Optical glass detection method and system based on image enhancement

Abstract

The invention relates to the technical field of optical glass detection, and discloses an optical glass detection method and system based on image enhancement. According to the method, optical glass surface images are shot in real time through a linear array camera arranged obliquely above an optical glass detection table and light sources at symmetrical positions, and multi-angle shooting is achieved through a rotation control module. And an adaptive histogram equalization algorithm with limited contrast is adopted to perform enhancement processing on the image, so that the image quality and the definition of defect features are improved. And performing automatic detection and classification on the enhanced image by using the trained optical glass detection model, and outputting confidence, category and position information of defects. And when the confidence coefficient of the defect of at least one image is greater than a preset threshold value, judging that the optical glass has the defect, and generating a detection report. According to the method, the detection efficiency and accuracy are improved, tiny and complex defects can be effectively identified, powerful support is provided for subsequent processing, and the method has important practical application value.

Description

Technical Field

[0001] The present invention relates to the technical field of optical glass detection, and in particular to an optical glass detection method and system based on image enhancement. Background Art

[0002] Optical glass may produce various surface defects during the manufacturing process, such as scratches, bubbles, stains, etc. These defects seriously affect the quality and performance of optical glass. Traditional detection methods usually rely on manual visual inspection, which is not only time-consuming and labor-intensive, but also easily affected by subjective factors, making it difficult to ensure the accuracy and consistency of the detection results. In recent years, with the continuous development of image processing technology, automated detection methods based on image processing have gradually been applied to the field of optical glass detection. However, the surface defects of optical glass are often relatively small and complex, and the image quality requirements are high. Traditional image processing algorithms may have poor results when processing such images. Therefore, how to effectively improve the accuracy and efficiency of optical glass surface defect detection has become a problem that needs to be solved urgently. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide an optical glass detection method and system based on image enhancement to solve the problems of low accuracy and efficiency in optical glass defect detection in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an optical glass detection method based on image enhancement, the method comprising: Step S100: using a linear array camera disposed obliquely above the optical glass inspection platform and a light source disposed at a symmetrical position to the linear array camera to capture an image of a surface of a first optical glass located on the optical glass inspection platform in real time; Step S200: after each image is taken, the optical glass inspection platform is rotated clockwise by a preset degree, and the next image is taken, until the optical glass inspection platform rotates for one circle or more; Step S300: performing image enhancement on the first optical glass surface image using a contrast-limited adaptive histogram equalization algorithm to obtain a second optical glass surface image; Step S400: inputting the second optical glass surface image into the optical glass detection model to output the confidence level, defect type and defect location of defects on the optical glass surface; Step S500: According to the output result of the optical glass detection algorithm, when the confidence level of at least one second optical glass surface image defect is greater than a preset threshold, it is determined that the optical glass has a defect.

[0005] Preferably, in a possible implementation manner of the first aspect, the shooting angle of the line array camera is at a preset angle with the optical glass to be inspected, and the light emitted by the light source is at the same preset angle with the optical glass to be inspected.

[0006] Preferably, in a possible implementation of the first aspect, the contrast-limited adaptive histogram equalization algorithm specifically includes: The first optical glass surface image is divided into a plurality of non-overlapping rectangular blocks, and a histogram is calculated for each rectangular block area; The histogram is cropped using a preset threshold, and the cropped part is evenly distributed to other parts of the histogram for histogram reconstruction. Based on the reconstructed histogram, the cumulative distribution function is calculated; A transformation function is calculated based on the cumulative distribution function, and the grayscale value of each pixel is mapped to a new grayscale value using the transformation function to obtain a second optical glass surface image.

[0007] Preferably, in a possible implementation manner of the first aspect, the optical glass detection model is trained by an improved YOLO v8 algorithm and uses an optical glass surface grayscale dataset.

[0008] Preferably, in a possible implementation of the first aspect, the improved yolo v8 algorithm specifically includes: The backbone network uses a convolutional additive self-attention mechanism to perform global context modeling on the input second optical glass surface image and obtain a feature map of the second optical glass surface image; The neck network uses a scale-compensated feature pyramid network to resize feature maps of different scales. It uses an image feature alignment mechanism to align the features of optical glass images at different rotation angles. Combined with variable convolution, it performs adaptive sampling and image feature extraction for each position in the feature map. The detection head uses the Anchor-Free detection method to predict the defect bounding box and combines the convolution layer to predict the defect category and confidence level. The output layer outputs the location, confidence and defect category of optical glass defects.

