Training method, device, equipment and medium for wind turbine blade bolt hole recognition model

Abstract

The present application discloses a training method, device, equipment and medium for a wind turbine blade bolt hole recognition model. The training method for the wind turbine blade bolt hole recognition model includes: obtaining a data set including a plurality of wind turbine blade bolt hole sample images and corresponding label images, wherein the label images are used to indicate the labels of the wind turbine blade bolt holes; dividing the images and their label images in the data set to obtain a first data set and a second data set; loading the model parameters of the first recognition model obtained by training the preset model according to the first data set into the preset model to obtain a second recognition model; fixing the network parameters of the first target layer structure of the second recognition model, and training the second recognition model according to the second data set to determine the network parameters of the second target layer structure of the second recognition model to obtain a wind turbine blade bolt hole recognition model. According to the embodiments of the present application, the model training efficiency and model accuracy can be improved.

Description

Technical Field

[0001] The present application belongs to the field of deep learning technology, and in particular relates to a training method, device, equipment and medium for a wind turbine blade bolt hole recognition model. Background Art

[0002] Wind turbine blades have bolt holes evenly distributed along their root ends, allowing them to be bolted together with the wind turbine generator. To facilitate bolt assembly and disassembly of wind turbine blades, bolt hole image recognition is required. However, the recognition model used in related technologies for bolt hole recognition in wind turbine blades is relatively inaccurate, resulting in poor bolt hole recognition accuracy. Summary of the Invention

[0003] The embodiments of the present application provide a training method, device, and information push method for a wind turbine blade bolt hole recognition model, which can ensure the accuracy of the wind turbine blade bolt hole recognition model.

[0004] In a first aspect, an embodiment of the present application provides a method for training a wind turbine blade bolt hole recognition model, comprising:

[0005] Acquire a data set including a plurality of wind turbine blade bolt hole sample images and corresponding label images, wherein the label images are used to indicate labels of the wind turbine blade bolt holes;

[0006] Dividing the images and their label images in the data set to obtain a first data set and a second data set;

[0007] Processing the first data set and the second data set so that data samples in the two data sets meet a preset similarity condition;

[0008] Training a preset model according to the processed first data set to obtain a first recognition model;

[0009] Loading the model parameters of the first recognition model into the preset model to obtain a second recognition model;

[0010] The network parameters of the first target layer structure of the second recognition model are fixed, and the second recognition model is trained according to the second data set to determine the network parameters of the second target layer structure of the second recognition model to obtain a wind turbine blade bolt hole recognition model.

[0011] In a second aspect, an embodiment of the present application provides a training device for a wind turbine blade bolt hole recognition model, the device comprising:

[0012] An acquisition module, configured to acquire a data set including a plurality of sample images of bolt holes of wind turbine blades and corresponding label images, wherein the label images are used to indicate labels of the bolt holes of the wind turbine blades;

[0013] A division module, configured to divide the images and their label images in the data set to obtain a first data set and a second data set;

[0014] A processing module, configured to process the first data set and the second data set so that data samples in the two data sets meet a preset similarity condition;

[0015] A first training module is used to train a preset model according to the processed first data set to obtain a first recognition model;

[0016] A loading module, configured to load the model parameters of the first recognition model into the preset model to obtain a second recognition model;

[0017] The second training module is used to fix the network parameters of the first target layer structure of the second recognition model, and train the second recognition model according to the second data set to determine the network parameters of the second target layer structure of the second recognition model to obtain the wind turbine blade bolt hole recognition model.

[0018] In a third aspect, an embodiment of the present application provides an electronic device comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the training method for the wind turbine blade bolt hole recognition model as described in any embodiment of the first aspect are implemented.

[0019] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the steps of the training method for the wind turbine blade bolt hole recognition model as described in any embodiment of the first aspect are implemented.

