Defect detection method and system for charger power supply, and storage medium

Abstract

The invention discloses a defect detection method and system for a charger power source and a storage medium, and belongs to the field of charger detection.The method comprises the following steps that the interior and the exterior of a power source outer shell are shot through a binocular camera, picture data shot by the two cameras of the power source outer shell are obtained, and the power source outer shell is obtained according to different parallax errors of the two cameras; the depth of each pixel point is calculated through the triangulation principle, so that depth feature maps of depth channels corresponding to RGB channels inside and outside the power supply outer shell are obtained, pixel deviation in depth and texture space is obtained, segmentation is achieved through a segmentation network, a training result is evaluated according to a label, and the depth feature maps of the depth channels corresponding to the RGB channels inside and outside the power supply outer shell are obtained. And finally obtaining internal and external shooting data and reference information. Performing RGB restoration on each pixel point by using the trained model to obtain a final restored power supply pixel point; and the repaired information is transmitted to the defect detection module to judge whether defects exist or not.

Description

Technical Field

[0001] The present invention belongs to the field of charger detection, and in particular relates to a defect detection method and system for a charger power supply, and a storage medium. Background Art

[0002] Machine vision is a comprehensive technology, which includes image processing, mechanical engineering technology, control, electric light source lighting, optical imaging, sensors, analog and digital video technology, computer software and hardware technology. A 5178-type machine vision application system includes image capture, light source system, image digitization module, digital image processing module, intelligent judgment and decision module and mechanical control execution module. The most basic feature of the machine vision system is to improve the flexibility and automation of production.

[0003] In some dangerous working environments that are not suitable for manual work or where artificial vision cannot meet the requirements, machine vision is often used to replace artificial vision. At the same time, in large-scale repetitive industrial production processes, the use of machine vision inspection methods can greatly improve production efficiency and automation.

[0004] However, in the existing technology, when identifying the power charger, most of them are through manual or other types of detection methods (such as ultrasonic and other non-destructive detection methods) to detect surface defects and internal defects. However, due to the characteristics of the power supply itself, its two sides are non-metallic materials with metal conductive materials in the middle, which is easy to produce errors during detection, making it difficult to detect defects, and the detection accuracy is low, which cannot meet the existing usage requirements. Summary of the invention

[0005] In view of one or more of the above defects or improvement requirements of the prior art, the present invention provides a defect detection method and system for a charger power supply, and a storage medium, wherein

[0006] To achieve the above object, the present invention provides a defect detection method for a charger power supply, the method comprising the following steps:

[0007] The inside and outside of the power supply housing are photographed simultaneously by using binocular cameras to obtain image data of the power supply housing taken by the two cameras;

[0008] According to the different parallaxes of the two cameras during shooting, the depth of each pixel is calculated by the triangulation principle to obtain the depth feature map of the depth channel corresponding to the RGB channels inside and outside the power supply shell. Then, the external depth feature map is obtained through the binocular camera, and the RGB image and the depth image are used for coordinate transformation to obtain a pixel deviation in the depth and texture space.

[0009] After obtaining the deviation of each pixel point between the inner and outer surfaces, segmentation is achieved through the segmentation network, and then the training results are evaluated according to the labels, and finally the internal and external shooting data and reference information are obtained;

[0010] Use the trained model to restore the RGB of each pixel to obtain the final repaired power pixel;

[0011] The repaired information is then transmitted to the defect detection module to determine whether there are defects.

[0012] As a further improvement of the present invention, the acquisition of pixel deviation specifically includes the following steps: calibrating the camera before shooting to obtain the camera's intrinsic parameter matrix, and obtaining the camera's distortion matrix by shooting a chessboard;

[0013] According to the calibrated parameters, the pictures taken by the camera are corrected to eliminate the camera distortion, so that the images taken by the two cameras have the same size and viewing angle.