[0009] Preferably, in a possible implementation manner of the first aspect, the optical glass surface grayscale dataset is formed by a historical second optical glass surface image.

[0010] Preferably, in a possible implementation manner of the first aspect, step S500 specifically includes: Summarize the optical glass defect positions, optical glass defect confidences and optical glass defect categories of all second optical glass surface images of the optical glass; When there is at least one image of the second optical glass surface with a defect confidence greater than a preset threshold, the optical glass is marked as a defective product, and the optical glass defect position and optical glass defect category corresponding to the highest confidence are recorded as the detection result; Generate a test report containing the test results, and store the test report in association with the corresponding image data.

[0011] In a second aspect, the present invention provides an optical glass detection system based on image enhancement, the system comprising: An image acquisition module is used to deploy a linear array camera obliquely above the optical glass inspection platform and a light source deployed at a symmetrical position of the linear array camera to capture a surface image of the first optical glass in real time; A rotation control module, used to control the optical glass inspection platform to rotate clockwise by a preset degree after each image is taken, until the optical glass inspection platform rotates more than or equal to one circle; An image enhancement module, used for performing image enhancement on the first optical glass surface image using a contrast-limited adaptive histogram equalization algorithm to obtain a second optical glass surface image; A defect detection module, used for inputting the second optical glass surface image into the optical glass detection model, and outputting the confidence level, defect category and defect location of the defects on the optical glass surface; The result judgment module is used to judge that the optical glass has defects according to the output result of the optical glass detection model when the confidence of at least one second optical glass surface image defect is greater than a preset threshold.

[0012] The beneficial effects of the present invention are as follows: by deploying a linear array camera obliquely above the optical glass inspection platform and a light source in a symmetrical position, the surface image of the optical glass is captured in real time, and the multi-angle shooting of the optical glass is achieved through a rotation control module, which effectively improves the comprehensiveness and accuracy of image acquisition. At the same time, the image is enhanced by using a contrast-limited adaptive histogram equalization algorithm to further improve the image quality and clarity of defect features. In addition, the present invention also realizes the automatic detection and classification of optical glass surface defects by training an optical glass detection model, greatly improving the detection efficiency and accuracy. This method can not only effectively identify various small and complex defects, but also output the confidence, category and location information of the defects, providing strong support for subsequent defect analysis and processing. At the same time, the implementation cost of this method is relatively low, and it is easy to promote and apply on optical glass production lines, and has important practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 A flow chart of an optical glass detection method based on image enhancement is provided for this application.

[0015] Figure 2 A structural diagram of an optical glass detection system based on image enhancement is provided for this application.

[0016] Explanation of the accompanying drawings: 1-image acquisition module, 2-rotation control module, 3-image enhancement module, 4-defect detection module, 5-result determination module. DETAILED DESCRIPTION

[0017] 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.

[0018] Embodiment 1: Figure 1 As shown, the present invention provides an optical glass detection method based on image enhancement, the method comprising: Step S100: using a linear array camera disposed obliquely above the optical glass inspection platform and a light source disposed at a symmetrical position to the linear array camera, to capture a surface image of a first optical glass located on the optical glass inspection platform in real time.

[0019] Specifically, the shooting angle of the linear array camera is at a preset angle to the optical glass to be inspected, and the light emitted by the light source is at the same preset angle to the optical glass to be inspected.

[0020] In this embodiment, the line array camera adopts the Sony ILX6946R-CE model, the light source adopts the OSRAM OptoSemiconductors OSLON SSL 150, the line array camera is deployed obliquely above the optical glass inspection table, the shooting angle forms an angle of 60 degrees with the optical glass to be inspected, the light source and the line array camera are symmetrical based on the center of the inspection table, and the emitted light forms an angle of 60 degrees with the optical glass to be inspected. The Sony ILX6946R-CE model line array camera and the control center transmit data through the network. When the optical glass to be inspected enters the optical glass inspection table, the line array camera and the light source are started.

