Single-slit method stress detection cut identification and evaluation method based on improved LIME algorithm

Abstract

This invention discloses a method for identifying and evaluating cuts in single-slit stress detection based on an improved LIME algorithm, belonging to the field of cut recognition technology. The method includes acquiring a detection image of a cut in single-slit stress detection and performing distortion correction; converting the distortion-corrected image into a grayscale image and enhancing the grayscale image using the improved LIME algorithm; extracting the cut contour using an edge detection method, then performing minimum bounding rectangle processing on the extracted cut contour to calculate the cut width and length; calculating the deviation rate of the length and width based on the cut length and width, and evaluating the cut quality using the product of the two deviation rates. The identification and evaluation method provided by this solution can achieve rapid and accurate identification of cuts under different lighting conditions, extract relevant parameters from the identified image, and evaluate the cut quality, providing a reliable guarantee for single-slit stress detection.

Description

Technical Field

[0001] This invention relates to the field of cut recognition technology, specifically to a single-cut stress detection cut recognition and evaluation method based on an improved LIME algorithm. Background Technology

[0002] The single-slit method is a method for testing concrete stress. It only requires cutting a single slit to achieve the testing objective, which not only reduces testing costs but also improves accuracy. Currently, the single-slit method is widely used in stress testing of concrete and rock.

[0003] During application, it was found that the length or width of the slit is not stable to the ideal value when making the cut, which can introduce certain errors into the test results. Traditional manual measurement is difficult to define the edge of the slit, is time-consuming and has low accuracy; while using image recognition to identify the slit often faces problems such as low illumination, high noise and loss of detail. The LIME algorithm has certain advantages for image enhancement, but when using the traditional LIME algorithm for image processing, there are problems such as the illumination estimation process easily producing excessive smoothing at the slit end, fixed parameter values ​​being unable to adapt to different scenes, and serious noise accumulation during iterative enhancement.

[0004] Therefore, a single-slit stress test cut identification and evaluation method based on the improved LIME algorithm is proposed. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, the present invention provides a single-slit stress detection cut identification and evaluation method based on an improved LIME algorithm. This method solves the problems that existing methods cannot quickly and accurately identify cuts under different lighting conditions, nor can they evaluate the quality of the cuts.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] A method for stress detection cut identification and evaluation based on an improved LIME algorithm is provided, comprising the following steps:

[0008] S1. Obtain the detection image of the stress detection cut using the single-slit method and perform distortion correction on it;

[0009] S2. Convert the distortion-corrected image into a grayscale image, and use the improved LIME algorithm to enhance the grayscale image.

[0010] S3. Use edge detection to extract the contour of the cut, then process the extracted cut contour with the minimum bounding rectangle, and calculate the width and length of the cut.

[0011] S4. Calculate the deviation rates of length and width based on the cut length and width, and evaluate the cut quality by multiplying the two deviation rates.

[0012] Furthermore, the expressions for calculating the deviation rates of length and width are as follows:

[0013] ,

[0014] in, This refers to the length deviation rate; The width deviation rate is represented by L and W, which are the theoretical kerf length and width, respectively. and These represent the calculated length and width of the cut, respectively.

[0015] Furthermore, step S2 further includes:

[0016] S21. Convert the distortion-corrected image into a grayscale image, and then perform bilateral filtering on the grayscale image to reduce noise, thus obtaining a denoised image.

[0017] S22. Based on the enhanced image of the previous iteration, a guide image for generating the denoised image is generated using a single-slit saliency weighting mechanism. The enhanced image of the first iteration is the denoised image.

[0018] S23. Based on the guide image and the enhanced image from the previous iteration, an improved spatial variational guide filter is used to estimate the illumination component at each location in the guide image;

[0019] S24. Based on the guide image and illumination components, extract the reflection components at each location in the enhanced image of the previous iteration, and then use the gamma correction function to correct and enhance the enhanced image of the previous iteration.

