CN117576137A - Rock slice image edge detection method based on improved canny algorithm - Google Patents

Abstract

The invention discloses a rock slice image edge detection method based on an improved canny algorithm, which comprises the steps of collecting an original rock slice image and performing image processing; establishing and training a canny algorithm edge intelligent detection model based on bird waiting algorithm optimization; searching an optimal threshold of the target image by using the model, so as to obtain an edge image; the Gaussian noise and the spiced salt noise contained in the rock slice image are effectively removed through the median-wiener mixed filter, the interference of the noise on an edge detection result is reduced, meanwhile, the mixed filter is used for replacing the traditional Gaussian filter, the edge of the image can be well reserved, the problem that the Gaussian filter causes blurring of the edge of the image is effectively solved, and as many edge characteristics as possible are reserved; the accuracy, the effectiveness and the self-adaptability of the rock slice image edge detection can be improved; provides a new idea for the research and engineering application of the subsequent rock slice image edge detection.

Description

An edge detection method for rock slice images based on improved canny algorithm

Technical field

The invention relates to the field of image detection technology, and specifically relates to an edge detection method for rock slice images based on an improved canny algorithm.

Background technique

In the petroleum geological exploration industry, accurate edge identification and particle segmentation of rock thin section sequence images are the prerequisite for the analysis and identification of rock mineral components. The pores, dissolution and regional structures in rock minerals are complex and irregular, which makes edge extraction and particle segmentation of rock minerals difficult. In order to better distinguish the particles and edges of rock minerals, digital image processing technology is used to perform edge extraction and particle segmentation on rock thin section image sequences. The accurate extraction of particle edges is an important prerequisite for later particle segmentation and rock mineral analysis, which directly affects the accuracy of the next step of rock mineral properties research, oil and gas reservoir deposition, and reservoir comprehensive evaluation.

According to the literature reviewed, most of the existing image edge detection methods are based on traditional detection methods, such as the gradient operator method, which only predicts edges based on color, text and other low-level features. Although it can extract image edges to a certain extent, it has There are still problems with noise suppression, edge localization and handling of fine edges. Among them, the canny algorithm is one of the most commonly used traditional detection methods. After improvements by many scholars, the canny algorithm can achieve higher-precision image edge detection. However, there are still problems that cannot be ignored if used for edge detection of rock thin slice images, such as: The primary task of the canny algorithm is to remove noise. The surface texture of rock particles in rock thin section images is complex, and there are many noise points inside the particles and near the edges. In particular, the canny algorithm is susceptible to interference from salt-and-pepper noise and Gaussian noise. Some improved methods in the literature only target One of the noises is filtered, causing the results to be interfered by noise and reducing the edge detection accuracy; for setting thresholds in the process of connecting edges in dual-threshold detection, some literature uses a manual threshold setting method that lacks adaptability and is based on experience, and some literature uses OTSU, etc. Global threshold method, but this method cannot obtain clear and effective edge detection results for images with uneven brightness distribution such as rock slice images.

In response to the above problems, an edge detection method for rock slice images based on the improved canny algorithm is proposed, which can improve the accuracy, effectiveness and adaptability of edge detection for rock slice images.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings of the existing technology and provide an edge detection method for rock slice images based on an improved canny algorithm, which includes collecting original rock slice images and performing image processing; and an intelligent edge detection model based on the canny algorithm optimized by the migratory bird algorithm. Establish and train; use the model to find the optimal threshold of the target image to obtain the edge image; use the median-Wiener hybrid filter to effectively remove Gaussian noise and salt-and-pepper noise contained in rock slice images, reducing noise for edge detection At the same time, using this hybrid filter to replace the traditional Gaussian filter can also well preserve the edges of the image, effectively solving the problem that the Gaussian filter will cause blurred image edges, and retaining as many edge features as possible; It can improve the accuracy, effectiveness and adaptability of edge detection in rock thin section images; it provides new ideas for subsequent research on edge detection in rock thin section images and engineering applications.

