A defogging method for icing wind tunnel test images

Abstract

This invention relates to a method for dehazing icy wind tunnel images, comprising the following steps: 1) acquiring foggy test images through a monitoring camera on the icy wind tunnel test section; 2) estimating atmospheric light intensity A using a dark channel; 3) obtaining the scene depth map d corresponding to the image captured by the camera at the current moment; 4) establishing a linear regression model between β and LWC, and fitting it using a dataset to obtain a prediction model M; 5) using the regression model M to predict the β value under the current test conditions; 6) solving for the dehazed image J using an atmospheric scattering model. This invention proposes a method to estimate the parameter β in the atmospheric scattering model using wind tunnel test parameters. Compared with other existing technologies using random or fixed values, this method can obtain a more accurate estimate of the scattering rate β that reflects the actual cloud and fog field conditions, thus achieving a better dehazing effect.

Description

Technical fields:

[0001] This invention belongs to the field of icing wind tunnel test observation technology, specifically a defogging method that can be used for icing wind tunnel test observation images. Background technology:

[0002] In aircraft icing and de-icing research, icing wind tunnel testing is the most important ground simulation test method. Icing wind tunnel testing provides a large amount of data support for the study of aircraft icing mechanism, the design of de-icing devices, and the airworthiness certification of icing.

[0003] During the experiment, the presence of clouds and fog significantly interferes with observations of the icing wind tunnel test section. Without clouds and fog, light from the light source illuminates the test component surface and is reflected to the imaging sensor of the observation camera, resulting in a clear image. However, with clouds and fog, numerous tiny suspended droplets scatter, attenuate, and absorb light, causing the originally clear image to become very blurry. The contrast and saturation of the image are severely reduced, and the overall image appears grayish-white, diminishing image detail. This seriously affects the observation of icing conditions during the experiment, as well as the determination of icing type, icing onset time, and detachment within the test section. It hinders the acquisition of experimental information and reduces experimental efficiency and effectiveness. Therefore, defogging the observation images from the icing wind tunnel experiment can assist researchers in obtaining information about the test section, improving experimental efficiency and effectiveness.

[0004] In past research on single-image dehazing, image dehazing algorithms can be broadly categorized into two types based on whether or not a physical model is used. The first type involves image restoration by solving a physical model of atmospheric scattering, while the second type relies on image enhancement techniques. Physical model-based restoration methods establish a physical imaging degradation model by analyzing the causes of image degradation during imaging under foggy conditions. These algorithms require solving equations between a theoretically clear image J and a known foggy, blurred image I to obtain the clear image J. Before solving, unknown parameters related to the environment in the equations need to be estimated, such as atmospheric illuminance and transmittance (depth, scattering coefficient). The quality of these location parameter estimations directly affects the dehazing effect. A commonly used estimation method in the past was based on dark channel priors. This method generally yields reasonable estimates and ideal dehazing results. However, for images containing many bright and highlighted areas, the dark channel prior fails. In practical applications, it has been found that this method leads to color distortion in the dark channel of wind tunnel fog images. Another type of image enhancement-based dehazing aims to improve image contrast through contrast enhancement, such as image enhancement methods based on histogram enhancement and homomorphic filtering algorithms. This type of method can achieve some effect when the fog is light, but it will fail in dense or uneven fog, resulting in significant distortion or loss of detail.

[0005] In icing wind tunnel experiments, many experimental conditions produce very dense water mist. Current defogging methods cannot achieve ideal defogging results for dense fog, which can significantly affect the observations of the experimenters. Summary of the Invention:

[0006] The purpose of this invention is to address the shortcomings of the prior art by proposing a defogging method for icing wind tunnel images based on depth and scattering coefficient estimation. This method fully utilizes the depth information of the scene and the experimental parameters of supercooled water content in the icing wind tunnel for defogging, thereby obtaining a more accurate transmission image estimation and achieving better defogging effect under dense fog.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] 1) Acquire foggy test images I using the monitoring camera in the icing wind tunnel test section;

[0009] 2) Estimating atmospheric light intensity A using the dark channel;

[0010] 3) Obtain the scene depth map d corresponding to the image captured by the camera at the current moment;

[0011] 4) Establish a linear regression model between β and LWC, and fit it using the dataset to obtain the prediction model M;

[0012] 5) Use regression model M to predict the β value under the current experimental conditions;

[0013] 6) Solve for the dehazed image J using an atmospheric scattering model;

[0014] In this invention, step 2) includes the following sub-steps:

[0015] 21) Find the minimum value in the RGB channels corresponding to each pixel position in the experimental image I to form a grayscale image I. gray ;

[0016] 22) to I gray Minimum filtering is performed to obtain the dark channel image I. dark ;

[0017] 23) From Dark Channel Diagram I dark The top 0.1% of pixels are selected based on their brightness.