[0020] S25. Perform bilateral filtering noise reduction on the corrected and enhanced image to obtain the enhanced image of the current iteration number. Then determine whether the current iteration number has reached the maximum iteration number. If yes, proceed to step S26; otherwise, return to step S22.

[0021] S26. Calculate the final enhanced image based on the enhanced image from the last iteration. :

[0022]

[0023] in, Let K be the enhanced image of the Kth iteration, where K is the maximum number of iterations.

[0024] Furthermore, the expression for the single-slit significance weighting mechanism is:

[0025] ,

[0026] in, Let p be the pixel value of the guide image in the k-th iteration; The pixel value at position p in the denoised image; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; It uses a 5×5 Gaussian convolution kernel; * indicates a convolution operation; The mixing coefficient; is the dynamic mixed weight; k is the current iteration number, which takes a value from 1 to K, and K is the total number of iterations.

[0027] Furthermore, the expression for estimating the illumination component at each location in the guide image using an improved spatial variational guided filter is as follows:

[0028]

[0029]

[0030]

[0031] in, The illumination component at position p in the k-th iteration; A local window of a preset size centered at p, the preset size is equal to (2r+1)×(2r+1), where r is 1 / 30-1 / 50 of the minimum size of the enhanced image in the (k-1)th iteration; For local windows All other positions except p; For local windows The number of pixels within; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; and All are locally linear functions; The guiding image for the k-th iteration in the local window The pixel value at position i; For the enhanced image of the (k-1)th iteration in the local window The pixel value at position i; is the spatial variational regularization parameter.

[0032] Furthermore, the spatial variational regularization parameter The expression is:

[0033]

[0034] in, This is the basic regularization parameter, with a value of 0.01; is the spatial variational intensity coefficient, with a value of 0.1; and These are the Sobel operator's convolution kernels in the X direction [[-1,0,1],[-2,0,2],[-1,0,1]] and Y direction [[-1,-2,-1],[0,0,0],[1,2,1]]; and This indicates that convolution is performed to obtain the entire gradient image, taking the gradient at pixel i in the X and Y directions; * indicates the convolution operation.

[0035] Furthermore, the expression for extracting the reflection component at each location in the enhanced image from the previous iteration is:

[0036]

[0037] in, The reflection component at position p in the k-th iteration; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; Let be the illumination component at position p in the k-th iteration; δ is a minimal constant to prevent division by zero;

[0038] The expression for the gamma correction function is:

[0039] ,

[0040] in, The pixel intensity at position p after gamma correction in the k-th iteration; The spatial adaptive gamma value at position p in the k-th iteration; For gamma correction intensity parameters; The maximum illumination component in the guided image during the k-th iteration.

[0041] Furthermore, the expression for bilateral filtering noise reduction in step S21 is as follows:

[0042]

[0043]

[0044] in, The pixel value at position p in the denoised image; Let p be the pixel intensity. Let q be the pixel intensity at position q within the neighborhood Ω1 (9*9 pixels); The spatial Gaussian kernel function used for denoising grayscale images has a standard deviation of . The standard deviation is 75. The range of Gaussian kernel functions used for denoising grayscale images has a standard deviation of . The standard deviation is 75. For spatial Euclidean distance; This represents the absolute value of the intensity difference. This is the normalization factor used for denoising grayscale images;

[0045] The expression for bilateral filtering noise reduction in step S25 is as follows:

[0046]

[0047]

[0048] in, Let p be the pixel intensity after the k-th iteration; The pixel intensity at position p after gamma correction in the k-th iteration; The pixel intensity at position q within the neighborhood Ω2 (5*5 pixels) is the gamma-corrected pixel intensity. Let be the spatial Gaussian kernel function corresponding to the iteration process, and its standard deviation be . The standard deviation is 5. Let be the range of Gaussian kernel functions during the iteration process, with a standard deviation of . The standard deviation is 0.08. This represents the absolute value of the intensity difference corresponding to the current iteration number. This is the iteration normalization factor corresponding to the iteration process.

