Self-adaptive crack identification and quantification method for periphery of open stope

Abstract

The invention discloses a self-adaptive crack identification and quantification method for the periphery of an open stope. The method comprises an image preprocessing step and a crack identification step. According to the method, aiming at the complexity of an open stope environment, a Gaussian pyramid is adopted to carry out multi-scale analysis on an image, and gray scale equalization is realized through gamma transformation so as to adapt to crack identification requirements under different illumination conditions; according to the crack recognition part, an improved difference entropy is used for optimizing a local threshold segmentation algorithm, high-efficiency and high-precision detection of the stope crack is achieved, the segmentation threshold can be dynamically determined according to gray level distribution of an image local area, and the accuracy and robustness of crack recognition are effectively improved; according to the method, the automation level of crack detection is greatly improved, powerful technical support is provided for safety management and risk early warning of the open stope, and the method has important practical application value and wide market prospects.

Description

Technical Field

[0001] The invention belongs to the technical field of computer vision, and in particular relates to an adaptive crack identification and quantification method around an open pit. Background Art

[0002] Mine slope safety is one of the core issues in open-pit mining operations. The stability of the slope is directly related to the safety and production efficiency of mining operations. Slope instability may lead to geological disasters such as landslides and mud-rock flows, which not only threaten the lives of workers, but also may cause huge economic losses and environmental damage. Therefore, effective stability monitoring and management of mine slopes is an important measure to ensure safe production in mining areas.

[0003] As an early sign of slope instability, crack detection is crucial for preventing geological disasters. Although crack detection based on image processing technology has improved detection efficiency to a certain extent, these technologies often lack sufficient adaptive capabilities and are difficult to cope with complex and changeable conditions in mining environments, and the recognition accuracy still needs to be improved. Although deep learning technology has made significant progress in the field of image recognition, its model cannot guarantee generalization and accuracy in the application of crack detection in mining slopes due to the complexity of data and the variability of the environment. In addition, the dependence of deep learning models on a large amount of labeled data and the high demand for computing resources are also a major difficulty in their application in mining scenarios. For example, Zhao Yiqing et al. discussed the application of deep learning in geotechnical engineering crack identification, but also pointed out the dependence of deep learning models on a large amount of labeled data, high consumption of computing resources, and low generalization ability in complex environments. Jing Ying et al. proposed an intelligent crack detection algorithm based on an improved Mask R-CNN. The article pointed out that when traditional image processing algorithms process crack images in open-pit mining slopes, they will have unsatisfactory detection results due to problems such as complex background textures, diverse noise, and irregular crack distribution. Although deep learning methods perform well in recognition accuracy, algorithm robustness, and generalization ability, in practical applications, model training requires a lot of computing resources and a complex training process, which is impractical in resource-constrained field environments.

[0004] In view of this, it is particularly important to develop a detection method that can automatically adapt to the mining environment and accurately identify cracks. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide an adaptive crack identification and quantification method around an open-pit mine. Through high-accuracy crack detection, an efficient and reliable solution is provided for the stability monitoring and management of the slopes in the mining area. This method does not require human intervention, can monitor the slope status in real time, and detect cracks in time, thereby providing strong technical support for safe production in the mining area.

