CN117522719B - Bronchoscope image auxiliary optimization system based on machine learning - Google Patents

Abstract

The invention relates to the technical field of image processing, in particular to a bronchoscope image auxiliary optimization system based on machine learning, which comprises the following steps: the image acquisition module is used for acquiring a bronchus image; the reference region acquisition module is used for acquiring a reference region of each pixel in the bronchial image according to the bronchial image; the correction confusion degree acquisition module is used for obtaining the initial confusion degree of each pixel according to the reference area of each pixel and obtaining the correction confusion degree of each pixel according to the initial confusion degree; the image enhancement module is used for obtaining corrected clustering distances of every two pixels according to the corrected confusion degree and obtaining a plurality of independent texture areas according to the corrected clustering distances; an enhanced bronchial image is obtained from each independent texture region. Thereby realizing the enhancement of focus areas in the bronchial image by analyzing the condition that each area in the bronchial image accords with focus characteristics.

Description

Bronchoscope image-assisted optimization system based on machine learning

Technical field

The present invention relates to the field of image processing technology, and in particular to a bronchoscope image-assisted optimization system based on machine learning.

Background technique

Bronchoscopy allows direct observation of the inside of the bronchi and parts of the lungs, providing the first visual evidence of disease or abnormalities. Since the bronchus is an internal structure of the human body, and natural light cannot penetrate into the human body, there is a fill light in the bronchoscope to illuminate local areas of the bronchus, and then the built-in camera in the bronchoscope collects images of the bronchus.

Since bronchial images are collected under a fill light environment, there will be highlighted areas in the bronchial images, and the grayscale values of the highlighted areas are generally larger. At the same time, inflammatory lesions in the bronchus are generally white, so the gray value of the inflammatory lesion area is also larger. Therefore, the light in the bronchial image will interfere with the extraction of the lesion area. Therefore, in order to eliminate light interference, the bronchial image needs to be enhanced.

The traditional enhancement method will compress smaller grayscale values smaller and stretch larger grayscale values larger. Therefore, traditional enhancement methods cannot distinguish the gray value of the lesion area from the gray value of the highlight area generated by light. Therefore, it is necessary to design an enhancement method to distinguish the gray value of the lesion area from other areas, so as to facilitate the extraction of the lesion area.

Contents of the invention

The present invention provides a bronchoscope image-assisted optimization system based on machine learning to solve the existing problem: how to distinguish the gray value of the lesion area from other areas through image enhancement.

The machine learning-based bronchoscope image-assisted optimization system of the present invention adopts the following technical solutions:

One embodiment of the present invention provides a machine learning-based bronchoscope image-assisted optimization system, which includes the following modules:

Image acquisition module, used to acquire bronchial images;

The reference area acquisition module is used to obtain the reference area of each pixel in the bronchial image based on the difference in gray value of the pixels in the bronchial image;

The modified confusion degree acquisition module is used to obtain the initial confusion degree of each pixel based on the area and edge complexity of the reference area of each pixel, and obtain the initial confusion degree of each pixel based on the initial confusion degree of each pixel and the gradient of each pixel in the reference area. The degree of correction confusion per pixel;

The image enhancement module is used to obtain the corrected clustering distance of every two pixels based on the corrected degree of confusion of each pixel, and perform cluster analysis on the pixels in the bronchial image to obtain multiple independent texture areas based on the corrected clustering distance of every two pixels. ; The bronchial image is enhanced according to the corrected confusion degree of each pixel in each independent texture area to obtain the enhanced bronchial image.

