Flange surface defect detection method based on deep learning
Through the deep learning-based flange surface defect detection method, the Yolov8 model and high-precision light source are used to realize the automatic detection of surface defects of semiconductor heat conductors, solving the problems of low manual detection efficiency and high error detection rate, and improving the detection efficiency and Accuracy reduces costs.
Patent Information
- Application Number
- CN202411054261.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-11-19
AI Technical Summary
In the prior art, defect detection of heat conduction sheet surfaces in the semiconductor field relies on manual detection, and there are problems such as low efficiency, high cost, high error detection rate and high labor intensity, making it difficult to meet the needs of large-scale production.
Using the flange surface defect detection method based on deep learning, the Yolov8 instance segmentation model combines high-precision blue circumferential light and coaxial light combination light source to perform image preprocessing and parallel processing, identify and locate defect areas, and calculate the collapse width. Implement automated detection.
It greatly improves detection efficiency and accuracy, reduces missed inspections and missed inspections, reduces labor costs, realizes full process automation, and meets real-time inspection needs.
Smart Images

Figure CN118982522A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer vision and image processing, and in particular to a flange surface defect detection method based on deep learning. Background Art
[0002] With the rapid development of industrial technology, computer vision, artificial intelligence and deep learning technology, more and more intelligent detection systems are being used in the field of industrial manufacturing to replace traditional manual detection methods and improve detection efficiency and accuracy. In the semiconductor manufacturing process, the flange is used as a chip heat spreader, and its surface quality detection is particularly important. Surface defects of the heat spreader, such as cracks, scratches, pits, etc., will not only affect the heat dissipation performance of the chip, but may also cause the chip to fail, thereby affecting the performance and reliability of the entire electronic device.
[0003] In the prior art, defect detection of the surface of thermal conductive sheets in the semiconductor field mainly relies on manual detection. Manual detection has the following significant disadvantages: Low efficiency: Manual inspection is slow and difficult to adapt to the needs of large-scale production in the semiconductor manufacturing process.
[0004] High cost: A large number of professionals are required to carry out the inspection, which increases the labor cost.
[0005] High false detection rate: Manual detection is easily affected by subjective factors, resulting in missed detection and false detection.
[0006] High labor intensity: Long-term high-intensity testing work can easily lead to fatigue of testers, thus affecting the accuracy of test results. Summary of the invention
[0007] According to an embodiment of the present invention, a flange surface defect detection method based on deep learning is provided, comprising the following steps: Acquire images and preprocess the images; The preprocessed image is processed using the Yolov8 instance segmentation model to identify and locate the defective area of the image; Obtain defect detection results and generate detection reports.
[0008] Further, obtaining the image includes the following steps: High-precision blue coaxial light and ring light combined light source is used to illuminate the flange; Use a high-resolution camera to capture the flange surface and obtain high-definition images.
[0009] Further, the image preprocessing includes the following steps: The image is grayscaled, denoised and contrast enhanced in sequence; The flange area in the image is extracted and segmented to obtain multiple image blocks.
[0010] Furthermore, multiple Yolov8 instance segmentation models are run simultaneously to process multiple image blocks in parallel to identify and locate the defective area of each image block.
[0011] Furthermore, the preprocessed image is processed using the Yolov8 instance segmentation model to identify and locate the defective area of the image, which includes the following steps: The preprocessed image is divided into several grid units, and each grid unit predicts multiple candidate boxes; Perform bounding box prediction and classification prediction for each candidate box, and perform mask prediction for each detected defect area; Perform NMS processing on the predicted candidate boxes to remove redundancy; Obtain the bounding box, category, and instance mask for each defect.
[0012] Furthermore, the Sigmoid function is used for classification prediction and mask prediction.
[0013] Furthermore, the method further comprises the following steps: determining whether there is a chipping defect in the defect category; if so, obtaining a chipping area in the defect area, and calculating a chipping width.
[0014] Further, calculating the chip width includes the following steps: Get the minimum enclosing rectangle of the collapse area; Extract the chipped area through threshold segmentation; Extract the nearest boundary contour of the collapse area; The maximum distance from all points of the chipping area contour to the adjacent boundary contour is calculated as the chipping width.
