Detection method for identifying bad appearance of product by using AI image

Abstract

The invention discloses a detection method for identifying bad appearance of a product by using an AI image, which belongs to the field of appearance detection and comprises the following steps of: acquiring a product surface image by using a high-resolution industrial camera under a control illumination condition; preprocessing the collected image to improve the image quality; carrying out AI image recognition on the preprocessed image, and analyzing whether flaws exist on the surface of the product or not; evaluating the bad appearance of the product according to the analysis result, and outputting an evaluation result; the specific process of analyzing whether flaws exist on the surface of the product comprises the following steps: constructing a texture analysis model, importing an image into the texture analysis model for AI analysis, and outputting a first analysis result; constructing a shape analysis model, importing the image into the shape analysis model for AI analysis, and outputting a second analysis result; and constructing a color analysis model, importing the image into the color analysis model for AI analysis, and outputting a third analysis result. According to the invention, the appearance of the product can be comprehensively identified, and the identification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of appearance inspection, and in particular to a method for inspecting product appearance defects using AI image recognition. Background Art

[0002] Product appearance inspection is an indispensable part of quality control. By selecting appropriate inspection methods and equipment and strictly implementing inspection standards and processes, we can ensure that the product appearance quality meets the requirements and enhance the company's market competitiveness and customer satisfaction. Currently, the most common and most intuitive inspection method is to directly observe the product appearance with human eyes to determine whether there are defects or undesirable conditions. This method is suitable for the initial inspection of most products, but it is easily affected by the subjectivity and experience level of personnel.

[0003] In order to reduce the workload of workers and improve detection accuracy, people use image recognition to identify defects on the surface of products and then determine the poor appearance of the products. However, the existing image recognition technology for product poor appearance detection has certain defects and cannot comprehensively identify the product appearance, resulting in insufficient recognition accuracy.

[0004] Therefore, those skilled in the art provide a method for detecting poor appearance of products using AI image recognition to solve the problems raised in the above background technology. Summary of the invention

[0005] The purpose of the present invention is to provide a method for detecting poor appearance of products using AI image recognition, which can comprehensively identify the appearance of products and improve the accuracy of recognition to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for detecting poor appearance of products using AI image recognition, comprising the following steps:

[0008] Use high-resolution industrial cameras to capture product surface images under controlled lighting conditions;

[0009] Preprocess the collected images to improve image quality;

[0010] Perform AI image recognition on the pre-processed images to analyze whether there are defects on the product surface;

[0011] Based on the analysis results, the product's poor appearance is evaluated and the evaluation results are output.

[0012] As a further solution of the present invention: the specific process of analyzing whether there are defects on the surface of the product is:

[0013] Build a texture analysis model, import the image into the texture analysis model for AI analysis, and output the first analysis result;

[0014] Construct a shape analysis model, import the image into the shape analysis model for AI analysis, and output the second analysis result;

[0015] Build a color analysis model, import the image into the color analysis model for AI analysis, and output the third analysis result;

[0016] Build a brightness analysis model, import the image into the brightness analysis model for AI analysis, and output the fourth analysis result;

[0017] If any one of the first analysis result, the second analysis result, the third analysis result, and the fourth analysis result is a defect, then the output product surface has a defect.

[0018] As a further solution of the present invention: the specific analysis process of the texture analysis model is:

[0019] Divide the product image into several sub-images, marked as Pi, i = 1···n, where n is a positive integer;

[0020] Perform image recognition on the texture structure on Pi. If there is a break in the texture on Pi, Pi is planned to the first sequence. If there is no break in the texture on Pi, Pi is planned to the second sequence.

[0021] Extract the texture end features of Pi in the second sequence, marked as Bi, i=1···j, where j is a positive integer and less than or equal to n;

[0022] Mark the preset texture end feature as bi, i = 1...k, where k is a positive integer;

[0023] Compare Bi with bi. If the comparison result of Bi is consistent with any bi, Bi is planned to the third sequence. If the comparison result of Bi is inconsistent with any bi, Bi is planned to the fourth sequence.

[0024] The number of Bi in the fourth sequence is counted and marked as A1. If A1>0, it means that there are defects in the surface texture of the product. If A1=0, it means that there are no defects in the surface texture of the product.

