A grinding wheel grinding accuracy quality evaluation method based on multi-angle visual inspection

Abstract

The present invention discloses a grinding wheel grinding precision quality evaluation method based on multi-angle visual detection, and belongs to the field of image processing technology. After the grinding wheel grinding is completed, the present invention shoots the grinding workpiece at low, medium and high fill light from a shooting angle to obtain the corresponding grinding image; the first photosensitivity of each pixel is calculated according to the difference between the medium fill light grinding image and the low fill light grinding image, and the second photosensitivity of each pixel is calculated according to the difference between the high fill light grinding image and the low fill light grinding image; then the suspected grinding degree of each pixel is calculated according to the first and second photosensitivity intensities, and a suspected grinding image is constructed; finally, the grinding weight is calculated for the suspected grinding image at each angle, and the grinding precision quality score is obtained based on the grinding precision quality evaluation network. This method effectively solves the problem of low evaluation accuracy in the existing evaluation and improves the accuracy of grinding quality evaluation.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a grinding wheel grinding precision quality evaluation method based on multi-angle visual detection. Background Art

[0002] In modern industrial production, grinding wheel grinding is a key link in the mechanical processing process, and its accuracy directly affects the surface quality and dimensional accuracy of the workpiece. With the continuous improvement of the manufacturing industry's requirements for product quality, higher requirements are also placed on the quality evaluation methods of grinding wheel grinding accuracy. At present, the existing grinding wheel grinding accuracy quality evaluation methods mainly rely on manual inspection or single-angle machine vision inspection. These methods have some limitations, such as low efficiency and strong subjectivity of manual inspection, and single-angle machine vision inspection cannot fully capture the details of the workpiece surface, resulting in inaccurate evaluation results. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a grinding wheel grinding accuracy quality evaluation method based on multi-angle visual inspection to solve the problem of low grinding quality evaluation accuracy of the grinding workpiece in the prior art.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a grinding wheel grinding accuracy quality evaluation method based on multi-angle visual inspection, comprising the following steps:

[0005] S1. After the grinding wheel is polished, a shooting angle is selected to shoot the polished workpiece using low fill light, medium fill light and high fill light, respectively, to obtain a low fill light polishing image, a medium fill light polishing image and a high fill light polishing image;

[0006] S2. Calculate the first light sensitivity of each pixel according to the difference between the medium fill light polishing image and the low fill light polishing image;

[0007] S3, calculating the second photosensitivity of each pixel according to the difference between the high fill light polishing image and the low fill light polishing image;

[0008] S4, calculating the suspected polishing degree of each pixel point according to the first photosensitivity and the second photosensitivity of each pixel point, and constructing a suspected polishing image;

[0009] S5. For each suspected polishing image at each angle, the polishing weight is calculated, and the polishing accuracy quality score is obtained based on the polishing accuracy quality evaluation network.

[0010] Further, S2 includes the following sub-steps:

[0011] S21, subtracting the R channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the R channel photosensitivity value;

[0012] S22, subtracting the G channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the G channel photosensitivity value;

[0013] S23, subtracting the B channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the B channel photosensitivity value;

[0014] S24, adding the R channel light sensitivity value, the G channel light sensitivity value and the B channel light sensitivity value at the same pixel position to obtain a first light sensitivity intensity.

[0015] Furthermore, S3 includes the following sub-steps:

[0016] S31, subtracting the R channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the R channel photosensitivity value;

[0017] S32, subtracting the G channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the G channel photosensitivity value;

[0018] S33, subtracting the B channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the B channel photosensitivity value;

[0019] S34, adding the R channel light sensitivity value, the G channel light sensitivity value and the B channel light sensitivity value at the same pixel position to obtain a second light sensitivity intensity.

[0020] Further, S4 includes the following sub-steps:

[0021] S41, calculating a light sensitivity intensity change value according to a first light sensitivity intensity and a second light sensitivity intensity at a same pixel position;

[0022] S42, calculating the suspected polishing degree of each pixel point according to the difference between the photosensitivity change value and the photosensitivity change reference value;

[0023] S43, replacing the original pixel value of the same pixel with the suspected polishing degree to construct a suspected polishing image.

[0024] Furthermore, the formula for calculating the photosensitivity change value in S41 is: , where γ i is the change in light intensity of the i-th pixel, q i,1 is the first light sensitivity of the i-th pixel, q i,2 is the second light sensitivity of the i-th pixel, I H For high fill light, I M I is the medium fill light. L is low fill light, i is a positive integer.