Preferably, the reference area of each pixel in the bronchial image is obtained based on the difference in gray value of the pixels in the bronchial image, and the specific method includes:

Default window side length , with each pixel as the center, obtain the L*L window, recorded as the reference window of each pixel; obtain the variance of the gray value of all pixels in the reference window of each pixel, recorded as the reference window of each pixel Gray-level variance, the mean value of the gray-level variance of all pixels in the bronchial image is recorded as the reference variance;

For any pixel, the reference window of each pixel is recorded as the first extended window, and the difference between the grayscale variance of the first extended window and the reference variance is recorded as the deviation of the first extended window; according to the deviation of the first extended window Obtain the second extended window, obtain the grayscale variance of the second extended window, record the difference between the grayscale variance of the second extended window and the reference variance as the deviation of the second extended window, and obtain the third extended window based on the deviation of the second extended window. Expand the window; and so on, until the number of pixels in the expansion window is greater than or equal to the preset maximum critical value B, less than or equal to the preset minimum critical value A, or the deviation of the expansion window is equal to 0, and several expansion windows are obtained;

Among all the extended windows of each pixel, the extended window with the smallest deviation is selected as the reference area of each pixel.

Preferably, the second expansion window is obtained according to the deviation of the first expansion window, including the specific method:

When the deviation of the first extended window is greater than 0, the outermost circle of pixels in the first extended window is obtained and recorded as the peripheral pixels of the first extended window. Any pixel is selected and removed from the peripheral pixels of the first extended window to obtain the second extended window;

When the deviation of the first expansion window is less than 0, obtain the peripheral pixels of the first expansion window, obtain the pixels adjacent to the peripheral pixels of the first expansion window and do not belong to the first expansion window, and record them as the first expansion window. For outer adjacent pixels, any outer adjacent pixel of the first extended window is added to the first extended window to obtain a second extended window.

Preferably, the initial degree of confusion of each pixel is obtained based on the area of the reference area of each pixel and the complexity of the edges, including the specific method:

Get the edge complexity of the reference area for each pixel;

The calculation method to obtain the initial confusion level of each pixel based on the edge complexity and area of the reference area of each pixel is:

in, Represents the area of the reference area of the i-th pixel, /> Represents linear normalization processing,/> Represents the edge complexity of the reference area of the i-th pixel, /> Represents the initial confusion level of the i-th pixel.

Preferably, the specific method of obtaining the edge complexity of the reference area of each pixel includes:

Obtain the peripheral pixels of the reference area of each pixel, perform chain code analysis on all peripheral pixels of the reference area of each pixel, and obtain the chain code sequence of the reference area of each pixel; convert the chain code sequence of the reference area of each pixel The variance of all data in , denoted as the edge complexity of the reference region for each pixel.

Preferably, the method of obtaining the corrected degree of confusion of each pixel based on the initial degree of confusion of each pixel and the gradient of each pixel in the reference area includes the following specific methods:

Obtain the gradient eccentricity angle of the reference pixel of each pixel according to the gradient of each pixel in the reference area of each pixel;

The calculation method for obtaining the corrected confusion degree of each pixel based on the gradient eccentricity angle of the reference pixel and the initial confusion degree of each pixel is:

in, Represents the Euclidean distance between the i-th pixel and the q-th pixel of the reference area, /> Represents the gradient eccentricity angle of the q-th reference pixel of the i-th pixel, /> Represents the number of pixels in the reference area of the i-th pixel, /> Represents the initial confusion level of the i-th pixel,/> Indicates the degree of correction confusion of the i-th pixel.

Preferably, the specific method of obtaining the gradient eccentricity angle of the reference pixel of each pixel based on the gradient of each pixel in the reference area of each pixel is:

Record the pixels in the reference area of each pixel as the reference pixel of each pixel, record any reference pixel of each pixel as the target reference pixel of each pixel; obtain the vector composed of each pixel and the target reference pixel, The angle with the gradient direction of the target reference pixel is recorded as the gradient eccentricity angle between each pixel and the target reference pixel;

Get the gradient eccentricity angle of the reference pixel for each pixel.

Preferably, the corrected clustering distance of every two pixels is obtained according to the corrected confusion degree of each pixel, including the specific method:

in, Indicates the degree of correction confusion of the i-th pixel,/> Indicates the degree of correction confusion of the j-th pixel,/> Represents the Euclidean distance between the i-th pixel and the j-th pixel,/> Represents the modified clustering distance between the i-th pixel and the j-th pixel.