[0025] As a further solution of the present invention: the specific analysis process of the shape analysis model is:

[0026] By performing image recognition on the product in the image, the contour of the product is marked as C, and each corner in the contour is marked as Di, i=1···n, where n is a positive integer;

[0027] The preset product outline is marked as c, and each corner in the preset product outline is marked as di, i = 1···n;

[0028] Compare C and c for similarity. If the similarity between the two is greater than the preset similarity, it means that there is no defect in the product shape. If the similarity between the two is less than or equal to the preset similarity, it means that there is a defect in the product shape and proceed to the next step.

[0029] Compare Di and di one by one. If the comparison results of Di and di are consistent, Di is planned to the fifth sequence. If the comparison results of Di and di are inconsistent, Di is planned to the sixth sequence.

[0030] The number of Di in the sixth sequence is counted and marked as A2. If there is no defect in the product shape, output A2=0.

[0031] As a further solution of the present invention: the specific analysis process of the color analysis model is:

[0032] The distribution characteristics, density characteristics and change characteristics of the product color in the picture are extracted through image recognition technology, and the distribution characteristics, density characteristics and change characteristics are marked as X1, X2 and X3 respectively;

[0033] The distribution characteristics, density characteristics, and variation characteristics of the preset product color are marked as x1, x2, and x3 respectively;

[0034] Compare X1 with x1, X2 with x2, and X3 with x3. If the comparison results of any of X1 with x1, X2 with x2, and X3 with x3 are inconsistent, it means that the product color has defects and proceed to the next step. If the comparison results of any of X1 with x1, X2 with x2, and X3 with x3 are consistent, it means that the product color has no defects;

[0035] The number of items with inconsistent comparison results between X1 and x1, X2 and x2, and X3 and x3 is counted and marked as A3. If there is no defect in the product color, output A3=0.

[0036] As a further solution of the present invention: the specific analysis process of the brightness analysis model is:

[0037] Using image recognition technology, the areas with inconsistent product brightness in the image are marked as Yi, i = 1···n, and the brightness of Yi is extracted and marked as Zi;

[0038] The value with the largest brightness in Zi is marked as Zmax, and the value with the smallest brightness in Zi is marked as Zmin;

[0039] Calculate the contrast ΔZ = (Zmax - Zmin);

[0040] Compare ΔZ with the preset value Δz. If ΔZ>Δz, it means that the product brightness has defects and proceed to the next step. If ΔZ≤Δz, it means that the product brightness has no defects.

[0041] The brightness defect weight value A4=(ΔZ-Δz) is calculated, and if there is no defect in the product brightness, A4=0 is output.

[0042] As a further solution of the present invention: the specific process of evaluating the poor appearance of the product according to the analysis results is:

[0043] Extract A1 from the first analysis result, A2 from the second analysis result, A3 from the third analysis result, and A4 from the fourth analysis result;

[0044] Calculate the bad weight value Q = (A1 / a1) + (A2 / a2) + (A3 / a3) + (A4 / a4), where a1, a2, a3, and a4 are all preset values ​​and can be set according to needs.

[0045] As a further solution of the present invention: the specific process of the pretreatment is:

[0046] Denoising: using filtering algorithms to remove noise from images and improve image quality;

[0047] Enhancement: Enhance the image through histogram equalization and contrast adjustment methods to make the defect features more prominent;

[0048] Grayscale, converting color images into grayscale images to simplify subsequent processing.

[0049] As a further solution of the present invention: after the surface analysis of the product shows defects, the cause of the defects can be determined by reanalyzing the defect characteristics, specifically:

[0050] If any single item of the product surface texture, product shape, product color, and product brightness on the sub-image Pi has a defect, it means that the defect on the product surface on the sub-image Pi is caused individually, and the records of the corresponding defects in the past are retrieved, and the defect cause with the most repetitions is selected as the final defect cause;

[0051] If there are defects in both the product surface texture and product color on the sub-image Pi, it means that the defects on the product surface on the sub-image Pi are concurrent, and the records of the texture and color defects that appeared simultaneously in the past are retrieved, and the defect cause with the most repetitions is selected as the final defect cause;

[0052] If there are defects in both the product surface texture and the product brightness on the sub-image Pi, it means that the defects on the product surface on the sub-image Pi are concurrent. Retrieve the records of the texture and brightness defects that appeared simultaneously in the past, and select the defect cause with the most repetitions as the final defect cause;

[0053] If there are defects in the surface texture, product brightness and product color of the product on the sub-image Pi at the same time, it means that the defects on the surface of the product on the sub-image Pi are concurrent. The records of the texture, brightness and color defects that appeared simultaneously in the past are retrieved, and the defect cause with the most repetitions is selected as the final defect cause.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. This application uses AI image recognition to comprehensively identify the product appearance and improve the accuracy of recognition. It analyzes and determines whether there are defects on the product surface from four angles: texture, shape, color, and brightness, thereby further improving the detection accuracy.

