Method, device, equipment and storage medium for detecting icing area of airfoil in ice wind tunnel

Abstract

The present invention relates to the field of icing detection technology, and specifically to a method, device, equipment, and storage medium for detecting icing areas on airfoils in ice wind tunnels. The method first photographs an experimental airfoil in an ice wind tunnel test section, obtains the original image, and processes it to obtain training and test image samples in an icing detection model training dataset; uses the training dataset to train a semantic segmentation network for icing area detection; then starts the ice wind tunnel, obtains icing images of the airfoil in the test section in real time, and inputs the processed images into a trained icing area detection model, outputting the position coordinates of several icing areas in the target image; finally, determines the actual icing area, obtains the icing area on the airfoil, and calculates the sum of the areas of each icing area on the airfoil. Combined with the computer's built-in timer, the icing speed of the airfoil under experimental conditions can also be calculated. This patent uses machine vision to determine the icing area of the airfoil in real time, and can accurately determine the icing area and icing speed of the airfoil during the experiment in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of icing detection, and in particular to a method, device, equipment and storage medium for detecting icing areas on airfoils in ice wind tunnels. Background Art

[0002] Since the early days of aviation, icing has been one of the greatest threats to safe flight, accounting for over half of all military and civilian aircraft accidents worldwide. Icing on various parts of an aircraft can affect safe flight. For example, icing on the wings, tail, and control surfaces can dramatically deteriorate an aircraft's aerodynamic performance and maneuverability. During the aircraft design phase, ice wind tunnel testing can reveal the ice areas and types on various aircraft components, allowing the aircraft's anti-icing zones to be determined and providing a basis for anti-icing (de-icing) design.

[0003] Due to the harsh experimental environment in ice wind tunnel experiments, researchers typically only analyze the iced airfoil in the test section after the experiment, recording the ice type and mass. However, they lack the means to calculate the ice area and speed during the experiment. Therefore, an image-based region detection algorithm can be introduced, using deep learning methods to obtain the ice area on the airfoil surface in real time and calculate the ice speed during the ice wind tunnel experiment.

[0004] Due to the high cost of ice wind tunnel experiments, fewer original images are obtained. In traditional deep learning models, underfitting may occur when there are few samples, that is, the accuracy of model inference is low and the ideal effect cannot be achieved. The UNet network proposed by Ciresan et al. uses patches as input training data, which can effectively solve the problem of fewer images. Based on the problem that the UNet network training model is slow, the present invention modifies the UNet network, optimizes the UNet training model, and improves the training speed. At the same time, it can still maintain a high accuracy rate for the detection of ice areas, providing great help to scientific researchers in analyzing the results of ice wind tunnel experiments. Summary of the Invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a method, device, equipment and storage medium for detecting icing areas on airfoils in an ice wind tunnel.

[0006] In a first aspect, the present invention provides a method for detecting icing areas on airfoils in an ice wind tunnel. The technical solution includes:

[0007] Step 1: Place a calibration plate in the ice wind tunnel test section to ensure that the calibration plate is clearly imaged in the camera. By adjusting the position of the calibration plate, record multiple images of the calibration plate at different orientations and angles. Calculate the camera's intrinsic parameter matrix, the translation and rotation matrices between each camera, and the distortion coefficient of each camera.

[0008] Step 2: Use the arranged cameras to shoot the experimental airfoil to obtain the original image. Based on the image preprocessing algorithm, the image is subjected to distortion correction, dehazing, and histogram equalization. Then, the annotation tool is used to mark the outline of the ice area in the ice image of the airfoil surface and mark the area with the corresponding label to obtain the training and test image samples in the ice detection model training dataset;

[0009] Step 3: Using the training dataset, training a semantic segmentation network for ice area detection;

[0010] Step 4: Start the ice wind tunnel and obtain the icing image of the airfoil in the test section in real time. After distortion correction, defogging and histogram equalization processing, input it into the trained icing area detection model and output the position coordinates (x i ,y i ), where x i 、y i is the coordinate in the pixel coordinate system;

[0011] Step 5: Determine the actual icing area. Using the calibrated camera parameters, convert the output pixel coordinates of the icing area into actual coordinates in the world coordinate system and output them to obtain the icing area on the airfoil. Calculate the sum of the areas of each icing area on the airfoil. Combined with the computer's built-in timer, the icing speed of the airfoil under experimental conditions can also be calculated.