[0015] According to the flange surface defect detection method based on deep learning in an embodiment of the present invention, the Yolov8 deep learning model is used to realize real-time detection of flange surface defects, greatly improving the detection efficiency; the model inference algorithm is developed and optimized, and multiple identical models are used for parallel inference to further improve the processing speed and system stability; based on the Yolov8 instance segmentation model, combined with a combined light source of high-precision blue ring light and coaxial light, the interference of the flange gold flower pattern is removed, and a high-precision chip width calculation method is used to accurately identify and locate various types of surface defects, reducing missed detections and false detections; the entire process is automated, from image acquisition to defect detection and result output, without the need for manual intervention, reducing labor costs.
[0016] It is to be understood that both the foregoing general description and the following detailed description are exemplary, and are intended to provide further explanation of the technology as claimed. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1It is a flow chart of a flange surface defect detection method based on deep learning according to an embodiment of the present invention; Figure 2 A flowchart of acquiring an image in a flange surface defect detection method based on deep learning according to an embodiment of the present invention; Figure 3 It is a flowchart of preprocessing an image according to a flange surface defect detection method based on deep learning according to an embodiment of the present invention; Figure 4 A flowchart of using the Yolov8 instance segmentation model to process the preprocessed image, identify and locate the defect area of the image according to the flange surface defect detection method based on deep learning according to an embodiment of the present invention; Figure 5 The present invention is a flowchart of calculating chip width according to a flange surface defect detection method based on deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION
[0018] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings to further illustrate the present invention.
[0019] First, combine Figures 1 to 5 A flange surface defect detection method based on deep learning according to an embodiment of the present invention is described, which is used to detect flange surface defects and has a wide range of application scenarios.
[0020] like Figures 1 to 5 As shown, the flange surface defect detection method based on deep learning of the embodiment of the present invention comprises the following steps: acquiring an image and preprocessing the image; processing the preprocessed image using the Yolov8 instance segmentation model to identify and locate the defect area of the image; obtaining defect detection results, generating a test report, and visualizing the test results. The present invention is based on the Yolov8 instance segmentation model for training, and the model is trained through a large number of labeled defect image data sets, so that it can accurately identify and locate various defects on the flange surface. Using Yolov8 has the following two advantages: higher detection accuracy and faster reasoning speed. First, the Yolov8 model performs well in the instance segmentation task, and can accurately identify and locate various defects on the flange surface. Compared with the traditional convolutional neural network, Yolov8 has higher detection accuracy and lower false detection rate and missed detection rate. Secondly, Yolov8 optimizes the network structure, which significantly improves the reasoning speed and can meet the needs of real-time detection.
[0021] Furthermore, in this embodiment, acquiring the image includes the following steps: using a high-precision blue coaxial light and ring light combined light source to illuminate the flange, which can provide uniform lighting, reduce reflections and shadows, and effectively remove the interference of the flange gold flower pattern, thereby ensuring image quality; using a high-resolution camera to photograph the flange surface, specifically a high-precision dual telecentric lens can be used, which can provide high-resolution and low-distortion imaging effects, ensure the accuracy and consistency of the image, and acquire a high-definition image.
[0022] Furthermore, in this embodiment, the image is preprocessed and includes the following steps: graying, denoising and contrast enhancing the image in sequence to improve the image quality and the accuracy of subsequent processing; extracting and segmenting the flange area in the image to obtain multiple image blocks for subsequent Yolov8 instance segmentation model processing.
[0023] Furthermore, in this embodiment, multiple Yolov8 instance segmentation models are run simultaneously to process multiple image blocks in parallel, and the defective area of each image block is identified and located. Through parallel processing, multiple Yolov8 models are inferred at the same time, which greatly improves the efficiency of image processing. Compared with the serial processing method, parallel reasoning significantly shortens the detection time and achieves the purpose of real-time detection. Moreover, parallel reasoning can effectively balance the computing load, reduce the processing pressure of a single model, and improve the stability and reliability of the overall system.
[0024] Further, in this embodiment, the preprocessed image is processed using the Yolov8 instance segmentation model, and identifying and locating the defect area of the image includes the following steps: dividing the preprocessed image into a number of grid units, and predicting multiple candidate boxes for each grid unit; performing bounding box prediction and classification prediction on each candidate box, and performing mask prediction on each detected defect area; the bounding box is a rectangular box used to select defects in the image, and the bounding box prediction means that the model will learn how to adjust the position, size and confidence of the candidate box to generate a final bounding box to more accurately locate the defects in the image; the classification prediction means that the model will predict a category vector, which contains the probability that the defect belongs to each category. In Yolov8, the Sigmoid function is usually used to calculate these probabilities; the mask prediction means that the model will calculate the Sigmoid function of each column in parallel for the detected defect area to generate a probability map, and then convert this probability map into a binary mask through upsampling and refinement processing to represent the precise shape of the target object; the predicted candidate box is processed by NMS (non-maximum suppression) to remove redundancy; and the bounding box, category and instance mask of each defect are obtained.
