CN117011531A - Method, device and equipment for extracting adhesion bubbles from dynamic ice image based on watershed segmentation - Google Patents

Abstract

The embodiment of the application discloses a method, a device and equipment for extracting adhesion bubbles from dynamic ice images based on watershed segmentation. The method comprises the following steps: performing distance transformation on the preprocessed binary image, and obtaining a distance image by the opposite number of the minimum distance from the pixel point of the central part of the bubble in the binary image to the background; performing histogram equalization on the distance image to ensure that the brightness of the adhered air bubble is as high as that of the central part of the larger air bubble existing nearby, thereby obtaining an adjustment image; obtaining a local minimum value from the adjustment image, and marking the point where the local minimum value is located to obtain a marked image; according to the marking points, watershed transformation is carried out on the pretreatment binary image to obtain a watershed dividing line, the watershed dividing line is inverted, and OR operation is carried out on the watershed dividing line and the pretreatment binary image to obtain a dividing image of the adhesion bubble. The method can effectively solve the problem that the effect of dividing the adhesion bubbles is poor in the prior art when the adhesion bubbles are extracted from the dynamic ice image.

Description

A method to extract adhesion bubbles from dynamic ice images based on watershed segmentation Methods, devices and equipment

This application is a divisional application of the applicant's earlier application. The application number of the earlier application is CN202210638584.X, and the application title is a method, device and equipment for extracting adhesion bubbles from dynamic ice images.

Technical field

The present application relates to the field of image processing technology, and more specifically, to a method, device and equipment for extracting adhesion bubbles from dynamic ice images based on watershed segmentation.

Background technique

When an aircraft passes through clouds, when supercooled water droplets hit the aircraft body, a phase change may occur and lead to icing. Icing will change the shape and flow field of the aircraft, destroy aerodynamic performance, reduce maneuverability and stability, threaten flight safety, and in serious cases lead to air crashes. Aircraft icing is essentially a dynamic icing process of supercooled water droplets. Supercooled water hits the low-temperature substrate and freezes, and the ice with non-uniformly distributed bubbles formed is dynamic ice. Bubbles in dynamic ice are the fundamental factor that determines the physical properties of dynamic ice. By analyzing the bubble content, distribution, pore size and other characteristics in the microstructure of dynamic ice and studying the physical properties of dynamic ice, scientific and effective icing protection methods can be established to ensure the safety of aircraft flight.

To study the bubble content, distribution, pore size and other characteristics of dynamic ice microstructure, it is necessary to obtain bubble images in dynamic ice. However, there are adhesive bubbles in dynamic ice. When extracting bubbles from dynamic ice images, it is difficult to extract adhesive bubbles. Currently, the watershed algorithm based on mark control can be used to segment adhering bubbles or particles. This method has better segmentation results when the bubbles or particles are of the same size. However, when the sizes of adhering bubbles in dynamic ice vary greatly, it is difficult to mark the small bubbles, and therefore it is difficult to separate multiple adhering small bubbles. Furthermore, when there are large bubbles with an area larger than that of the adhesion bubbles in the vicinity of the adhesion bubbles in the dynamic ice, it is difficult to mark the adhesion bubbles and therefore it is difficult to divide the adhesion bubbles.

Therefore, when extracting adhesion bubbles from dynamic ice images, the existing technology has a problem of poor performance in segmenting adhesion bubbles.

Contents of the invention

The inventor of the present application discovered through long-term practice that the watershed algorithm based on mark control obtains a distance image based on the minimum distance from the bubble pixels in the preprocessed binary image to the background, and then marks the distance image. Take marking the point where the local minimum value in the distance image is located as an example. The value in the distance matrix of the distance image is related to the size of the bubble in the distance image. The larger the bubble, the farther the minimum distance between the center of the bubble and the background is. , after inversion, the smaller the value of the bubble center part in the distance matrix, the lower the brightness of the center of the large bubble in the distance image. When marking local minima in the distance image, in the local area, the value of the central part of the large bubble is smaller than the value of the central part of the small bubble. Therefore, the central part of the small bubble will not be marked, and it is difficult to detect the stuck small bubbles. segmentation. Based on this, this application proposes a method for extracting adhesion bubbles from dynamic ice images. According to the minimum distance from the bubble pixels in the preprocessed binary image to the background, a distance image is obtained; the distance image is subjected to histogram equalization. , adjust the value of the center part of each bubble in the distance matrix of the distance image to obtain an adjustment image; obtain the values that satisfy the preset conditions from the image matrix of the adjustment image, and add the values that meet the preset conditions in the adjustment image. Assume that the points where the numerical values of the conditions are located are marked to obtain a marked image. The marked image includes the marked points of the central parts of all bubbles. According to the marked points in the marked image, the preprocessed binary image is subjected to watershed transformation to obtain Segmentation image of adhesion bubbles. In this way, the problem of poor segmentation effect of adhesion bubbles existing in the existing technology when extracting adhesion bubbles from dynamic ice images can be effectively solved.