[0006] The object of the present invention is achieved by comprising the following steps: S100, image preprocessing: first convert the original image into a grayscale image, then construct a Gaussian image pyramid, use Gaussian filtering to smooth the image, and use adaptive gamma transform to enhance the image; S200, crack identification: S201, improved difference entropy: fitting the parameters of the local threshold segmentation algorithm based on the improved difference entropy method, specifically: S2011. Calculate the weighted difference between the center pixel and other pixels based on the sliding window and Gaussian weight matrix: Assume that the image uses a sliding window of arbitrary radius to traverse all pixels of the image, and obtain the difference between the grayscale values ​​of the center pixel of the window and the remaining pixels, that is, the grayscale difference matrix (DMij) of a single pixel; wherein the grayscale difference value range is [0,255], and the difference value of the center element of the grayscale difference matrix is ​​0; based on the probability density function of the two-dimensional Gaussian distribution, generate a weight matrix of size n×n; since the center of the difference matrix is ​​0, the weight of the center of the matrix is ​​distributed to other elements according to the size of the weights of the remaining elements; the probability density function of the two-dimensional Gaussian distribution is: Where x=[x,y]T is a two-dimensional random variable, μ is the mean vector, Σ is the covariance matrix, and |Σ| represents the determinant of the covariance matrix; S2012, establishing a distribution law of differences: using the grayscale difference matrix (DM) and weight matrix of each pixel, recording the weight number of each difference in the grayscale difference matrix (DM), that is, the weighted number of times, and then obtaining the distribution X of the grayscale value difference between each pixel of the image and its surrounding pixels; S2013. Substitute the improved difference entropy formula: The calculation formula of the improved difference entropy H(x) is: Where i is the value of the difference, n is the number of gray levels, and P(x) is the distribution rate of the difference; S202, parameter fitting: S2021. Through the local threshold segmentation parameter analysis experiment, obtain sufficient optimal sensitivity and optimal minimum standard deviation threshold data when the crack recognition effect is best; the local threshold segmentation parameter analysis experiment is specifically performed by applying the local threshold segmentation algorithm to a set of experimental images, and finding the optimal segmentation parameters for each image when the segmentation effect is best.

[0007] S2022. Based on the least squares fitting method, the difference entropy of each experimental image is fitted with the optimal parameters of the image, and a functional relationship is established; S203, adaptive threshold segmentation based on local threshold segmentation algorithm: S2031, using a sliding window of definable size, calculating the mean and standard deviation of the grayscale values ​​of pixels in the window; S2032, calculating the mean and mode of the entire image, and comparing the mean and mode. If the mean is smaller, the foreground grayscale value is lower than the background, and if the mean is larger, the foreground grayscale value is higher than the background, thereby distinguishing whether the image foreground has a higher grayscale value or a lower grayscale value; S2033. Segment the foreground and background in the sliding window using standard deviation segmentation, brightness parameters, and sensitivity parameters, where the foreground is the crack.

[0008] In step S202, the sensitivity parameter is the sensitivity of the local threshold segmentation algorithm based on the standard deviation when segmenting the image; the minimum calibration difference parameter is the minimum threshold value of the local threshold segmentation algorithm based on the standard deviation when segmenting the image; since the standard deviation is used to measure the degree of discreteness of a set of pixel values ​​and is closely related to the difference entropy, the optimal sensitivity of a sufficient amount of images is fitted with the optimal minimum standard deviation threshold and the difference entropy based on the least squares fitting method, and a functional relationship is established.

[0009] Preferably, in step S100, in constructing the Gaussian image pyramid, the grayscale image is downsampled three times to generate a four-layer pyramid structure, and the image of each layer of the structure is divided into the same size.

[0010] The grayscale mean values ​​of different parts of the original grayscale image are uneven. The present invention uses adaptive gamma transformation to perform grayscale stretching, and realizes the adaptation of gamma transformation parameters based on the average grayscale value of the image; preferably, in step S100, when using adaptive gamma transformation for image enhancement, the adaptive gamma transformation is used to stretch the grayscale of all images, and the gamma transformation formula is: Among them, P out (x, y) is the pixel value at (x, y) of the output; A is a constant (gain) used to adjust the range of output brightness; P in (x, y) is the pixel value at the input (x, y); γ is the gamma value, which is a positive real number, 0≤γ≤5, used to control the shape of the gamma curve. When γ=1, the gamma curve is a straight line, and the grayscale value of the image will not change after mapping; Calculate the adaptive gamma value according to the average gray value, the formula is: Among them, μ is the average gray value of the image, T is the standard gray value; when T=μ, the image does not change; t is a control parameter used to prevent the image from being over-enhanced; when the average gray value of the image is too low, use The calculated gamma value is too small, resulting in too high average gray level of the image after gamma transformation. Therefore, a parameter t is introduced. When μ < t, calculate the gamma value.

[0011] Preferably, it further includes the post-processing step of S300, specifically including the following steps: S301, Connected component analysis and screening: Perform connected component analysis on the foreground segmented in step S2033, and perform connected component screening based on the region area and the regional average gray value, where the region area threshold is positively correlated with the image size; S302, Construct a shadow area recognition method to remove the shadow area: Select any area and determine its skeleton; Select evenly spaced seed points on its skeleton, set the gray difference threshold, and perform directional growth along the normal; Calculate the shadow areas on both sides of the region through the growth degree of the seed points. If the area of the grown region is too large, then this region is a shadow. The calculation formula for the shadow area S is: where l i is the growth length of the i-th seed, and d is the set interval of the seeds; Remove the shadow area.