[0025] Furthermore, the formula for calculating the suspected polishing degree of each pixel in S42 is: , where s i is the suspected polishing degree of the i-th pixel, γ ref is the reference value of photosensitivity change, γ i is the change in photosensitivity of the i-th pixel, where i is a positive integer.

[0026] Further, S5 includes the following sub-steps:

[0027] S51, calculating a polishing weight for each suspected polishing image at each angle;

[0028] S52, inputting the suspected polishing images at multiple angles into the polishing accuracy quality evaluation network, and obtaining the polishing accuracy quality score based on the attention applied by the polishing weights.

[0029] Furthermore, the formula for calculating the polishing weight in S51 is: , where α is the polishing weight, N is the number of pixels with a suspected polishing degree greater than 0 on the suspected polishing image, and N total It is the total number of pixels on the suspected polished image.

[0030] Furthermore, the polishing accuracy quality evaluation network in S52 includes: multiple image processing channels, adder A and a fully connected layer;

[0031] The input end of each image processing channel is used to input a suspected polishing image at an angle;

[0032] The input end of the adder A is connected to the output ends of the multiple image processing channels respectively, and the output end thereof is connected to the input end of the fully connected layer;

[0033] The output end of the fully connected layer serves as the output end of the polishing accuracy quality evaluation network.

[0034] Further, each image processing channel includes: a convolution layer, a multiplier M, and a pooling layer;

[0035] The input end of the convolutional layer is used as the input end of the image processing channel, and its output end is connected to the first input end of the multiplier M;

[0036] The second input terminal of the multiplier M is used to input the polishing weight;

[0037] The input end of the pooling layer is connected to the output end of the multiplier M, and its output end serves as the output end of the image processing channel;

[0038] The expression of the multiplier M is: ,in, is the output of the multiplier M, X is the feature of the convolutional layer output, and α is the polishing weight.

[0039] The beneficial effects of the present invention are:

[0040] 1. The present invention sets three fill light degrees at a shooting angle, and collects images of the polished workpiece respectively. The first photosensitivity is calculated by the difference between the medium fill light polishing image and the low fill light polishing image, and then the second photosensitivity is calculated by the difference between the high fill light polishing image and the low fill light polishing image, so as to reflect the imaging difference of the polished workpiece for different fill light degrees. The suspected polishing degree of each pixel is calculated by combining the first photosensitivity and the second photosensitivity, highlighting the polishing condition at the pixel, thereby improving the accuracy of polishing quality assessment.

[0041] 2. The present invention calculates the polishing weight for the suspected polishing image at each angle, uses a polishing accuracy quality evaluation network to process the suspected polishing images at each angle, obtains a polishing accuracy quality score based on the attention applied by the polishing weight, and realizes the integration of suspected polishing images at multiple angles, thereby improving the polishing quality assessment accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of a grinding wheel grinding accuracy quality evaluation method based on multi-angle visual inspection;

[0043] Figure 2 Schematic diagram of shooting angles;

[0044] Figure 3 This is a schematic diagram of the structure of the grinding accuracy quality evaluation network. DETAILED DESCRIPTION

[0045] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0046] like Figure 1 As shown, a grinding wheel grinding accuracy quality evaluation method based on multi-angle visual inspection includes the following steps:

[0047] S1. After the grinding wheel is polished, a shooting angle is selected to shoot the polished workpiece using low fill light, medium fill light and high fill light, respectively, to obtain a low fill light polishing image, a medium fill light polishing image and a high fill light polishing image;

[0048] S2. Calculate the first light sensitivity of each pixel according to the difference between the medium fill light polishing image and the low fill light polishing image;

[0049] S3, calculating the second photosensitivity of each pixel according to the difference between the high fill light polishing image and the low fill light polishing image;

[0050] S4, calculating the suspected polishing degree of each pixel point according to the first photosensitivity and the second photosensitivity of each pixel point, and constructing a suspected polishing image;

[0051] S5. For each suspected polishing image at each angle, the polishing weight is calculated, and the polishing accuracy quality score is obtained based on the polishing accuracy quality evaluation network.

[0052] In this embodiment, the low fill light sets the brightness to 30% of the fill light device, the medium fill light sets the brightness to 60% of the fill light device, and the high fill light sets the brightness to 90% of the fill light device.