In the first aspect, embodiments of the present application provide a method for extracting adhesion bubbles from dynamic ice images. The method includes: S110. Obtain a distance image based on the minimum distance from the bubble pixels in the preprocessed binary image to the background; S120. Perform histogram equalization on the distance image to obtain an adjusted image; S130. Obtain the values that satisfy the preset conditions from the image matrix of the adjusted image, and add the values that satisfy the preset conditions in the adjusted image. Mark the points where the numerical values are located to obtain a marked image; S140. According to the marked points in the marked image, perform a watershed transformation on the preprocessed binary image to obtain a segmented image of adhering bubbles.

In a second aspect, embodiments of the present application also provide a system for extracting adhesion bubbles from dynamic ice images. The system includes a distance acquisition unit for determining the minimum distance from the bubble pixels in the preprocessed binary image to the background. Obtain a distance image; an adjustment unit is used to perform histogram equalization on the distance image to obtain an adjustment image; a marking unit is used to obtain a value that satisfies the preset condition from the image matrix of the adjustment image, and in the Mark the points in the adjusted image where the values that meet the preset conditions are located to obtain a marked image; a watershed transformation unit is used to perform watershed transformation on the marked image according to the marked points in the marked image to obtain adhesion bubbles segmented image.

In a third aspect, embodiments of the present application also provide an electronic device, which includes one or more processors; a memory; a screen for displaying images in the foregoing method; and one or more application programs, wherein the One or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs configured to perform the aforementioned methods.

To sum up, this application has at least the following technical effects:

1. Taking the distance transformation and inversion of the preprocessed binary image to obtain a distance image as an example, this application performs histogram equalization on the distance image that needs to be marked, so that the center brightness of the large bubbles in the distance image becomes higher and the center brightness of the small bubbles becomes higher. The brightness of the center of the distance matrix becomes lower, that is, the value of the center part of the large bubble in the distance matrix becomes larger, and the value of the center part of the small bubble becomes smaller. When marking the point where the local minimum value is located, the difference due to different bubble sizes is avoided. As a result, the value of the central part of the bubble in the local area is different, and the value of the central part of the small bubble is not recognized as a local minimum value, so that the point where the central part of the small bubble is located cannot be marked. Using the method provided by this application to extract adhesion bubbles from dynamic ice images, when the adhesion bubbles in the dynamic ice have a large difference in size, and when there are larger bubbles near the adhesion bubbles in the dynamic ice that have an area larger than the adhesion bubbles, the method can be used to extract adhesion bubbles from dynamic ice images. The point where the center part of the small adhering bubbles is located can be marked, so that the effect of dividing the adhering bubbles is better.

2. This application uses a preset neural network model to directly extract bubbles from the original image, extracts the first intermediate image containing complete large bubbles, then segments the original image, and uses the preset neural network model to extract the first intermediate image from the second segmented image block. Extract bubbles and obtain a second intermediate image containing complete small bubbles. OR the first intermediate image and the second intermediate image to obtain a preprocessed binary image containing both complete large bubbles and complete small bubbles, thus avoiding the need to combine dynamic The texture in the ice image is identified as bubbles, the bubble boundary identification is not clear, a large number of small bubbles will be missed, and there are holes in the extracted large bubbles. This makes the bubbles extracted from the preprocessed binary image better and improves segmentation. The effect of sticking bubbles establishes the basis.

Therefore, the solution provided by this application can effectively solve the problem of poor segmentation of adhesion bubbles in the existing technology when extracting adhesion bubbles from dynamic ice images.