[0012] Preferably, it further includes S400 crack quantification, specifically including: (1) Crack classification: Based on the number of crack intersections and branches, distinguish the crack category as strip cracks or mesh cracks; (2) Direction and length calculation: Calculate the crack skeleton direction and length as the direction and length of the crack; (3) Width calculation: Select evenly spaced seed points along the skeleton, perform directional growth along the normal, and calculate the average growth length, which is the crack width; The above (1), (2), and (3) are not in a specific order.

[0013] Compared with the prior art, the present invention has the following technical effects: 1. In view of the complexity of the open-pit stope environment, the present invention uses a Gaussian pyramid to perform multi-scale analysis on the image and realizes gray level equalization through gamma transformation to meet the crack recognition requirements under different lighting conditions; 2. In the crack recognition part, the present invention uses an improved differential entropy to optimize the local threshold segmentation algorithm, aiming to achieve high-efficiency and high-precision detection of stope cracks. This method can dynamically determine the segmentation threshold according to the gray distribution of the local area of the image, effectively improving the accuracy and robustness of crack recognition; 3. The present invention also includes the post-processing after crack recognition. By means of connected component analysis and skeleton extraction, etc., the crack features are further refined to provide support for risk assessment and area demarcation; 4. The method of the present invention not only greatly improves the automation level of crack detection, but also provides strong technical support for the safety management and risk warning of open-pit mines. It has important practical application value and broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 Schematic diagram of two types of cracks: light and dark; Figure 3 Identify the schematic diagram for the shaded area; Figure 4 This is a schematic diagram for calculating crack width; Figure 5 This is the crack identification effect diagram. DETAILED DESCRIPTION

[0015] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0016] Example 1 As attached Figure 1 As shown, the adaptive crack identification and quantification method around the open pit in this embodiment includes the following steps: S100, image preprocessing: first convert the original image into a grayscale image, then construct a Gaussian image pyramid, use Gaussian filtering to smooth the image, and use adaptive gamma transform to enhance the image; S200, crack identification: S201, improved difference entropy: fitting the parameters of the local threshold segmentation algorithm based on the improved difference entropy method, specifically: S2011. Calculate the weighted difference between the center pixel and other pixels based on the sliding window and Gaussian weight matrix: Assume that the image uses a sliding window of arbitrary radius to traverse all pixels of the image, and obtain the difference between the grayscale values ​​of the center pixel of the window and the remaining pixels, that is, the grayscale difference matrix of a single pixel; wherein the grayscale difference value range is [0,255], and the difference value of the center element of the grayscale difference matrix is ​​0; based on the probability density function of the two-dimensional Gaussian distribution, generate a weight matrix of size n×n; since the center of the difference matrix is ​​0, the weight of the center of the matrix is ​​distributed to other elements according to the size of the weights of the remaining elements; the probability density function of the two-dimensional Gaussian distribution is: Where x=[x,y]T is a two-dimensional random variable, μ is the mean vector, Σ is the covariance matrix, and |Σ| represents the determinant of the covariance matrix; S2012, establishing a distribution law of differences: using the grayscale difference matrix and weight matrix of each pixel, recording the weight number of each difference in the grayscale difference matrix, that is, the weighted number of times, and then obtaining the distribution X of the grayscale value difference between each pixel of the image and its surrounding pixels; S2013. Substitute the improved difference entropy formula: The calculation formula of the improved difference entropy H(x) is: Where i is the value of the difference, n is the number of gray levels, and P(x) is the distribution rate of the difference; S202, parameter fitting: S2021. Through the local threshold segmentation parameter analysis experiment, obtain sufficient optimal sensitivity and optimal minimum standard deviation threshold data when the crack recognition effect is best; the local threshold segmentation parameter analysis experiment is specifically performed by applying the local threshold segmentation algorithm to a set of experimental images, and finding the optimal segmentation parameters for each image when the segmentation effect is best.