[0021] Step S200: After each image is taken, the optical glass inspection platform is rotated clockwise by a preset degree and the next image is taken, until the optical glass inspection platform rotates one circle or more.

[0022] Specifically, after the line array camera completes capturing the first optical glass surface image, the control system starts the rotation mechanism of the optical glass inspection table.

[0023] In this embodiment, the preset degree is set to 30 degrees to ensure that the overlapping area between images is moderate, avoid repeated detection, and ensure that different details of the optical glass surface are captured during one rotation. As the optical glass inspection table rotates clockwise, the linear array camera and the light source maintain their relative positions and continue to capture the next image. During the rotation, the control system will track and record the rotation angle in real time. When the optical glass inspection table rotates for more than one rotation, the rotation and shooting are stopped, and the first optical glass surface image that has been captured is transmitted to the control center.

[0024] Step S300: performing image enhancement on the first optical glass surface image using a contrast-limited adaptive histogram equalization algorithm to obtain a second optical glass surface image.

[0025] Specifically, the contrast-limited adaptive histogram equalization algorithm specifically includes: dividing the first optical glass surface image into multiple non-overlapping rectangular blocks, and calculating the histogram of each rectangular block area; using a preset threshold to crop the histogram, and evenly distributing the cropped part to other parts of the histogram for histogram reconstruction, and calculating the cumulative distribution function based on the reconstructed histogram; calculating the transformation function based on the cumulative distribution function, and using the transformation function to map the grayscale value of each pixel to a new grayscale value to obtain the second optical glass surface image.

[0026] In this embodiment, after receiving the transmitted first optical glass surface image, the control center uses a contrast-limited adaptive histogram equalization algorithm to perform image enhancement, and the first optical glass surface image is input into the algorithm according to the shooting order.

[0027] For the contrast-limited adaptive histogram equalization algorithm, first define the function clip_and_redistribute_histogram, which is used to clip the histogram and evenly distribute the clipped part to other parts. It receives two parameters: the input histogram histogram and the clipping threshold clip_threshold. The return value of the function is the clipped and reconstructed histogram.

[0028] Inside the function clip_and_redistribute_histogram, first calculate the total number of pixels total_pixels of the input histogram, then use the np.clip function to clip the histogram between 0 and the clipping threshold clip_threshold to obtain the clipped histogram clipped_hist, and then calculate the number of clipped pixels clipped_pixels. Next, find the index non_zero_indices of the non-zero entries in the clipped histogram, and calculate the number of pixels increment that should be increased for each non-zero entry. This increment is the number of pixels clipped divided by the number of non-zero entries, and then add this increment to the clipped histogram, and use the np.clip function to ensure that the result is not less than 0. Finally, convert the result to np.uint32 type, and adjust the value of the last entry to ensure that the total number of pixels remains unchanged.

[0029] Next, another function adaptive_histogram_equalization is defined. This function is used to perform adaptive histogram equalization on the image. It receives three parameters: the input grayscale image image, the rectangular block size block_size used when segmenting the image, and the threshold clip_threshold used when clipping the histogram. The return value of the function is the equalized image. Inside the adaptive_histogram_equalization function, the height h and width w of the image are first calculated, and then the height pad_h and width pad_w that need to be filled are calculated so that the image can be evenly divided into blocks of the specified size, and then the np.pad function is used to fill the image. A zero matrix equalized_image of the same size as the input image is initialized to store the equalized image. The filled image is traversed, and a block of size block_size×block_size is taken out each time. For each block, its histogram is calculated, and the clip_and_redistribute_histogram function is called to clip and reconstruct the histogram. Then the cumulative distribution function of the reconstructed histogram is calculated and normalized to the range of 0-255. Then, the zero value in the CDF is processed using a mask to avoid division by zero errors.

[0030] Finally, the normalized CDF is applied as a transformation function to each pixel in the block to obtain the equalized block, and the equalized block is copied back to the corresponding position of the original image.

[0031] In this embodiment, a grayscale image of the optical glass surface is read, and the block size is set to 16 and the cropping threshold is set to 100. Then, the adaptive_histogram_equalization function is called to perform adaptive histogram equalization on the image. Finally, the equalized image is saved to a file. After the image enhancement is completed, a corresponding number of second optical glass surface images are obtained.