Preferably, the cluster analysis of pixels in the bronchial image based on the modified clustering distance of every two pixels is performed to obtain multiple independent texture areas, including the specific method:

Set the clustering parameters of the pixels. Based on the clustering parameters and the modified clustering distance between pixels, use the ISODATA algorithm to cluster the pixels to obtain several independent texture areas.

Preferably, the bronchial image is enhanced according to the correction degree of confusion of each pixel in each independent texture area to obtain an enhanced bronchial image, which includes the following specific methods:

Obtain the mean value of the corrected confusion degree of all pixels in each independent texture area, and record it as the corrected confusion degree of each independent texture area; multiply the grayscale value of each pixel in each independent texture area in the bronchial image by the independent texture area Image enhancement is achieved by correcting the degree of confusion to obtain enhanced bronchial images.

The beneficial effects of the technical solution of the present invention are:

Obtain the bronchial image and obtain the reference area of each pixel based on the gray value difference. Since the gray level difference of the lesion area in the bronchial image is large and the edge of the lesion area is irregular, the edge of the reference area based on each pixel is complex. The initial confusion level of each pixel can be obtained by using the initial confusion level of each pixel and the area of the reference area. The distinguishing characteristics of the lesion can be described by the initial confusion level. Since the highlight area and the lesion area cannot be distinguished by the initial confusion level of each pixel, the initial confusion level of each pixel can be used to distinguish the highlight area and the lesion area. Analyze the gray value change of the reference area of each pixel to match the situation of the lesion area, correct the initial degree of confusion to obtain the corrected degree of confusion of each pixel, and cluster the pixels in the bronchial image according to the corrected degree of confusion of each pixel. The enhanced bronchial image is obtained through the enhancement processing, and the enhanced bronchial image can highlight the lesion area to facilitate subsequent lesion analysis.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without exerting creative efforts.

Figure 1 is a structural block diagram of the bronchoscope image-assisted optimization system based on machine learning of the present invention;

Figure 2 is an image of a bronchus containing inflammatory lesions provided by the present invention.

Detailed ways

In order to further elaborate on the technical means and effects adopted by the present invention to achieve the intended purpose of the invention, the following is a detailed description of the machine learning-based bronchoscope image auxiliary optimization system proposed by the present invention in conjunction with the drawings and preferred embodiments, and its specific implementation methods. , structure, characteristics and functions are described in detail below. In the following description, different terms "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, the specific features, structures, or characteristics of one or more embodiments may be combined in any suitable combination.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which the invention belongs.

The specific scheme of the machine learning-based bronchoscope image assisted optimization system provided by the present invention will be described in detail below with reference to the accompanying drawings.

Please refer to Figure 1, which shows a structural block diagram of a machine learning-based bronchoscope image-assisted optimization system provided by one embodiment of the present invention. The system includes the following modules:

The image acquisition module 101 is used to acquire bronchial images.

In order to implement the machine learning-based bronchoscope image-assisted optimization system proposed in this embodiment, bronchial images need to be collected first.

The specific process of collecting bronchial images is: use a bronchoscope to collect bronchial images. Figure 2 shows a bronchial image. The black rectangular frame of the bronchial image shows the inflammatory infiltration lesions, and the black circle shows the highlights generated by the light. The bronchial image is grayscaled to obtain a grayscale image of the bronchial image. For the convenience of description, the grayscale image of the bronchial image will still be called the bronchial image.

The reference area acquisition module 102 is used to obtain the reference area of each pixel according to the bronchial image.

It should be noted that since there is a grayscale difference between the lesion area and the non-lesional area in the bronchial image, the regions can be divided first according to the grayscale value of the pixels.

