Steel coil loose abnormality detection and quantification method based on active vision imaging

Abstract

The application patent is based on the steel coil band loosening anomaly detection and quantification method of active vision imaging, first, the saliency detection model is constructed to obtain the brightness saliency map of the laser stripe, second, the laser stripe image is segmented by using the serialized threshold to obtain the Boolean graph, the weighted sum of different binary images is obtained to obtain the region stability image. The brightness saliency map and the region stability saliency map are fused by wavelet transform, the mean fusion is adopted for the brightness saliency map, the maximum value fusion mode is adopted for the region stability saliency map, and the adaptive maximum entropy segmentation is carried out on the fused saliency map, finally, the stripe normal field is obtained by calculating the stripe gradient vector, and the stripe center line is extracted based on the stripe distribution normal and the fusion gray gravity center method. The quantification of the loosening band is realized by calculating the neighborhood difference value and the curvature radius of the stripe center, and the application prospect is wide.

Description

TECHNICAL FIELD

[0001] The present application relates to a steel coil band loosening abnormality recognition and quantification method based on active vision imaging, which is to judge the abnormal state of the band during the hoisting process of the steel coil. The present application mainly aims at the accurate segmentation and center line extraction of the laser stripe under the influence of complex background interference, uneven illumination and noise, as well as the quantification of the band abnormality features. The present application patent is related to the fields of non-destructive testing of industrial products, automatic detection of steel plate surface, abnormal detection of conveyor belt, product shape detection (height, diameter, irregularity, etc.), automatic recognition and geometric size measurement of mechanical parts, surface roughness and surface defects, and real-time control based on structured light vision. BACKGROUND

[0002] With the development of structured light active vision imaging technology, perceiving external information through vision is one of the important research fields at present. The traditional band recognition is mainly completed by manual work, which is low in efficiency and long in process. The fusion of structured light vision improves the measurement accuracy of the system. At present, the development trend of intelligentization and high efficiency is shown in the field of industrial measurement. Since it is difficult to recognize the band features in weak light environment directly through the CCD / CMOS image sensor in a non-structured environment, a laser beam is projected onto the steel coil to assist in feature measurement, so as to complete the discrimination of the band abnormality state.

[0003] In computer vision, the main task of saliency detection is to detect the most salient object features in the scene, and the obtained saliency features are used for object segmentation. Due to the influence of uneven illumination and noise in the industrial environment, the laser stripe image inevitably has blur, uneven gray scale and noise, and even under the condition of background reflection, the laser stripe image will appear laser speckle. When processing the laser stripe image segmentation, problems such as incomplete segmentation region, under-segmentation and over-segmentation are prone to occur, which is easily disturbed by noise and has poor stability. For the problem of laser stripe image segmentation, the saliency feature detection model is introduced in the present application.

[0004] It has been proved in practice that in the process of unmanned logistics transportation, if the band is torn or loosened, safety accidents are prone to occur during hoisting. Therefore, the condition of the band must be detected in time, otherwise serious economic losses will be caused. The active measurement technology represented by structured light has many advantages such as high measurement accuracy and high automation degree. In order to quickly and accurately detect the abnormal condition of the steel coil band, the present application designs an auxiliary measurement band abnormality detection system based on structured light vision. The influence of illumination and background noise can be effectively avoided, and the system has strong robustness.

[0005] Therefore, a set of low-cost, simple structure, real-time online measurement can meet the abnormal feature detection and processing system of hot-rolled steel coil binding belt, provide technical support for intelligent unmanned heavy load process, realize full-process control, and has great market potential. The multi-scale salient feature fusion, maximum entropy segmentation, abnormal area extraction and quantization of the binding belt are the core technologies of the whole application. SUMMARY

[0006] The application aims to provide a steel coil binding belt loosening abnormality recognition and quantization method based on CCD structured light active vision imaging, to solve the existing problems as follows: judging the abnormal state of the binding belt during the steel coil hoisting process, mainly aiming at the accurate segmentation of the laser stripe and the extraction of the center line under the influence of complex background interference, uneven illumination and noise, and the quantization of the binding belt abnormal features.