[0053] In this embodiment, a shooting angle perpendicular to the workpiece to be polished and a 45-degree side angle shooting angle are set. Figure 2 shown.

[0054] In this embodiment, S2 includes the following sub-steps:

[0055] S21, subtracting the R channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the R channel photosensitivity value;

[0056] S22, subtracting the G channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the G channel photosensitivity value;

[0057] S23, subtracting the B channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the B channel photosensitivity value;

[0058] S24, adding the R channel light sensitivity value, the G channel light sensitivity value and the B channel light sensitivity value at the same pixel position to obtain a first light sensitivity intensity.

[0059] The present invention performs subtraction operations on the R, G, and B channel values ​​of the medium fill light polished image and the low fill light polished image respectively, and then adds the photosensitivity values ​​of each channel of the same pixel point to obtain the first photosensitivity intensity, which can accurately reflect the difference between the two images at each pixel point.

[0060] The present invention obtains a low fill light polishing image, a medium fill light polishing image and a high fill light polishing image at the same shooting angle. Therefore, the pixel points with the same coordinates on the low fill light polishing image, the medium fill light polishing image and the high fill light polishing image represent the imaging results of the same spatial position of the polishing workpiece under different fill light conditions.

[0061] In this embodiment, S3 includes the following sub-steps:

[0062] S31, subtracting the R channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the R channel photosensitivity value;

[0063] S32, subtracting the G channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the G channel photosensitivity value;

[0064] S33, subtracting the B channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the B channel photosensitivity value;

[0065] S34, adding the R channel light sensitivity value, the G channel light sensitivity value and the B channel light sensitivity value at the same pixel position to obtain a second light sensitivity intensity.

[0066] The present invention performs subtraction operations on the R, G, and B channel values ​​of the high-fill light polished image and the low-fill light polished image respectively, and then adds the photosensitivity values ​​of each channel of the same pixel point to obtain the second photosensitivity intensity, which can accurately reflect the difference between the two images at each pixel point.

[0067] Because different polishing degrees will lead to different reflection and absorption characteristics of the surface to light, and the imaging differences under different fill light levels can just reflect these differences. Even extremely subtle changes in the degree of polishing can be reflected through the difference in the light sensitivity of the pixel points, so that scratches, wear, unevenness and other problems on the surface of the polished workpiece can be detected more accurately.

[0068] In this embodiment, S4 includes the following sub-steps:

[0069] S41, calculating a light sensitivity intensity change value according to a first light sensitivity intensity and a second light sensitivity intensity at a same pixel position;

[0070] S42, calculating the suspected polishing degree of each pixel point according to the difference between the photosensitivity change value and the photosensitivity change reference value;

[0071] S43, replacing the original pixel value of the same pixel with the suspected polishing degree to construct a suspected polishing image.

[0072] In this embodiment, the formula for calculating the photosensitivity change value in S41 is: , where γ i is the change in light intensity of the i-th pixel, q i,1 is the first light sensitivity of the i-th pixel, q i,2 is the second light sensitivity of the i-th pixel, I H For high fill light, I M I is the medium fill light. L is low fill light, i is a positive integer.

[0073] The present invention calculates the fill light ratio according to the difference between the high fill light degree and the low fill light degree, and the difference between the medium fill light degree and the low fill light degree. , combined with the ratio of the first photosensitivity and the second photosensitivity, the relative change relationship of the photosensitivity of the pixels under different fill light conditions is highlighted. For example, the better polished area will appear as a mirror surface, which has a stronger reflection of light and a larger photosensitivity change value.

[0074] In this embodiment, the low fill light sets the brightness to 30% of the fill light device, the medium fill light sets the brightness to 60% of the fill light device, and the high fill light sets the brightness to 90% of the fill light device. Equals 2.

[0075] In this embodiment, the formula for calculating the suspected polishing degree of each pixel in S42 is: , where s i is the suspected polishing degree of the i-th pixel, γ ref is the reference value of photosensitivity change, γ i is the change in photosensitivity of the i-th pixel, where i is a positive integer.

[0076] The photosensitivity variation reference value is a standard value set for the photosensitivity variation value. By calculating the photosensitivity variation by applying S41 of the present invention to the polished workpiece that meets the quality standard, a reference benchmark for the photosensitivity variation is established.