[0012] Furthermore, the specific implementation steps of step three are:

[0013] Step 3.1: Use a structure composed of convolution and pooling to complete feature extraction of the preprocessed airfoil surface image. Repeat this process multiple times to form the encoding part of the network model. The number of downsampled feature channels is doubled each time.

[0014] In step 3.2, in the decoding part, the obtained feature map is deconvolved, the number of feature channels is halved, and the size of the feature map is doubled. The deconvolution result is spliced with the result of the corresponding step in the encoding part, and the spliced feature map is convolved twice again. This is repeated multiple times, and finally the feature map is converted into the result of a specific number of categories;

[0015] Step 3.3: Using a multi-supervision approach, input all the outputs of each part of the decoding stage into the loss function, perform backpropagation calculations, and finally generate an ice area detection model. The loss function requirement is:

[0016]

[0017] Where A is the prior mask and B is the predicted mask.

[0018] Furthermore, in step 3.1, principal component analysis is used to determine the initial value of the convolution kernel in each convolution layer. The specific steps are:

[0019] Step 3.1.1. For the k×k convolution kernel required for the encoding part, extract each input sample using a k×k patch and pack it into a row to form a data sample, and then calculate its covariance matrix;

[0020] Step 3.1.2: For each N convolutional layer in the encoding part, solve the first N eigenvectors of the covariance matrix. The eigenvector is 1×kk. The values in the N eigenvectors correspond to the values of the M convolution kernels in row-first order.

[0021] Furthermore, the specific implementation steps of step 5 are:

[0022] Step 5.1: Obtain the scale factor k between the world coordinate system and the pixel coordinate system by using the pre-calibrated unit length on the airfoil. x 、k y :

[0023]

[0024]

[0025] Among them, w x 、w x is the coordinate in the world coordinate system, x and y are the coordinates in the pixel coordinate system;

[0026] Step 5.2: Due to the high wind speed in the ice wind tunnel, newly formed ice in certain areas of the airfoil is easily blown away by the wind, and these areas have no significant impact on the airfoil. Therefore, based on the video display and a series of parameters from the ice wind tunnel experiment, a certain time threshold is set. Only when the threshold is met and ice still exists on the airfoil is it considered to be actual ice.

[0027] Step 5.3: The coordinates of the ice area obtained in step 4 (x i ,y i ) is converted into coordinates in the world coordinate system:

[0028] (w x ,w x )=(k x x i ,k y y i )

[0029] Step 5.4: Find the outermost coordinates of the ice area by binarizing the image, and use cubic spline interpolation to draw the outline of the ice area;

[0030] Step 5.5: Calculate the sum of the areas of all iced areas on the airfoil, S:

[0031]

[0032] Where n is the number of icing areas on the airfoil, S i is the ice area of a region;

[0033] Step 5.6: Since the present invention judges the icing area based on a two-dimensional plane, the icing area that increases per unit time is used as the icing speed v of the airfoil:

[0034]

[0035] Where ΔS is the growing area and t is the time.

[0036] In a second aspect, the present invention provides a device for detecting icing areas on airfoils in an ice wind tunnel, the device comprising:

[0037] The calibration module is used to calibrate the cameras installed in the test section to obtain the camera intrinsic parameter matrix, the translation and rotation matrices between each camera, and the distortion coefficient of each camera;

[0038] An acquisition module is used to obtain the original image of airfoil icing in the ice wind tunnel;

[0039] A preprocessing module is used to perform data processing on the original image, use the distortion coefficient of the camera to perform distortion correction on the original image, and perform defogging and histogram equalization processing on the original image to obtain a data set after data preprocessing;

[0040] a training module, configured to construct an icing area detection model, using the data set as input to construct the icing area detection model, optimizing model parameters, and obtaining an inference model for icing area detection;

[0041] The detection module is used to obtain images during the airfoil experiment in real time, input them into the icing area detection model, and obtain the coordinates of the icing area.

[0042] In a third aspect, the present invention provides an apparatus for detecting icing areas on airfoils in an ice wind tunnel, the apparatus comprising:

[0043] One or more cameras and light sources are arranged in the ice wind tunnel test section to collect airfoil test conditions;

[0044] a storage device for storing one or more programs and a data set required by an icing area detection model;

[0045] One or more processors, when the one or more programs are executed by the processors, can implement the method for detecting icing areas on airfoils in ice wind tunnels as provided in the first aspect of the present invention.