[0007] In order to achieve the above-mentioned purpose, the application adopts the following technical scheme:

[0008] The steel coil binding belt loosening abnormality detection and quantization method based on active vision imaging comprises the following steps:

[0009] A high-precision CCD / CMOS image sensor is used, and a filter with a specific wavelength is installed in front of the lens, which can ensure the high quality of image acquisition and prevent external strong light interference. In order to facilitate the installation of the acquisition device, in the space allowed by the measurement range, considering the size of the steel coil, a height-adjustable mounting bracket and a mounting angle adjusting device for the camera and the laser emitter are designed;

[0010] The brightness saliency map of the laser stripe is obtained by constructing a saliency detection model, so as to reduce the interference of complex background, uneven illumination and noise on the laser stripe;

[0011] The laser stripe image is segmented by using a serialized threshold to obtain a Boolean graph, which aims to expose the features of the laser stripe image under different threshold levels, and a region stability image is obtained by calculating the weighted sum of different binary images, so as to highlight the difference between the laser stripe and the background;

[0012] The brightness saliency map and the region stability saliency map are fused by wavelet transform, the brightness saliency map is fused by mean value, and the region stability saliency map is fused by maximum value;

[0013] The fused saliency map is adaptively segmented by maximum entropy, and the final segmentation result is obtained based on the stability measurement;

[0014] The stripe normal field is obtained by calculating the stripe gradient vector, and the stripe center line is extracted based on the distribution normal of the stripe.

[0015] Further preferably, the brightness saliency map of the laser stripe is obtained by a saliency detection model, mainly comprising:

[0016] The saliency detection module is introduced to distinguish the target and the background, and the specific method is as follows:

[0017] The mean value of the RGB channel image is calculated respectively, and the initial brightness saliency map is obtained by subtracting the RGB channel mean value and normalizing the RGB channel image:

[0018]

[0019] Where I c (x,y) is the input image, is the mean value of the input image, C represents the color channel of the input image, and c∈{R,G,B}.

[0020] Further preferably, the laser stripe image is segmented by calculating a serial threshold to obtain a Boolean map, mainly comprising:

[0021] The stable saliency region of the laser stripe image is extracted by calculating the Boolean map, and the Boolean map under different segmentation thresholds is defined as BM={BM1,…,BM n}, and the function expression is:

[0022] BM=Thr(I,θ)

[0023] Where Thr(·) represents the threshold function, I represents the feature map of the input image, and θ=δ / 255 is the segmentation threshold, δ is increased by 16 as a step and δ∈[δ / 2:δ:255-δ / 2];

[0024] After obtaining a series of Boolean maps, the sum of the weights of each Boolean map is calculated to obtain the stable saliency map of the laser stripe region, and the function expression is:

[0025]

[0026] Where θ i is the different segmentation threshold normalized to [0,1], and BM i is the Boolean map under different segmentation thresholds.

[0027] Further preferably, the brightness saliency map and the region stable saliency map are fused by wavelet transform, the mean fusion is adopted for the brightness saliency map, and the maximum value fusion is adopted for the region stable saliency map, and the expression is as follows:

[0028]

[0029] In the formula, H r , G rand H c , G c respectively represent one-dimensional mirror filter operators H and G acting on rows and columns respectively, for two-dimensional image, the operator H r G c is equivalent to two-dimensional low-pass filter. represents the vertical high-frequency component of C j , represents the horizontal high-frequency component of C j , represents the diagonal high-frequency component of C j . For a wavelet transform of an image X, the wavelet coefficients and scale coefficients of the j+1th level are denoted as and C j (X) respectively.

[0030] The corresponding wavelet transform reconstruction expression is:

[0031]

[0032] In the formula, H * , G * are denoted as the conjugate transpose matrices of H and G respectively.