[0056] 2. The texture analysis model of the present application determines whether the texture on the product is normal by comparing the collected product texture fracture end structure with the preset texture fracture end normal structure, while the shape analysis model first performs an initial analysis through the similarity of the contours, and then quickly analyzes whether there are any defects in the shape of the product appearance through comparison analysis of each corner of the contour. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The present invention is a flowchart of a method for detecting product appearance defects using AI image recognition. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] As mentioned in the background technology of this application, it has been found through research that the existing image recognition technology for product defect detection cannot fully identify the product appearance, resulting in insufficient recognition accuracy and certain defects.

[0060] In order to solve the above-mentioned defects, the present application discloses a method for detecting poor appearance of products using AI image recognition, which can comprehensively identify the appearance of the product and improve the accuracy of recognition.

[0061] The following will describe in detail how the solution of the present application solves the above technical problems in conjunction with the accompanying drawings.

[0062] See also Figure 1 In an embodiment of the present invention, a method for detecting poor appearance of a product using AI image recognition includes the following steps: using a high-resolution industrial camera to capture a product surface image under controlled lighting conditions; preprocessing the captured image to improve image quality; performing AI image recognition on the preprocessed image to analyze whether there are defects on the product surface; evaluating the poor appearance of the product based on the analysis results, and outputting the evaluation results. This application uses AI image recognition to comprehensively identify the appearance of the product and improve the accuracy of recognition.

[0063] In this embodiment, the specific process of analyzing whether there are defects on the surface of the product is as follows: construct a texture analysis model, import the image into the texture analysis model for AI analysis, and output the first analysis result; construct a shape analysis model, import the image into the shape analysis model for AI analysis, and output the second analysis result; construct a color analysis model, import the image into the color analysis model for AI analysis, and output the third analysis result; construct a brightness analysis model, import the image into the brightness analysis model for AI analysis, and output the fourth analysis result; if any one of the first analysis result, the second analysis result, the third analysis result, and the fourth analysis result is a defect, then output that there are defects on the product surface. This application analyzes and determines whether there are defects on the product surface from four perspectives: texture, shape, color, and brightness, and further improves the detection accuracy.

[0064] In this embodiment, the specific analysis process of the texture analysis model is: divide the product image into several sub-images, marked as Pi, i=1···n, where n is a positive integer; perform image recognition on the texture structure on Pi, if there is a break in the texture on Pi, Pi is planned to the first sequence, if there is no break in the texture on Pi, Pi is planned to the second sequence; extract the texture end feature of Pi in the second sequence, marked as Bi, i=1···j, where j is a positive integer and less than or equal to n; mark the preset texture end feature as bi, i=1···k, where k is a positive integer; compare Bi with bi, if the comparison result of Bi with any bi is consistent, Bi is planned to the third sequence, if the comparison result of Bi with any bi is inconsistent, Bi is planned to the fourth sequence; count the number of Bi in the fourth sequence and mark it as A1, if A1>0, it means that there are defects in the surface texture of the product, if A1=0, it means that there are no defects in the surface texture of the product. Analyze whether the fracture direction is consistent with the direction of the surrounding texture. If the fracture direction is chaotic or obviously inconsistent with the direction of the surrounding texture, this is a sign of a defect. Carefully observe whether the fracture edge is neat and smooth or has irregularities such as burrs and cracks. Irregular fracture edges often mean that the product is defective or has been poorly processed. This application compares the collected product texture fracture end structure with the preset texture fracture end normal structure to determine whether the texture on the product is normal.