[0049] Furthermore, the contour of the cut is extracted using the Canny edge detection method.

[0050] The beneficial effects of this invention are as follows: The single-slit stress detection cut identification and evaluation method provided by this solution is based on the improved LIME algorithm for image enhancement, which can ensure the enhancement effect of images acquired under different lighting backgrounds; then, the cut length and width parameters are obtained on the enhanced image, and the deviation rate of the cut length and width is calculated through the cut length and width parameters to quickly detect and evaluate the single-slit cut. This solves the limitations of the current traditional cut detection method based on manual detection and the inapplicability of current visual algorithms, realizes timely judgment and comprehensive analysis of cut shape, and ensures the reliability and authenticity of the single-slit stress test analysis results.

[0051] Strong image detail preservation capability: Compared with traditional recognition technology, which is easily affected by uneven lighting or background interference, this solution introduces a saliency weighting mechanism for slits. By dynamically strengthening the edge response of slender structures, it can effectively avoid the problem of feature blurring or loss, and achieve accurate preservation of slit morphological details under complex imaging conditions, thereby improving recognition reliability.

[0052] Strong adaptability: Through innovative spatial variational guided filtering and dynamic gamma correction mechanisms, key processing parameters can be adjusted in real time based on local image features. Among these, the regularization parameter ε... q Based on adaptive optimization of local gradient intensity, gamma correction parameters By dynamically adjusting the brightness distribution across regions, the algorithm significantly improves its adaptability under varying lighting conditions and texture complexity. This hierarchical adaptive strategy greatly reduces the need for manual parameter tuning, making the algorithm more robust and practical in complex environments such as field surveys and industrial inspections.

[0053] Effective noise control: This scheme uses a noise-aware update design (corresponding to steps S22 to S24) to synchronously monitor and suppress noise accumulation during image enhancement. This dynamic balancing mechanism can significantly reduce random noise and artifacts in the image and also avoid image distortion caused by over-enhancement. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the principle of a single-slit stress detection cut identification and evaluation method based on the improved LIME algorithm.

[0055] Figure 2 This is a portion of the original images that were captured.

[0056] Figure 3 The image after distortion correction.

[0057] Figure 4 This is a diagram showing the effect of image enhancement using the improved LIME algorithm.

[0058] Figure 5 The image shows the cut contour edge image extracted using the methods corresponding to steps S1 to S3 of this scheme.

[0059] Figure 6 The minimum bounding rectangle image is determined by the methods corresponding to steps S1 to S3 of this scheme. Detailed Implementation

[0060] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0061] refer to Figure 1 , Figure 1A flowchart of a single-slit stress detection cut identification and evaluation method based on an improved LIME algorithm is shown; Figure 1 As shown, the method S includes steps S1 to S4. The single-slit stress detection cut identification and evaluation method provided by this solution can effectively solve the technical problems of poor detection accuracy and low detection efficiency of existing manual methods for measuring concrete cut parameters, as well as the problem of poor extraction results of cut areas under complex lighting backgrounds in existing technologies.

[0062] In step S1, the detection image of the cut seam in the single-slit stress detection method is acquired. This scheme uses a camera to acquire cut seam images in a natural environment, such as... Figure 2 As shown; then, distortion correction is performed on the detected image. Distortion correction includes image calibration of the acquisition device to obtain the device's distortion parameters, and then distortion correction is performed on the acquired image based on the distortion parameters. The corrected image is shown below. Figure 3 As shown.

[0063] In step S2, the distortion-corrected image is converted into a grayscale image, and the improved LIME algorithm is used to enhance the grayscale image.