In order to achieve the above technical effects, the following technical solutions are adopted:

An edge detection method for rock slice images based on an improved canny algorithm, including the following steps:

Step S1: Collect original rock slice images, establish a feature database, and select training set and test set data;

Step S2: Original rock slice image processing, including the following steps:

S21: Grayscale the collected original thin section image;

S22: Perform image space transformation on the grayscale image, unify the original image size to 512×512 pixels, and obtain the transformed image;

S23: Perform image enhancement on the transformed image to obtain an enhanced image;

Step S3: Establish a canny algorithm edge intelligent detection model based on the migratory bird algorithm optimization:

Input the training set and test set data, use the migratory bird algorithm to optimize and set the high and low thresholds of the canny edge detection algorithm, and obtain the trained canny algorithm edge intelligent detection model optimized based on the migratory bird algorithm; specifically including:

S31: Filter the image using a hybrid filter;

S32: Calculate the module and gradient direction of the gradient;

S33: Perform non-maximum suppression on gradient images;

S34: Use the optimal double threshold after optimization for edge connection;

Step S4: Obtain the edge image based on the best high and low thresholds obtained by the model.

Further, in step S21, a weighted average method is used to weight and average the three RGB components with different weights to obtain a grayscale image, specifically as follows:

S211: Read the pixel value of the original rock slice image;

S212: Perform a weighted average of the red, green, and blue components of each pixel to obtain the grayscale value;

S213: Assign the gray value to the corresponding pixel;

f(x,y)=0.299R(x,y)+0.578G(x,y)+0.114B(x,y)

Among them: f (x, y) represents the pixel located at the spatial position (x, y). The R component, G component, and B component values of the pixel are R (x, y), G (x, y), and B respectively. (x, y);

S214: Output grayscale image.

Further, the image space transformation method in step S22 includes an image deformation processing method and a displacement processing method: the displacement processing includes image translation transformation, image mirror transformation and image rotation transformation; the deformation processing includes image miscut transformation and cropping transformation. , scaling transformation.

Further, the scaling transformation will change the size of the image, and the grayscale image size is unified to 512×512 pixels through the bicubic interpolation method. The steps are as follows:

S221: Assume that the size of the source image A is m×n, m and n are the length and width of the image respectively, the size of the scaled target image B is 512×512, and the coordinates B(x, y) on the scaled image B are obtained according to the ratio. ) corresponds to the coordinates on A as A(x, y)=A[X×(m/512,), Y×(n/512,)];

S222: The coordinate position P corresponding to image A in S221 may have a decimal part. At this time, find the integer coordinate of the pixel closest to it. Let the coordinate be P(i+u, j+v), where i and j are integers. u, v are positive or negative decimals. At this time, P is the position of image B in the corresponding source image at (X, Y). The 16 pixels closest to P are selected as parameters for calculating the pixel value of the target image;

S223: For example, assume a 4×4 matrix composed of 16 pixel points. The target point P found in S222 is located at (2, 2). Then the value ranges of the horizontal and vertical coordinates of the 16 points are [i-1, i+2], [j-1, j+2], the subsequent calculation results are as follows;

S224: Calculate the pixel value of point P, the formula is as follows:

Among them: w(x) is the weight calculation formula, the value of a is generally -1, -0.75 or -0.5; row is the abscissa value range of 16 pixels; col is the ordinate value range of 16 pixels ; f(i+row, j+col) is the original pixel value of 16 pixels; F(i+u, j+v) is the interpolated pixel value of P point; w(row-u) is the lateral distance Weight; w(col-v) is the weight of vertical distance.

Further, in step S23, the image enhancement uses a histogram equalization method to enhance the contrast of the image while preventing overfitting. The steps are as follows:

S231: Count the histogram of the original image, calculate the number of occurrences of each pixel value, normalize it to [0, 1], and obtain the frequency of each pixel value;

S232: Calculate the cumulative distribution function (CDF), which represents the probability of each pixel value appearing in the original image. The formula is as follows:

Among them: CDF(i) is the cumulative distribution of gray value i; P(j) represents the frequency of pixels with gray value j appearing in the image.

S233: Calculate the equalized pixel value, map each pixel value in the original image to a new pixel value, so that the equalized histogram is approximately a uniformly distributed histogram; this mapping function can be calculated by the following formula calculate:

Among them: H(i) represents the mapped pixel value; L represents the range of pixel values; CDF(min) represents the cumulative distribution of the minimum pixel value in the original image; CDF(i) is the cumulative distribution of the gray value i; round Used for rounding; CDF(i) is the cumulative distribution of gray value i.