[0065] In this embodiment, the specific analysis process of the shape analysis model is as follows: by performing image recognition on the product in the image, the contour of the product is marked as C, and each corner in the contour is marked as Di, i=1···n, where n is a positive integer; the preset product contour is marked as c, and each corner in the preset product contour is marked as di, i=1···n; C and c are compared for similarity, if the similarity between the two is greater than the preset similarity, it means that there is no defect in the product shape, if the similarity between the two is less than or equal to the preset similarity, it means that there is a defect in the product shape and enter the next step; Di and di are compared one by one, if the comparison results of Di and di are consistent, Di is planned to the fifth sequence, if the comparison results of Di and di are inconsistent, Di is planned to the sixth sequence; the number of Di in the sixth sequence is counted and marked as A2, if there is no defect in the product shape, A2=0 is output. This setting first performs an initial analysis through the similarity of the contour, and then compares and analyzes each corner of the contour to quickly analyze whether there is a defect in the shape of the product appearance.

[0066] In this embodiment, the specific analysis process of the color analysis model is as follows: extract the distribution characteristics, density characteristics and change characteristics of the product color in the picture through image recognition technology, and mark the distribution characteristics, density characteristics and change characteristics as X1, X2 and X3 respectively; mark the distribution characteristics, density characteristics and change characteristics of the preset product color as x1, x2 and x3 respectively; compare X1 with x1, X2 with x2, and X3 with x3. If the comparison results of any one of X1 with x1, X2 with x2, and X3 with x3 are inconsistent, it means that the product color has defects and enters the next step. If the comparison results of any one of X1 with x1, X2 with x2, and X3 with x3 are consistent, it means that the product color has no defects; count the number of items with inconsistent comparison results among X1 with x1, X2 with x2, and X3 with x3 and mark them as A3. If the product color has no defects, output A3=0. This setting can quickly determine whether the color of the product appearance has defects.

[0067] In this embodiment, the specific analysis process of the brightness analysis model is as follows: by using image recognition technology, the areas with inconsistent product brightness in the picture are marked as Yi, i=1···n, and the brightness of Yi is extracted and marked as Zi; the value with the largest brightness in Zi is marked as Zmax, and the value with the smallest brightness in Zi is marked as Zmin; the contrast ΔZ=(Zmax-Zmin); ΔZ is compared with the preset value Δz. If ΔZ>Δz, it means that there is a defect in the product brightness and enters the next step. If ΔZ≤Δz, it means that there is no defect in the product brightness; the brightness defect weight value A4=(ΔZ-Δz) is calculated. If there is no defect in the product brightness, A4=0 is output. The brightness and contrast of the product surface are also key factors in identifying defects. The contrast is the brightness ratio of the brightest part to the darkest part of the image, which is used to describe the degree of difference between different brightness areas in the image. This setting can quickly determine whether there is a defect in the brightness of the product appearance.

[0068] In this embodiment, the specific process of evaluating the poor appearance of the product according to the analysis results is: extracting A1 in the first analysis result, A2 in the second analysis result, A3 in the third analysis result, and A4 in the fourth analysis result; calculating the poor weight value Q = (A1 / a1) + (A2 / a2) + (A3 / a3) + (A4 / a4), where a1, a2, a3, and a4 are all preset values ​​and can be set according to needs. This setting can quickly evaluate and determine the degree of poor appearance of the product, where the larger the poor weight value Q, the more appearance defects the product has, and the product is a defective product.

[0069] In this embodiment, the specific process of preprocessing is: denoising, using filtering algorithms (such as median filtering, Gaussian filtering) to remove noise in the image and improve image quality; enhancing, enhancing the image through histogram equalization and contrast adjustment methods to make the defect features more prominent; grayscale, converting the color image into a grayscale image to simplify the subsequent processing process. Among them, median filtering is a commonly used image processing technology, which reduces noise and smoothes the image by replacing the central pixel with the median of the pixel value in the window. Gaussian filtering is a linear smoothing filtering technology widely used in the field of image processing and computer vision. Its basic principle is to use Gaussian functions to perform convolution operations on the image to achieve the purpose of smoothing the image and reducing noise. Histogram equalization is a method of transforming the original image to obtain a new image with a uniformly distributed grayscale histogram. Its purpose is to enhance the contrast of the image. In particular, when the contrast of the useful data of the image is quite close, the brightness can be better distributed on the histogram through histogram equalization, thereby enhancing the local contrast without affecting the overall contrast. Contrast adjustment is an important technique in image processing, which can be achieved in a variety of ways to enhance the visual effect and layering of images.