[0020] The training method, device, equipment and medium of the wind turbine blade bolt hole recognition model in the embodiment of the present application divides the data set into a first data set and a second data set based on a data set composed of multiple acquired wind turbine blade bolt hole sample images and corresponding label images, and further processes the two data sets so that they meet a certain degree of correlation; then, the first recognition model parameters obtained by training the processed first data set are directly loaded into the second recognition model, and the second recognition model is trained according to the second data set to obtain a wind turbine blade bolt hole recognition model. In this way, the existing trained model is used and retrained on its basis, which can greatly speed up the training process, improve the model training efficiency, and improve the accuracy of the wind turbine blade bolt hole recognition model, thereby ensuring the accuracy of wind turbine blade bolt hole recognition based on the recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 This is a flow chart of a method for training a wind turbine blade bolt hole recognition model provided in an embodiment of the present application;

[0023] Figure 2 This is a flow chart of another method for training a wind turbine blade bolt hole recognition model provided in an embodiment of the present application;

[0024] Figure 3 This is a structural diagram of a training device for a wind turbine blade bolt hole recognition model provided by an embodiment of the present application;

[0025] Figure 4 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The features and exemplary embodiments of various aspects of the present application will be described in detail below. 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 conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0027] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

[0028] It should be noted that the acquisition, storage, use and processing of data in the embodiments of this application are in compliance with the relevant provisions of national laws and regulations.

[0029] Wind turbine blades have bolt holes evenly distributed along their root ends, allowing them to be bolted together with the wind turbine generator. To facilitate bolt assembly and disassembly of wind turbine blades, bolt hole image recognition is required. However, the recognition model used in related technologies for bolt hole recognition in wind turbine blades is relatively inaccurate, resulting in poor bolt hole recognition accuracy.

[0030] In order to solve the problems of the related art, the embodiments of the present application provide a training method, device, equipment and medium for a wind turbine blade bolt hole recognition model.

[0031] The following describes in detail the training method for the wind turbine blade bolt hole recognition model provided by the embodiment of the present application through specific embodiments and application scenarios in conjunction with the accompanying drawings.

[0032] Figure 1 FIG. 1 is a flow chart showing a method 100 for training a wind turbine blade bolt hole recognition model according to an embodiment of the present application. Figure 1 As shown, the wind turbine blade bolt hole recognition model training method 100 may specifically include the following steps:

[0033] S101, acquiring a data set including a plurality of wind turbine blade bolt hole sample images and corresponding label images, wherein the label images are used to indicate labels of the wind turbine blade bolt holes;

[0034] S102, dividing the images in the data set and their label images to obtain a first data set and a second data set;

[0035] S103, processing the first data set and the second data set so that data samples in the two data sets meet a preset similarity condition;

[0036] S104: training a preset model based on the processed first data set to obtain a first recognition model;

[0037] S105, loading the model parameters of the first recognition model into the preset model to obtain a second recognition model;

[0038] S106. Fix the network parameters of the first target layer structure of the second recognition model, and train the second recognition model according to the second data set to determine the network parameters of the second target layer structure of the second recognition model, and obtain a wind turbine blade bolt hole recognition model.

[0039] Therefore, based on the data set composed of the multiple acquired wind turbine blade bolt hole sample images and the corresponding label images, the data set is divided into a first data set and a second data set, and the two data sets are further processed to satisfy a certain degree of correlation; then, the first recognition model parameters obtained by training the processed first data set are directly loaded into the second recognition model, and the second recognition model is trained according to the second data set to obtain a wind turbine blade bolt hole recognition model. In this way, the existing trained model is used and retrained on its basis, which can greatly speed up the training process, improve the model training efficiency, and improve the accuracy of the wind turbine blade bolt hole recognition model, thereby ensuring the accuracy of wind turbine blade bolt hole recognition based on the recognition model.

[0040] The specific implementation methods of the above steps are introduced below.

[0041] In some embodiments, in step S101, a 3D camera may be used to capture images of the bolt holes of a wind turbine blade. Thus, the multiple sample images of the bolt holes of the wind turbine blade may be high-precision 3D image data. Furthermore, to achieve similar distributions among the sample data obtained through subsequent processing of the multiple sample images of the bolt holes of the wind turbine blade, the camera may be calibrated before each image capture of different bolt holes of the wind turbine blade, so that the captured images of the different bolt holes of the wind turbine blade are all in the same reference coordinate system.