[0014] As a further improvement of the present invention, the step of implementing segmentation by segmenting the network specifically includes the following steps:

[0015] The same image is photographed by a binocular stereo camera to obtain two images with different perspectives in the same space, as well as a baseline between the two images;

[0016] The world coordinate system is converted to the camera coordinate system, and then the pixel coordinate systems of the two images taken from different perspectives are converted to the camera coordinate system respectively. The camera coordinate system conversion between the images taken by the two cameras requires a homography matrix;

[0017] Through the two images taken by the camera, the three-dimensional coordinates of each point on the image in the camera coordinate system are obtained, and the three-dimensional coordinates of each point in the camera coordinate system are calculated through triangulation. Then, the corresponding point of each point in the other camera coordinate system can be calculated through the homography matrix. The points on the two images are triangulated, and finally the depth information of all points on the image is calculated to obtain the depth feature maps of the two images;

[0018] Convert the depth feature map to the world coordinate system, and convert the two depth feature maps belonging to the two images to the same world coordinate system;

[0019] After the depth feature map is calculated, each pixel point corresponds to a height H. Through this height H, it is converted into a distance transform, and then the distances d1, d2 and the distance between the two cameras are calculated based on the baseline between the two cameras; the pixel deviation is calculated and the position of each camera is corrected.

[0020] As a further improvement of the present invention, the following steps are also included: collecting image data including the inside and outside of the charger shell and annotating them; selecting a network architecture to implement semantic segmentation, using an encoder-decoder structure, the encoder extracts high-level features, and the decoder gradually restores the original image resolution for segmentation; using a cross entropy loss function and a Dice coefficient loss function as training loss functions to ensure that the model strikes a balance between segmentation accuracy and recall rate; and training the model using a training data set;

[0021] According to the labels of different regions, the IoU value of each region is calculated to obtain the segmentation ability of the model for each region; the segmentation result map is generated to show the labels of different regions for subsequent processing and analysis;

[0022] The segmentation results are used to extract the internal and external information of the charger casing and generate a label map containing the defective areas, providing basic data for subsequent defect detection. The segmentation results are compared with the reference standards to analyze the differences in different areas to ensure the accuracy and completeness of the detection.

[0023] As a further improvement of the present invention, the performance of the model is regularly evaluated, and the precision, recall and F1 score are calculated using the validation set;

[0024] The segmentation effect of the model is evaluated using standard evaluation metrics to determine the performance of the model on different categories.

[0025] As a further improvement of the present invention, in the step of restoring RGB for each pixel, the following steps are specifically included:

[0026] Use the trained semantic segmentation model to restore the RGB value of each pixel to ensure that each pixel gets the appropriate color value according to its classification;

[0027] Perform a linear transformation on the restored RGB values ​​to adjust the color saturation and brightness to match the real appearance. The linear transformation can be fitted by the parameters obtained after multiple trainings.

[0028] Generate the final power pixel map based on the repaired RGB values ​​to ensure that the color of each pixel accurately reflects its actual appearance;

[0029] Output RGB restored and repaired images to provide clear image data for subsequent defect detection and quality assessment.

[0030] As a further improvement of the present invention, the step of transmitting the repaired information to the defect detection module specifically includes the following steps:

[0031] The RGB restored and repaired image is integrated with the previously extracted segmentation information to form a complete input dataset, which includes the repaired RGB image, the comparison information of the internal and external segmentation results and the reference image;

[0032] Convert the integrated data into a format suitable for processing by the defect detection module, such as image matrix, annotation information, etc., to ensure data compatibility and validity;

[0033] Transmit the integrated information to the defect detection module through API, message queue or other data transmission protocol;

[0034] Use the trained defect detection model to analyze the transmitted data and identify the defect areas;

[0035] The defect detection module will generate defect detection results, which include scratches, cracks and color differences, location information of each defect, and defect severity assessment.