[0064] In one embodiment of the present invention, step S2 further includes:

[0065] S21. Convert the distortion-corrected image to a grayscale image, then perform bilateral filtering on the grayscale image to obtain a denoised image; the expression for bilateral filtering is:

[0066]

[0067]

[0068] in, The pixel value at position p in the denoised image; Let p be the pixel intensity. Let q be the pixel intensity at position q within the neighborhood Ω1 (9*9 pixels); The spatial Gaussian kernel function used for denoising grayscale images has a standard deviation of . The standard deviation is 75. The range of Gaussian kernel functions used for denoising grayscale images has a standard deviation of . The standard deviation is 75. For spatial Euclidean distance; This represents the absolute value of the intensity difference. This is the normalization factor used for denoising grayscale images.

[0069] S22. Based on the enhanced image of the previous iteration, a guide image for generating the denoised image is generated using a single-slit saliency weighting mechanism. The enhanced image of the first iteration is the denoised image.

[0070] In implementation, the preferred expression for the single-suture significance weighting mechanism in this scheme is:

[0071] ,

[0072] in, Let p be the pixel value of the guide image in the k-th iteration; The pixel value at position p in the denoised image; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; It uses a 5×5 Gaussian convolution kernel; * indicates a convolution operation; This is the mixing coefficient, with a value of 0.1. The weights are dynamically mixed and increase linearly with each iteration; k is the current iteration number, ranging from 1 to K, and K is the total number of iterations.

[0073] S23. Based on the guiding image and the enhanced image from the previous iteration, an improved spatial variational guided filter is used to estimate the illumination component at each location in the guiding image:

[0074]

[0075]

[0076]

[0077] in, The illumination component at position p in the k-th iteration; A local window of a preset size centered at p, the preset size is equal to (2r+1)×(2r+1), where r is 1 / 30-1 / 50 of the minimum size of the enhanced image in the (k-1)th iteration; For local windows All other positions except p; For local windows The number of pixels within; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; and All are locally linear functions; The guiding image for the k-th iteration in the local window The pixel value at position i; For the enhanced image of the (k-1)th iteration in the local window The pixel value at position i; Let be the spatial variational regularization parameter, and its expression is:

[0078]

[0079] in, This is the basic regularization parameter, with a value of 0.01; is the spatial variational intensity coefficient, with a value of 0.1; and These are the Sobel operator's convolution kernels in the X direction [[-1,0,1],[-2,0,2],[-1,0,1]] and Y direction [[-1,-2,-1],[0,0,0],[1,2,1]]; and This indicates that convolution is performed to obtain the entire gradient image, taking the gradient at pixel i in the X and Y directions; * indicates the convolution operation.

[0080] S24. Based on the guide image and illumination components, extract the reflection component at each location in the enhanced image from the previous iteration:

[0081]

[0082] in, The reflection component at position p in the k-th iteration; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; Let be the illumination component at position p in the k-th iteration; δ is a minimal constant to prevent division by zero;

[0083] The enhanced image from the previous iteration is then corrected and enhanced using a gamma correction function, the expression of which is:

[0084] ,

[0085] in, The pixel intensity at position p after gamma correction in the k-th iteration; The reflection component at position p in the k-th iteration; The spatial adaptive gamma value at position p in the k-th iteration; For gamma correction intensity parameters; The maximum illumination component in the guided image during the k-th iteration.

[0086] S25. Perform bilateral filtering noise reduction on the corrected and enhanced image to obtain the enhanced image for the current iteration number. This step suppresses the accumulation of noise in the image. Its calculation expression is the same as that in step S21, except that the values ​​of each parameter are the parameters during the iteration process, rather than the parameters before the iteration starts.

[0087] The expression for bilateral filtering noise reduction in step S25 is as follows:

[0088]

[0089]

[0090] in, Let p be the pixel intensity after the k-th iteration; The pixel intensity at position p after gamma correction in the k-th iteration; The pixel intensity at position q within the neighborhood Ω2 (5*5 pixels) is the gamma-corrected pixel intensity. Let be the spatial Gaussian kernel function corresponding to the iteration process, and its standard deviation be . The standard deviation is 5. Let be the range of Gaussian kernel functions during the iteration process, with a standard deviation of . The standard deviation is 0.08. This represents the absolute value of the intensity difference corresponding to the current iteration number. This is the iteration normalization factor corresponding to the iteration process.