[0033] The high-frequency part of wavelet transform corresponds to the edges and contour features of the image with sharp changes, and the low-frequency part reflects the overall gray value distribution of the image. The wavelet transform fusion should retain as many details of the image as possible while retaining the overall contour of the image. For the high-frequency features in the laser stripe image, the contrast changes of the edges and other features in the image are reflected, so for the high-frequency features, the maximum fusion rule is adopted in this paper. For laser stripe images A and B, the high-frequency fusion function expression is:

[0034]

[0035] In the formula, H(x, y) represents the image fusion coefficient, (x, y) is the coefficient coordinate, H A (x, y) and H B (x, y) represent the high-frequency sub-band coefficients of images A and B respectively.

[0036] For image low-frequency part fusion, it is necessary to retain the overall features of the image as much as possible, so the mean weighted processing is adopted for the fusion method, and its expression is:

[0037] L(x, y) = (L A (x, y) + L B (x, y)) / 2

[0038] In the formula, L(x, y) is the image fusion coefficient; L A (x, y) and LB (x,y) represent the low frequency sub-band coefficients of images A and B respectively.

[0039] Further preferably, the adaptive maximum entropy segmentation is performed on the fused saliency map, and a final segmentation result is obtained based on the stability measure, and the expression is as follows:

[0040] According to Shannon theory, the entropy is expressed as follows:

[0041]

[0042] Where p(x) is the probability of event x occurring;

[0043] Using an image to describe the above formula, x is a certain gray level of the image, and p(x) is the probability of the gray value being x. If the image is N gray levels, the above formula can be expressed as:

[0044]

[0045] Assuming T as a threshold, the gray level less than T as a target area, and the gray level greater than T as a background area. The probability of the gray level of the target area and the background area is expressed as follows:

[0046]

[0047]

[0048] The entropy of the target area and the background area is defined as:

[0049]

[0050]

[0051] The entropy function of the image is defined as:

[0052] H(t)=H0(t)+H b (t)

[0053] The threshold can be expressed as:

[0054] T=arg max H(t).

[0055] Further preferably, the fringe normal field is obtained by calculating the fringe gradient vector, and the fringe center line is extracted based on the fringe distribution normal by using the fused gray center method. The loosening of the bundle belt is quantified by calculating the domain difference value and the curvature radius of the fringe center, and the expression of the bundle belt loosening anomaly is as follows:

[0056] d=y i+2 -y i

[0057] In the formula, d represents the difference value of adjacent pixels, y i The center longitudinal coordinate of a stripe at a position is represented, if |d|≥T1, it is considered that the area can exist loose condition of the bundle belt.

[0058] As shown in the figure is the curvature calculation schematic diagram, assuming that the curve C is smooth, the arc length of the point M to M' on the curve C is Δs, the tangent angle is Δα, and the average curvature of the arc segment MM' is represented as The curvature of the curve C at the point M is represented as If the following condition is met Then

[0059]

[0060] In the formula, K is the curvature radius, y' is the first derivative, y'' is the second derivative, if K≥T2, it is considered that the area can exist loose condition of the bundle belt.

[0061] The present application has at least the following beneficial effects:

[0062] The present application adopts the saliency detection model to obtain the initial brightness saliency map, so as to reduce the interference of complex background, uneven light and noise on the laser stripe; and the quality of subsequent image segmentation can be effectively improved.

[0063] The present application adopts the serialized threshold to segment the laser stripe gray scale image, obtains the laser stripe Boolean graph, further highlights the contrast of the laser stripe, and calculates the weight of each Boolean graph to obtain the area stability image, further highlights the difference between the laser stripe and the background, and strengthens the segmentation effect.

[0064] The present application adopts the wavelet feature fusion mode, adopts the mean fusion of the brightness saliency map, and adopts the maximum value fusion mode of the area stability saliency map, can effectively locate the region of interest in the image, improve the image retrieval speed, complementary advantages complete the segmentation task in the corresponding image processing field, and has strong robustness.

[0065] The present application adopts the neighborhood difference value and the curvature radius of the quantized stripe center, compares with the threshold value, and reflects the tightness degree of the steel coil bundle belt, and has strong anti-interference ability.