[0036] As a further improvement of the present invention, after the defect detection module produces the defect detection result, the following steps are also included:

[0037] Feedback of defect detection results to system operators for further quality control measures

[0038] On this basis, the present invention also provides a defect detection system for a charger power supply, and the defect detection system is used to implement the above-mentioned defect detection method.

[0039] On this basis, the present invention further provides a storage medium, which executes the steps in the above-mentioned defect detection method when running on a computer.

[0040] The above-mentioned improved technical features can be combined with each other as long as they do not conflict with each other.

[0041] In general, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:

[0042] The defect detection method and system for a charger power supply of the present invention, and a storage medium, use two cameras to simultaneously photograph the power supply from different positions to obtain two pictures of the power supply, and use stereo matching to calculate the parallax between the two pictures. And according to the principle of triangulation, the depth information of the corresponding pixel point is obtained. Finally, the image is transformed into coordinates according to the binocular camera and the depth information, and then combined with the depth information to obtain three-dimensional information, and then the coordinate position and the color space are corresponded by the coordinate transformation, so as to repair the photographed picture data. At the same time, because of the depth information of the three-dimensional position, the internal and external defects of the outer shell can be detected simultaneously by stereo matching, and the detection accuracy is relatively high. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is an overall flow chart of a defect detection method and system for a charger power supply and a storage medium in an embodiment of the present invention; DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0045] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in each embodiment of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0046] See also Figure 1 In the preferred embodiment of the present invention, the defect detection method for the charger power supply uses two cameras to simultaneously shoot the power supply from different positions to obtain two pictures of the power supply, and uses stereo matching to calculate the parallax between the two pictures. And according to the principle of triangulation, the depth information of the corresponding pixel point is obtained. Finally, the image is transformed into coordinates according to the binocular camera and the depth information, and then combined with the depth information to obtain three-dimensional information, and then the coordinate position and the color space are corresponded by the coordinate transformation, so as to repair the captured picture data. At the same time, because of the depth information of the three-dimensional position, the internal and external defects of the outer shell can be detected simultaneously by stereo matching, and the detection accuracy is relatively high.

[0047] Specifically, the outer shell defect detection method for a charger power supply in a preferred embodiment of the present invention comprises the following steps:

[0048] S100, photographing the inside and outside of the power supply housing simultaneously with binocular cameras to obtain image data of the power supply housing taken by the two cameras.

[0049] In actual use, use a binocular camera and ensure that the camera has high resolution and a wide field of view to capture the internal and external details of the charger housing. Set an adjustable light source, such as an LED light, next to the camera to ensure clear shooting under different lighting conditions. At the same time, fix the binocular camera in an appropriate position to ensure that the internal and external views of the charger housing can be captured at the same time. The camera should be aimed at the opening of the charger to ensure that it can enter the interior for shooting.

[0050] Furthermore, in actual use, it is preferred to use a retractable robotic arm to flexibly adjust the position of the camera to facilitate all-round shooting inside.

[0051] S200. According to the different parallaxes of the two cameras during shooting, the depth of each pixel is calculated by the triangulation principle to obtain the depth feature map of the depth channel corresponding to the RGB channels inside and outside the power supply shell, and then the external depth feature map is obtained by the binocular camera, and the RGB image and the depth image are used for coordinate transformation to obtain a pixel deviation in the depth and texture space.

[0052] Specifically, in step S200, the following steps are also included:

[0053] S201, calibrating the camera before shooting to obtain the camera's intrinsic parameter matrix, and obtaining the camera's distortion matrix by shooting a chessboard.