[0046] In a fourth aspect, the present invention provides a storage medium for detecting icing areas on airfoils in an ice wind tunnel, wherein the storage medium is a computer-readable medium storing one or more programs, which, when executed by a processor, can implement the method for detecting icing areas on airfoils in an ice wind tunnel provided in the first aspect of the present invention.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) The airfoil icing area detection in ice wind tunnel experiments was realized, and the icing area and icing speed of the airfoil during the experiment were determined in real time;

[0049] (2) Improve the existing ice area detection model, which can accelerate the model training process and reduce the training time while retaining the high accuracy of small sample training;

[0050] (3) A new method for determining the icing area of an airfoil in an ice wind tunnel experiment is proposed, which can determine the icing area of an airfoil in real time through machine vision. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0052] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0053] Figure 2 This is a schematic diagram of the process of ice area detection according to the present invention;

[0054] Figure 3 This is a diagram showing the effect of determining the icing area in the ice wind tunnel. DETAILED DESCRIPTION

[0055] The present invention will be further described below with reference to the accompanying drawings:

[0056] In a first aspect, the present invention provides a method for detecting icing areas on airfoils in an ice wind tunnel. The technical solution includes:

[0057] Step 1: Place a calibration plate in the ice wind tunnel test section to ensure a clear image of the calibration plate in the camera. By adjusting the position of the calibration plate, record multiple images of the calibration plate at different orientations and angles. Calculate the camera's intrinsic parameter matrix, the translation and rotation matrices between each camera, and the distortion coefficient of each camera.

[0058] The requirement for recording multiple images of the calibration plate at different positions and angles is that the calibration images taken by the camera at different positions, angles, and postures are at least 3, preferably about 20.

[0059] The camera's intrinsic parameter matrix and the translation and rotation matrices seen by each camera are obtained by calculation. The requirements are: for each calibration image, convert it into a grayscale image, extract the corner information, and perform sub-pixel corner optimization. Use the Opencv open source program to calculate the internal and external parameters and distortion coefficients of the camera that has been set up.

[0060] In a specific embodiment, 20 photos of the calibration plate in different poses are taken using a camera that has been set up. All calibration photos are imported into the Opencv open source program to calculate and export the camera's internal and external parameters and distortion coefficients.

[0061] Step 2: Use the arranged camera to shoot the experimental airfoil to obtain the original image. Based on the image preprocessing algorithm, the image is subjected to distortion correction, dehazing and histogram equalization. Then, the annotation tool is used to mark the outline of the ice area in the ice image of the airfoil surface and mark the area with the corresponding label to obtain the training and test image samples in the ice detection model training dataset.

[0062] The experimental airfoil is photographed using arranged cameras to obtain the original image. The requirements are as follows: the number and placement of the cameras are determined according to the size, structure and position of the observation window of the ice wind tunnel test section, the focal length of the camera is determined by measuring the distance between the camera and the observed airfoil, and then a suitable camera is selected.

[0063] The camera should be a color camera with a resolution of at least 572×572×3. The lens should be equipped with a thermal insulation device to prevent the lens from fogging during the ice wind tunnel experiment and causing unclear imaging.

[0064] Use the installed camera to shoot the experimental airfoil, obtain the original image, convert the camera image into a grayscale image, and determine the optimal installation position and number of light sources in the test section by comparing the distribution of grayscale values between the ice area and other areas of the airfoil. Select the appropriate light source installation method from direct lighting, coaxial lighting, dark field lighting, diffuse lighting, and backlighting.

[0065] The selection of a suitable camera requires that: the distortion correction of the acquired airfoil icing image requires that: the correction formulas for radial distortion and tangential distortion are:

[0066] x corrected_径向 =x(1+k1r 2 +k2r 4 +k3r 6 )

[0067] y corrected_径向 =y(1+k1r 2 +k2r 4 +k3r 6 )

[0068] xcorrected_切向 =x+(2p1y+p2(r 2 +2x 2 )

[0069] y corrected_切向 =y+(2p2x+p1(r 2 +2y 2 )

[0070] Among them, x and y are the coordinates of the position after image distortion, and r is the distance between the point and the imaging center.