[0042] In specific implementation, the camera and projector can be calibrated in the structured light system through a calibration plate. The calibration plate has a known three-dimensional shape and feature points. By photographing and analyzing it, the spatial relationship between the camera and the projector, as well as the deformation law of the light pattern or light spot under the camera's field of view, can be calculated; after the calibration is completed, the 3D camera enters the preset position, that is, the target position corresponding to the target bolt hole to be collected, to collect the image and depth information of the edge of the target bolt hole.

[0043] In some embodiments, before step S102 , the collected wind turbine blade bolt hole image data is preprocessed to improve data quality, thereby improving processing effects in subsequent steps.

[0044] In practice, this approach optimizes and integrates traditional filters (such as Gaussian, median, and conditional filters) to remove noise, using adaptive filters to streamline data for improved computational performance and real-time performance. Specifically, the parameter ratios of Gaussian and median denoising can be adjusted based on the acquired lighting conditions, and pixel denoising is performed based on certain conditions to clean the acquired image data. This results in improved robustness and accuracy in the actual wind turbine blade scenario.

[0045] In practice, because color channels reflect the dimensionality of the image array, the captured wind turbine blade bolt hole image data is a color (RGB) image consisting of three color channels, while grayscale images have only one channel. As you can see, the more complex the color channels, the more complex the dataset and the longer it takes to train the model. Therefore, the wind turbine blade bolt hole image data can be further converted into grayscale images.

[0046] It should be understood that for the preset model, the size of the input image data must be appropriate. If the image size is too small, the model will be unable to recognize the image's features, while if the image size is too large, the required computing resources will increase. Therefore, after denoising, the images in the dataset and their labeled images are further resized to obtain images of a first size; these images of the first size are then randomly cropped to obtain images of a second size. In this way, through resizing and cropping, the image data is adapted to the preset model, that is, it meets the size accepted by the preset model.

[0047] Furthermore, the cropped image of the second size is converted into tensor data; and then the tensor data is normalized to obtain normalized tensor data.

[0048] In specific implementation, you can call the transforms.Resize function in torchvision to change the input image size to the first size (for example, 256×256), then call the transforms.CenterCrop function to randomly crop the image to the second size (for example, 224×224); then call the transforms.ToTensor function to convert the cropped image into a Tensor tensor, and finally call the transforms.Normalize function to normalize the image.

[0049] In some embodiments, the sample images of wind turbine blade bolt holes are acquired at different acquisition time intervals. Therefore, in step S102, the images in the dataset and their labeled images can be divided based on the acquisition time intervals, and the acquisition time intervals of the first dataset and the second dataset are adjacent time intervals.

[0050] In this way, due to the fact that data has a large correlation in adjacent time intervals before and after the production process, sample migration can be achieved, thereby enabling data reuse to avoid the problem of low accuracy in final model training due to a small number of image data samples.

[0051] Furthermore, in some embodiments, in step S103, based on the first data set and the second data set, the similarity between each data sample in the first data set and the second data set is determined; based on the similarity, the initialization weight of each data sample in the first data set is determined; and for each data sample in the first data set, initialization is performed according to the initialization weight to obtain a processed first data set.

[0052] In another embodiment, the VeryFastKMM algorithm can be combined to complete sample migration and achieve reuse between limited data samples to enrich the sample size, thereby improving the model training accuracy. Specifically, the first data set is divided into a first training set and a first test set according to a preset division ratio, and the second data set is divided into a second training set and a second test set; the first training set and the second training set are combined into a target training set. The first training set is defined as X auxiliary , define the second training set as X target , the second test set is defined as S, and the weight of the sample data in the first data set is defined as β.

[0053] In addition, the data in the target training set can be obtained by random sampling of the first training set and the second training set respectively, so as to reduce the computational complexity and computing resources. In addition, the sampling sample size p, parameter θ, and tolerance η are defined.