[0017] S2022. Based on the least squares fitting method, the difference entropy of each experimental image is fitted with the optimal parameters of the image, and a functional relationship is established; S203, adaptive threshold segmentation based on local threshold segmentation algorithm: S2031, using a sliding window of definable size, calculating the mean and standard deviation of the grayscale values ​​of pixels in the window; S2032, calculate the mean and mode of the entire image, compare the mean and mode, if the mean is smaller, the gray value of the foreground (crack) is lower than the background, and if the mean is larger, the gray value of the foreground (crack) is higher than the background, so as to distinguish whether the image foreground has a higher gray value or a lower gray value; see the attached Figure 2 As shown; S2033, using standard deviation segmentation, brightness parameter and sensitivity parameter, segment the foreground and background in the sliding window, the foreground is the crack, see the attached Figure 5 As shown, Figure 5 There are 4 groups of identification results in total, and the colored lines are the identified cracks.

[0018] Example 2 The adaptive crack identification and quantification method around an open pit in this embodiment is based on the embodiment 1. In step S100, in constructing a Gaussian image pyramid, the grayscale image is downsampled three times to generate a four-layer pyramid structure, and the image of each layer of the structure is divided into the same size.

[0019] Example 3 The adaptive crack identification and quantification method around the open pit in this embodiment is based on the embodiment 2. In step S100, the adaptive gamma transform is used to perform grayscale stretching on all images during image enhancement. The gamma transform formula is: Among them, P out (x, y) is the pixel value at (x, y) of the output; A is a constant used to adjust the range of output brightness; P in (x, y) is the pixel value at the input (x, y); γ is the gamma value, 0≤γ≤5, which is used to control the shape of the gamma curve. When γ=1, the gamma curve is a straight line, and the grayscale value of the image will not change after mapping; Calculate the adaptive gamma value according to the average gray value, the formula is: Among them, μ is the average gray value of the image; T is the standard gray value, and t is the control parameter; when T=μ, the image does not change.

[0020] Example 4 The adaptive crack identification and quantification method around the open pit in this embodiment is based on the embodiment 3, and further includes a post-processing step S300, specifically including the following steps: S301, connected domain analysis and screening: Perform connected domain analysis on the foreground segmented in step S2033, and screen the connected domain based on the area and average gray value of the region, wherein the area threshold is positively correlated with the image size; S302, construct a shadow area recognition method to remove the shadow area: select an arbitrary area and determine its skeleton; select evenly spaced seed points on its skeleton, set the grayscale difference threshold, and perform directional growth along the normal; calculate the shadow area on both sides of the area according to the growth degree of the seed points. If the area of ​​the grown area is too large, the area is a shadow. The calculation formula of the shadow area S is: Among them, l i is the growth length of the ith seed, and d is the setting interval of the seeds; Remove the shadow area; see the attached Figure 3 shown.

[0021] Example 5 As attached Figure 4 As shown, the adaptive crack identification and quantification method around the open pit in this embodiment is based on the embodiment 4 and also includes S400 crack quantification, which specifically includes: (1) Crack classification: Based on the number of crack intersections and branches, the cracks are classified into strip cracks or network cracks; (2) Direction and length calculation: The direction and length of the crack skeleton are calculated as the direction and length of the crack; (3) Width calculation: uniformly spaced seed points are selected along the skeleton, and directional growth is performed along the normal line. The average growth length is calculated, which is the crack width. The above (1), (2), and (3) are in no particular order.

