A box type steel member welding weld appearance visual identification system

Abstract

The application discloses a kind of weld appearance visual identification systems for box type steel member welding, comprising: image acquisition module, image preprocessing module, image recognition module;It is related to the technical field of weld identification, solves the technical problem that it is difficult to accurately identify weld area under the condition that the background of weld image exists and the weld surface texture is relatively complex;By the processing of the obtained weld image is rotated and cut, obtain several processing images, and mark the position of each pixel in the image before the weld image processing, then identify the pixel in the processing image based on the weld detection model, obtain the weld pixel position information, finally based on the weld pixel position information corresponding to all processing images restores weld image, by this kind of way, it can avoid the interference of other contents in image to weld identification leads to the situation that weld is missed identification occurs, to improve the accuracy of weld identification.

Description

Technical Field

[0001] This invention belongs to the field of weld seam recognition technology, specifically a visual recognition system for the appearance of weld seams in box-shaped steel components. Background Technology

[0002] In the traditional welding process of box-type steel components, the inspection of weld appearance quality mainly relies on manual visual inspection or simple mechanical inspection tools. This method is not only time-consuming and labor-intensive, but also easily affected by subjective factors, resulting in inaccurate and inconsistent inspection results. With the development of industrial automation and intelligence, there is an urgent need for a system that can automatically, quickly, and accurately identify weld appearance quality.

[0003] Most weld inspection systems currently on the market employ single image processing methods, such as edge detection and template matching. While these methods can identify welds to some extent, they are easily affected by noise or other elements in complex backgrounds, leading to incorrect weld identification. This is especially true when the weld image contains a background or the weld surface texture is complex, making accurate identification of the weld area difficult. Summary of the Invention

[0004] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a visual recognition system for weld appearance in box-shaped steel components, which solves the technical problem of difficulty in accurately identifying weld areas when the weld image has a background and the weld surface texture is relatively complex. This invention obtains several processed images by rotating and cropping the acquired welding image. Before processing the welding image, the position of each pixel in the image is marked. Then, the pixels in the processed image are identified based on the weld detection model to obtain the weld pixel position information. Finally, the weld image is restored based on the weld pixel position information corresponding to all processed images to solve the above-mentioned problem.

[0005] To achieve the above objectives, a first aspect of the present invention provides a visual recognition system for weld appearance of box-shaped steel components, comprising: an image acquisition module, an image preprocessing module, and an image recognition module;

[0006] The image acquisition module is used to capture welding images of the welding area using an imaging device.

[0007] The image preprocessing module is used to mark the pixels in the welding image and then process the welding image to obtain several processed images.

[0008] The image recognition module: inputs several images into the weld detection model to obtain weld pixel position information corresponding to the several processed images; wherein, the weld detection model is constructed based on an artificial intelligence model; and,

[0009] The weld seam locations corresponding to several processed images are processed to obtain weld seam images.

[0010] Preferably, the processing of the welding image after marking the pixels in the welding image includes:

[0011] A reference rectangular coordinate system is established with the center point of the welding image as the origin. The pixel position information of each pixel is obtained based on the reference rectangular coordinate system, and the origin of the reference rectangular coordinate system is marked as the reference point.

[0012] Rotate the welding image n times at random angles with the origin of the rectangular coordinate system as the center, and then crop the n rotated welding images to a specified size with the origin of the rectangular coordinate system as the center, to obtain n rotated images; where n is a positive integer and n≤360;

[0013] The welding image is cropped to obtain several cropped images; the rotated image and the cropped image are marked as processed images.

[0014] It should be noted that even after cropping and rotating the welding image, the position information of the pixels in the original welding image remains unchanged; that is, the processing of the welding image will not affect the pixel position information of the pixel marker.

[0015] Preferably, the cropping process for the welding image includes:

[0016] Step 1: Using the x-axis of the rectangular coordinate system as the dividing line, rotate the x-axis at random angles n times, and then use the x-axis as the dividing line to crop the welding image, resulting in 2n cropped images;

[0017] Step 2: Determine if the size of each cropped image conforms to the specified size; if yes, mark the cropped image as a processed image; otherwise, proceed to Step 3.