[0018] 24) At these locations, find the value of the point with the highest brightness in the original image as an estimate of A.

[0019] Furthermore, the scene depth map d in step 3) is obtained as follows:

[0020] Input the experimental image I into the depth estimation neural network model, and output the corresponding depth map d.

[0021] Alternatively, step 3) can also obtain the scene depth map d in the following way:

[0022] Replace the camera in step 1) with a depth camera to acquire the experimental image I and its corresponding depth map d.

[0023] Furthermore, step 4) includes the following sub-steps:

[0024] 41) Establish the following linear regression model:

[0025] β=θ0+θ1×LWC

[0026] Where β is the scattering coefficient of the icy wind tunnel cloud field, θ i is the regression coefficient, and LWC is the supercooled water content of the cloud field.

[0027] 42) Within the commonly used LWC test parameter range, 20 different values ​​were selected for testing. For each test, the LWC of the wind tunnel test and one frame image before and after the start of the spray were recorded. The fog-laden image after the start of the spray was recorded as I. i The image without fog before spraying is denoted as J. i At the same time, the atmospheric light intensity A is estimated using the method in step 2). iAnd estimate the scene depth map d using the method in step 3). i For each experiment, the following formula can be derived from the atmospheric scattering model to calculate the scattering coefficient β for that experiment. i :

[0028]

[0029] The LWC values ​​for each experiment and the calculated β i Each data point is recorded as a data point, and the set of all data points obtained from the 20 trials constitutes the final dataset D;

[0030] 43) Apply the least squares method to the regression model from 41) on dataset D to obtain the regression coefficients. and The final regression model M is obtained as follows:

[0031]

[0032] Furthermore, the calculation formula used in step 6) is:

[0033]

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] This invention proposes a method to estimate the parameter β in an atmospheric scattering model using wind tunnel test parameters. Compared with other existing technologies that use random or fixed values, this method can obtain a more accurate estimate of the scattering rate β that better reflects the actual cloud and fog field conditions, thereby achieving a better defogging effect. Attached image description:

[0036] Figure 1 This is a flowchart of the defogging method for icing wind tunnel test images proposed in this invention;

[0037] Figure 2 These are images obtained in the embodiments;

[0038] Figure 3 This is the scene depth map corresponding to the image captured by the camera in the embodiment;

[0039] Figure 4 This is the image after dehazing in the example. Detailed implementation method:

[0040] The invention will now be further described with reference to the accompanying drawings.

[0041] Reference Figure 1 The specific implementation steps of this invention are as follows:

[0042] 1) Acquire foggy test images I using the monitoring camera in the icing wind tunnel test section;

[0043] In this embodiment, the monitoring camera is fixed near the center of the outer wall of the icing wind tunnel test section. Figure 2 Image I was acquired.

[0044] 2) Estimate atmospheric light intensity A using the dark channel; including the following sub-steps:

[0045] 21) The test image is I. Store the minimum value of the RGB channel corresponding to each pixel in I into a grayscale image of the same size as I. gray middle;

[0046] 22) to I gray Minimum filtering with a filter radius of 7 is performed to obtain the dark channel image I. dark ;

[0047] 23) From Dark Channel Diagram I dark The top 0.1% of pixels are selected based on their brightness.

[0048] 24) At these locations, find the value of the point with the highest brightness in the original image as an estimate of A.

[0049] 3) Obtain the scene depth map d corresponding to the image captured by the camera at the current moment;

[0050] In this embodiment, the depth estimation network model BinsFormer is used to estimate the depth of I, resulting in a depth map d, as shown below. Figure 3 As shown.