[0071] The histogram equalization processing of the image requires: widening the grayscale values with a large number of pixel formats in the image (i.e., the pixel values that play a major role in the picture), and merging the grayscale values with a small number of pixels (i.e., the grayscale values that do not play a major role in the picture), thereby increasing the contrast, making the image clear, and achieving the purpose of enhancement.

[0072] The method uses a labeling tool to mark the outline of the ice area in the ice image of the airfoil surface and puts a corresponding label on the area. The requirement is: semantic segmentation is essentially pixel-level classification, which predicts which category the target pixel point belongs to. Therefore, segmentation labeling needs to mark which pixels belong to which category. Open source labeling tools can be used to assist in marking the outline of the object.

[0073] In a specific embodiment, a full-color camera with a focal length of 10mm and a resolution of 1920×1080×3 was selected for testing. It was placed outside the ice wind tunnel test section, perpendicular to the side of the experimental airfoil. The camera captured images through glass to prevent lens fogging due to low temperatures and ensure clear images. A total of 16 LEDs were selected as the illumination source, using a combination of backlighting and direct lighting, with eight LEDs for each lighting method.

[0074] Once the number and position of cameras and the installation position of the light source are determined, the image of the test section airfoil is obtained through the camera. After conversion to a grayscale image, it can be found that the iced area and non-iced area on the airfoil surface can be well distinguished, so that the subsequent steps can be carried out smoothly. The camera is used to take about 1,000 pictures of the airfoil icing conditions in the experiment, which are used as the original images and the camera distortion coefficient is imported for distortion correction.

[0075] For the original image that has been corrected for distortion and dehazed, it is imported into the Opencv open source program for histogram equalization processing, and the histogram of the original image is transformed into a uniform distribution. It is then imported into LabelMe to perform pixel-level annotation of the ice area contours in the ice dataset image. The ice label is added after the outline of the ice area is marked, and the remaining areas are marked with background labels to obtain the training set and verification set of the ice area detection model.

[0076] Step 3: Use the training dataset to train a semantic segmentation network for ice area detection.

[0077] The semantic segmentation network for training ice area detection includes:

[0078] Step 3.1: Use a structure composed of convolution and pooling to complete feature extraction of the preprocessed airfoil surface image. Repeat this process multiple times to form the encoding part of the network model. The number of downsampled feature channels is doubled each time.

[0079] In step 3.2, in the decoding part, the obtained feature map is deconvolved, the number of feature channels is halved, and the size of the feature map is doubled. The deconvolution result is spliced with the result of the corresponding step in the encoding part, and the spliced feature map is convolved twice again. This is repeated multiple times, and finally the feature map is converted into the result of a specific number of categories;

[0080] Step 3.3: Using a multi-supervision approach, input all the outputs of each part of the decoding stage into the loss function, perform backpropagation calculations, and finally generate an ice area detection model. The loss function requirement is:

[0081]

[0082] Where A is the prior mask and B is the predicted mask.

[0083] In step 3.1, the initial value of the convolution kernel in each convolution layer is determined by principal component analysis, including:

[0084] Step 3.1.1. For the k×k convolution kernel required for the encoding part, extract each input sample using a k×k patch and pack it into a row to form a data sample, and then calculate its covariance matrix;

[0085] Step 3.1.2: For each N convolutional layer in the encoding part, solve the first N eigenvectors of the covariance matrix. The eigenvector is 1×kk. The values in the N eigenvectors correspond to the values of the M convolution kernels in row-first order.

[0086] The result of the deconvolution in step 3.2 is spliced with the result of the corresponding step of the encoding part. The requirement is: if the feature map of the encoding part is large, it needs to be cropped and then spliced.

[0087] In a specific embodiment, the architecture of the encoding part is composed of four repeated structures: two 3×3 convolutional layers with a padding strategy of valid and stride=1, a nonlinear ReLU layer and a 2×2 maximum pooling layer with a stride of 2 and a padding strategy of valid; in the principal component analysis to extract the initial convolution value of the convolution layer, the selected patch size is 3×3, the first two eigenvectors of the mean square error matrix are taken, and the size of the eigenvector is 1×9; similar to the encoding layer, the deconvolution of the decoding part is also composed of four repeated structures, and deconvolution is used before each repeated structure, with a convolution kernel size of 2×2. After deconvolution, the feature map of the corresponding step of the encoding part is cropped and then spliced with the result of deconvolution. The spliced feature map is then subjected to two 3×3 convolutions, and the last layer is a convolution with a convolution kernel of 1×1. Sigmoid is used to convert the 64-channel feature map into an ice area and background; the output of each step of the decoding part is brought into the loss function, backpropagated and calculated, and finally an ice area detection model is generated.