Description of the drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 shows a schematic flow chart of a method for extracting adhesion bubbles from dynamic ice images provided in Embodiment 1 of the present application;

Figure 2 shows a micrograph of adhering bubbles in dynamic ice provided in Example 1 of the present application;

Figure 3 shows the original image of the dynamic ice microstructure provided in Example 1 of the present application;

Figure 4 shows the distance image obtained by distance transforming and inverting the preprocessed binary image provided in Embodiment 1 of the present application;

Figure 5 shows the adjusted image obtained by performing histogram equalization on the distance image provided in Embodiment 1 of the present application;

Figure 6 shows a marked image obtained by marking the point where the local minimum value is located in the adjusted image provided in Embodiment 1 of the present application;

Figure 7 shows a marked image obtained without histogram equalization provided in Embodiment 1 of the present application;

Figure 8 shows a segmented image of adhesion bubbles obtained using histogram equalization provided in Embodiment 1 of the present application;

Figure 9 shows a segmented image of adhering bubbles obtained without histogram equalization provided in Embodiment 1 of the present application;

Figure 10 shows a block diagram of a system for extracting adhesion bubbles from dynamic ice images provided in Embodiment 2 of the present application;

Figure 11 shows a block diagram of an electronic device for executing the method of extracting adhesion bubbles from dynamic ice images according to the embodiment of the present application provided in Embodiment 3 of the present application.

Detailed ways

In order to enable those in the technical field to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only These are part of the embodiments of this application, but not all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

Currently, the watershed algorithm based on mark control can be used to segment adhering bubbles or particles. This method has better segmentation results when the bubbles or particles are of the same size. However, when the sizes of adhering bubbles in dynamic ice vary greatly, it is difficult to mark the small bubbles, and therefore it is difficult to separate multiple adhering small bubbles. Furthermore, when there are large bubbles with an area larger than that of the adhesion bubbles in the vicinity of the adhesion bubbles in the dynamic ice, it is difficult to mark the adhesion bubbles and therefore it is difficult to divide the adhesion bubbles.

Therefore, in order to solve the above defects, embodiments of the present application provide a method for extracting adhesive bubbles from dynamic ice images. The method includes: obtaining a distance image based on the minimum distance from the bubble pixels in the preprocessed binary image to the background; The distance image is subjected to histogram equalization, and the value in the center part of each bubble is adjusted in the distance matrix of the distance image to obtain an adjusted image; the value that satisfies the preset conditions is obtained from the image matrix of the adjustment image, and is Mark the points where the values that meet the preset conditions are located in the adjustment image to obtain a marked image. The marked image includes the marked points of the central parts of all bubbles. According to the marked points in the marked image, the preset Process the binary image and perform watershed transformation to obtain the segmented image of adhering bubbles. In this way, the problem of poor segmentation effect of adhesion bubbles existing in the existing technology when extracting adhesion bubbles from dynamic ice images can be effectively solved.

The method of extracting adhesion bubbles from dynamic ice images involved in this application is introduced below.

Example 1

Please refer to Figures 1 and 2. Figure 1 is a schematic flow chart of a method for extracting adhesion bubbles from dynamic ice images provided in Embodiment 1 of the present application. Figure 2 is a microscopic schematic diagram of adhesion bubbles in dynamic ice. In this embodiment, the method for extracting adhesion bubbles from dynamic ice images may include the following steps:

Step S110: Obtain a distance image based on the minimum distance from the bubble pixels in the preprocessed binary image to the background.

As shown in Figure 3, Figure 3 is an original image of the dynamic ice microstructure. By preprocessing the original image, the preprocessed binary image can be obtained.

In Figure 3, the upper box shows the situation where the sizes of adhesion bubbles in dynamic ice are quite different. Specifically, three bubbles of different sizes are adhering together. Among these three adhesion bubbles, the topmost bubble is the largest. , the bubble in the middle is the second largest, and the bubble at the bottom is the smallest.

In Figure 3, the lower box shows the situation where larger bubbles exist near the adhesion bubbles in dynamic ice. Specifically, three bubbles are adhered together, and the diameter of the larger bubbles existing nearby is larger than the three adhesion bubbles. Any one of the bubbles. In this case, the sizes of the three adhesion bubbles can be the same or different. For convenience of description, Figure 3 shows the situation where the three adhesion bubbles are the same size.

As an optional implementation, step S110 also includes sub-step S111.

Sub-step S111: Perform distance transformation on the preprocessed binary image, and obtain the distance image based on the minimum distance from the bubble pixels in the preprocessed binary image to the background.

In an exemplary embodiment, the distance transformation may be a Euclidean distance transformation, a Manhattan distance transformation, or other distance transformation methods.