[0070] In this embodiment, after the product surface is analyzed for defects, the defect characteristics can be re-analyzed to determine the cause of the defect, specifically: if any single item of the product surface texture, product shape, product color, and product brightness on the sub-image Pi has defects, it means that the defect on the product surface on the sub-image Pi is caused separately, and the records of the corresponding defects in the past are retrieved, and the defect cause with the most repetitions is screened out as the final defect cause; if the product surface texture and product color on the sub-image Pi have defects at the same time, it means that the defects on the product surface on the sub-image Pi are concurrent, and the records of the texture and color defects that appeared simultaneously in the past are retrieved, and the defect cause with the most repetitions is screened out. Select the defect cause with the most repetitions as the final defect cause; if there are defects in both the product surface texture and product brightness on the sub-image Pi, it means that the defects on the product surface on the sub-image Pi are concurrent, retrieve the records of the previous simultaneous occurrence of texture and brightness defects, and select the defect cause with the most repetitions as the final defect cause; if there are defects in the product surface texture, product brightness and product color on the sub-image Pi, it means that the defects on the product surface on the sub-image Pi are concurrent, retrieve the records of the previous simultaneous occurrence of texture, brightness and color defects, and select the defect cause with the most repetitions as the final defect cause. This setting can quickly determine the cause of product surface defects, making it easier for staff to quickly optimize and remedy.

[0071] The present invention uses AI image recognition to comprehensively identify the appearance of the product, thereby improving the accuracy of recognition. Among them, it analyzes and determines whether there are defects on the product surface from four perspectives: texture, shape, color, and brightness, thereby further improving the detection accuracy. The texture analysis model of the present application compares the collected product texture fracture end structure with the preset texture fracture end normal structure to determine whether the texture on the product is normal, while the shape analysis model first performs an initial analysis through the similarity of the contour, and then quickly analyzes whether there are defects in the shape of the product appearance through the comparison analysis of each corner of the contour.

[0072] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0073] What has been described above are only preferred specific implementations of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting product appearance defects using AI image recognition, characterized in that: The following steps are involved: Use high-resolution industrial cameras to capture product surface images under controlled lighting conditions; Preprocess the collected images to improve image quality; Perform AI image recognition on the pre-processed images to analyze whether there are defects on the product surface; Based on the analysis results, the product's poor appearance is evaluated and the evaluation results are output.

2. The method for detecting product appearance defects using AI image recognition according to claim 1, characterized in that: The specific process of analyzing whether there are defects on the product surface is as follows: Build a texture analysis model, import the image into the texture analysis model for AI analysis, and output the first analysis result; Construct a shape analysis model, import the image into the shape analysis model for AI analysis, and output the second analysis result; Build a color analysis model, import the image into the color analysis model for AI analysis, and output the third analysis result; Build a brightness analysis model, import the image into the brightness analysis model for AI analysis, and output the fourth analysis result; If any one of the first analysis result, the second analysis result, the third analysis result, and the fourth analysis result is a defect, then the output product surface has a defect.

3. The method for detecting product appearance defects using AI image recognition according to claim 2, characterized in that: The specific analysis process of the texture analysis model is as follows: Divide the product image into several sub-images, marked as Pi, i = 1···n, where n is a positive integer; Perform image recognition on the texture structure on Pi. If there is a break in the texture on Pi, Pi is planned to the first sequence. If there is no break in the texture on Pi, Pi is planned to the second sequence. Extract the texture end features of Pi in the second sequence, marked as Bi, i=1···j, where j is a positive integer and less than or equal to n; Mark the preset texture end feature as bi, i = 1...k, where k is a positive integer; Compare Bi with bi. If the comparison result of Bi is consistent with any bi, Bi is planned to the third sequence. If the comparison result of Bi is inconsistent with any bi, Bi is planned to the fourth sequence. The number of Bi in the fourth sequence is counted and marked as A1. If A1>0, it means that there are defects in the surface texture of the product. If A1=0, it means that there are no defects in the surface texture of the product.