[0088] Step 4: Start the ice wind tunnel and obtain the icing image of the airfoil in the test section in real time. After distortion correction, defogging and histogram equalization processing, the image is input into the trained icing area detection model, and the position coordinates of the icing area in the target image are output.

[0089] The real-time acquisition of the icing image of the airfoil in the test section requires: using a set-up camera to record a video of the icing of the airfoil in the test section during the experiment, and processing it frame by frame to obtain the icing image of the airfoil.

[0090] In a specific embodiment, the video of the airfoil experiment in the ice wind tunnel captured by the camera is subjected to distortion correction, defogging and histogram equalization processing frame by frame, and then imported into the UNet detection model trained in step 4 to obtain the pixel coordinates of the ice area.

[0091] Step 5: Determine the actual icing area. Using the calibrated camera parameters, convert the output pixel coordinates of the icing area into actual coordinates in the world coordinate system and output them to obtain the icing area on the airfoil. Calculate the sum of the areas of each icing area on the airfoil. Combined with the computer's built-in timer, the icing speed of the airfoil under experimental conditions can also be calculated.

[0092] The step of obtaining the icing area on the airfoil comprises:

[0093] Step 5.1: Obtain the scale factor k between the world coordinate system and the pixel coordinate system by using the pre-calibrated unit length on the airfoil. x 、k y :

[0094]

[0095]

[0096] Among them, w x 、w x is the coordinate in the world coordinate system, x and y are the coordinates in the pixel coordinate system;

[0097] Step 5.2: Due to the high wind speed in the ice wind tunnel, newly formed ice in certain areas of the airfoil is easily blown away by the wind, and these areas have no significant impact on the airfoil. Therefore, based on the video display and a series of parameters from the ice wind tunnel experiment, a certain time threshold is set. Only when the threshold is met and ice still exists on the airfoil is it considered to be actual ice.

[0098] Step 5.3: The coordinates of the ice area obtained in step 4 (x i ,y i ) is converted into coordinates in the world coordinate system:

[0099] (w x ,w x )=(k x x i ,k y y i )

[0100] Step 5.4: Find the outermost coordinates of the ice area by binarizing the image, and use cubic spline interpolation to draw the outline of the ice area;

[0101] The calculation above obtains the sum of the areas of the icing regions on the airfoil, which is required to be:

[0102]

[0103] Where S is the sum of the areas of all icing regions on the airfoil, n is the number of icing regions on the airfoil, S i The ice area of a region.

[0104] The calculation obtains the icing speed of the airfoil under experimental conditions, which is required to be: Since the present invention judges the icing area based on a two-dimensional plane, the icing speed v of the airfoil is calculated by taking the icing area increased per unit time:

[0105]

[0106] Where ΔS is the growing area and t is the time.

[0107] In a specific embodiment, Figure 3 As shown, the proportional coefficient k between the world coordinate system and the pixel coordinate system is obtained x 、k y :

[0108]

[0109]

[0110] Since the wind speed in this experiment is 50m / s, MVD is 20um, LWC is 1, and the experimental temperature is -15℃, it is assumed that the airfoil surface is actually frozen only after the ice crystals exist for 5s. The actual ice area on the airfoil surface is calculated to be 113.789mm 2 , combined with the ice area in the previous second video, the real-time ice speed on the airfoil surface is 15.264mm 2 / s.

[0111] Based on the camera parameters obtained in step 1, the proportional relationship between the world coordinate system and the pixel coordinate system is calculated. The pixel coordinates of the iced area in step 4 are then substituted into the actual coordinates of the airfoil ice. The area of the airfoil iced area and the ice formation speed can then be calculated.

