising: memory 303 and one or more processors 301.

[0099] The memory 303 has one or more application programs stored therein, and the one or more application programs are adapted to be executed by the one or more processors 301 to realize the detection method of surface defect of pultruded plate for wind turbine blade described in the method embodiment.

[0100] The detection system 30 of surface defect of pultruded plate for wind turbine blade includes: processor 301 and memory 303.

[0101] The structure of the detection system 30 of surface defect of pultruded plate for wind turbine blade does not constitute limitation to the embodiments of the present application.

[0102] The processor 301 can be CPU, general processor, DSP, ASIC, FPGA or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can realize or execute the various exemplary logic blocks, modules and circuits described in combination with the disclosure. The processor 301 can also be a combination of computing functions, such as one or more microprocessor combinations, DSP and microprocessor combinations, etc.

[0103] The bus 302 can include a passage for transmitting information between the above-mentioned components. The bus 302 can be PCI bus or EISA bus, etc. The bus 302 can be divided into address bus, data bus, control bus, etc. For the convenience of indication, only one thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0104] The memory 303 can be ROM or other types of static storage devices that can store static information and instructions, RAM or other types of dynamic storage devices that can store information and instructions, and can also be EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and capable of being accessed by a computer, but not limited to this.

[0105] It should be noted that the wind turbine blade pultruded plate surface defect detection system 30 can realize the wind turbine blade pultruded plate surface defect detection method described above, and can realize the same or similar technical effects, to avoid repetition, the present application will not be described again.

[0106] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects:

[0107] In the present application, by acquiring the cross stroboscopic images of the pultruded plate surface under the reflection field and the backlight field, and splitting the cross stroboscopic images to acquire the reflection field image and the backlight field image, the translucent characteristics of the pultruded plate for wind turbine blades are fully considered, which helps to effectively extract internal defects under the traditional reflected light field, and by respectively constructing the visual detection model based on deep learning under the reflection field and the backlight field, the pultruded plate defect detection no longer faces the problems of noise interference, high equipment and maintenance cost, high technical requirements for operators and high technical difficulty, by constructing the reflection field dataset based on the reflection field image, constructing the backlight field dataset based on the backlight field image, training the visual detection model under the reflection field using the reflection field dataset, and training the visual detection model under the backlight field using the backlight field dataset, the accuracy of the pultruded plate surface defect detection is improved.

[0108] The present application also provides a computer readable storage medium having a computer program stored thereon, which can be loaded and executed by a processor to perform the wind turbine blade pultruded plate surface defect detection method of the first aspect.

[0109] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0110] The following points need to be explained:

[0111] (1) The drawings of the embodiments of the present application only involve the structures involved in the embodiments of the present application, and other structures can be referred to the usual design.

[0112] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present application, the thickness of the layer or region is magnified or reduced, that is, the drawings are not drawn according to the actual proportion. It can be understood that when an element such as a layer, a film, a region or a substrate is referred to as being located "on" or "under" another element, the element can be "directly" located "on" or "under" another element or there can be an intermediate element.

[0113] (3) In the case of no conflict, the embodiments and features in the embodiments can be combined to obtain new embodiments.