4. The method for detecting product appearance defects using AI image recognition according to claim 3, characterized in that: The specific analysis process of the shape analysis model is as follows: By performing image recognition on the product in the image, the contour of the product is marked as C, and each corner in the contour is marked as Di, i=1···n, where n is a positive integer; The preset product outline is marked as c, and each corner in the preset product outline is marked as di, i = 1···n; Compare C and c for similarity. If the similarity between the two is greater than the preset similarity, it means that there is no defect in the product shape. If the similarity between the two is less than or equal to the preset similarity, it means that there is a defect in the product shape and proceed to the next step. Compare Di and di one by one. If the comparison results of Di and di are consistent, Di is planned to the fifth sequence. If the comparison results of Di and di are inconsistent, Di is planned to the sixth sequence. The number of Di in the sixth sequence is counted and marked as A2. If there is no defect in the product shape, output A2=0.

5. The method for detecting product appearance defects using AI image recognition according to claim 4, characterized in that: The specific analysis process of the color analysis model is as follows: The distribution characteristics, density characteristics and change characteristics of the product color in the picture are extracted through image recognition technology, and the distribution characteristics, density characteristics and change characteristics are marked as X1, X2 and X3 respectively; The distribution characteristics, density characteristics, and variation characteristics of the preset product color are marked as x1, x2, and x3 respectively; Compare X1 with x1, X2 with x2, and X3 with x3. If the comparison results of any of X1 with x1, X2 with x2, and X3 with x3 are inconsistent, it means that the product color has defects and proceed to the next step. If the comparison results of any of X1 with x1, X2 with x2, and X3 with x3 are consistent, it means that the product color has no defects; The number of items with inconsistent comparison results between X1 and x1, X2 and x2, and X3 and x3 is counted and marked as A3. If there is no defect in the product color, output A3=0.

6. The method for detecting product appearance defects using AI image recognition according to claim 5, characterized in that: The specific analysis process of the brightness analysis model is as follows: Using image recognition technology, the areas with inconsistent product brightness in the image are marked as Yi, i = 1···n, and the brightness of Yi is extracted and marked as Zi; The value with the largest brightness in Zi is marked as Zmax, and the value with the smallest brightness in Zi is marked as Zmin; Calculate the contrast ΔZ = (Zmax - Zmin); Compare ΔZ with the preset value Δz. If ΔZ>Δz, it means that the product brightness has defects and proceed to the next step. If ΔZ≤Δz, it means that the product brightness has no defects. The brightness defect weight value A4=(ΔZ-Δz) is calculated, and if there is no defect in the product brightness, A4=0 is output.

7. The method for detecting product appearance defects using AI image recognition according to claim 6, characterized in that: The specific process of evaluating the product's poor appearance based on the analysis results is as follows: Extract A1 from the first analysis result, A2 from the second analysis result, A3 from the third analysis result, and A4 from the fourth analysis result; Calculate the bad weight value Q = (A1 / a1) + (A2 / a2) + (A3 / a3) + (A4 / a4), where a1, a2, a3, and a4 are all preset values.

8. The method for detecting product appearance defects using AI image recognition according to claim 7, characterized in that: The specific process of the pretreatment is: Denoising: using filtering algorithms to remove noise from images and improve image quality; Enhancement: Enhance the image through histogram equalization and contrast adjustment methods to make the defect features more prominent; Grayscale, converting color images into grayscale images to simplify subsequent processing.

9. The method for detecting product appearance defects using AI image recognition according to claim 8, characterized in that: If the product surface analysis shows defects, the cause of the defects can be determined by reanalyzing the defect characteristics, specifically: If any single item of the product surface texture, product shape, product color, and product brightness on the sub-image Pi has a defect, it means that the defect on the product surface on the sub-image Pi is caused individually, and the records of the corresponding defects in the past are retrieved, and the defect cause with the most repetitions is selected as the final defect cause; If there are defects in both the product surface texture and product color on the sub-image Pi, it means that the defects on the product surface on the sub-image Pi are concurrent, and the records of the texture and color defects that appeared simultaneously in the past are retrieved, and the defect cause with the most repetitions is selected as the final defect cause; If there are defects in both the product surface texture and the product brightness on the sub-image Pi, it means that the defects on the product surface on the sub-image Pi are concurrent. Retrieve the records of the texture and brightness defects that appeared simultaneously in the past, and select the defect cause with the most repetitions as the final defect cause; If there are defects in the surface texture, product brightness and product color of the product on the sub-image Pi at the same time, it means that the defects on the surface of the product on the sub-image Pi are concurrent. The records of the texture, brightness and color defects that appeared simultaneously in the past are retrieved, and the defect cause with the most repetitions is selected as the final defect cause.

Images