[0077] The present invention compares the photosensitivity change reference value with the photosensitivity change value. When the photosensitivity change value is less than or equal to the photosensitivity change reference value, the suspected polishing degree is set to 0. When the photosensitivity change value is greater than the photosensitivity change reference value, the suspected polishing degree is calculated according to the difference between the photosensitivity change value and the photosensitivity change reference value to distinguish the unpolished area from the polished area.

[0078] In this embodiment, S5 includes the following sub-steps:

[0079] S51, calculating a polishing weight for each suspected polishing image at each angle;

[0080] S52, inputting the suspected polishing images at multiple angles into the polishing accuracy quality evaluation network, and obtaining the polishing accuracy quality score based on the attention applied by the polishing weights.

[0081] In this embodiment, the formula for calculating the polishing weight in S51 is: , where α is the polishing weight, N is the number of pixels with a suspected polishing degree greater than 0 on the suspected polishing image, and N total It is the total number of pixels on the suspected polished image.

[0082] The present invention calculates the polishing weight by the number of pixels with suspected polishing degree greater than 0, determines the proportion of the polishing area, and thus applies corresponding attention in the polishing accuracy quality evaluation network to improve the assessment accuracy of the polishing quality.

[0083] like Figure 3 As shown, the polishing accuracy quality evaluation network in S52 includes: multiple image processing channels, adder A and a fully connected layer;

[0084] The input end of each image processing channel is used to input a suspected polishing image at an angle;

[0085] The input end of the adder A is connected to the output ends of the multiple image processing channels respectively, and the output end thereof is connected to the input end of the fully connected layer;

[0086] The output end of the fully connected layer serves as the output end of the polishing accuracy quality evaluation network.

[0087] In the present invention, the number of image processing channels is equal to the number of selected angles.

[0088] In this embodiment, each image processing channel includes: a convolution layer, a multiplier M and a pooling layer;

[0089] The input end of the convolutional layer is used as the input end of the image processing channel, and its output end is connected to the first input end of the multiplier M;

[0090] The second input terminal of the multiplier M is used to input the polishing weight;

[0091] The input end of the pooling layer is connected to the output end of the multiplier M, and its output end serves as the output end of the image processing channel;

[0092] The expression of the multiplier M is: ,in, is the output of the multiplier M, X is the feature of the convolutional layer output, and α is the polishing weight.

[0093] In this embodiment, the polishing accuracy quality evaluation network can be trained using the gradient descent method.

[0094] The present invention applies polishing weights to the convolutional layer and pooling layer of the image processing channel. After the convolutional layer, the polishing weights are multiplied by the image features, so that the network can pay more attention to the features of key areas in the suspected polished image according to the weight size, effectively enhancing the ability to capture key information reflecting the polishing quality and avoiding interference from secondary information.

[0095] The present invention calculates a polishing weight for each suspected polishing image at each angle. When the suspected polishing image is processed by an image processing channel, the multiplier M in the image processing channel inputs the polishing weight of the suspected polishing image at the corresponding angle.

[0096] The present invention uses an image processing channel to process a suspected polishing image at an angle, adds the output features of each image processing channel through adder A to achieve the fusion of image features at each angle, and then uses a fully connected layer to output the polishing accuracy quality score.

[0097] The present invention sets three fill light degrees at a shooting angle, and respectively collects images of the polished workpiece. The first photosensitivity is calculated by the difference between the medium fill light polishing image and the low fill light polishing image, and the second photosensitivity is calculated by the difference between the high fill light polishing image and the low fill light polishing image, so as to reflect the imaging difference of the polished workpiece for different fill light degrees. The suspected polishing degree of each pixel point is calculated by combining the first photosensitivity and the second photosensitivity, so as to highlight the polishing condition at the pixel point and improve the accuracy of polishing quality assessment.

[0098] The present invention calculates the polishing weight for the suspected polishing image at each angle, uses a polishing accuracy quality evaluation network to process the suspected polishing images at each angle, obtains a polishing accuracy quality score based on the attention applied by the polishing weight, realizes the integration of suspected polishing images at multiple angles, and improves the polishing quality assessment accuracy.