Further, the specific method of using a hybrid filter to filter the image in step S31 is:

The median-Wiener hybrid filter is used to filter out the salt-and-pepper noise and Gaussian noise contained in the image. The median filter can filter out the salt-and-pepper noise well and retain the edge information of the image. The Wiener filter can filter out the salt-and-pepper noise well and retain the edge information of the image. Remove Gaussian noise; therefore, the gray value based on salt and pepper noise is the characteristic of the gray extreme point in its field. Distinguish the salt and pepper noise from the effective signal points, perform median filtering on the salt and pepper noise, retain the signal points, and finally use Wiener Filter filters the entire image.

Further, in step S32, the sobel operator is used to calculate the module of the gradient. The specific formula is as follows:

In the formula: S _x and S _y are sobel operators; G _x is the pixel gradient matrix in the x direction of the image; G _y is the pixel gradient matrix in the y direction of the image; I is the grayscale image matrix; * here means mutual Correlation operation, and the origin of the image matrix coordinate system is at the upper left corner, and the positive x direction is from left to right, and the positive y direction is from top to bottom;

In the formula: M (x, y) represents the gradient strength of the pixel; θ (x, y) represents the gradient direction of the pixel; G _x is the pixel gradient matrix in the x direction of the image; G _y is the y direction of the image pixel gradient matrix.

Further, the specific method for performing non-maximum suppression on the gradient image in step S33 is:

Compare the gradient modulus value of the current pixel with the gradient modulus value of its two neighbor pixels along the direction. The direction is 8 areas, each area is a range of 45°: If the gradient modulus value of the current pixel is greater than the neighborhood in the same direction If the pixel gradient modulus value indicates that the point may be an edge point, the pixel value of the point is retained; otherwise, it is a pseudo edge, and non-maximum values are suppressed and the pixel value of the point is marked as 0.

Further, the specific method of using the optimal double threshold after optimization to perform edge connection in step S34 is:

After non-maximum suppression, non-edge points are marked as 0, and possible edge points retain their pixel values. However, the results still contain many false edges caused by noise or other reasons. At this time, the migratory bird optimization algorithm is used to The double thresholds of the algorithm are optimized and set. The steps are as follows:

S341: Initialize the population, initialize the leading bird individual, the following bird individual, the number of tours K, set the maximum number of iterations U _max and the maximum number of tours K _max , and divide the population structure into the leading bird and the left and right queues. Assume that the leading bird is Initial high and low thresholds _TH and _TL , where the fixed high and low threshold ratio is _TH : _TL = 3:1;

S342: The leader bird evolves. The leader bird searches for its own neighborhood solution and replaces itself with a double threshold that can obtain the best edge detection effect. The remaining unused domain solutions are passed to the next individual;

S343: Follow Feiyao to evolve, search with Feiyao for its own neighborhood solutions and the unused neighborhood solution sets generated by the individuals ahead of itself during the last search process, and find the best edge in these neighborhood solution sets. The double-threshold optimal solution of the detection effect is used to replace itself. Similarly, each unused neighborhood solution generated by the flying bird is passed to the next individual;

S344: Determine whether the number of tours reaches the maximum number of tours: if the number of tours K ≤ K _max , go to step S342; if the number of tours K > K _max , go to step S345;

S345: Replacement of the leading bird. When the number of tours reaches the maximum, the initial leading bird is moved to the end of the team. According to the edge detection effect of the selected dual threshold, the following bird in the left queue or the right queue behind the initial leading bird is selected as the new One leader leads the bird, initializes the number of tours, and then starts the next search process;

S346: Determine whether the number of iterations reaches the maximum number of iterations: if the number of iterations U≤U _max , go to step S342; if the number of iterations U>U _max , it means that the double threshold of the leader bird reference in the population is global at this time The optimal high threshold T _H and the global optimal low threshold T _L satisfy the termination condition;

S347: Select the best high and low thresholds T _H and T _L selected in S346, and judge: Let the point (x, y), g _N (x, y) be the image midpoint (x, y) after non-maximum suppression of the image. y), if the pixel value g _N (x, y) ≥ T _H , the point is marked as a strong edge point, that is, the point must be a point on the edge; g _N (x, y) < T _L , then the point must not be a point on the edge, and suppress it; T _L ≤ g _N (x, y) < T _H , then it is marked as a weak edge point, and it is judged whether there are high-level points in the 8 neighborhoods of the point. If there is a pixel with a high threshold, then the point is an edge point, otherwise it is not an edge point.