[0091] Next, determine whether the current iteration count has reached the maximum iteration count. If so, proceed to step S26; otherwise, return to step S22.

[0092] S26. Calculate the final enhanced image based on the enhanced image from the last iteration. :

[0093]

[0094] in, Let K be the enhanced image of the Kth iteration, where K is the maximum number of iterations.

[0095] The image enhancement results after steps S21-S26 can be referenced. Figure 4 .

[0096] In step S3, the Canny edge detection method is used to extract the contour of the cut. The extracted cut contour can be referenced. Figure 5 Then, the extracted kerf contour is processed using the minimum bounding rectangle method, and the resulting image can be used as a reference. Figure 6 Then calculate the width and length of the cut.

[0097] This plan Figure 2 , Figure 3 , Figure 5 and Figure 6 The board at the top is a calibration board, which is used to determine the conversion relationship between physical dimensions and pixels.

[0098] In step S4, the deviation rates of length and width are calculated based on the kerf length and width:

[0099] ,

[0100] Where L and W are the theoretical kerf length and width, respectively; and These are the calculated length and width of the cut, respectively; This refers to the length deviation rate; , The closer to 0, the lower the deviation rate of the kerf length. A value greater than 0 indicates that the kerf length is greater than the theoretical length. A value less than 0 indicates that the cut length is less than the theoretical length; This is the width deviation rate, which is the same as the length deviation rate. The closer it is to 0, the lower the deviation rate of the cut width. A value greater than 0 indicates that the cut width is greater than the theoretical width, and a value less than 0 indicates that the cut width is less than the theoretical width.

[0101] The product of the two deviation rates was then used as the cutting standard to evaluate the cutting quality.

[0102]

[0103] in, This represents the standard rate of the cut. The closer the value is to 0, the lower the deviation, the more standard the cut, and the closer it is to the ideal cut shape.

[0104] In summary, the identification and evaluation method provided by this solution can achieve rapid and accurate identification of cuts under different lighting conditions, extract relevant parameters from the identified images, evaluate the quality of the cuts, and provide reliable assurance for stress detection using the single-cut method.