[0054] S202, correcting the pictures taken by the camera according to the parameters calibrated in S201 to eliminate the camera distortion, so that the pictures taken by the two cameras have the same size and the same viewing angle;

[0055] S203, photographing the same image with a binocular stereo camera to obtain two images with different viewing angles in the same space, where the distance between the two images is the baseline;

[0056] S204, calculating the homography matrix: at the same time, the world coordinate system is converted into the camera coordinate system, and then the pixel coordinate systems of the two images taken from different perspectives are respectively converted into the camera coordinate system. The camera coordinate system conversion between the images taken by the two cameras requires the homography matrix h, which is a 3x3 matrix;

[0057] S205. The two images captured by the camera have depth, that is, each point on the image corresponds to a unique three-dimensional coordinate in the camera coordinate system. Therefore, the three-dimensional coordinates of each point in the camera coordinate system can be calculated by triangulating the points on the image. Then, the corresponding point of each point in the other camera coordinate system can be calculated by the homography matrix h. The points on the two images are triangulated to finally calculate the depth information of all points on the images, and the depth feature maps of the two images are obtained.

[0058] S206, converting the depth feature map to the world coordinate system: converting the two depth feature maps belonging to the two images to the same world coordinate system. For any depth feature map, the depth information of each point thereon has unique coordinates in the camera coordinate system and the world coordinate system.

[0059] S207, after the depth feature map is calculated, each pixel point corresponds to a height H, and the height H is converted into a distance transform, and then the distances d1, d2 and the distance between the two cameras are calculated according to the baseline between the two cameras;

[0060] S208. After calculating the above three parameters, the pixel deviation can be calculated to correct the position of each camera.

[0061] S300, after obtaining the deviation of each pixel point between the inner and outer surfaces, segmentation is achieved through a segmentation network, and then the training results are evaluated according to the labels, and finally the internal and external shooting data and reference information are obtained.

[0062] Specifically, in step S300, the following steps are included:

[0063] S301: Collect image data including the inside and outside of the charger housing, and annotate them to ensure that different areas (such as internal structure, external surface, defective area, etc.) in each image are accurately marked.

[0064] S302: Select a suitable network architecture (such as U-Net, DeepLab or SegNet) to achieve efficient semantic segmentation.

[0065] S303: adopting an encoder-decoder structure, the encoder extracts high-level features, and the decoder gradually restores the original image resolution for segmentation.

[0066] S304: The cross entropy loss function and the Dice coefficient loss function are used as training loss functions to ensure that the model achieves a balance between segmentation accuracy and recall rate.

[0067] S305: Use the training data set to train the model, monitor the training loss and validation loss, and use early stopping to prevent overfitting.

[0068] S306: Regularly evaluate the performance of the model and calculate indicators such as accuracy, recall, and F1 score using the validation set.

[0069] S307: Use a standard evaluation index (Dice coefficient) to evaluate the segmentation effect of the model to determine the performance of the model on different categories.

[0070] S308: According to the labels of different regions (internal defects, external defects, etc.), the IoU value of each region is calculated to obtain the segmentation capability of the model for each region.

[0071] S309: Generate a segmentation result map, displaying the marks of different regions, to facilitate subsequent processing and analysis.

[0072] S310: Extract internal and external information of the charger housing using the segmentation result, generate a label map containing defective areas, and provide basic data for subsequent defect detection.

[0073] S311: Compare the segmentation results with the reference standard, analyze the differences in different regions, and ensure the accuracy and completeness of the detection.

[0074] S400, using the trained model to restore the RGB of each pixel to obtain a final restored power pixel.

[0075] Step S400 also includes the following steps:

[0076] S401: Using the trained semantic segmentation model, restore the RGB value of each pixel to ensure that each pixel obtains a suitable color value according to its classification.

[0077] S402: Performing a linear transformation on the restored RGB values ​​to adjust the color saturation and brightness to match the real appearance. The linear transformation can be fitted by parameters obtained after multiple trainings.

[0078] S403: Generate a final power pixel map according to the repaired RGB values ​​to ensure that the color of each pixel accurately reflects its actual appearance.

[0079] S404: Output the image after RGB restoration and repair, and provide clear image data for subsequent defect detection and quality assessment.

[0080] Through the steps of S300 and S400, semantic segmentation and image restoration of the inside and outside of the charger shell can be effectively achieved. This not only improves the accuracy of defect detection, but also provides high-quality image support for subsequent quality assessment.