[0112] In a second aspect, the present invention provides a device for detecting icing areas on airfoils in an ice wind tunnel, the device comprising:

[0113] The calibration module is used to calibrate the cameras installed in the test section to obtain the camera intrinsic parameter matrix, the translation and rotation matrices between each camera, and the distortion coefficient of each camera;

[0114] An acquisition module is used to obtain the original image of airfoil icing in the ice wind tunnel;

[0115] A preprocessing module is used to perform data processing on the original image, use the distortion coefficient of the camera to perform distortion correction on the original image, and perform defogging and histogram equalization processing on the original image to obtain a data set after data preprocessing;

[0116] a training module, configured to construct an icing area detection model, using the data set as input to construct the icing area detection model, optimizing model parameters, and obtaining an inference model for icing area detection;

[0117] The detection module is used to obtain images during the airfoil experiment in real time, input them into the icing area detection model, and obtain the coordinates of the icing area.

[0118] In this embodiment, the device for detecting icing areas on airfoils in ice wind tunnels can execute the method for detecting icing areas on airfoils in ice wind tunnels provided in any embodiment of the present invention, has all the functional modules for executing the method, and can achieve corresponding effects.

[0119] In a third aspect, the present invention provides an apparatus for detecting icing areas on airfoils in an ice wind tunnel, the apparatus comprising:

[0120] One or more cameras and light sources are arranged in the ice wind tunnel test section to collect airfoil test conditions;

[0121] a storage device for storing one or more programs and a data set required by an icing area detection model;

[0122] One or more processors, when the one or more programs are executed by the processors, can implement the method for detecting icing areas on airfoils in ice wind tunnels as provided in the first aspect of the present invention.

[0123] In this embodiment, a computer device is proposed, including a processor, a memory, an input device, and an output device, all of which are connected via a bus.

[0124] The memory is used to store program software, modules corresponding to the icing area detection method in this example, and data sets required by the model.

[0125] The processor needs to meet the memory and video memory requirements of the running program, and implements the above-mentioned ice area detection method by executing a series of program instructions stored in the memory.

[0126] The input device mainly includes the camera used in the experiment, the light source equipment for auxiliary imaging, and other devices such as mouse, keyboard, microphone, etc. for receiving user command input.

[0127] The output device is used to display the ice area results in the above example, including but not limited to a display, a printer and a microphone.

[0128] In a fourth aspect, the present invention provides a storage medium for detecting icing areas on airfoils in an ice wind tunnel, wherein the storage medium is a computer-readable medium storing one or more programs, which, when executed by a processor, can implement the method for detecting icing areas on airfoils in an ice wind tunnel provided in the first aspect of the present invention.

[0129] The term "storage device" refers to a computer-readable storage medium, including electrical, magnetic, optical, infrared, and other media. More specific examples include, but are not limited to, magnetic disks, hard disks, RAM, ROM, optical storage devices, magnetic storage devices, and any combination thereof. In some instances, the memory device may further include a memory device remote from the processor. Such remote memory devices may be connected to the processor device via a network, including, but not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0130] In this embodiment, a storage medium for detecting icing areas on airfoils in an ice wind tunnel is provided. The program of the method for detecting icing areas on airfoils in an ice wind tunnel according to the first aspect of the present invention stored in the storage medium can be executed by a processor.

[0131] It should be noted that, for the device, equipment and storage medium example solutions, since they are basically similar to the method example solutions, the description is relatively simple, and the relevant parts can be referred to the partial description of the method example solution.

[0132] The above examples are merely exemplary examples for illustrating the principles of the present disclosure, but the present disclosure is not limited thereto. Those skilled in the art may make various modifications and improvements without departing from the spirit and substance of the present disclosure, and such modifications and improvements are also considered to be within the scope of protection of the present disclosure.