Among them, the larger the bubble in the preprocessed binary image is, the greater the minimum distance between the pixels in the center of the bubble and the background is. In the distance matrix of the distance image, the value of the center part of the bubble is larger. In the distance image, the value of the bubble is larger. The center brightness is higher.

As another optional implementation, step S110 also includes sub-step S112.

Sub-step S112: Perform distance transformation and inversion on the preprocessed binary image, and obtain the distance image based on the inverse of the minimum distance between the bubble pixels in the preprocessed binary image and the background.

Among them, the larger the bubble in the preprocessed binary image is, the smaller the inverse number of the minimum distance from the pixel in the center of the bubble to the background is. In the distance matrix of the distance image, the smaller the value in the center of the bubble is. In the distance image The center of the bubble is less bright. As shown in Figure 4, Figure 4 is a distance image obtained by distance transforming and inverting the preprocessed binary image.

As can be seen from Figure 4, among the three adhesion bubbles in the upper box, the brightness of the center part of the uppermost bubble is the lowest, the brightness of the center part of the middle bubble is the second lowest, and the brightness of the center part of the bottom bubble is the highest. , that is, in the distance matrix of the distance image, the value of the center part of the uppermost bubble is the smallest, the value of the center part of the middle bubble is the second smallest, and the value of the center part of the bottom bubble is the largest.

As can be seen from Figure 4, among the three adhering bubbles in the box below and the larger bubbles nearby, the brightness of the center parts of the three adhering bubbles is equally high, and both are higher than the center parts of the larger bubbles nearby. The brightness of , that is, in the distance matrix of the distance image, the values of the central parts of the three adhesion bubbles are equally large, and are all higher than the values of the central parts of the larger bubbles that exist nearby.

In the embodiment of the present application, the negation may also be performed after step S120.

Step S120: Perform histogram equalization on the distance image to obtain an adjusted image.

Taking the distance transformation and inversion of the preprocessed binary image to obtain a distance image as an example, perform histogram equalization on the distance image in this case, so that the center brightness of the large bubbles in the distance image becomes higher, and the center brightness of the small bubbles becomes higher. The brightness of the center becomes lower, that is, the value of the center part of the large bubble in the distance matrix becomes larger, and the value of the center part of the small bubble becomes smaller, thereby obtaining the adjusted image. As shown in Figure 5, Figure 5 is an adjusted image obtained by performing histogram equalization on the distance image.

As can be seen from Figure 5, among the three adhesion bubbles in the box above, the brightness of the center parts of the three adhesion bubbles is the same. That is, in the image matrix of the adjusted image, the values of the center parts of the three adhesion bubbles are the same. big.

As can be seen from Figure 5, among the three adhering bubbles in the box below and the larger bubbles nearby, the brightness of the center parts of the three adhering bubbles is as high as the brightness of the center parts of the larger bubbles nearby. That is, in the image matrix of the adjusted image, the value of the center part of the three stuck bubbles is the same as the value of the center part of the larger bubble existing nearby.

Step S130: Obtain the values that satisfy the preset conditions from the image matrix of the adjustment image, and mark the points where the values that meet the preset conditions are located in the adjustment image to obtain a marked image.

As an optional implementation manner, if step S110 includes sub-step S111 and the preset condition is a local maximum, then step S130 includes sub-step S131.

Sub-step S131: Obtain the local maximum value from the image matrix of the adjustment image, and mark the point where the local maximum value is located in the adjustment image to obtain the marked image.

The local maximum value can be: the maximum value of the value in the local area in the image matrix of the adjusted image. The local area can be an area that just covers the largest bubble in the adjusted image, such as the box shown in Figure 5, or it can be a slightly larger area than the area that just covers the largest bubble in the adjusted image. This application does not limit this.

Iteratively calculate the image matrix of the local area in the adjusted image, and obtain the local maximum value in each iterative calculation.

After performing histogram equalization on the distance image to obtain the adjusted image, the brightness of the central part of the bubble in the local area of the adjusted image is consistent, and the value of the central part of the bubble in the local area of the image matrix of the adjusted image is consistent. Therefore, each The values in the center of each bubble are the maximum values in the local area, and the point where the center of each bubble is located can be marked.

If histogram equalization is not performed, the local maximum value is obtained directly from the distance matrix of the distance image obtained by distance transforming the preprocessed binary image, and the point where the local maximum value is located is marked in the distance image. Since the brightness of the central part of the bubble in the local area of the distance image is inconsistent, the value of the central part of the bubble in the local area of the distance matrix of the distance image is also inconsistent. Therefore, only the bubble with the largest value in the central part can be marked. Bubbles whose central value is not the largest will be missed.