[0066] In addition, the application is applied to nondestructive testing of industrial products on a production line, and relates to the fields of nondestructive testing of industrial products, automatic detection of a steel plate surface, detection of an abnormality of a conveyor belt, detection of a product shape (height, diameter, irregularity, etc.), automatic identification and geometric size measurement of a mechanical part, surface roughness and surface defects, and real-time control based on structured light vision. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0068] Figure 1 Fig. 1 is a schematic diagram of a steel coil and a loose abnormality test;

[0069] Figure 2 Fig. 2 is a schematic diagram of fusion of multi-scale salient features;

[0070] Figure 3 Fig. 3 is a schematic diagram of stripe segmentation in different scenes;

[0071] Figure 4 Fig. 4 is a schematic diagram of center line extraction of a coil in different states;

[0072] Figure 5 Fig. 5 is a schematic diagram of quantitative features of a coil in different states;

[0073] Figure 6 Fig. 6 is a schematic diagram of curvature calculation; DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0075] In the process of judging the loose abnormality of the coil, the brightness saliency map and the region stability saliency map are fused, and the fused saliency map is adaptively maximum entropy segmented on the basis of complementary advantages, so as to detect the loose condition of the coil by quantifying the abnormal region of the laser stripe.

[0076] In order to achieve the above purpose, the present application adopts the following design scheme:

[0077] Firstly, the brightness saliency map of laser stripe is obtained by constructing a saliency detection model to reduce the interference of complex background, uneven illumination and noise on the laser stripe. Secondly, the laser stripe image is segmented by using a serialized threshold to obtain a Boolean map. The region stability image is obtained by calculating the weighted sum of different binary images. The brightness saliency map and the region stability saliency map are fused by using the mean fusion for the brightness saliency map and the maximum fusion for the region stability saliency map. The fused saliency map is adaptively segmented by the maximum entropy. Finally, the stripe normal field is obtained by calculating the stripe gradient vector, and the stripe center line is extracted based on the stripe distribution normal and the gray centroid fusion method. The loose bundle belt is quantified by calculating the field difference and the curvature radius of the stripe center.

[0078] The steel coil bundle belt loose abnormal detection and quantification method based on active vision imaging is described as follows:

[0079] ①A high-precision CCD / CMOS image sensor and a high-precision non-diffractive line laser transmitter are installed. The laser transmitter is vertically irradiated on the steel coil bundle belt, and the CCD / CMOS image sensor and the laser transmitter are installed at a certain inclination angle, as shown in FIG. 1. Figure 1 FIG. 1 is a schematic diagram of a steel coil bundle belt loose abnormal test

[0080] ②The mean values of the RGB channel images are calculated respectively. The initial brightness saliency map is obtained by subtracting the RGB channel images from the RGB channel mean values and normalizing them. The function expression is as follows:

[0081]

[0082] where I c (x,y) is the input image, is the mean value of the input image, C represents the color channel of the input image, and c∈{R,G,B}.

[0083] ③The stable saliency region of the laser stripe is extracted by calculating the stripe Boolean map.

[0084] The function expression of BM n = {BM1,…,BM

[0085] BM = Thr(I, θ) (2)

[0086] where Thr(·) represents the threshold function, I represents the feature map of the input image, θ = δ / 255 is the segmentation threshold, and δ is increased by 16 as a step and δ∈[δ / 2:δ:255-δ / 2].

[0087] After obtaining a series of Boolean maps, the sum of the weights of each Boolean map is calculated to obtain the saliency map of the laser stripe region stability. The calculation formula is as follows:

[0088]

[0089] where θ i is the different segmentation threshold normalized to [0, 1], and BM i is the Boolean map at different segmentation threshold.

[0090] ④ The luminance saliency map and the region stable saliency map are fused by wavelet transform. The mean fusion is used for the luminance saliency map, and the maximum value fusion is used for the region stable saliency map. The expression is as follows:

[0091]

[0092] In the formula, H r , G r and H c , G c respectively represent one-dimensional mirror filter operators H and G acting on rows and columns, respectively. For a two-dimensional image, the operator H r G c corresponds to a two-dimensional low-pass filter. represents the vertical direction high-frequency component of C j , represents the horizontal direction high-frequency component of C j , represents the diagonal direction high-frequency component of C j . For an image X, the wavelet coefficients and scale coefficients of the j+1th layer are represented as and C j (X) respectively.