[0081] S500, transmitting the repaired information to the defect detection module to determine whether there are defects.

[0082] Specifically, in step S500, the following steps are also included:

[0083] S501: Integrate the RGB restored and repaired image with the previously extracted segmentation information to form a complete input data set, which includes the repaired RGB image, the comparison information of the internal and external segmentation results and the reference image.

[0084] S502: Convert the integrated data into a format suitable for processing by the defect detection module, such as image matrix, annotation information, etc., to ensure data compatibility and validity.

[0085] S503: Transmit the integrated information to the defect detection module via an API, a message queue, or other data transmission protocols.

[0086] Ensure data integrity and accuracy during transmission, using verification mechanisms when necessary.

[0087] S504: In the defect detection module, the transmitted data is analyzed using a trained defect detection model (such as a convolutional neural network). The model will identify possible defect areas.

[0088] S505: The defect detection module will generate defect detection results, which include scratches, cracks and color differences, location information of each defect, and defect severity assessment (such as slight, medium, severe).

[0089] S506: Feedback the defect detection results to the system operator so that further quality control measures can be taken.

[0090] On this basis, the present invention also provides a defect detection system for a charger power supply, which is used to implement the defect detection method in the above preferred embodiment.

[0091] On this basis, the present invention further provides a storage medium, which, when running on a computer program, executes the defect detection method in the above preferred embodiment.

[0092] The defect detection method and system for a charger power supply of the present invention, and a storage medium, use two cameras to simultaneously photograph the power supply from different positions to obtain two pictures of the power supply, and use stereo matching to calculate the parallax between the two pictures. And according to the principle of triangulation, the depth information of the corresponding pixel point is obtained. Finally, the image is transformed into coordinates according to the binocular camera and the depth information, and then combined with the depth information to obtain three-dimensional information, and then the coordinate position and the color space are corresponded by the coordinate transformation, so as to repair the photographed picture data. At the same time, because of the depth information of the three-dimensional position, the internal and external defects of the outer shell can be detected simultaneously by stereo matching, and the detection accuracy is relatively high.

[0093] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A defect detection method for a charger power supply, characterized in that: The method comprises the following steps: The inside and outside of the power supply housing are photographed simultaneously by using binocular cameras to obtain image data of the power supply housing taken by the two cameras; According to the different parallaxes of the two cameras during shooting, the depth of each pixel is calculated by the triangulation principle to obtain the depth feature map of the depth channel corresponding to the RGB channels inside and outside the power supply shell. Then, the external depth feature map is obtained through the binocular camera, and the RGB image and the depth image are used for coordinate transformation to obtain a pixel deviation in the depth and texture space. After obtaining the deviation of each pixel point between the inner and outer surfaces, segmentation is achieved through the segmentation network, and then the training results are evaluated according to the labels, and finally the internal and external shooting data and reference information are obtained; Use the trained model to restore the RGB of each pixel to obtain the final repaired power pixel; The repaired information is then transmitted to the defect detection module to determine whether there are defects.

2. The defect detection method for a charger power supply according to claim 1, characterized in that: The acquisition of pixel deviation specifically includes the following steps: Before shooting, the camera is calibrated to obtain the camera's intrinsic parameter matrix, and the camera's distortion matrix is ​​obtained by shooting a chessboard. According to the calibrated parameters, the pictures taken by the camera are corrected to eliminate the camera distortion, so that the images taken by the two cameras have the same size and viewing angle.