Specifically, preset a window side length , with each pixel as the center, obtain the L*L window, which is recorded as the reference window of each pixel. The variance of the gray value of all pixels in the reference window of each pixel is obtained, recorded as the gray variance of the reference window of each pixel, and the mean value of the gray variance of all pixels in the bronchial image is recorded as the reference variance.

In this embodiment, L is set to 7 as an example for description. In other embodiments, other values may be used, and there is no specific limitation in this embodiment.

It should be noted that since the above process only artificially divides some windows, it is not possible to separate different pixels and separate similar pixels together at this time, so adjustments need to be made based on the window.

Further, for any pixel, the reference window of each pixel is recorded as the first extended window, and the difference between the grayscale variance of the first extended window and the reference variance is recorded as the deviation of the first extended window. The second extended window is obtained according to the deviation of the first extended window, and the grayscale variance of the second extended window is obtained. The difference between the grayscale variance of the second extended window and the reference variance is recorded as the deviation of the second extended window. According to the second extended window The deviation of the expanded window results in a third expanded window. By analogy, it ends when the number of pixels in the expansion window is greater than or equal to the preset maximum critical value B, less than or equal to the preset minimum critical value A, or the deviation of the expansion window is equal to 0, and several expansion windows are obtained.

This embodiment uses A as 4 and B as 196 as an example for description. In other embodiments, other values may be used and are not specifically limited in this embodiment.

Further, the method of obtaining the second expansion window based on the deviation of the first expansion window is:

When the deviation of the first extended window is greater than 0, the outermost circle of pixels in the first extended window is obtained and recorded as the peripheral pixels of the first extended window. Any pixel is selected and removed from the peripheral pixels of the first extended window to obtain the second Expand window.

When the deviation of the first expansion window is less than 0, obtain the peripheral pixels of the first expansion window, obtain the pixels adjacent to the peripheral pixels of the first expansion window and do not belong to the first expansion window, and record them as the first expansion window. Outer adjacent pixels. Select any outer adjacent pixel of the first extended window and add it to the first extended window to obtain a second extended window.

Further, among all the extended windows of each pixel, the extended window with the smallest deviation is selected as the reference area of each pixel.

It should be noted that for pixels at the edge of the bronchial image, when the size of the pixel window does not meet the requirements, just obtain the largest window possible within the size range.

At this point, the reference area of each pixel is obtained. The gray value of the reference area of each pixel is relatively similar. It generally describes an object with the same attribute. The object with the same attribute can be an independent focus area, or an independent highlight area, or an independent non-lesion area and non-highlight area.

The corrected confusion degree acquisition module 103 is used to obtain the initial confusion degree of each pixel according to the reference area of each pixel, and obtain the corrected confusion degree of each pixel according to the initial confusion degree of each pixel.

Specifically, peripheral pixels of the reference area of each pixel are obtained, chain code analysis is performed on all peripheral pixels of the reference area of each pixel, and a chain code sequence of the reference area of each pixel is obtained. The variance of all data in the chain code sequence of the reference area of each pixel is recorded as the edge complexity of the reference area of each pixel.

The initial confusion level for each pixel is:

in, Indicates the area of the reference area of the i-th pixel. The larger the value, the greater the number of pixels required to achieve the baseline variance, which in turn indicates that the grayscale difference around the i-th pixel is smaller, so the initial chaos of the i-th pixel is smaller. , Represents linear normalization processing,/> Indicates the edge complexity of the reference area of the i-th pixel. The larger the value, the greater the direction change of the edge of the i-th reference area. The complexity of the outer edge of the texture where the i-th pixel is located is greater, so the i-th The initial confusion of pixels is relatively large,/> Represents the initial confusion level of the i-th pixel.

It should be noted that when calculating the initial confusion level, the grayscale difference of the surrounding pixels of each pixel and the complexity of the outer edge of the texture where each pixel is located are mainly considered. However, in bronchial images, not only the lesion area has this characteristic, but also the highlight area. Therefore, the highlight area and the lesion area cannot be distinguished by the initial degree of confusion.