The beneficial effects of the present invention are:

The present invention provides an edge detection method for rock slice images based on an improved canny algorithm. Compared with the existing technical solutions, the invention has the following characteristics: the Gaussian noise contained in the rock slice image is detected through a median-Wiener hybrid filter. The effective removal of salt and pepper noise reduces the interference of noise on edge detection results. At the same time, using this hybrid filter to replace the traditional Gaussian filter can also well retain the edges of the image, effectively solving the problem that the Gaussian filter will cause blurred image edges. solve the problem and retain as many edge features as possible; the migratory bird algorithm is used to automatically optimize and set the double threshold of the canny algorithm to enhance the adaptability of the algorithm and improve the accuracy of edge detection results.

Description of the drawings

Figure 1 is a flow chart of a rock thin section image edge detection method based on an improved canny algorithm provided by an embodiment of the present invention;

Figure 2 is a training flow chart of the canny algorithm edge intelligent detection model based on the optimization of the migratory bird algorithm provided by the embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with examples. It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention.

It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It should be noted that the terms used herein are for the purpose of describing specific embodiments only, and are not intended to limit the exemplary embodiments according to the present invention. As used herein, the singular forms are also intended to include the plural forms unless the context clearly indicates otherwise. Furthermore, it will be understood that when the terms "comprises" and/or "includes" are used in this specification, they indicate There are features, steps, operations and/or combinations thereof.

Example 1:

As shown in Figure 1, the present invention provides a rock slice image edge detection method based on an improved canny algorithm, including the following steps:

Step S1: Collect original rock slice images, establish a feature database, and select training set and test set data;

Step S2: Original rock slice image processing, including the following steps:

S21: Grayscale the collected original thin section image;

In step S21, a weighted average method is used to weight and average the three RGB components with different weights to obtain a grayscale image, specifically as follows:

S211: Read the pixel value of the original rock slice image;

S212: Perform a weighted average of the red, green, and blue components of each pixel to obtain the grayscale value;

S213: Assign the gray value to the corresponding pixel;

f(x,y)=0.299R(x,y)+0.578G(x,y)+0.114B(x,y)

Among them: f (x, y) represents the pixel located at the spatial position (x, y). The R component, G component, and B component values of the pixel are R (x, y), G (x, y), and B respectively. (x, y);

S214: Output grayscale image.

S22: Perform image space transformation on the grayscale image, unify the original image size to 512×512 pixels, and obtain the transformed image;

The image space transformation method in step S22 includes an image deformation processing method and a displacement processing method: the displacement processing includes image translation transformation, image mirror transformation and image rotation transformation; the deformation processing includes image miscut transformation, cropping transformation, and scaling transformation. .

The scaling transformation will change the size of the image, and the grayscale image size is unified to 512×512 pixels through the bicubic interpolation method. The steps are as follows:

S221: Assume that the size of the source image A is m×n, m and n are the length and width of the image respectively, the size of the scaled target image B is 512×512, and the coordinates B(x, y) on the scaled image B are obtained according to the ratio. ) corresponds to the coordinates on A as A(x, y)=A[X×(m/512,), Y×(n/512,)];

S222: The coordinate position P corresponding to image A in S221 may have a decimal part. At this time, find the integer coordinate of the pixel closest to it. Let the coordinate be P(i+u, j+v), where i and j are integers. u, v are positive or negative decimals. At this time, P is the position of image B in the corresponding source image at (X, Y). The 16 pixels closest to P are selected as parameters for calculating the pixel value of the target image;

S223: For example, assume a 4×4 matrix composed of 16 pixel points. The target point P found in S222 is located at (2, 2). Then the value ranges of the horizontal and vertical coordinates of the 16 points are [i-1, i+2], [j-1, j+2], the subsequent calculation results are as follows;

S224: Calculate the pixel value of point P, the formula is as follows:

Among them: w(x) is the weight calculation formula, the value of a is generally -1, -0.75 or -0.5; row is the abscissa value range of 16 pixels; col is the ordinate value range of 16 pixels ; f(i+row, j+col) is the original pixel value of 16 pixels; F(i+u, j+v) is the interpolated pixel value of P point; w(row-u) is the lateral distance Weight; w(col-v) is the weight of vertical distance.