[0054] Specifically, the weight of the first data set can be iteratively adjusted and updated using the following formula.

[0055]

[0056]

[0057]

[0058] β=aggeregate(β *(i) ); (4)

[0059] From the above formulas (1) to (4), it can be seen that the number of sampling samples is determined by the above formula (1), and based on formula (2), the number of sampling samples is determined from the first training set X auxiliary Randomly select p samples from Then, through formula (3) in the sampling sample Apply the KMM algorithm and adopt the sampling method with replacement; then, the β of the integrated sampling sample set is *(i) As β. Repeat this iterative cycle to get the updated weight.

[0060] It should be noted that there is no limit on the number of iterative cycles, so as to minimize the distribution difference between the data samples in the two data sets by continuously adjusting the weights.

[0061] The weight allocation strategy of the Tradaboost algorithm used in the related art enables the samples in the first data set to obtain the same initial weights during weight initialization. However, since the similarity between each sample in the first data set and the second data set is very different, the same initial weight allocation cannot ensure that the processed first data set and the second data set meet the preset similarity conditions.

[0062] Therefore, in this embodiment, based on the weight initialization strategy of the VeryFastKMM algorithm, samples in the second data set with higher similarity to the first data set are given higher weights during weight initialization, and samples with lower similarity are given lower weights. Compared with the related art in which the weight of each sample is the same, the weight adjustment can ensure that the similarity or correlation conditions are met between the samples in sample migration, thereby improving the quality and efficiency of sample migration.

[0063] Thus, in step S104, the preset model is trained according to the processed first data set to obtain a first recognition model, wherein the first recognition model can be a weak learner; thus, in steps S105 to S106, the model parameters of the weak learner generated by the training of the first data set are directly loaded into the new weak learner model, the first several layers of the original model are fixed, the new layers behind the model are redefined, and the second data set is loaded to fine-tune some of its layers, thereby forming a new weak learner, that is, obtaining a wind turbine blade bolt hole recognition model.

[0064] In another embodiment, the wind turbine blade bolt hole recognition model includes a third target layer structure, which is obtained based on the first target layer structure and the second target layer structure. Specifically, the network parameters of the first target layer structure and the network parameters of the second target layer structure can be input into the third target layer structure; based on the weights of the first target layer structure and the second target layer structure, the weights of the third target layer structure are obtained by training according to the second data set.

[0065] That is to say, after a basic network (i.e., the first recognition model) has been learned on the first data set, a network with the same architecture as the first recognition model but with different parameters can be trained to fit the second data set. That is, the first recognition model is expanded and a new BiGRU layer is trained based on the last BiGRU layer of the first recognition model. The two layers are connected to the shared layer SharedLayer, and the outputs of the original layer and the new layer are weighted by the weight matrix U of the first recognition model and the second recognition model W. In other words, the input of the shared layer is the fusion of the output information of the historical neural network layer and the output information of the current neural network layer.

[0066] In this way, the output of the original network layer (i.e., the recognition model obtained by training the first data set) and the output of the current network layer (i.e., the recognition model obtained by training the second data set) are input into the third target layer structure together. The third target layer structure can be defined as a shared layer and combined with a certain weight ratio to extract the information of the original neural network and fuse it with the current information. Compared with the model migration with fine-tuning processing of the model, the final wind turbine blade bolt hole recognition model has higher accuracy.

[0067] In addition, the present application provides another training method 200 for a wind turbine blade bolt hole recognition model, that is, another specific implementation of the training of a wind turbine blade bolt hole recognition model, which can be referred to Figure 2 .

[0068] like Figure 2 As shown, in step S201, the first data set is divided into a first training set and a first test set according to a preset division ratio, and the second data set is divided into a second training set and a second test set; in step S202, the first training set and the second training set are aggregated into a target training set; in step S203, the target training set is input into a preset model for training, and then the second test set is input into the preset model for testing, and the network parameters of the preset model are adjusted until the preset conditions are met to obtain a wind turbine blade bolt hole recognition model; the preset condition is that the number of times the test loss value remains continuously without decreasing reaches a preset number threshold, and the test loss value is obtained based on the second test set and the loss function.