[0018] Step 3: Compare the size of the cropped image with the specified size; if the size of the cropped image is smaller than the specified size, repair the cropped image and proceed to Step 2; if the size of the cropped image is larger than the specified size, crop the image again and proceed to Step 2.

[0019] It should be noted that when repairing the cropped image, a white image is used to repair the cropped image to conform to the input specifications of the subsequent model. When cropping, the center point of the cropped image is used as the reference point to crop it to the specified size.

[0020] Preferably, the weld inspection model is constructed based on an artificial intelligence model, including:

[0021] Several welding images are extracted from the database. A Cartesian coordinate system is established with the center of the welding images, and the positions of the welding pixels in the welding images are obtained to obtain the welding pixel position information of the corresponding welding images. A training set and a test set are constructed based on several welding images and the corresponding welding pixel position information.

[0022] A weld detection model to be trained is constructed based on the YOLO model; the weld detection model to be trained is trained using a training set, and the trained weld detection model is tested using a test set; the trained weld detection model is adjusted according to the test accuracy; finally, a weld detection model is obtained with the welding image as input and the welding pixel position information as output.

[0023] Preferably, the processing of the weld positions corresponding to the plurality of processed images includes:

[0024] Extract the welding pixel position information corresponding to several processed images and create a blank image of a specified size;

[0025] Extract the pixel information of the pixels corresponding to the welding pixel positions, and assign the pixel information of the corresponding pixels to the corresponding pixels in the blank image according to the welding pixel positions to obtain the initial image; correct the initial image to obtain the weld image; wherein, the pixel information includes, but is not limited to, the RGB color value, contrast, and grayscale value of the pixel.

[0026] Preferably, the correction of the initial image includes:

[0027] In the initial image, count the number of times each pixel appears in several welding pixel position information and mark it as the pixel repetition number;

[0028] The number of pixel repetitions Li of several pixels is compared with the number threshold. If the number of pixel repetitions Li of a pixel is less than the number threshold, the corresponding pixel is marked as a weak pixel. Otherwise, the corresponding pixel is marked as a strong pixel. Here, Li is the number of times the i-th pixel appears.

[0029] Weak pixels in the initial image are removed to obtain the weld image.

[0030] It should be noted that the quantity threshold is determined based on the total number of processed images containing the pixel, and the quantity threshold can be 20% of the total number of processed images containing the pixel.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: by rotating and cropping the acquired welding image to obtain several processed images, and marking the position of each pixel in the image before processing the welding image, the pixels in the processed image are identified based on the weld detection model to obtain the weld pixel position information, and finally the weld image is restored based on the weld pixel position information corresponding to all processed images. This method can avoid interference from other content in the image to weld recognition, which may lead to missed weld recognition, thereby improving the accuracy of weld recognition. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a schematic diagram of the process of the present invention;

[0034] Figure 2 This is a logic block diagram of the welding image cropping process of the present invention;

[0035] Figure 3 This is a schematic diagram of the module connections in this invention. Detailed Implementation

[0036] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Please see Figures 1-3 The first aspect of the present invention provides a visual recognition system for weld appearance of box-shaped steel components, including: an image acquisition module, an image preprocessing module, and an image recognition module;

[0038] Please see Figure 3 As shown, the image acquisition module, image preprocessing module, and image recognition module are connected in a progressive manner;

[0039] Please see Figure 1 As shown, the image acquisition module is used to capture welding images of the welding area through an imaging device;

[0040] Image preprocessing module: This module is used to process the welding image after marking the pixels in the welding image, resulting in several processed images.

[0041] Image recognition module: Inputs several images into the weld detection model to obtain weld pixel position information corresponding to the processed images; wherein, the weld detection model is built based on an artificial intelligence model; and,

[0042] The weld seam locations corresponding to several processed images are processed to obtain weld seam images.

[0043] The information marked on each pixel remains unchanged during the processing of the welding image.

[0044] By processing the welding image, multiple processed images based on the welding image are obtained. Then, the position information of the weld pixels in the processed image is identified by the trained weld recognition model. Finally, the weld pixel position information of all processed images is integrated to obtain the weld image. This method can avoid background drying and improve accuracy.