S23: Perform image enhancement on the transformed image to obtain an enhanced image;

In step S23, the image enhancement uses a histogram equalization method to enhance the contrast of the image while preventing overfitting. The steps are as follows:

S231: Count the histogram of the original image, calculate the number of occurrences of each pixel value, normalize it to [0, 1], and obtain the frequency of each pixel value;

S232: Calculate the cumulative distribution function (CDF), which represents the probability of each pixel value appearing in the original image. The formula is as follows:

Among them: CDF(i) is the cumulative distribution of gray value i; P(j) represents the frequency of pixels with gray value j appearing in the image.

S233: Calculate the equalized pixel value, map each pixel value in the original image to a new pixel value, so that the equalized histogram is approximately a uniformly distributed histogram; this mapping function can be calculated by the following formula calculate:

Among them: H(i) represents the mapped pixel value; L represents the range of pixel values; CDF(min) represents the cumulative distribution of the minimum pixel value in the original image; CDF(i) is the cumulative distribution of the gray value i; round Used for rounding; CDF(i) is the cumulative distribution of gray value i.

Step S3: Establish a canny algorithm edge intelligent detection model based on the migratory bird algorithm optimization:

Input the training set and test set data, use the migratory bird algorithm to optimize and set the high and low thresholds of the canny edge detection algorithm, and obtain the trained canny algorithm edge intelligent detection model optimized based on the migratory bird algorithm; specifically including:

S31: Filter the image using a hybrid filter;

The specific method of filtering the image using a hybrid filter in step S31 is:

The median-Wiener hybrid filter is used to filter out the salt-and-pepper noise and Gaussian noise contained in the image. The median filter can filter out the salt-and-pepper noise well and retain the edge information of the image. The Wiener filter can filter out the salt-and-pepper noise well and retain the edge information of the image. Remove Gaussian noise; therefore, the gray value based on salt and pepper noise is the characteristic of the gray extreme point in its field. Distinguish the salt and pepper noise from the effective signal points, perform median filtering on the salt and pepper noise, retain the signal points, and finally use Wiener Filter filters the entire image.

S32: Calculate the module and gradient direction of the gradient;

In step S32, the sobel operator is used to calculate the module of the gradient. The specific formula is as follows:

In the formula: S _x and S _y are sobel operators; G _x is the pixel gradient matrix in the x direction of the image; G _y is the pixel gradient matrix in the y direction of the image; I is the grayscale image matrix; * here means mutual Correlation operation, and the origin of the image matrix coordinate system is at the upper left corner, and the positive x direction is from left to right, and the positive y direction is from top to bottom;

In the formula: M (x, y) represents the gradient strength of the pixel; θ (x, y) represents the gradient direction of the pixel; G _x is the pixel gradient matrix in the x direction of the image; G _y is the y direction of the image pixel gradient matrix

S33: Perform non-maximum suppression on gradient images;

The specific method for non-maximum suppression of the gradient image in step S33 is:

Compare the gradient modulus value of the current pixel with the gradient modulus value of its two neighbor pixels along the direction. The direction is 8 areas, each area is a range of 45°: If the gradient modulus value of the current pixel is greater than the neighborhood in the same direction If the pixel gradient modulus value indicates that the point may be an edge point, the pixel value of the point is retained; otherwise, it is a pseudo edge, and non-maximum values are suppressed and the pixel value of the point is marked as 0.

S34: Use the optimal double threshold after optimization for edge connection;

The specific method of using the optimal double threshold after optimization to perform edge connection in step S34 is:

After non-maximum suppression, non-edge points are marked as 0, and possible edge points retain their pixel values. However, the results still contain many false edges caused by noise or other reasons. At this time, the migratory bird optimization algorithm is used to The double thresholds of the algorithm are optimized and set. The steps are as follows:

S341: Initialize the population, initialize the leading bird individual, the following bird individual, the number of tours K, set the maximum number of iterations U _max and the maximum number of tours K _max , and divide the population structure into the leading bird and the left and right queues. Assume that the leading bird is Initial high and low thresholds _TH and _TL , where the fixed high and low threshold ratio is _TH : _TL = 3:1;