[0051] 4) Establish a linear regression model between β and LWC, and fit it using the dataset to obtain the prediction model M; including the following sub-steps:

[0052] 41) Establish the following linear regression model:

[0053] β=θ0+θ1×LWC

[0054] Where β is the scattering coefficient of the icy wind tunnel cloud field, θ i is the regression coefficient, and LWC is the supercooled water content of the cloud field.

[0055] 42) Within the commonly used LWC test parameter range, 20 different values ​​were selected for testing. For each test, the LWC of the wind tunnel test and one frame image before and after the start of the spray were recorded. The fog-laden image after the start of the spray was recorded as I. i The image without fog before spraying is denoted as J. i At the same time, the atmospheric light intensity A is estimated using the method in step 2). iAnd estimate the scene depth map d using the method in step 3). i For each experiment, the following formula can be derived from the atmospheric scattering model to calculate the scattering coefficient β for that experiment. i :

[0056]

[0057] The LWC values ​​for each experiment and the calculated β i Each data point is recorded as a data point, and the set of all data points obtained from the 20 trials constitutes the final dataset D;

[0058] 43) Apply the least squares method to the regression model from 41) on dataset D to obtain the regression coefficients. and The final regression model M is obtained as follows:

[0059]

[0060] 5) Use regression model M to predict the β value under the current experimental conditions;

[0061] In this embodiment, the LWC content in the experimental conditions was 2 g / m³. 3 From model M, we can obtain

[0062] 6) Solve for the dehazed image J using an atmospheric scattering model; the calculation formula used is as follows:

[0063]

[0064] The resulting J is the dehazed image, such as... Figure 4 As shown.

Claims

1. A method for dehazing images from an icing wind tunnel test, characterized in that, Includes the following steps: S1: Acquire foggy test images I through the monitoring camera in the icing wind tunnel test section; S2: Estimating atmospheric light intensity A using the dark channel; S3: Obtain the scene depth map d corresponding to the image captured by the camera at the current moment; S4: Establish a linear regression model between β and LWC, and fit it using the dataset to obtain the prediction model M; 41) Establish the following linear regression model: β=θ0+θ1×LWC Where β is the scattering coefficient of the icy wind tunnel cloud field, θ i LWC is the supercooled water content of the cloud field, representing the regression coefficient. 42) Within the commonly used LWC test parameter range, 20 different values ​​were selected for testing. For each test, the LWC of the wind tunnel test and one frame image before and after the start of the spray were recorded. The fog-laden image after the start of the spray was recorded as I. i The image without fog before spraying is denoted as J. i At the same time, the atmospheric light intensity A is estimated using the method in step 2). i And estimate the scene depth map d using the method in step 3). i For each experiment, the following formula can be derived from the atmospheric scattering model to calculate the scattering coefficient β for that experiment. i : The LWC values ​​for each experiment and the calculated β i Each data point is recorded as a data point, and the set of all data points obtained from the 20 trials constitutes the final dataset D; 43) Apply the least squares method to the regression model from 41) on dataset D to obtain the regression coefficients. and The final regression model M is obtained as follows: S5: Use regression model M to predict the β value under the current experimental conditions; S6: Solve for the dehazed image J using an atmospheric scattering model.

2. A method for defogging icing wind tunnel test images as described in claim 1, characterized in that, Step S2 includes the following sub-steps: 21) Find the minimum value in the RGB channels corresponding to each pixel position in the experimental image I to form a grayscale image I. gray ; 22) to I gray Minimum filtering is performed to obtain the dark channel image I. dark ; 23) From Dark Channel Diagram I dark The top 0.1% of pixels are selected based on their brightness. 24) At these locations, find the value of the point with the highest brightness in the original image as an estimate of A.

3. A method for defogging icing wind tunnel test images as described in claim 1, characterized in that, The method for obtaining the scene depth map d in step S3 is as follows: Input the experimental image I into the depth estimation neural network model, and output the corresponding depth map d.

4. A method for dehazing icing wind tunnel test images as described in claim 1, characterized in that, Step S3 obtains the scene depth map d in the following manner: Replace the camera in step 1) with a depth camera to acquire the experimental image I and its corresponding depth map d.

5. A method for dehazing icing wind tunnel test images as described in claim 1, characterized in that, The calculation formula used in step S6 is:

Images