[0045] In this embodiment, the welding image is processed after pixel marking in the welding image, including:

[0046] The center point of the welding image is taken as the reference point, i.e. the origin, and a reference rectangular coordinate system is established with the straight line perpendicular to the edge of the image as the X-axis and Y-axis. The pixel position information of each pixel in the welding image is obtained and marked based on the reference rectangular coordinate system.

[0047] The welding image is rotated 36 times at random angles with the origin of the rectangular coordinate system as the center, and the 36 rotated welding images are cropped to a specified size with the origin of the rectangular coordinate system as the center, resulting in 36 rotated images.

[0048] The welding image is cropped to obtain several cropped images; the rotated image and the cropped image are marked as processed images.

[0049] During implementation, even after cropping and rotating the welding image, the position information of the pixels in the original welding image remains unchanged; that is, the processing of the welding image will not affect the pixel position information of the pixel marker.

[0050] By using pixel marking, even after rotation and cropping, the positional information of pixels in the original welding image can be kept unchanged, thus ensuring the accuracy of subsequent integration. Furthermore, the various processed images generated by rotation and cropping can increase the diversity and robustness of the image recognition model and improve the accuracy of weld recognition.

[0051] Please see Figure 2As shown, in this embodiment, in order to effectively avoid the background in the welding image from interfering with subsequent recognition, the welding image is cropped, including:

[0052] Step 1: Using the x-axis of the rectangular coordinate system as the dividing line, rotate the x-axis 36 times at random angles, and then crop the welding image using the x-axis as the dividing line to obtain 72 cropped images;

[0053] Step 2: Determine if the size of each cropped image conforms to the specified size; if yes, mark the cropped image as a processed image; otherwise, proceed to Step 3.

[0054] Step 3: Compare the size of the cropped image with the specified size; if the size of the cropped image is smaller than the specified size, repair the cropped image and proceed to Step 2; if the size of the cropped image is larger than the specified size, crop the image again and proceed to Step 2.

[0055] When repairing the cropped image as described above, a white image is used to repair the cropped image to conform to the input specifications of the subsequent model. When cropping, the center point of the cropped image is used as the reference point to crop it to the specified size.

[0056] The specified dimensions mentioned above can be set according to requirements. In this embodiment, the specified dimensions can be set to 512×512.

[0057] The weld inspection model described in this example is built based on an artificial intelligence model, including:

[0058] Several welding images are extracted from the database. A Cartesian coordinate system is established with the center pixel of the welding image as the reference point, and the position of the welding pixel in the welding image is obtained to obtain the welding pixel position information of the corresponding welding image. A training set and a test set are constructed based on several welding images and the corresponding welding pixel position information. The ratio of the training set to the test set is 8:2.

[0059] A weld detection model to be trained is constructed based on the YOLO model; the weld detection model to be trained is trained using a training set, and the trained weld detection model is tested using a test set; the trained weld detection model is adjusted according to the test accuracy; finally, a weld detection model is obtained with the welding image as input and the welding pixel position information as output.

[0060] In this embodiment, the weld positions corresponding to several processed images are processed, including:

[0061] Extract the welding pixel position information corresponding to several processed images and create a blank image of a specified size;

[0062] Extract the RGB color value, contrast, and grayscale value of the pixels corresponding to the welding pixel position information. Assign the pixel information of the corresponding pixel to the corresponding pixel in the blank image according to the welding pixel position to obtain the initial image. Correct the initial image to obtain the weld image.

[0063] Removing weak pixels can reduce the impact of noise, making the weld image clearer and facilitating subsequent weld identification and analysis.

[0064] Furthermore, the initial image is corrected, including:

[0065] In the initial image, count the number of times each pixel appears in several welding pixel position information and mark it as the pixel repetition number;

[0066] The number of pixel repetitions Li of several pixels is compared with the number threshold. If the number of pixel repetitions Li of a pixel is less than the number threshold, the corresponding pixel is marked as a weak pixel. Otherwise, the corresponding pixel is marked as a strong pixel. Here, Li is the number of times the i-th pixel appears.

[0067] Weak pixels in the initial image are removed to obtain the weld image.

[0068] The quantity threshold mentioned above can be 5% of the total number of images processed, i.e. (36+72)×5%=5.4. At this time, 5.4 can be rounded up, i.e., the quantity threshold is 6.