S342: The leader bird evolves. The leader bird searches for its own neighborhood solution and replaces itself with a double threshold that can obtain the best edge detection effect. The remaining unused domain solutions are passed to the next individual;

S343: Follow Feiyao to evolve, search with Feiyao for its own neighborhood solutions and the unused neighborhood solution sets generated by the individuals ahead of itself during the last search process, and find the best edge in these neighborhood solution sets. The double-threshold optimal solution of the detection effect is used to replace itself. Similarly, each unused neighborhood solution generated by the flying bird is passed to the next individual;

S344: Determine whether the number of tours reaches the maximum number of tours: if the number of tours K ≤ K _max , go to step S342; if the number of tours K > K _max , go to step S345;

S345: Replacement of the leading bird. When the number of tours reaches the maximum, the initial leading bird is moved to the end of the team. According to the edge detection effect of the selected dual threshold, the following bird in the left queue or the right queue behind the initial leading bird is selected as the new One leader leads the bird, initializes the number of tours, and then starts the next search process;

S346: Determine whether the number of iterations reaches the maximum number of iterations: if the number of iterations U≤U _max , go to step S342; if the number of iterations U>U _max , it means that the double threshold of the leader bird reference in the population is global at this time The optimal high threshold T _H and the global optimal low threshold T _L satisfy the termination condition;

S347: Select the best high and low thresholds T _H and T _L selected in S346, and judge: Let the point (x, y), g _N (x, y) be the image midpoint (x, y) after non-maximum suppression of the image. y), if the pixel value g _N (x, y) ≥ T _H , then the point is marked as a strong edge point, that is, the point must be a point on the edge; the pixel value g _N (x , y)＜T _L , then the point must not be a point on the edge, suppress it; T _L ≤ g _N (x, y)＜T _H , then it is marked as a weak edge point, and the 8-neighborhood of the point is determined Is there a pixel in the pixel that is higher than the high threshold? If so, then the point is an edge point, otherwise it is not an edge point.

Step S4: Obtain the edge image based on the best high and low thresholds obtained by the model.

In summary, the present invention discloses an edge detection method for rock slice images based on an improved canny algorithm, which includes collecting original rock slice images and performing image processing; establishing and training an intelligent edge detection model of the canny algorithm based on the optimization of the migratory bird algorithm; Use the model to find the optimal threshold of the target image to obtain the edge image; use the median-Wiener hybrid filter to effectively remove the Gaussian noise and salt-and-pepper noise contained in the rock slice image, reducing the interference of noise on the edge detection results. At the same time, using this hybrid filter to replace the traditional Gaussian filter can also well retain the edges of the image, effectively solving the problem that the Gaussian filter will cause blurred image edges, retaining as many edge features as possible; it can improve rock thin sections The accuracy, effectiveness and adaptability of image edge detection provide new ideas for subsequent research on edge detection of rock thin section images and engineering applications.

At this point, those skilled in the art realize that although the embodiments of the present invention have been shown and described in detail herein, without departing from the spirit and scope of the present invention, the content of the disclosure of the present invention can still be directly determined or deduced. There are many other variations or modifications of the principles of the invention. Accordingly, the scope of the invention should be understood and deemed to cover all such other variations or modifications.

Claims (9)