[0069] In the specific implementation, it is assumed that the first data set D auxiliary The above training generates a weak learner Learner auxiliary , the target training set is T∈{(X auxiliary ∪X target )×Y}, where Y is the true value set of all training data in the target training set. Divide all training data T in the target training set into two data sets:

[0070]

[0071]

[0072]

[0073] Among them, y (x) represents the true value of the sample data x, T auxiliary and T target The difference is that T target and the test data S are identically distributed, while T auxiliary The distribution of the test data S is different from that of the original data.

[0074] Therefore, a new weak learner Learner is trained on the target dataset T target , so that its loss on the test dataset S is minimized.

[0075] In addition, in some embodiments, after training a wind turbine blade bolt hole recognition model, the recognition model can be used to identify misaligned bolt holes so that they can be corrected and replaced. It is understood that during the operation of a large wind turbine generator set, the blades are impacted by strong gusts of wind and the alternating loads caused by the rotation of the impeller. The blades and the hub pitch bearing connection flange surface will experience relative slippage. Under the shear force of the flange surface, the blade bolts may be damaged or even break and fail. In this case, the damaged bolts need to be removed and replaced to correct the misaligned bolt holes. Therefore, the wind turbine blade bolt hole recognition model can be used to identify misaligned bolt holes, so that the bolts in some of the misaligned bolt holes of the misaligned blade can be removed.

[0076] It should be noted that the above description is limited to some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0077] Based on the same technical concept, corresponding to the wind turbine blade bolt hole recognition model training method of any of the above embodiments, the present application also provides a wind turbine blade bolt hole recognition model training device 300. Furthermore, the information push model includes multiple sub-models corresponding to multiple fields, each of which includes an encoder and a decoder.

[0078] like Figure 3 As shown, the training device 300 for the wind turbine blade bolt hole recognition model may include:

[0079] An acquisition module 301 is configured to acquire a data set including a plurality of wind turbine blade bolt hole sample images and corresponding label images, wherein the label images are used to indicate labels of the wind turbine blade bolt holes;

[0080] A division module 302 is used to divide the images and their label images in the data set to obtain a first data set and a second data set;

[0081] A processing module 303 is configured to process the first data set and the second data set so that data samples in the two data sets meet a preset similarity condition;

[0082] A first training module 304 is configured to train a preset model based on the processed first data set to obtain a first recognition model;

[0083] The loading module 305 is used to load the model parameters of the first recognition model into the preset model to obtain a second recognition model;

[0084] The second training module 306 is used to fix the network parameters of the first target layer structure of the second recognition model and train the second recognition model according to the second data set to determine the network parameters of the second target layer structure of the second recognition model to obtain a wind turbine blade bolt hole recognition model.

[0085] In some embodiments, the wind turbine blade bolt hole identification model includes a third target layer structure, and the third target layer structure is obtained according to the first target layer structure and the second target layer structure.

[0086] In some embodiments, the wind turbine blade bolt hole recognition model training device 300 further includes an input module ( Figure 3 (not shown), used to input the network parameters of the first target layer structure and the network parameters of the second target layer structure into a third target layer structure; based on the weight of the first target layer structure and the weight of the second target layer structure, the weight of the third target layer structure is obtained by training according to the second data set.

[0087] In some embodiments, the collection time intervals of the first data set and the second data set are adjacent time intervals.

[0088] In some embodiments, the wind turbine blade bolt hole recognition model training device 300 further includes a first adjustment module ( Figure 3 ), which is used to resize the images and their label images in the dataset to obtain images of a first size; randomly crop the images of the first size to obtain images of a second size; convert the images of the second size into tensor data; and normalize the tensor data to obtain normalized tensor data.