[0069] The working principle of this invention is as follows: The welding image of the welding area is captured by the imaging device, the position of the pixels in the welding image is marked, and the welding image is processed to obtain several processed images. The several images are input into the weld detection model to obtain the weld pixel position information corresponding to the several processed images. The welding pixel position information corresponding to the several processed images is extracted respectively to establish a blank image of a specified size.

[0070] Extract the pixel information of the pixels corresponding to the welding pixel positions, and assign the pixel information of the corresponding pixels to the corresponding pixels in the blank image according to the welding pixel positions to obtain the initial image; correct the initial image to obtain the weld image.

[0071] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A visual recognition system for weld appearance in box-shaped steel components, characterized in that, include: Image acquisition module, image preprocessing module, image recognition module; The image acquisition module is used to capture welding images of the welding area using an imaging device. The image preprocessing module is used to mark the pixels in the welding image and then process the welding image to obtain several processed images. The image recognition module: inputs several images into the weld detection model to obtain weld pixel position information corresponding to the several processed images; wherein, the weld detection model is constructed based on an artificial intelligence model; and, The weld seam positions corresponding to several processed images are processed to obtain weld seam images; The process of marking pixels in the welding image and then processing the welding image includes: A reference rectangular coordinate system is established with the center point of the welding image as the origin. The pixel position information of each pixel is obtained based on the reference rectangular coordinate system, and the origin of the reference rectangular coordinate system is marked as the reference point. Rotate the welding image n times at random angles with the origin of the Cartesian coordinate system as the center, and then crop the n rotated welding images to a specified size with the origin of the Cartesian coordinate system as the center, to obtain n rotated images; where n is a positive integer; The welding image is cropped to obtain several cropped images; the rotated image and the cropped image are marked as processed images; The cropping process for the welding image includes: Step 1: Using the x-axis of the rectangular coordinate system as the dividing line, rotate the x-axis at random angles n times, and then use the x-axis as the dividing line to crop the welding image, resulting in 2n cropped images; Step 2: Determine if the size of each cropped image conforms to the specified size; if yes, mark the cropped image as a processed image; otherwise, proceed to Step 3. Step 3: Compare the size of the cropped image with the specified size; if the size of the cropped image is smaller than the specified size, repair the cropped image and proceed to Step 2; if the size of the cropped image is larger than the specified size, crop the image again and proceed to Step 2. The processing of the weld positions corresponding to several processed images includes: Extract the welding pixel position information corresponding to several processed images and create a blank image of a specified size; Extract the pixel information of the pixels corresponding to the welding pixel positions, and assign the pixel information of the corresponding pixels to the corresponding pixels in the blank image according to the welding pixel positions to obtain the initial image; correct the initial image to obtain the weld image; wherein, the pixel information includes, but is not limited to, the RGB color value, contrast, and grayscale value of the pixel.

2. The visual recognition system for weld appearance of box-type steel components according to claim 1, characterized in that, The correction of the initial image includes: In the initial image, count the number of times each pixel appears in several welding pixel position information and mark it as the pixel repetition number; The number of pixel repetitions Li of several pixels is compared with the number threshold. If the number of pixel repetitions Li of a pixel is less than the number threshold, the corresponding pixel is marked as a weak pixel. Otherwise, the corresponding pixel is marked as a strong pixel. Here, Li is the number of times the i-th pixel appears. Weak pixels in the initial image are removed to obtain the weld image.

3. The weld appearance visual recognition system for box-type steel components according to claim 1, characterized in that, The weld inspection model is built based on an artificial intelligence model and includes: Several welding images are extracted from the database. A Cartesian coordinate system is established with the center of the welding images, and the positions of the welding pixels in the welding images are obtained to obtain the welding pixel position information of the corresponding welding images. A training set and a test set are constructed based on several welding images and the corresponding welding pixel position information. A weld detection model to be trained is constructed based on the YOLO model; the weld detection model to be trained is trained using a training set, and the trained weld detection model is tested using a test set; the trained weld detection model is adjusted according to the test accuracy; finally, a weld detection model is obtained with the welding image as input and the welding pixel position information as output.

Images