Claims

1. An adaptive crack identification and quantification method around an open pit, characterized in that The following steps are involved: S100, image preprocessing: first convert the original image into a grayscale image, then construct a Gaussian image pyramid, use Gaussian filtering to smooth the image, and use adaptive gamma transform to enhance the image; S200, crack identification: S201, improved difference entropy: fitting the parameters of the local threshold segmentation algorithm based on the improved difference entropy method, specifically: S2011. Calculate the weighted difference between the center pixel and other pixels based on the sliding window and Gaussian weight matrix: Assume that the image uses a sliding window of arbitrary radius to traverse all pixels of the image, and obtain the difference between the grayscale values ​​of the center pixel of the window and the remaining pixels, that is, the grayscale difference matrix of a single pixel; wherein the grayscale difference value range is [0,255], and the difference value of the center element of the grayscale difference matrix is ​​0; based on the probability density function of the two-dimensional Gaussian distribution, generate a weight matrix of size n×n; since the center of the difference matrix is ​​0, the weight of the center of the matrix is ​​distributed to other elements according to the size of the weights of the remaining elements; the probability density function of the two-dimensional Gaussian distribution is: Where x=[x,y]T is a two-dimensional random variable, μ is the mean vector, Σ is the covariance matrix, and |Σ| represents the determinant of the covariance matrix; S2012, establishing a distribution law of differences: using the grayscale difference matrix and weight matrix of each pixel, recording the weight number of each difference in the grayscale difference matrix, that is, the weighted number of times, and then obtaining the distribution X of the grayscale value difference between each pixel of the image and its surrounding pixels; S2013. Substitute the improved difference entropy formula: The calculation formula of the improved difference entropy H(x) is: Where i is the value of the difference, n is the number of gray levels, and P(x) is the distribution rate of the difference; S202, parameter fitting: S2021. Through local threshold segmentation parameter analysis experiments, obtain sufficient optimal sensitivity and optimal minimum standard deviation threshold data when the crack identification effect is best; S2022. Based on the least squares fitting method, the difference entropy of each experimental image is fitted with the optimal parameters of the image, and a functional relationship is established; S203, adaptive threshold segmentation based on local threshold segmentation algorithm: S2031, using a sliding window of definable size to calculate the mean and standard deviation of the grayscale values ​​of pixels in the window; S2032, calculating the mean and mode of the entire image, and comparing the mean and mode. If the mean is smaller, the foreground grayscale value is lower than the background, and if the mean is larger, the foreground grayscale value is higher than the background, thereby distinguishing whether the image foreground has a higher grayscale value or a lower grayscale value; S2033. Segment the foreground and background in the sliding window using standard deviation segmentation, brightness parameters, and sensitivity parameters, where the foreground is the crack.

2. The method for adaptive crack identification and quantification around an open pit according to claim 1, characterized in that In step S100, in constructing the Gaussian image pyramid, the grayscale image is downsampled three times to generate a four-layer pyramid structure, and the image of each layer of the structure is divided into the same size.

3. The method for adaptive crack identification and quantification around an open pit according to claim 1, characterized in that In step S100, in the process of image enhancement using adaptive gamma transform, the grayscale of all images is stretched using adaptive gamma transform. The gamma transform formula is: Among them, P out (x, y) is the pixel value at (x, y) of the output; A is a constant used to adjust the range of output brightness; P in (x, y) is the pixel value at the input (x, y); γ is the gamma value, 0≤γ≤5, which is used to control the shape of the gamma curve. When γ=1, the gamma curve is a straight line, and the grayscale value of the image will not change after mapping; Calculate the adaptive gamma value based on the average gray value. The formula is: Among them, μ is the average gray value of the image; T is the standard gray value, and t is the control parameter; when T=μ, the image does not change.

4. The method for adaptive crack identification and quantification around an open pit according to claim 1, characterized in that The step S300 also includes a post-processing step, which specifically includes the following steps: S301, connected domain analysis and screening: Perform connected domain analysis on the foreground segmented in step S2033, and screen the connected domain based on the area and average gray value of the region, wherein the area threshold is positively correlated with the image size; S302, construct a shadow area recognition method to remove the shadow area: select an arbitrary area and determine its skeleton; select evenly spaced seed points on its skeleton, set the grayscale difference threshold, and perform directional growth along the normal; calculate the shadow area on both sides of the area according to the growth degree of the seed points. If the area of ​​the grown area is too large, the area is a shadow. The calculation formula of the shadow area S is: Among them, l i is the growth length of the ith seed, and d is the setting interval of the seeds; Remove the shadow area.

5. The method for adaptive crack identification and quantification around an open pit according to claim 1, characterized in that Also included is S400 crack quantification, specifically including: (1) Crack classification: Based on the number of crack intersections and branches, the cracks are classified into strip cracks or network cracks; (2) Direction and length calculation: The direction and length of the crack skeleton are calculated as the direction and length of the crack; (3) Width calculation: uniformly spaced seed points are selected along the skeleton, and directional growth is performed along the normal line. The average growth length is calculated, which is the crack width. The above (1), (2), and (3) are in no particular order.

Images