As another optional implementation, if step S110 includes sub-step S112 and the preset condition is a local minimum, then step S130 includes sub-step S132.

Sub-step S132: Obtain the local minimum value from the image matrix of the adjustment image, and mark the point where the local minimum value is located in the adjustment image to obtain the marked image.

The local minimum can be: the minimum value of the value in the local area in the image matrix of the adjusted image. For the local area and the content of obtaining the local minimum value, please refer to the local area and the content of obtaining the local maximum value in sub-step S131, which will not be described again in this application.

After performing histogram equalization on the distance image to obtain the adjusted image, the brightness of the central part of the bubble in the local area of the adjusted image is consistent, and the value of the central part of the bubble in the local area of the image matrix of the adjusted image is consistent. Therefore, each The values in the center of each bubble are the minimum values in the local area, and the point where the center of each bubble is located can be marked.

As shown in Figure 6, Figure 6 is a marked image obtained using the method of the present application.

As can be seen from Figure 6, among the three adhesion bubbles in the upper box, the points where the central parts of the three adhesion bubbles are are marked.

It can be seen from Figure 6 that among the three adhering bubbles in the box below and the larger bubbles nearby, the points where the center parts of the three adhering bubbles are located and the points where the center parts of the larger bubbles are nearby are both. is marked out.

If histogram equalization is not performed, the local minimum value is obtained directly from the distance matrix of the distance image as shown in Figure 4, and the point where the local minimum value is located is marked in the distance image. The brightness of the central part of the bubble in the area is inconsistent, and the value of the central part of the bubble in the local area of the distance matrix of the distance image is also inconsistent. Therefore, only the bubble with the smallest value in the central part can be marked, and the value in the central part is not The smallest air bubble will be missed. If histogram equalization is not performed, the resulting labeled image is shown in Figure 7.

As can be seen from Figure 7, among the three adhesion bubbles in the upper box, only the central point of the largest bubble at the top is marked, the second largest bubble in the middle and the smallest bubble at the bottom are marked. The points where the central part is located are all missed.

As can be seen from Figure 7, among the three adhering bubbles in the box below and the larger bubbles nearby, only the point where the center part of the larger bubble exists nearby is marked, and the center part of the three adhering bubbles is points are missed.

Step S140: Perform watershed transformation on the preprocessed binary image according to the marker points in the marker image to obtain a segmented image of adhesion bubbles.

As an optional implementation manner, if step S110 includes sub-step S111, then step S140 includes sub-step S141.

Sub-step S141: Perform watershed transformation on the preprocessed binary image according to the marked points in the marked image to obtain a watershed dividing line based on the marked points, and perform an OR operation on the watershed dividing line and the preprocessed binary image. , obtain the segmented image of adhesion bubbles.

In the embodiment of this application, since the point where the central part of each bubble is located is marked, the watershed dividing line of each bubble can be obtained by performing watershed transformation on the preprocessed binary image according to the marked point of each bubble. The watershed dividing line of each bubble is ORed with the preprocessed binary map to segment the adhesive bubbles.

As another optional implementation manner, if step S110 includes sub-step S112, then step S140 includes sub-step S142.

Sub-step S142: Perform watershed transformation on the preprocessed binary image according to the marked points in the marked image to obtain a watershed dividing line based on the marked points, invert the watershed dividing line and combine it with the preprocessed binary image. Perform an OR operation on the graph to obtain the segmented image of the adhesion bubbles.

In the embodiment of this application, since the point where the central part of each bubble is located is marked, the watershed dividing line of each bubble can be obtained by performing watershed transformation on the preprocessed binary image according to the marked point of each bubble. The watershed dividing line of each bubble is inverted, and then ORed with the preprocessed binary map to segment the adhering bubbles.

As shown in Figure 8, Figure 8 is a segmented image of adhesion bubbles obtained using the method of the present application.

As can be seen from Figure 8, among the three adhesion bubbles in the upper box, all three adhesion bubbles are divided.

As can be seen from Figure 8, among the three adhesion bubbles in the box below, all three adhesion bubbles are divided.

If histogram equalization is not performed, the segmented image of the stuck bubbles obtained is shown in Figure 9.