1. A rock slice image edge detection method based on an improved canny algorithm, characterized in that the detection method includes the following steps: Step S1: Collect original rock slice images, establish a feature database, and select training set and test set data; Step S2: Original rock slice image processing, including the following steps: S21: Grayscale the collected original thin section images; S22: Perform image space transformation on the grayscale image, unify the original image size to 512×512 pixels, and obtain the transformed image; S23: Perform image enhancement on the transformed image to obtain an enhanced image; Step S3: Establish a canny algorithm edge intelligent detection model based on the migratory bird algorithm optimization: Input the training set and test set data, use the migratory bird algorithm to optimize and set the high and low thresholds of the canny edge detection algorithm, and obtain the trained canny algorithm edge intelligent detection model optimized based on the migratory bird algorithm; specifically including: S31: Filter the image using a hybrid filter; S32: Calculate the module and gradient direction of the gradient; S33: Perform non-maximum suppression on gradient images; S34: Use the optimal double threshold after optimization for edge connection; Step S4: Obtain the edge image based on the best high and low thresholds obtained by the model. 2. A rock slice image edge detection method based on an improved canny algorithm as claimed in claim 1, characterized in that, in step S21, a weighted average method is used to weight the three components of RGB with different weights. Get the grayscale image, specifically: S211: Read the pixel value of the original rock slice image; S212: Perform a weighted average of the red, green, and blue components of each pixel to obtain the grayscale value; S213: Assign the gray value to the corresponding pixel; f(x,y)＝0.299R(x,y)+0.578G(x,y)+0.114B(x,y) Among them: f (x, y) represents the pixel located at the spatial position (x, y). The R component, G component, and B component values of the pixel are R (x, y), G (x, y), and B respectively. (x, y); S214: Output grayscale image. 3. A rock slice image edge detection method based on an improved canny algorithm as claimed in claim 1, characterized in that the image space transformation method in step S22 includes an image deformation processing method and a displacement processing method: displacement processing. The processing includes image translation transformation, image mirror transformation and image rotation transformation; the deformation processing includes image miscut transformation, cropping transformation and scaling transformation. 4. A rock slice image edge detection method based on an improved canny algorithm as claimed in claim 3, characterized in that the scaling transformation changes the size of the image, and the grayscale image size is unified by the bicubic interpolation method. 512×512 pixels, the steps are as follows: S221: Assume that the size of the source image A is m×n, m and n are the length and width of the image respectively, and the size of the scaled target image B is 512x512. According to the ratio, the corresponding coordinates B(x, y) on the scaled image B are obtained. The coordinates on A are A(x, y)=A[X×(m/512,), Y×(n/512,)]; S222: The coordinate position P corresponding to image A in S221 may have a decimal part. At this time, find the integer coordinate of the pixel closest to it. Let the coordinate be P(i+u,j+v), where i and j are integers. u, v are positive or negative decimals. At this time, P is the position of image B in the corresponding source image at (X, Y). The 16 pixels closest to P are selected as parameters for calculating the pixel value of the target image; S223: For example, assume a 4×4 matrix composed of 16 pixel points. The target point P found in S222 is located at (2,2). Then the value ranges of the horizontal and vertical coordinates of the 16 points are [i-1, i+2],[j-1,j+2], where i and j are integers, and the subsequent calculation results are as follows; S224: Calculate the pixel value of point P, the formula is as follows: Among them: w(x) is the weight calculation formula, the value of a is generally -1, -0.75 or -0.5; row is the abscissa value range of 16 pixels; col is the ordinate value range of 16 pixels ; f(i+row,j+col) is the original pixel value of 16 pixels; F(i+u,j+v) is the interpolated pixel value of P point; w(row-u) is the lateral distance Weight; w(col-v) is the weight of vertical distance. 5. A rock slice image edge detection method based on an improved canny algorithm as claimed in claim 1, characterized in that the image enhancement in step S23 adopts a histogram equalization method to enhance the contrast of the image while preventing overfitting. , the steps are as follows: S231: Count the histogram of the original image, calculate the number of occurrences of each pixel value, normalize it to [0, 1], and obtain the frequency of each pixel value; S232: Calculate the cumulative distribution function (CDF), which represents the probability of each pixel value appearing in the original image. The formula is as follows: Among them: CDF(i) is the cumulative distribution of gray value i; P(j) represents the frequency of pixels with gray value j appearing in the image. S233: Calculate the equalized pixel value, map each pixel value in the original image to a new pixel value, so that the equalized histogram is approximately a uniformly distributed histogram; this mapping function can be expressed by the following formula calculate: Among them: H(i) represents the mapped pixel value; L represents the range of pixel values; CDF(min) represents the cumulative distribution of the minimum pixel value in the original image; CDF(i) is the cumulative distribution of the gray value i; round Used for rounding; CDF(i) is the cumulative distribution of gray value i. 6. A rock slice image edge detection method based on an improved canny algorithm as claimed in claim 1, characterized in that the specific method of filtering the image using a hybrid filter in step S31 is: The median-Wiener hybrid filter is used to filter out the salt-and-pepper noise and Gaussian noise contained in the image. The median filter can filter out the salt-and-pepper noise well and retain the edge information of the image. The Wiener filter can filter out the salt-and-pepper noise well and preserve the edge information of the image. Remove Gaussian noise; therefore, the gray value based on salt and pepper noise is the characteristic of the gray extreme point in its field. Distinguish the salt and pepper noise from the effective signal points, perform median filtering on the salt and pepper noise, retain the signal points, and finally use Wiener Filter filters the entire image. 