3. The defect detection method for a charger power supply according to claim 1, characterized in that: In the step of implementing segmentation by segmenting the network, the following steps are specifically included: The same image is photographed by a binocular stereo camera to obtain two images with different perspectives in the same space, as well as a baseline between the two images; The world coordinate system is converted to the camera coordinate system, and then the pixel coordinate systems of the two images taken from different perspectives are converted to the camera coordinate system respectively. The camera coordinate system conversion between the images taken by the two cameras requires a homography matrix; Through the two images taken by the camera, the three-dimensional coordinates of each point on the image in the camera coordinate system are obtained, and the three-dimensional coordinates of each point in the camera coordinate system are calculated through triangulation. Then, the corresponding point of each point in the other camera coordinate system can be calculated through the homography matrix. The points on the two images are triangulated, and finally the depth information of all points on the image is calculated to obtain the depth feature maps of the two images; Convert the depth feature map to the world coordinate system, and convert the two depth feature maps belonging to the two images to the same world coordinate system; After the depth feature map is calculated, each pixel corresponds to a height H. Through this height H, it is converted into a distance transform, and then the distance d1, d2 and the distance between the two cameras are calculated based on the baseline between the two cameras; The pixel deviation is calculated and the position of each camera is corrected.

4. The defect detection method for a charger power supply according to claim 1, characterized in that: In the step of segmentation through the segmentation network, specifically The following steps are involved: Collect image data including the inside and outside of the charger housing and annotate them; Select a network architecture to achieve semantic segmentation, using an encoder-decoder structure, where the encoder extracts high-level features and the decoder gradually restores the original image resolution for segmentation; The cross entropy loss function and Dice coefficient loss function are used as training loss functions to ensure that the model achieves a balance between segmentation accuracy and recall rate; Train the model using the training dataset; According to the labels of different regions, the IoU value of each region is calculated to obtain the segmentation ability of the model for each region; Generate segmentation result map, showing the marks of different areas, which is convenient for subsequent processing and analysis; The segmentation results are used to extract the internal and external information of the charger shell and generate a label map containing the defect area, providing basic data for subsequent defect detection. Compare the segmentation results with the reference standard and analyze the differences in different regions to ensure the accuracy and completeness of the detection.

5. The defect detection method for a charger power supply according to claim 4, characterized in that: Regularly evaluate the performance of the model and calculate the accuracy, recall, and F1 score using the validation set; The segmentation effect of the model is evaluated using standard evaluation metrics to determine the performance of the model on different categories.

6. The defect detection method for a charger power supply according to claim 1, characterized in that: The step of restoring the RGB of each pixel includes the following steps: Use the trained semantic segmentation model to restore the RGB value of each pixel to ensure that each pixel gets the appropriate color value according to its classification; Perform a linear transformation on the restored RGB values ​​to adjust the color saturation and brightness to match the real appearance. The linear transformation can be fitted by the parameters obtained after multiple trainings. Generate the final power pixel map based on the repaired RGB values ​​to ensure that the color of each pixel accurately reflects its actual appearance; Output RGB restored and repaired images to provide clear image data for subsequent defect detection and quality assessment.

7. The defect detection method for a charger power supply according to claim 1, characterized in that: The step of transmitting the repaired information to the defect detection module specifically includes the following steps: The RGB restored and repaired image is integrated with the previously extracted segmentation information to form a complete input dataset, which includes the repaired RGB image, the comparison information of the internal and external segmentation results and the reference image; Convert the integrated data into a format suitable for processing by the defect detection module, such as image matrix, annotation information, etc., to ensure data compatibility and validity; Transmit the integrated information to the defect detection module through API, message queue or other data transmission protocol; Use the trained defect detection model to analyze the transmitted data and identify the defect areas; The defect detection module will generate defect detection results, which include scratches, cracks and color differences, location information of each defect, and defect severity assessment.

8. The defect detection method for a charger power supply according to claim 7, characterized in that: After the defect detection module produces the defect detection results, the following steps are also included: Feedback of defect detection results to system operators for further quality control measures.

9. A defect detection system for a charger power supply, characterized in that: The defect detection system is used to implement the defect detection method described in any one of claims 1 to 8.

10. A storage medium, characterized in that: When the storage medium is run on a computer, the steps of the defect detection method according to any one of claims 1 to 8 are executed.

Images