In order to further distinguish the highlight area and the lesion area, it is necessary to further analyze the distinguishing characteristics of these two areas. The highlight area generally has the phenomenon of gray value decreasing from the highlight center to the surrounding areas, but the lesion area does not have this phenomenon, so the two areas can be distinguished based on this feature.

Further, the pixels in the reference area of each pixel are recorded as the reference pixels of each pixel, and any reference pixel of each pixel is recorded as the target reference pixel of each pixel. Obtain the angle between the vector composed of each pixel and the target reference pixel and the gradient direction of the target reference pixel, and record it as the gradient eccentricity angle between each pixel and the target reference pixel. Get the gradient eccentricity angle of the reference pixel for each pixel.

The corrected clutter per pixel is calculated as:

in, Represents the Euclidean distance between the i-th pixel and the q-th pixel of the reference area, /> Represents the gradient eccentricity angle of the q-th reference pixel of the i-th pixel. The larger the value is, the less centripetal the gradient direction of the q-th reference pixel of the i-th pixel is, while the grayscale value of the highlighted area has from The center tends to decrease toward the surroundings, so the gradient direction of the highlight area is generally from the highlight center to the surroundings. Therefore, when the reference area of the i-th pixel is the highlight area, the gradient eccentricity angle of the reference pixel of the pixel should be smaller. , therefore, when the gradient eccentricity angle is larger, it means that the i-th pixel is less likely to be a highlight area and more likely to be a lesion area, because its corresponding correction degree is greater,/> Represents the number of pixels in the reference area of the i-th pixel, /> Represents the distance weight. The larger the value, the smaller the distance between the i-th pixel and the q-th pixel, so the i-th pixel is more affected by the q-th pixel./> Represents the initial confusion level of the i-th pixel. /> Indicates the degree of correction confusion of the i-th pixel.

At this point, the corrected confusion level of each pixel is obtained. The corrected confusion level is an index obtained by describing how each pixel conforms to the characteristics of the lesion area. This index can distinguish the lesion area from the non-lesion area.

The image enhancement module 104 is configured to perform clustering processing according to the corrected confusion degree of each pixel to obtain several independent texture areas, and perform enhancement processing on the bronchial image according to each independent texture area and the corrected confusion degree to obtain an enhanced bronchial image.

It should be noted that in order to separate the lesion area and non-lesion area in the bronchial image, the bronchial image needs to be clustered. The traditional ISODATA algorithm method uses the gray value between pixels and the distance between pixels to set the clustering Class distance, this clustering method does not consider the characteristics of the lesion area, so this clustering method cannot separate the lesion area and non-lesion area. Therefore, in order to better separate the lesion area and non-lesion area, the clustering distance in the algorithm needs to be corrected.

Specifically, the modified clustering distance between pixels is calculated as:

in, Indicates the corrected confusion level of the i-th pixel, /> represents the corrected confusion level of the jth pixel, /> represents the Euclidean distance between the i-th pixel and the j-th pixel,/> represents the corrected clustering distance between the i-th pixel and the j-th pixel.

Further, set the clustering parameters: the number of cluster centers , among which,/> Represents the number of pixels in the bronchial image, L represents the preset window side length, /> Indicates the number of cluster centers. The minimum number of category elements is 5, the intra-category difference is 10, the category merging threshold is 8, and the maximum number of pairs of cluster centers that can be merged in one iteration is/> , the maximum number of iterations is 30.

It should be noted that the clustering parameters are parameters that need to be manually set in the ISODATA algorithm. Since the ISODATA algorithm is an existing algorithm, its setting process is relatively routine and will not be described again here.

Furthermore, based on the clustering parameters and the modified clustering distance between pixels, the ISODATA algorithm is used to cluster the pixels to obtain several independent texture areas.