As can be seen from Figure 9, among the three adhesion bubbles in the upper box, since only the point where the center part of the uppermost bubble is located is marked, the points where the center part of the middle bubble and the bottom bubble are are not marked. Marked so that only the top bubble is divided, the middle bubble and the bottom bubble are not divided and are still stuck together.

As can be seen from Figure 9, among the three adhesion bubbles in the box below, since the points where the central parts of the three adhesion bubbles are are not marked, the three adhesion bubbles have not been divided and are still stuck together.

By performing histogram equalization on the distance image that needs to be marked, this application avoids the different numerical sizes of the central parts of the bubbles in the local area due to different bubble sizes, and does not identify the numerical values of the central parts of the small bubbles as satisfying the preset The value of the condition makes it impossible to mark the point where the center part of the small bubble is located. Using the method provided by this application to extract adhesion bubbles from dynamic ice images, when the adhesion bubbles in the dynamic ice have a large difference in size, and when there are larger bubbles near the adhesion bubbles in the dynamic ice that have an area larger than the adhesion bubbles, the method can be used to extract adhesion bubbles from dynamic ice images. The point where the center part of the small adhering bubbles is located can be marked, so that the effect of dividing the adhering bubbles is better.

In an exemplary embodiment, before step S110, steps S101 to S104 are also included.

Step S101: Extract bubbles from the original image through a preset neural network model to obtain a first intermediate image.

In the embodiment of the present application, the original image of the dynamic ice may be a microscopic image or an image obtained by photographing the dynamic ice with a mobile phone camera. The original image may be a color image or a grayscale image.

As an optional implementation, if the size of the original image is equal to the preset size, bubbles are extracted from the original image through the preset neural network model to obtain the first intermediate image.

As another optional implementation, if the size of the original image is not equal to the preset size, the size of the original image is adjusted to the preset size, and the preset neural network model is used to adjust the size from Extract bubbles from the original image of the preset size to obtain a first extracted image, and restore the first extracted image to the size of the original image to obtain the first intermediate image.

Wherein, if the size of the original image is smaller than the preset size, the size of the original image is enlarged to the preset size, and bubbles are extracted from the original image enlarged to the preset size through a preset neural network model to obtain the first extraction image, and reduce the first extracted image to the size of the original image to obtain a first intermediate image.

If the size of the original image is larger than the preset size, reduce the size of the original image to the preset size, and extract bubbles from the original image reduced to the preset size through the preset neural network model to obtain the first extracted image, The first extracted image is enlarged to the size of the original image to obtain a first intermediate image.

As another optional implementation, if the size of the original image is smaller than the preset size, bubbles are extracted from the original image through a preset neural network model to obtain the first intermediate image.

The default size may be a size set according to computer resource limitations, such as 256*256 or 512*512.

The method of enlarging or reducing the original image may be an interpolation method, a method of reducing the first extracted image to the size of the original image or enlarging the first extracted image to the size of the original image, or an interpolation method.

By adjusting the original image to the preset size when the size of the original image is not equal to the preset size, avoiding the inability to perform image recognition on the original image due to limitations of computer resources, and restoring the first extracted image to the size of the original image, Avoid loss of accuracy of the first intermediate image due to resizing.

In the embodiment of this application, the preset neural network model may be a U-net network model with an attention mechanism added, or it may be an R2U-net model, or it may be other neural network models, which this application does not limit.

This application extracts bubbles from dynamic ice images through a preset neural network model, avoiding problems such as identifying textures in dynamic ice images as bubbles and unclear bubble boundary identification, and provides a basis for extracting bubbles with better effects.

As an optional implementation, if the shape of the original image does not satisfy the input shape of the preset neural network model, fill the original image with the input shape of the preset neural network model, and then perform steps S101.

Step S102: Divide the original image into n second divided image blocks, and n≥2, and extract bubbles from the n second divided image blocks respectively through the preset neural network model to obtain n Secondly extract image blocks.

This application divides the dynamic ice image into second segmented image blocks with smaller sizes, and uses a preset neural network model to extract bubbles from each second segmented image block respectively to avoid missing small ones when the original image size is too large and the bubbles are too small. Bubbles make the extracted bubbles better.

Step S103: Splice the n second extracted image blocks into a second intermediate image according to the order of division positions. When the original image is divided into the n second divided image blocks, The arrangement order of the position of each second segmented image block in the original image.