[0099] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for evaluating the grinding wheel grinding accuracy quality based on multi-angle visual inspection, characterized in that: The following steps are involved: S1. After the grinding wheel is polished, a shooting angle is selected to shoot the polished workpiece using low fill light, medium fill light and high fill light, respectively, to obtain a low fill light polishing image, a medium fill light polishing image and a high fill light polishing image; S2. Calculate the first light sensitivity of each pixel according to the difference between the medium fill light polishing image and the low fill light polishing image; S3, calculating the second photosensitivity of each pixel according to the difference between the high fill light polishing image and the low fill light polishing image; S4, calculating the suspected polishing degree of each pixel point according to the first photosensitivity and the second photosensitivity of each pixel point, and constructing a suspected polishing image; S5. For each suspected polishing image at each angle, the polishing weight is calculated, and the polishing accuracy quality score is obtained based on the polishing accuracy quality evaluation network; The S4 comprises the following sub-steps: S41, calculating a light sensitivity intensity change value according to a first light sensitivity intensity and a second light sensitivity intensity at a same pixel position; S42, calculating the suspected polishing degree of each pixel point according to the difference between the photosensitivity change value and the photosensitivity change reference value; S43, replacing the original pixel value of the same pixel with the suspected polishing degree to construct a suspected polishing image; The formula for calculating the photosensitivity change value in S41 is: , where γ i is the change in light intensity of the i-th pixel, q i,1 is the first light sensitivity of the i-th pixel, q i,2 is the second light sensitivity of the i-th pixel, I H For high fill light, I M is the medium fill light, I L is low fill light, i is a positive integer; The formula for calculating the suspected polishing degree of each pixel in S42 is: , where s i is the suspected polishing degree of the i-th pixel, γ ref is the reference value of photosensitivity change, γ i is the change in photosensitivity of the i-th pixel, where i is a positive integer.

2. The grinding wheel grinding precision quality evaluation method based on multi-angle visual inspection according to claim 1 is characterized in that: The S2 comprises the following sub-steps: S21, subtracting the R channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the R channel photosensitivity value; S22, subtracting the G channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the G channel photosensitivity value; S23, subtracting the B channel values ​​of the medium fill light polished image and the low fill light polished image at the same pixel position to obtain the B channel photosensitivity value; S24, adding the R channel light sensitivity value, the G channel light sensitivity value and the B channel light sensitivity value at the same pixel position to obtain a first light sensitivity intensity.

3. The grinding wheel grinding precision quality evaluation method based on multi-angle visual inspection according to claim 1 is characterized in that: The S3 comprises the following sub-steps: S31, subtracting the R channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the R channel photosensitivity value; S32, subtracting the G channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the G channel photosensitivity value; S33, subtracting the B channel values ​​of the high fill light polished image and the low fill light polished image at the same pixel position to obtain the B channel photosensitivity value; S34, adding the R channel light sensitivity value, the G channel light sensitivity value and the B channel light sensitivity value at the same pixel position to obtain a second light sensitivity intensity.

4. The grinding wheel grinding precision quality evaluation method based on multi-angle visual inspection according to claim 1 is characterized in that: The S5 comprises the following sub-steps: S51, calculating a polishing weight for each suspected polishing image at each angle; S52, inputting the suspected polishing images at multiple angles into the polishing accuracy quality evaluation network, and obtaining the polishing accuracy quality score based on the attention applied by the polishing weights.

5. The grinding wheel grinding precision quality evaluation method based on multi-angle visual inspection according to claim 4 is characterized in that: The formula for calculating the polishing weight in S51 is: , where α is the polishing weight, N is the number of pixels with a suspected polishing degree greater than 0 on the suspected polishing image, and N total It is the total number of pixels on the suspected polished image.

6. The grinding wheel grinding precision quality evaluation method based on multi-angle visual inspection according to claim 4 is characterized in that: The polishing accuracy quality evaluation network in S52 includes: multiple image processing channels, adder A and a fully connected layer; The input end of each of the image processing channels is used to input a suspected polishing image at an angle; The input end of the adder A is connected to the output ends of the multiple image processing channels respectively, and the output end thereof is connected to the input end of the fully connected layer; The output end of the fully connected layer serves as the output end of the polishing accuracy quality evaluation network.

7. The grinding wheel grinding precision quality evaluation method based on multi-angle visual inspection according to claim 6 is characterized in that: Each of the image processing channels comprises: a convolution layer, a multiplier M and a pooling layer; The input end of the convolutional layer is used as the input end of the image processing channel, and the output end thereof is connected to the first input end of the multiplier M; The second input terminal of the multiplier M is used to input the polishing weight; The input end of the pooling layer is connected to the output end of the multiplier M, and its output end serves as the output end of the image processing channel; The expression of the multiplier M is: ,in, is the output of the multiplier M, X is the feature of the convolutional layer output, and α is the polishing weight.

Images