[0114] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for detecting surface defects in pultruded sheet metal for wind turbine blades, characterized in that, include: S1: Acquire cross-stroboscopic images of the pultruded sheet surface under reflected and backlit fields; S2: Decompose the cross-strobe image to obtain the reflection field image and the backlight field image; S3: Based on the reflected field image, construct a reflected field dataset; based on the backlight field image, construct a backlight field dataset. S4: Construct deep learning-based visual detection models under reflected and backlit fields respectively; S5: Train the visual detection model in the reflected field using the reflected field dataset, and train the visual detection model in the backlight field using the backlight field dataset; S6: Acquire cross-stroboscopic images of the surface of the pultruded sheet to be tested under reflected and backlight fields; S7: Decompose the cross-stroboscopic image of the surface of the pultruded sheet to be tested to obtain the reflection field image and the backlight field image to be tested; S8: Input the image of the reflection field to be detected into the trained visual detection model under the reflection field for detection to obtain the reflection field detection result; input the image of the backlight field to be detected into the trained visual detection model under the backlight field for detection to obtain the backlight field detection result. S9: Based on the reflection field detection results and the backlight field detection results, the final detection result is obtained; Specifically, S2 is: The cross-stroboscopic image is split using a grayscale threshold-based image splitting method to obtain the reflection field image and the backlight field image; Specifically, the step of using a grayscale threshold-based image segmentation method to segment the cross-stroboscopic image to obtain the reflection field image and the backlight field image includes: The cross-stroboscopic image is converted to grayscale to obtain a grayscale image; Obtain the global pixel grayscale value information of the grayscale image and determine the grayscale threshold; The grayscale image is divided into multiple image rows, using image rows as the splitting unit. Calculate the average grayscale value of each row of pixels in the image; Image rows whose average pixel grayscale value is greater than the grayscale threshold are selected as reflectance field image rows, and image rows whose average pixel grayscale value is less than or equal to the grayscale threshold are selected as backlight field image rows. The selected rows of reflection field images and rows of backlight field images are stitched together according to the arrangement order of the rows of the cross strobe image to obtain the reflection field image and the backlight field image respectively. The section following S2 and preceding S3 also includes: Using an image-based region extraction method, the reflection field image and the backlight field image are cropped to obtain the reflection field detection region image and the backlight field detection region image. Specifically, the step of using an image-based region extraction method to crop the reflection field image and the backlight field image to obtain the reflection field detection region image and the backlight field detection region image includes: The reflection field image is binarized to obtain a binary image; Opening and dilation operations are performed on the binary image to extract the black and white pixel regions in the binary image; Determine the cropping boundary coordinates based on the coordinates of the dividing line of the black and white pixel regions; Based on the cropping boundary coordinates, the reflection field image and the backlight field image are cropped respectively to obtain the reflection field detection area image and the backlight field detection area image.

2. The method for detecting surface defects in pultruded sheet for wind turbine blades according to claim 1, characterized in that, Specifically, S1 is: A time-division stroboscopic strategy based on encoder trigger signals is used to acquire cross-stroboscopic images of the pultruded sheet surface under reflected and backlight fields.

3. The method for detecting surface defects in pultruded sheet for wind turbine blades according to claim 1, characterized in that, After S3 and before S4, it also includes: The reflection field dataset and the backlight field dataset are augmented using a random global brightness noise image enhancement method.

4. The method for detecting surface defects in pultruded sheet for wind turbine blades according to claim 3, characterized in that, The method of using random global brightness noise image enhancement to perform data enhancement processing on the reflection field dataset and the backlight field dataset specifically includes: The brightness value distribution of all images in the reflection field dataset and the backlight field dataset are obtained respectively. Based on the brightness value distribution, the brightness ranges of the reflection field dataset and the backlight field dataset are determined respectively; With the goal of ensuring that the brightness value of the image falls within the brightness range, a brightness transformation of random size is performed on all images in the reflection field dataset and the backlight field dataset respectively, in order to perform data augmentation processing on the reflection field dataset and the backlight field dataset.

5. The method for detecting surface defects in pultruded sheet for wind turbine blades according to claim 1, characterized in that, Specifically, S9 is: The final detection result is obtained by performing an OR operation on the reflection field detection result and the backlight field detection result: ； Where Out represents the final detection result, OR represents the logical OR operator, Out1 and Out2 represent the reflection field detection result and the backlight field detection result, respectively. The values ​​of Out1 and Out2 are both 1 or 0, where 1 indicates that a defect was detected and 0 indicates that no defect was detected.

6. A system for detecting surface defects in pultruded sheet metal for wind turbine blades, characterized in that, include: Memory and one or more processors; The memory stores one or more application programs adapted to be executed by the one or more processors to implement the method for detecting surface defects in pultruded sheet metal for wind turbine blades as described in any one of claims 1 to 5.

Images