In the embodiment of the present application, the step of obtaining the first intermediate image may be performed first, and then the step of obtaining the second intermediate image may be performed, or the step of obtaining the second intermediate image may be performed first, and then the step of obtaining the first intermediate image may be performed. The steps of acquiring the first intermediate image and the steps of acquiring the second intermediate image may also be performed simultaneously.

Step S104: Perform an OR operation on the first intermediate image and the second intermediate image to obtain a preprocessed binary image.

Specifically, the small bubbles in the first intermediate image are missing, and the large bubbles in the second intermediate image have holes. The first intermediate image and the second intermediate image are ORed to fill in the missing small bubbles and large bubbles. holes, a preprocessed binary image is obtained, and the preprocessed binary image contains complete small bubbles and complete large bubbles.

As an optional implementation manner, perform an opening operation on the preprocessed binary image obtained in step S104 to obtain a preprocessed binary image after eliminating noise.

This application directly extracts bubbles from the original image through a preset neural network model, extracts the first intermediate image containing complete large bubbles, then segments the original image, and extracts bubbles from the second segmented image block through the preset neural network model. , obtain the second intermediate image containing complete small bubbles, perform an OR operation on the first intermediate image and the second intermediate image, and obtain a preprocessed binary image containing both complete large bubbles and complete small bubbles, so that the preprocessed binary image The extracted bubbles are better, which lays the foundation for improving the effect of segmenting stuck bubbles.

Example 2

Please refer to FIG. 10 , which is a structural block diagram of a system 1000 for extracting adhesion bubbles from dynamic ice images provided in Embodiment 2 of the present application. The system may include: a distance acquisition unit 1010, an adjustment unit 1020, a marking unit 1030, and a watershed transformation unit 1040.

The distance acquisition unit 1010 is used to obtain a distance image based on the minimum distance from the bubble pixels in the preprocessed binary image to the background.

The adjustment unit 1020 is used to perform histogram equalization on the distance image to obtain an adjusted image.

The marking unit 1030 is configured to obtain a value that satisfies a preset condition from the image matrix of the adjusted image, and mark the point in the adjusted image where the value that satisfies the preset condition is located, to obtain a marked image.

The watershed transformation unit 1040 is configured to perform watershed transformation on the marked image according to the marked points in the marked image to obtain segmented images of adhering bubbles.

As an optional implementation manner, the distance acquisition unit 1010 includes a first distance acquisition subunit, which is used to perform distance transformation on the preprocessed binary image. According to the bubble pixel points in the preprocessed binary image, Minimum distance from the background to obtain the distance image.

The marking unit 1030 includes a first marking subunit, which is used to obtain the local maximum value from the image matrix of the adjustment image, and mark the point where the local maximum value is located in the adjustment image, to obtain The labeled image.

The watershed transformation unit 1040 includes a first watershed transformation subunit, which is used to perform watershed transformation on the preprocessed binary image according to the marker points in the marker image to obtain a watershed dividing line based on the marker points, and convert the The watershed dividing line is ORed with the preprocessed binary image to obtain the segmented image of the adhesion bubbles.

As another optional implementation, the distance acquisition unit 1010 also includes a second distance acquisition subunit, which is used to distance transform and invert the preprocessed binary image. According to the preprocessed binary image, The distance image is obtained by taking the opposite number of the minimum distance from the bubble pixel to the background.

The marking unit 1030 also includes a second marking subunit, used to obtain the local minimum value from the image matrix of the adjustment image, and mark the point where the local minimum value is located in the adjustment image, Obtain the labeled image.

The watershed transformation unit 1040 includes a second watershed transformation subunit, which is used to perform watershed transformation on the preprocessed binary image according to the marker points in the marker image to obtain a watershed dividing line based on the marker points, and convert the The watershed dividing line is inverted and ORed with the preprocessed binary image to obtain the segmented image of the adhesion bubbles.

Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems and units can be referred to the corresponding processes in the foregoing method embodiments, and will not be described again here.

In addition, each functional module in each embodiment of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software function modules.

Example 3

Please refer to FIG. 11 , which is a structural block diagram of an electronic device 1100 provided in Embodiment 3 of the present application. The electronic device 1100 in this application may include one or more of the following components: a memory 1110, a processor 1120, a screen 1130, and one or more application programs, where one or more application programs may be stored in the memory 1110 and configured. For execution by one or more processors 1120, one or more programs are configured to perform the method as described in the preceding method embodiments.