[0093] The corresponding wavelet transform reconstruction expression is:

[0094]

[0095] In the formula, H * , G * are represented as the conjugate transpose matrices of H and G respectively.

[0096] The high-frequency part of the wavelet transform corresponds to the edges and contour features in the image that change sharply, and the low-frequency part reflects the overall gray value distribution of the image. The wavelet transform fusion should preserve as many image details as possible while preserving the overall contour of the image. For the high-frequency features in the laser stripe image, the contrast changes of the edges and other features in the image are reflected. Therefore, for the high-frequency features, the maximum value fusion rule is used in this paper. For the laser stripe images A and B, the high-frequency fusion function expression is as follows:

[0097]

[0098] where H(x, y) represents the image fusion coefficient, (x, y) is the coefficient coordinate, H A (x, y) and H B (x, y) represent the high frequency sub-band coefficients of images A and B respectively.

[0099] For the fusion of the low frequency part of the image, the overall characteristics of the image should be preserved as much as possible, therefore, the mean weighted processing is adopted for the fusion, and its expression is:

[0100] L(x, y) = (L A (x, y) + L B (x, y)) / 2 (7)

[0101] where L(x, y) is the image fusion coefficient; L A (x, y) and L B (x, y) represent the low frequency sub-band coefficients of images A and B respectively.

[0102] On the basis of the fused image, the maximum entropy is used for segmentation according to the Shannon theory, and the entropy is expressed as follows:

[0103]

[0104] where p(x) is the probability of event x.

[0105] The formula (8) is described by using the image, x is a certain gray level of the image, and p(x) is the probability of the gray value x. If the image is N gray levels, the formula (8) can be expressed as:

[0106]

[0107] Suppose T is the threshold value, the gray level less than T is the target region, and the gray level greater than T is the background region. The probability of the gray levels of the target region and the background region is expressed as follows:

[0108]

[0109]

[0110] The entropy of the target region and the background region is defined as:

[0111]

[0112]

[0113] The entropy function of the image is defined as:

[0114] H(t) = H0(t) + H b (t) (14)

[0115] The threshold value can be expressed as:

[0116] T = arg max H(t) (15)

[0117] 5. Steel coil strip abnormal area extraction and strip loosening quantification. The stripe normal field is obtained by calculating the stripe gradient vector, and the stripe center line is extracted based on the stripe distribution normal by fusing the gray center method. The loosening strip is quantified by calculating the field difference value and the curvature radius of the stripe center, and the expression of the strip loosening abnormality is as follows:

[0118] d = y i+2 -y i (16)

[0119] In the formula, d represents the difference value of adjacent pixels, y i represents the longitudinal coordinate of the stripe center at a certain position, and if |d|≥T1, it can be considered that the area may exist strip loosening condition.

[0120] As shown in the figure is a curvature calculation schematic diagram, assuming that the curve C is smooth, the arc length of the point M to M' on the curve C is Δs, the tangent angle is Δα, and the average curvature of the arc segment MM' is represented as The curvature of the curve C at the point M is represented as If then

[0121]

[0122] In the formula, K is the curvature radius, y' is the first derivative, and y'' is the second derivative, and if K≥T2, it can be considered that the area may exist strip loosening condition.

[0123] As Figure 2 shown is the fusion schematic diagram of the multiscale salient feature of the present application, as Figure 3 shown is a stripe segmentation schematic diagram for different scenes. As Figure 4 shown is the center extraction result of the strip in different states, as Figure 5 shown is the quantification result, as Figure 6 shown is a curvature calculation schematic diagram, and according to the algorithm simulation result of the present application, the algorithm of the present application can not only suppress the interference of background noise, but also highlight the contrast of the laser stripe, and has strong detection and quantification ability for the loosening characteristics of the steel coil strip in different states.