At this point, the clustering algorithm is used to complete the segmentation of the lesion area and the non-lesion area. Next, image enhancement processing needs to be performed based on the segmentation results.

Further, the mean value of the corrected confusion degree of all pixels in each independent texture area is obtained, and recorded as the corrected confusion degree of each independent texture area. The corrected confusion degree of each independent texture area is normalized using the maximum and minimum value normalization method to obtain the normalized corrected confusion degree of each independent texture area. For the convenience of description, the normalized corrected confusion degree of each independent texture area is subsequently recorded as the corrected confusion level of each independent texture area. Image enhancement is achieved by multiplying the grayscale value of each pixel of each independent texture area in the bronchial image by the corrected confusion level of the independent texture area to obtain the enhanced bronchial image.

It should be noted that the gray value of the pixel in each independent texture area in the bronchial image is multiplied by the corrected confusion level of the independent texture area, and the resulting gray value is rounded down.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention shall be included in the protection scope of the present invention. within.

Claims (8)

1. Bronchoscope image-assisted optimization system based on machine learning, characterized in that the system includes the following modules: Image acquisition module, used to acquire bronchial images; The reference area acquisition module is used to obtain the reference area of each pixel in the bronchial image based on the difference in gray value of the pixels in the bronchial image; The modified confusion degree acquisition module is used to obtain the initial confusion degree of each pixel based on the area and edge complexity of the reference area of each pixel, and obtain the initial confusion degree of each pixel based on the initial confusion degree of each pixel and the gradient of each pixel in the reference area. The degree of correction confusion per pixel; Among them, the initial degree of confusion of each pixel is obtained according to the area and edge complexity of the reference area of each pixel, including the specific methods of: obtaining the edge complexity of the reference area of each pixel; The calculation method to obtain the initial confusion level of each pixel based on the edge complexity and area of the reference area of each pixel is: H _i =norm(-S _i )×σ _i Among them, S _i represents the area of the reference area of the i-th pixel, norm() represents linear normalization processing, σ _i represents the edge complexity of the reference area of the i-th pixel, and H _i represents the initial chaos of the i-th pixel. degree; Among them, obtaining the edge complexity of the reference area of each pixel includes the following specific methods: obtaining the peripheral pixels of the reference area of each pixel, performing chain code analysis on all peripheral pixels of the reference area of each pixel, and obtaining each The chain code sequence of the reference area of the pixel; the variance of all data in the chain code sequence of the reference area of each pixel is recorded as the edge complexity of the reference area of each pixel; The image enhancement module is used to obtain the corrected clustering distance of every two pixels based on the corrected degree of confusion of each pixel, and perform cluster analysis on the pixels in the bronchial image to obtain multiple independent texture areas based on the corrected clustering distance of every two pixels. ; The bronchial image is enhanced according to the corrected confusion degree of each pixel in each independent texture area to obtain the enhanced bronchial image. 2. The bronchoscope image-assisted optimization system based on machine learning according to claim 1, characterized in that the reference area of each pixel in the bronchial image is obtained according to the gray value difference of the pixels in the bronchial image, including a specific method for: Preset a window side length L, with each pixel as the center, obtain the L*L window, recorded as the reference window of each pixel; obtain the variance of the gray value of all pixels in the reference window of each pixel, recorded as The gray level variance of the reference window for each pixel, the mean of the gray level variance of all pixels in the bronchial image is recorded as the reference variance; For any pixel, the reference window of each pixel is recorded as the first extended window, and the difference between the grayscale variance of the first extended window and the reference variance is recorded as the deviation of the first extended window; according to the deviation of the first extended window Obtain the second extended window, obtain the grayscale variance of the second extended window, record the difference between the grayscale variance of the second extended window and the reference variance as the deviation of the second extended window, and obtain the third extended window based on the deviation of the second extended window. Expand the window; and so on, until the number of pixels in the expansion window is greater than or equal to the preset maximum critical value B, less than or equal to the preset minimum critical value A, or the deviation of the expansion window is equal to 0, and several expansion windows are obtained; Among all the extended windows of each pixel, the extended window with the smallest deviation is selected as the reference area of each pixel. 