[0089] In some embodiments, the processing module 303 is specifically used to determine the similarity between each data sample in the first data set and the second data set based on the first data set and the second data set; determine the initialization weight of each data sample in the first data set based on the similarity; and initialize each data sample in the first data set according to the initialization weight to obtain a processed first data set.

[0090] In some embodiments, the wind turbine blade bolt hole recognition model training device 300 further includes a second adjustment module ( Figure 3 (not shown in the figure), which is used to divide the first data set into a first training set and a first test set according to a preset division ratio, and divide the second data set into a second training set and a second test set; summarize the first training set and the second training set into a target training set; input the target training set into a preset model for training, and then input the second test set into the preset model for testing, and adjust the network parameters of the preset model until the preset conditions are met to obtain a wind turbine blade bolt hole recognition model; the preset condition is that the number of times the test loss value remains continuously without decreasing reaches a preset number threshold, and the test loss value is obtained based on the second test set and the loss function.

[0091] It should be noted that, for the convenience of description, the above devices are described as being divided into various modules according to their functions. Of course, when implementing this application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0092] The device of the above embodiment is used to implement the training method of the wind turbine blade bolt hole recognition model corresponding to any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0093] Based on the same technical concept, corresponding to the training method of the wind turbine blade bolt hole recognition model in any of the above embodiments, the present application also provides an electronic device.

[0094] Figure 4 A more specific hardware structure diagram of an electronic device provided by this embodiment is shown.

[0095] The electronic device 400 may include a processor 401 and a memory 402 storing computer program instructions.

[0096] Specifically, the processor 401 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0097] Memory 402 may include a large capacity memory for data or instructions. By way of example and not limitation, memory 402 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 402 may include removable or non-removable (or fixed) media. Where appropriate, memory 402 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, memory 402 is a non-volatile solid-state memory.

[0098] In certain embodiments, the memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to an aspect of the present application.

[0099] The processor 401 reads and executes computer program instructions stored in the memory 402 to implement any one of the wind turbine blade bolt hole recognition model training methods in the above embodiments.

[0100] In some examples, the electronic device 400 may further include a communication interface 403 and a bus 410. Figure 4 As shown, the processor 401 , the memory 402 , and the communication interface 403 are connected via a bus 410 and communicate with each other.

[0101] The communication interface 403 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0102] Bus 410 includes hardware, software or both, and the components of online data flow metering equipment are coupled to each other. For example, but not limitation, bus 410 may include accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnect (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 410 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0103] Illustratively, the electronic device 400 may be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA).

[0104] Based on the same technical concept, corresponding to any of the above-mentioned embodiments and methods, the present application also provides a non-transitory computer-readable storage medium. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any of the wind turbine blade bolt hole recognition model training methods in the above-mentioned embodiments is implemented. Examples of computer-readable storage media include non-transitory computer-readable storage media, such as portable disks, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), portable compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, etc.

[0105] Based on the same technical concept, corresponding to any of the aforementioned embodiments of the wind turbine blade bolt hole identification model training method, this application also provides a computer program product comprising computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer to cause the computer and / or the processors to perform the aforementioned wind turbine blade bolt hole identification model training method. The processors executing the corresponding steps in each embodiment of the wind turbine blade bolt hole identification model training method can be the corresponding executing entities.

[0106] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0107] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium that can store or transmit information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0108] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0109] Aspects of the present application have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed via the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. This processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It is also understood that each box in the block diagram and / or the flowchart and the combination of the boxes in the block diagram and / or the flowchart can also be implemented by the dedicated hardware that performs the specified function or action, or can be implemented by the combination of dedicated hardware and computer instructions.

[0110] The above description is only a specific embodiment of the present application. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the scope of protection of the present application.