Claims

1. A method for stress detection, cut identification, and evaluation based on an improved LIME algorithm in single-slit stress detection, characterized in that: Including the following steps: S1. Obtain the detection image of the stress detection cut using the single-slit method and perform distortion correction on it; S2. Convert the distortion-corrected image into a grayscale image, and use the improved LIME algorithm to enhance the grayscale image. S3. Use edge detection to extract the contour of the cut, then process the extracted cut contour with the minimum bounding rectangle, and calculate the width and length of the cut. S4. Based on the cut length and width, calculate the deviation rate of the length and width, and use the product of the two deviation rates to evaluate the cut quality; Step S2 further includes: S21. Convert the distortion-corrected image into a grayscale image, and then perform bilateral filtering on the grayscale image to reduce noise, thus obtaining a denoised image. S22. Based on the enhanced image of the previous iteration, a guide image for generating the denoised image is generated using a single-slit saliency weighting mechanism. The enhanced image of the first iteration is the denoised image. S23. Based on the guide image and the enhanced image from the previous iteration, an improved spatial variational guide filter is used to estimate the illumination component at each location in the guide image; S24. Based on the guide image and illumination components, extract the reflection components at each location in the enhanced image of the previous iteration, and then use the gamma correction function to correct and enhance the enhanced image of the previous iteration. S25. Perform bilateral filtering noise reduction on the corrected and enhanced image to obtain the enhanced image of the current iteration number. Then determine whether the current iteration number has reached the maximum iteration number. If yes, proceed to step S26; otherwise, return to step S22. S26. Calculate the final enhanced image based on the enhanced image from the last iteration. : in, Let K be the enhanced image of the Kth iteration, where K is the maximum number of iterations. The expression for the single-slit significance weighting mechanism is: ， in, Let p be the pixel value of the guide image in the k-th iteration; The pixel value at position p in the denoised image; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; It uses a 5×5 Gaussian convolution kernel; * indicates a convolution operation; The mixing coefficient; For dynamic mixing weights; k is the current iteration number, ranging from 1 to K, where K is the total number of iterations; The expression for estimating the illumination component at each location in the guide image using an improved spatial variational guided filter is as follows: in, Let p be the illumination component at position p in the k-th iteration; A local window of a preset size centered at p; For local windows All other positions except p; For local windows The number of pixels within; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; and All are locally linear functions; The guiding image for the k-th iteration in the local window The pixel value at position i; For the enhanced image of the (k-1)th iteration in the local window The pixel value at position i; For spatial variational regularization parameters; The spatial variational regularization parameter The expression is: in, This is the basic regularization parameter, with a value of 0.01; This is the spatial variational intensity coefficient, with a value of 0.1; and These are the Sobel operator's convolution kernels in the X direction [[-1,0,1],[-2,0,2],[-1,0,1]] and Y direction [[-1,-2,-1],[0,0,0],[1,2,1]]; and This indicates that convolution is performed to obtain the entire gradient image, taking the gradient at pixel i in the X and Y directions; * indicates the convolution operation.

2. The method for stress detection, cut identification, and evaluation using the single-slit method according to claim 1, characterized in that, The expressions for calculating the deviation rates of length and width are as follows: ， in, This refers to the length deviation rate; The width deviation rate is represented by L and W, which are the theoretical kerf length and width, respectively. and These represent the calculated length and width of the cut, respectively.

3. The method for stress detection, cut identification, and evaluation using the single-slit method according to claim 1, characterized in that, The expression for extracting the reflection component at each location in the enhanced image from the previous iteration is: in, The reflection component at position p in the k-th iteration; Let p be the pixel value of the enhanced image at position p in the (k-1)th iteration; Let be the illumination component at position p in the k-th iteration; δ is a minimal constant to prevent division by zero; The expression for the gamma correction function is: ， in, The pixel intensity at position p after gamma correction in the k-th iteration; The spatial adaptive gamma value at position p in the k-th iteration; For gamma correction intensity parameters; The maximum illumination component in the guided image during the k-th iteration.

4. The single-slit stress detection cut identification and evaluation method according to claim 1 or 3, characterized in that, The expression for bilateral filtering noise reduction in step S21 is: in, The pixel value at position p in the denoised image; Let p be the pixel intensity. Let be the pixel intensity at position q within the neighborhood Ω1; The spatial Gaussian kernel function used for denoising grayscale images has a standard deviation of . ; The range of Gaussian kernel functions used for denoising grayscale images has a standard deviation of . ; For spatial Euclidean distance; This represents the absolute value of the intensity difference. This is the normalization factor used for denoising grayscale images; The expression for bilateral filtering noise reduction in step S25 is as follows: in, Let p be the pixel intensity after the k-th iteration; The pixel intensity at position p after gamma correction in the k-th iteration; The pixel intensity at position q within the neighborhood Ω2 is the gamma-corrected intensity. Let be the spatial Gaussian kernel function corresponding to the iteration process, and its standard deviation be . ; Let be the range of Gaussian kernel functions during the iteration process, with a standard deviation of . ; This represents the absolute value of the intensity difference corresponding to the current iteration number. This is the iteration normalization factor corresponding to the iteration process.

5. The single-slit stress detection cut identification and evaluation method according to claim 1, characterized in that, The contour of the cut was extracted using the Canny edge detection method.

Images