3. The bronchoscope image-assisted optimization system based on machine learning according to claim 2, characterized in that the second expansion window is obtained according to the deviation of the first expansion window, and the specific method includes: When the deviation of the first extended window is greater than 0, the outermost circle of pixels in the first extended window is obtained and recorded as the peripheral pixels of the first extended window. Any pixel is selected and removed from the peripheral pixels of the first extended window to obtain the second extended window; When the deviation of the first expansion window is less than 0, obtain the peripheral pixels of the first expansion window, obtain the pixels adjacent to the peripheral pixels of the first expansion window and do not belong to the first expansion window, and record them as the first expansion window. For outer adjacent pixels, any outer adjacent pixel of the first extended window is added to the first extended window to obtain a second extended window. 4. The machine learning-based bronchoscope image auxiliary optimization system according to claim 1, characterized in that the corrected degree of confusion of each pixel is obtained based on the initial degree of confusion of each pixel and the gradient of each pixel in the reference area. , the specific methods included are: Obtain the gradient eccentricity angle of the reference pixel of each pixel according to the gradient of each pixel in the reference area of each pixel; The calculation method for obtaining the corrected confusion degree of each pixel based on the gradient eccentricity angle of the reference pixel and the initial confusion degree of each pixel is: Among them, d _i,q represents the Euclidean distance between the i-th pixel and the q-th pixel in the reference area, represents the gradient eccentricity angle of the q-th reference pixel of the i-th pixel, Q _i represents the number of pixels in the reference area of the i-th pixel, H _i represents the initial disorder degree of the i-th pixel, H′ _i represents the i-th Corrected clutter of pixels. 5. The machine learning-based bronchoscope image auxiliary optimization system according to claim 4, characterized in that the gradient eccentricity angle of the reference pixel of each pixel is obtained according to the gradient of each pixel in the reference area of each pixel, Specific methods included are: Record the pixels in the reference area of each pixel as the reference pixel of each pixel, record any reference pixel of each pixel as the target reference pixel of each pixel; obtain the vector composed of each pixel and the target reference pixel, The angle with the gradient direction of the target reference pixel is recorded as the gradient eccentricity angle between each pixel and the target reference pixel; Get the gradient eccentricity angle of the reference pixel for each pixel. 6. The machine learning-based bronchoscope image auxiliary optimization system according to claim 1, characterized in that the corrected clustering distance of every two pixels is obtained according to the corrected degree of confusion of each pixel, and the specific method includes: Among them, H′ _i represents the corrected confusion degree of the i-th pixel, H′ _j represents the corrected confusion degree of the j-th pixel, d _i,j represents the Euclidean distance between the i-th pixel and the j-th pixel, D ′ _i,j represents the modified clustering distance between the i-th pixel and the j-th pixel. 7. The machine learning-based bronchoscope image auxiliary optimization system according to claim 1, characterized in that the pixels in the bronchial image are clustered and analyzed according to the corrected clustering distance of every two pixels to obtain multiple independent texture areas. , the specific methods included are: Set the clustering parameters of the pixels. Based on the clustering parameters and the modified clustering distance between pixels, use the ISODATA algorithm to cluster the pixels to obtain several independent texture areas. 8. The machine learning-based bronchoscope image auxiliary optimization system according to claim 1, characterized in that the bronchial image is enhanced according to the correction degree of confusion of each pixel in each independent texture area to obtain an enhanced bronchial image. , the specific methods included are: Obtain the mean value of the corrected confusion degree of all pixels in each independent texture area, and record it as the corrected confusion degree of each independent texture area; multiply the grayscale value of each pixel in each independent texture area in the bronchial image by the independent texture area Image enhancement is achieved by correcting the degree of confusion to obtain enhanced bronchial images.