Claims

1. A training method for a wind turbine blade bolt hole recognition model, characterized in that: include: Acquire a data set including a plurality of wind turbine blade bolt hole sample images and corresponding label images, wherein the label images are used to indicate labels of the wind turbine blade bolt holes; Dividing the images and their label images in the data set to obtain a first data set and a second data set; Processing the first data set and the second data set so that data samples in the two data sets meet a preset similarity condition; Training a preset model according to the processed first data set to obtain a first recognition model; Loading the model parameters of the first recognition model into the preset model to obtain a second recognition model; Fixing the network parameters of the first target layer structure of the second recognition model, and training the second recognition model according to the second data set to determine the network parameters of the second target layer structure of the second recognition model, and obtaining a wind turbine blade bolt hole recognition model; The wind turbine blade bolt hole sample images are acquired according to different acquisition time intervals; The collection time intervals of the first data set and the second data set are adjacent time intervals; The processing of the first data set and the second data set so that the data samples in the two data sets meet a preset similarity condition includes: Determining, based on the first data set and the second data set, a similarity between each data sample in the first data set and the second data set; Determining an initialization weight for each data sample in the first data set according to the similarity; Each data sample in the first data set is initialized according to the initialization weight to obtain a processed first data set.

2. The method according to claim 1, characterized in that The wind turbine blade bolt hole identification model includes a third target layer structure, and the third target layer structure is obtained according to the first target layer structure and the second target layer structure.

3. The method according to claim 2, characterized in that After fixing the network parameters of the first target layer structure of the second recognition model and training the second model according to the second data set to determine the network parameters of the second target layer structure of the second recognition model, the method further includes: inputting the network parameters of the first target layer structure and the network parameters of the second target layer structure into a third target layer structure; Based on the weight of the first target layer structure and the weight of the second target layer structure, the weight of the third target layer structure is obtained by training according to the second data set.

4. The method according to claim 1, wherein Before dividing the images and their label images in the data set to obtain the first data set and the second data set, the method further includes: Resize the images in the dataset and their label images to obtain images of a first size; Randomly cropping the image of the first size to obtain an image of a second size; Converting the image of the second size into tensor data; Normalization is performed on the tensor data to obtain normalized tensor data.

5. The method according to claim 1, wherein After dividing the images and their label images in the data set to obtain the first data set and the second data set, the method further includes: Dividing the first data set into a first training set and a first test set according to a preset division ratio, and dividing the second data set into a second training set and a second test set; Aggregating the first training set and the second training set into a target training set; Inputting the target training set into a preset model for training, then inputting the second test set into the preset model for testing, and adjusting the network parameters of the preset model until preset conditions are met, thereby obtaining a wind turbine blade bolt hole recognition model; The preset condition is that the number of times the test loss value remains continuously without decreasing reaches a preset number threshold, and the test loss value is obtained based on the second test set and the loss function.

6. A training device for a wind turbine blade bolt hole recognition model, characterized in that: The device comprises: An acquisition module, configured to acquire a data set including a plurality of sample images of bolt holes of wind turbine blades and corresponding label images, wherein the label images are used to indicate labels of the bolt holes of the wind turbine blades; A division module, configured to divide the images and their label images in the data set into a first data set and a second data set; A processing module, configured to process the first data set and the second data set so that data samples in the two data sets meet a preset similarity condition; A first training module is used to train a preset model according to the processed first data set to obtain a first recognition model; A loading module, configured to load the model parameters of the first recognition model into the preset model to obtain a second recognition model; A second training module is configured to fix the network parameters of the first target layer structure of the second recognition model, and train the second recognition model according to the second data set to determine the network parameters of the second target layer structure of the second recognition model, thereby obtaining a wind turbine blade bolt hole recognition model; The wind turbine blade bolt hole sample images are acquired according to different acquisition time intervals; The collection time intervals of the first data set and the second data set are adjacent time intervals; The processing module is specifically used for: Determining, based on the first data set and the second data set, a similarity between each data sample in the first data set and the second data set; An initialization weight of each data sample in the first data set is determined according to the similarity.

7. An electronic device, characterized in that: The device includes: a processor and a memory storing computer program instructions; when the processor calls the computer program instructions, it implements the training method of the wind turbine blade bolt hole recognition model according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when called by a processor, implement the training method for a wind turbine blade bolt hole recognition model according to any one of claims 1 to 5.

Images