The memory 1110 may include Random Access Memory (RAM) or Read-Only Memory (ROM). Memory 1110 may be used to store instructions, programs, codes, sets of codes, or sets of instructions. The memory 1110 may include a program storage area and a data storage area, where the program storage area may store instructions for implementing the operating system, instructions for implementing at least one function (such as a histogram equalization function, etc.), and instructions for implementing the following Instructions for various method embodiments, etc. The storage data area can also store data created during use of the electronic device 1100 (such as image matrix data, etc.).

Processor 1120 may include one or more processing cores. The processor 1120 uses various interfaces and lines to connect various parts of the entire electronic device 1100, and executes by running or executing instructions, programs, code sets or instruction sets stored in the memory 1110, and calling data stored in the memory 1110. Various functions and processing data of the electronic device 1100. Optionally, the processor 1120 may adopt at least one of digital signal processing (Digital Signal Processing, DSP), field-programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). implemented in hardware form. The processor 1120 may integrate one or a combination of a central processing unit (CPU) and a modem. Among them, the CPU mainly handles operating systems and applications, etc.; the modem is used to handle wireless communications. It can be understood that the above modem may not be integrated into the processor 1120 and may be implemented solely through a communication chip.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art will understand that: it can still Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent substitutions are made to some of the technical features; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions in the embodiments of the present application.

Claims (5)

1. A method for extracting adhesion bubbles from dynamic ice images based on watershed segmentation, characterized in that the method includes: S110. Obtain the distance image based on the minimum distance from the pixels in the center of the bubble in the preprocessed binary image to the background, including: Perform distance transformation on the preprocessed binary image, and obtain the distance image based on the inverse of the minimum distance from the pixels in the center of the bubble in the preprocessed binary image to the background; S120. Perform histogram equalization on the distance image, adjust the value of the center part of each bubble in the distance matrix of the distance image, so that the brightness of the stuck bubbles is as high as the center part of the larger bubbles existing nearby, and the adjusted image is obtained; S130. Obtain the local minimum value from the image matrix of the adjustment image, and mark the point where the local minimum value is located in the adjustment image to obtain a marked image; S140. According to the marked points in the marked image, perform watershed transformation on the preprocessed binary image to obtain a watershed dividing line based on the marked points, invert the watershed dividing line, and perform the same operation with the preprocessed binary image. Or operation to obtain the segmented image of adhesion bubbles. 2. The method of extracting adhesion bubbles from dynamic ice images based on watershed segmentation according to claim 1, characterized in that the distance transformation is Euclidean distance transformation or Manhattan distance transformation. 3. The method for extracting adhesion bubbles from dynamic ice images based on watershed segmentation according to claim 1, characterized in that before step S110, it also includes: Extract bubbles from the original image through a preset neural network model to obtain the first intermediate image; Divide the original image into n second segmented image blocks, and n≥2, and extract bubbles from the n second segmented image blocks respectively through the preset neural network model to obtain n second extractions image block; The n second extracted image blocks are spliced into a second intermediate image according to the order of division positions. When the original image is divided into the n second divided image blocks, each of the second intermediate images is The arrangement order of the positions of the binary segmented image blocks in the original image; Perform an OR operation on the first intermediate image and the second intermediate image to obtain a preprocessed binary image. 4. A system for extracting adhesion bubbles from dynamic ice images based on watershed segmentation, characterized by: The distance acquisition unit is used to obtain the distance image based on the minimum distance from the pixels in the center of the bubble in the preprocessed binary image to the background, including: Perform distance transformation on the preprocessed binary image, and obtain the distance image based on the inverse of the minimum distance from the pixels in the center of the bubble in the preprocessed binary image to the background; The adjustment unit is used to perform histogram equalization on the distance image, and adjust the value of the center part of each bubble in the distance matrix of the distance image so that the brightness of the adhesion bubble is as high as the center part of the larger bubble existing nearby, to obtain adjust image; A marking unit, configured to obtain the local minimum value from the image matrix of the adjustment image, and mark the point where the local minimum value is located in the adjustment image to obtain a marked image; A watershed transformation unit is configured to perform watershed transformation on the marked image according to the marked points in the marked image, obtain a watershed dividing line based on the marked points, invert the watershed dividing line, and compare it with the preprocessed binary image Perform an OR operation to obtain the segmented image of the stuck bubbles. 5. An electronic device, characterized in that it includes: one or more processors; memory; A screen for displaying images in the method according to any one of claims 1-3; one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs are configured to perform, e.g. The method described in any one of claims 1-3.