7. A kind of rock slice image edge detection method based on improved canny algorithm as claimed in claim 1, characterized in that, in the step S32, the sobel operator is used to calculate the module of the gradient, and the specific formula is as follows: In the formula: S _x and S _y are sobel operators; G _x is the pixel gradient matrix in the x direction of the image; G _y is the pixel gradient matrix in the y direction of the image; I is the grayscale image matrix; * here means mutual Correlation operation, and the origin of the image matrix coordinate system is at the upper left corner, and the positive x direction is from left to right, and the positive y direction is from top to bottom; In the formula: M (x, y) represents the gradient strength of the pixel; θ (x, y) represents the gradient direction of the pixel; G _x is the pixel gradient matrix in the x direction of the image; G _y is the y direction of the image pixel gradient matrix. 8. A rock slice image edge detection method based on an improved canny algorithm as claimed in claim 1, characterized in that the specific method for non-maximum suppression of the gradient image in step S33 is: Compare the gradient modulus value of the current pixel with the gradient modulus value of its two neighbor pixels along the direction. The direction is 8 areas, each area is a range of 45°: If the gradient modulus value of the current pixel is greater than the neighborhood in the same direction If the pixel gradient modulus value indicates that the point may be an edge point, the pixel value of the point is retained; otherwise, it is a pseudo edge, and non-maximum values are suppressed and the pixel value of the point is marked as 0. 9. A rock slice image edge detection method based on an improved canny algorithm as claimed in claim 1, characterized in that the specific method for edge connection using the optimal double threshold after optimization in step S34 is: After non-maximum suppression, non-edge points are marked as 0, and possible edge points retain their pixel values. However, the results still contain many false edges caused by noise or other reasons. At this time, the migratory bird optimization algorithm is used to The double thresholds of the algorithm are optimized and set. The steps are as follows: S341: Initialize the population, initialize the leading bird individual, the following bird individual, the number of tours K, set the maximum number of iterations U _max and the maximum number of tours K _max , and divide the population structure into the leading bird and the left and right queues. Assume that the leading bird is Initial high and low thresholds _TH and _TL , where the _{fixed high and low threshold ratio is TH:TL =} 3:1; S342: The leader bird evolves. The leader bird searches for its own neighborhood solution and replaces itself with a double threshold that can obtain the best edge detection effect. The remaining unused domain solutions are passed to the next individual; S343: Follow Feiyao to evolve, search with Feiyao for its own neighborhood solutions and the unused neighborhood solution sets generated by the individuals ahead of itself during the last search process, and find the best edge in these neighborhood solution sets. The double-threshold optimal solution of the detection effect is used to replace itself. Similarly, each unused neighborhood solution generated by the flying bird is passed to the next individual; S344: Determine whether the number of tours reaches the maximum number of tours: if the number of tours K ≤ K _max , go to step S342; if the number of tours K ≤ K _max , go to step S345; S345: Replacement of the leading bird. When the number of tours reaches the maximum value, the initial leading bird is moved to the end of the team. According to the edge detection effect of the selected dual threshold, the following bird in the left queue or the right queue behind the initial leading bird is selected as the new One leader leads the bird, initializes the number of tours, and then starts the next search process; S346: Determine whether the number of iterations reaches the maximum number of iterations: if the number of iterations U ≤ U _max , go to step S342; if the number of iterations U ≤ U _max , it means that the double threshold of the leading bird in the population is global at this time. The optimal high threshold T _H and the global optimal low threshold T _L satisfy the termination condition; S347: Select the best high and low thresholds T _H and T _L selected in S346, and judge: Let the points (x, y), g _{N (x, y)} be the image midpoint (x, y) after non-maximum suppression of the image. y), if the pixel value g _N (x, y) ≥ T _H , the point is marked as a strong edge point, that is, the point must be a point on the edge; g _N (x, y) < T _L , then the point must not be a point on the edge, and suppress it; T _L <g _N (x, y) < T _H , then it is marked as a weak edge point, and it is judged whether there is a high point in the 8 neighborhoods of the point If there is a pixel with a high threshold, then the point is an edge point, otherwise it is not an edge point.