[0025] Furthermore, in this embodiment, the following steps are also included: determining whether there is a chipping defect in the defect category, and if so, obtaining the chipping area in the defect area, and calculating the chipping width. For chipping defects, commonly used regional features, such as the minimum circumscribed rectangle width or the maximum inscribed circle diameter, cannot accurately obtain the maximum distance from the flange boundary, so the following method is used for calculation: obtaining the minimum circumscribed rectangle of the chipping area; extracting the chipping area by threshold segmentation; extracting the nearest boundary contour of the chipping area; calculating the maximum distance from all points of the chipping area contour to the adjacent boundary contour as the chipping width.
[0026] Above, refer to Figures 1 to 5 The present invention describes a flange surface defect detection method based on deep learning according to an embodiment of the present invention. The Yolov8 deep learning model is used to realize real-time detection of flange surface defects, greatly improving the detection efficiency. The model inference algorithm is developed and optimized, and multiple identical models are used in parallel to further improve the processing speed and system stability. Based on the Yolov8 instance segmentation model, a combined light source of high-precision blue ring light and coaxial light is combined to remove the interference of the flange gold flower pattern, and a high-precision chip width calculation method is used to accurately identify and locate various types of surface defects and reduce missed detections and false detections. The entire process is automated, from image acquisition to defect detection and result output, without the need for manual intervention, reducing labor costs.
[0027] It should be noted that, in this specification, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprises..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0028] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.
Claims
1. A flange surface defect detection method based on deep learning, characterized in that: The following steps are included: Acquire an image and preprocess the image; Using the Yolov8 instance segmentation model to process the preprocessed image, identify and locate the defect area of the image; Obtain defect detection results and generate detection reports.
2. The flange surface defect detection method based on deep learning as claimed in claim 1, characterized in that: The image acquisition comprises the following steps: High-precision blue coaxial light and ring light combined light source is used to illuminate the flange; Use a high-resolution camera to capture the flange surface and obtain high-definition images.
3. The flange surface defect detection method based on deep learning as claimed in claim 1 or 2, characterized in that: The preprocessing of the image comprises the following steps: graying, denoising and contrast enhancement are sequentially performed on the image; The flange area in the image is extracted and segmented to obtain a plurality of image blocks.
4. The flange surface defect detection method based on deep learning as claimed in claim 3 is characterized in that: Run multiple Yolov8 instance segmentation models simultaneously to process multiple image blocks in parallel and identify and locate the defect area of each image block.
5. The flange surface defect detection method based on deep learning as claimed in claim 1, characterized in that: The method of processing the preprocessed image using the Yolov8 instance segmentation model to identify and locate the defective area of the image comprises the following steps: The preprocessed image is divided into several grid units, and each grid unit predicts multiple candidate boxes; Perform bounding box prediction and classification prediction for each candidate box, and perform mask prediction for each detected defect area; Performing NMS processing on the predicted candidate boxes to remove redundancy; Obtain the bounding box, category, and instance mask for each defect.
6. The flange surface defect detection method based on deep learning as claimed in claim 5, characterized in that: The Sigmoid function is used for classification prediction and mask prediction.
7. The flange surface defect detection method based on deep learning as claimed in claim 5, characterized in that: The method also includes the following steps: determining whether there is a chipping defect in the defect category; if so, obtaining a chipping area in the defect area, and calculating a chipping width.
8. The flange surface defect detection method based on deep learning as claimed in claim 7, characterized in that: The calculation of chip width comprises the following steps: Get the minimum enclosing rectangle of the collapse area; Extract the chipped area through threshold segmentation; Extract the nearest boundary contour of the collapse area; The maximum distance from all points of the chipping area contour to the adjacent boundary contour is calculated as the chipping width.
Citation Information
Cited By
Plate edge sealing defect online detection method and device based on machine vision
CN120219294A