Claims

1. A method for detecting icing areas on airfoils in an ice wind tunnel, characterized in that: The steps include: Step 1: Place a calibration plate in the ice wind tunnel test section to ensure that the calibration plate is clearly imaged in the camera. By adjusting the position of the calibration plate, record multiple images of the calibration plate at different orientations and angles. Calculate the camera's intrinsic parameter matrix, the translation and rotation matrices between each camera, and the distortion coefficient of each camera. Step 2: Use the arranged cameras to shoot the experimental airfoil to obtain the original image. Based on the image preprocessing algorithm, the image is subjected to distortion correction, dehazing, and histogram equalization. Then, the annotation tool is used to mark the outline of the ice area in the ice image of the airfoil surface and mark the area with the corresponding label to obtain the training and test image samples in the ice detection model training dataset; Step 3: Use the training dataset to train a semantic segmentation network for ice area detection: Step 3.1: Use a structure composed of convolution and pooling to complete feature extraction of the preprocessed airfoil surface image. Repeat this process multiple times to form the encoding part of the network model. The number of downsampled feature channels is doubled each time. In step 3.2, in the decoding part, the obtained feature map is deconvolved, the number of feature channels is halved, and the size of the feature map is doubled. The deconvolution result is spliced with the result of the corresponding step in the encoding part, and the spliced feature map is convolved twice again. This is repeated multiple times, and finally the feature map is converted into the result of a specific number of categories; Step 3.3: Using a multi-supervision approach, input all the outputs of each part of the decoding stage into the loss function, perform backpropagation calculations, and finally generate an ice area detection model. The loss function requirement is: Where A is the prior mask and B is the predicted mask; Step 4: Start the ice wind tunnel and obtain the icing image of the airfoil in the test section in real time. After distortion correction, defogging and histogram equalization processing, the image is input into the trained icing area detection model, and the position coordinates of several icing areas in the target image are output. ,in is the coordinate in the pixel coordinate system; Step 5: Determine the actual icing area. Using the calibrated camera parameters, convert the output pixel coordinates of the icing area into actual coordinates in the world coordinate system and output them to obtain the icing area on the airfoil. Calculate the sum of the areas of each icing area on the airfoil. Combined with the computer's built-in timer, the icing speed of the airfoil under experimental conditions can also be calculated.

2. A method for detecting icing areas on airfoils in an ice wind tunnel according to claim 1, characterized in that: In step 3.1, principal component analysis is used to determine the initial value of the convolution kernel in each convolution layer. The specific steps are: Step 3.1.1, the size of the encoding part is The convolution kernel of each input sample is The patches are extracted and grouped into a row to form a data sample, and then the covariance matrix is calculated; Step 3.1.2, for each N convolutional layer in the encoding part, solve the first N eigenvectors of the covariance matrix, the eigenvector is , and the values in the N eigenvectors correspond to the values of the M convolution kernels in row-first order.

3. A method for detecting icing areas on airfoils in an ice wind tunnel according to claim 1, characterized in that: The specific implementation steps of step five are: Step 5.1: Obtain the scale factor between the world coordinate system and the pixel coordinate system by using the pre-calibrated unit length on the airfoil. : Where n is the number of icing areas on the airfoil, is the ice area of a region; Step 5.6: Since the present invention judges the icing area based on a two-dimensional plane, the icing area that increases per unit time is used as the icing speed v of the airfoil: in, is the growing area, and t is the time.

4. A device for detecting icing areas on airfoils in an ice wind tunnel, characterized in that: The device is used to implement the method for detecting icing areas on airfoils in an ice wind tunnel according to any one of claims 1 to 3, and the device comprises: The calibration module is used to calibrate the cameras installed in the test section to obtain the camera intrinsic parameter matrix, the translation and rotation matrices between each camera, and the distortion coefficient of each camera; An acquisition module is used to obtain the original image of airfoil icing in the ice wind tunnel; A preprocessing module is used to perform data processing on the original image, use the distortion coefficient of the camera to perform distortion correction on the original image, and perform defogging and histogram equalization processing on the original image to obtain a data set after data preprocessing; a training module, configured to construct an icing area detection model, using the data set as input to construct the icing area detection model, optimizing model parameters, and obtaining an inference model for icing area detection; The detection module is used to obtain images during the airfoil experiment in real time, input them into the icing area detection model, and obtain the coordinates of the icing area.

5. An airfoil icing area detection device in an ice wind tunnel, characterized by: include: One or more cameras and light sources are arranged in the ice wind tunnel test section to collect airfoil test conditions; a storage device for storing one or more programs and a data set required by an icing area detection model; One or more processors, when the one or more programs are executed by the processors, implement the method for detecting icing areas on airfoils in an ice wind tunnel according to any one of claims 1 to 3.

6. A storage medium for detecting icing areas on airfoils in an ice wind tunnel, characterized by: The storage medium thereon is a computer-readable medium, storing one or more programs, which, when executed by a processor, can implement the method for detecting icing areas on airfoils in an ice wind tunnel according to any one of claims 1 to 3.

Images