[0124] As can be known from the above:

[0125] The biggest advantage of the present application is that the laser stripe is accurately segmented based on wavelet feature saliency fusion and maximum entropy segmentation model for the interference of uneven illumination and background noise on the laser stripe. Meanwhile, the feature of the abnormal area of the steel coil is extracted and the abnormality of the coil belt is quantified, which has a good application background. The experimental results show that the method can effectively suppress the noise of the stripe image, and for the low resolution laser stripe image, the segmentation of the laser stripe can be effectively realized, and the anti-interference ability and accuracy in the abnormal feature detection of the coil belt are high. The steel coil belt loosening abnormality detection and quantification method based on active vision imaging has good engineering application prospect. It is embodied in the following points:

[0126] 1. The mean value of the RGB channel image is calculated respectively, the RGB channel image is subtracted from the RGB channel mean value and normalized to obtain the initial brightness saliency map.

[0127] 2. The laser stripe gray image is segmented by using a serialized threshold to obtain a laser stripe Boolean graph, and the weight and calculation of each Boolean graph are calculated to obtain a region stability image, which further highlights the difference between the laser stripe and the background and strengthens the segmentation effect.

[0128] 3. The abnormal area of the steel coil belt is identified and quantified, the position of the laser stripe is effectively located by the saliency model, which has strong anti-interference ability and segmentation precision, can complementarily complete the segmentation task in the corresponding image processing field and the quantification task of the abnormal coil of the steel coil, and provides theoretical and practical basis for unmanned and corresponding application scenarios.

[0129] 4. The method of the present application can be used for target feature extraction and segmentation in low-illumination environment, and the saliency detection model can effectively locate the region of interest in the image, improve the image retrieval speed, and can be used for flatness detection and defect detection for the extraction and quantification of the abnormal area of the coil belt.

[0130] 5. In addition to being applied to nondestructive testing of industrial products on the production line, the present application can also be used for flatness detection, belt damage detection, three-dimensional target reconstruction, military field and other real-time production and processing fields based on structure light active imaging visual auxiliary measurement technology.

[0131] The basic principles, main features and advantages of the present application are shown and described. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for detecting and quantifying steel coil strapping looseness anomalies based on active visual imaging, characterized in that, Includes the following steps: By constructing a saliency detection model, the brightness saliency map of the laser stripes is obtained to reduce the interference of complex background, uneven illumination and noise on the laser stripes; Boolean graphs are obtained by segmenting laser stripe images using serialized thresholding, aiming to expose the features of laser stripe images at different threshold levels. By calculating the weighted sum of different binary images, a region stability image is obtained, thereby highlighting the difference between laser stripes and the background. The brightness saliency map and the region stable saliency map are fused by wavelet transform. The brightness saliency map is fused by mean fusion, and the region stable saliency map is fused by maximum fusion. Adaptive maximum entropy segmentation is performed on the fused saliency map, and the final segmentation result is obtained based on the stability metric. The stripe normal field is obtained by calculating the stripe gradient vector. The stripe center line is extracted by fusing the gray-scale centroid method based on the stripe distribution normal. The loose strapping is quantified by calculating the neighborhood difference and radius of curvature of the stripe center. The fringe normal field is obtained by calculating the fringe gradient vector. The fringe center line is extracted by fusing the gray-scale centroid method based on the fringe distribution normal. The loose strapping is quantified by calculating the neighborhood difference and radius of curvature of the fringe center. The expression for strapping looseness anomaly is as follows: d=y i+2 -y i In the formula, d represents the difference between adjacent pixels, and y i The vertical coordinate of the stripe center at a certain location is represented by |d|. If |d| ≥ T1, then the area can be considered to have loose straps. Assuming curve C is smooth, the arc length from point M to M' on curve C is Δs, the tangent turn angle is Δα, and the average curvature of arc segment MM' is expressed as... The curvature of curve C at point M is expressed as: If satisfied but In the formula, K is the radius of curvature, y′ is the first derivative, and y′ is the second derivative. If K≥T2, then the strapping in this region can be considered to be loose.

2. The method for detecting and quantifying steel coil strapping looseness anomalies based on active vision imaging according to claim 1, characterized in that, in, The initial brightness saliency map is obtained through a saliency detection model, mainly including: Calculate the mean of each RGB channel image, subtract the mean of the RGB channel images from the mean of the RGB channels, and normalize to obtain the initial saliency map: Where I c (x,y) is the input image. It is the average value of the input image, and C represents the color channels of the input image, c∈{R,G,B}.

3. The method for detecting and quantifying steel coil strapping looseness anomalies based on active vision imaging according to claim 1, characterized in that, in, Boolean graphs are obtained by segmenting laser stripe images using a serialization threshold, which mainly includes: By calculating the Boolean graph, stable and salient regions of the laser stripe image are extracted. The Boolean graph under different segmentation thresholds is defined as BM={BM1,…,BM n }, its function expression is: BM = Thr(I,θ) Where Thr(·) represents the threshold function, I represents the feature map of the input image, θ=δ / 255 is the segmentation threshold, δ increases with a step size of 16 and δ∈[δ / 2:δ:255-δ / 2]; After obtaining a series of Boolean graphs, the saliency map of the stable laser stripe region is obtained by calculating the sum of the weights of each Boolean graph. Its functional expression is as follows: Where θ i These are different segmentation thresholds normalized to [0, 1], and BM i It is a Boolean graph under different segmentation thresholds.

4. The method for detecting and quantifying steel coil strapping looseness anomalies based on active vision imaging according to claim 1, characterized in that, in, The brightness saliency map and the region-stable saliency map are fused using wavelet transform. Mean fusion is applied to the brightness saliency map, and maximum fusion is applied to the region-stable saliency map. The expression is as follows: In the formula, H r G r and H c G c These represent the one-dimensional mirror filtering operators H and G acting on the rows and columns, respectively. For a two-dimensional image, operator H... r G c Equivalent to a two-dimensional low-pass filter, Indicate C j The high-frequency components in the vertical direction, Indicate C j High-frequency components in the horizontal direction, Indicate C j For the high-frequency components in the diagonal direction, the wavelet coefficients and scaling coefficients of the (j+1)th layer of an image X, after wavelet transform, are expressed as follows: and C j (X), The corresponding wavelet transform reconstruction expression is: In the formula, H * G * The subdivision is represented as the conjugate transpose of H and G. The high-frequency component of wavelet transform corresponds to the rapidly changing edge and contour features in the image, while the low-frequency component reflects the overall grayscale distribution of the image. Wavelet transform fusion should preserve as many detailed features as possible while maintaining the overall contour of the image. For the high-frequency features in the laser stripe image, which reflect features such as edges with significant contrast changes, this paper adopts a fusion rule based on maximum values. The high-frequency fusion function expression for laser stripe images A and B is as follows: In the formula, H(x,y) represents the image fusion coefficient, (x,y) is the coefficient coordinate, and H... A (x,y),H B (x, y) represent the high-frequency subband coefficients of images A and B, respectively. For low-frequency image fusion, it is necessary to preserve the overall image features as much as possible. Therefore, the fusion method adopts mean-weighted processing, and its expression is: L(x,y)=(L A (x,y)+L B (x,y)) / 2 In the formula, L(x,y) represents the image fusion coefficient; L A (x,y),L B (x,y) represent the low-frequency subband coefficients of images A and B, respectively.

5. The method for detecting and quantifying steel coil strapping looseness anomalies based on active vision imaging according to claim 1, characterized in that, in, Adaptive maximum entropy segmentation is performed on the fused saliency map, and the final segmentation result is obtained based on a stability metric, as shown in the following expression: According to Shannon's theory, entropy is expressed as follows: Where p(x) is the probability of event x occurring; Using an image to describe the above formula, x is a certain gray level of the image, and p(x) is the probability that the gray level is x. If the image has N gray levels, then the above formula can be expressed as: Assuming T is the threshold, gray levels less than T are considered target regions, and gray levels greater than T are considered background regions. The probabilities of gray levels in the target and background regions are expressed as follows: The entropy of the target region and the background region is defined as follows: The entropy function of an image is defined as: H(t)=H0(t)+H b (t) The threshold can be expressed as: T = argmaxH(t).

Images