[0032] Step S400: inputting the second optical glass surface image into the optical glass detection model to output the confidence level, defect category and defect location of defects on the optical glass surface.

[0033] Specifically, the optical glass inspection model is trained by the improved YOLO v8 algorithm and uses an optical glass surface grayscale dataset, which is composed of the historical second optical glass surface image. The historical second optical glass surface image is obtained from the control center and annotated with LabelImg. The annotated defect categories include bubbles, stripes, stone patterns, crystals, impurities, and scratches, and the optical glass surface grayscale dataset is divided into a training set and a validation set in a ratio of 8:2.

[0034] The improved YOLO V8 algorithm specifically includes: a backbone network that uses a convolutional additive self-attention mechanism to perform global context modeling on the input second optical glass surface image and obtain a feature map of the second optical glass surface image; a neck network that uses a scale-compensated feature pyramid network to scale feature maps of different scales, and uses an image feature alignment mechanism to align the features of optical glass images at different rotation angles, and combines variable convolution to adaptively sample and extract image features at each position in the feature map; a detection head that uses an Anchor-Free detection method to predict the bounding box of the defect and combines the convolution layer to predict the defect category; an output layer that outputs the position, confidence, and defect category of the optical glass defect.

[0035] In this embodiment, the second optical glass surface images are input into the trained optical glass detection model one by one. The optical glass detection model is trained using the training set using the improved YOLO V8 algorithm. After each round of training, the validation set is used for validation. Multiple rounds of iterations are performed until the accuracy on the validation set reaches a stable level.

[0036] For the improved YOLO v8 algorithm, the convolutional additive self-attention mechanism is first used as the core component of the backbone network. When constructing the backbone network, the second optical glass surface image is feature extracted through the convolution layer to obtain the initial feature map. Subsequently, the feature map is further processed using the convolutional self-attention module (ConvSelfAttention). This module contains three convolutional layers: query, key, and value, which are used to generate query features, key features, and value features, respectively. By calculating the dot product attention weights between the query features and the key features, and applying the weights to the value features, a weighted feature map is obtained. Finally, the weighted feature map is fused with the original feature map to obtain the output of the backbone network.

[0037] Next, the scale-compensation feature pyramid network (ScaleCompensationFPN) is used as the neck network to achieve multi-scale feature fusion. The network first further processes the output feature map of the backbone network through multiple convolutional layers to obtain feature maps of optical glass surface images of different scales. These feature maps are then fused using upsampling operations and lateral connections to obtain fused multi-scale feature maps. In order to improve the feature representation capability, a deformable convolution (DeformConv2D) is introduced on the fused feature map. The offset of the convolution kernel is predicted to achieve adaptive sampling of the feature map, thereby enhancing its ability to detect optical glass defects.

[0038] Finally, the Anchor-Free detection head is used to perform target detection on the fused multi-scale feature maps. The detection head consists of three parts: classification convolution layer, regression convolution layer, and target confidence convolution layer. The classification convolution layer is used to predict the target category, the regression convolution layer is used to predict the bounding box position of the target, and the target confidence convolution layer is used to evaluate the probability of the target existence. In order to improve the detection performance, feature maps of different scales are fused to obtain the final classification, regression, and target confidence prediction results.

[0039] Finally, the output layer integrates the results of the detection head and outputs the location, confidence and defect category of the optical glass defect.

[0040] The second optical glass surface image is input into the trained optical glass detection model, and the optical glass defect position, optical glass defect confidence and optical glass defect category are output. The optical glass defect category includes bubbles, stripes, stone patterns, crystals, impurities and scratches.

[0041] Step S500: According to the output result of the optical glass detection algorithm, when the confidence level of at least one second optical glass surface image defect is greater than a preset threshold, it is determined that the optical glass has a defect.

[0042] Specifically, the optical glass defect positions, optical glass defect confidences and optical glass defect categories of all second optical glass surface images of the optical glass are summarized; when there is at least one second optical glass surface image whose defect confidence is greater than a preset threshold, the optical glass is marked as a defective product, and the optical glass defect position and optical glass defect category corresponding to the highest confidence are recorded as the detection result; a detection report containing the detection results is generated, and the detection report is associated with the corresponding image data and stored.

[0043] In this embodiment, the defect confidence threshold is 0.6, which is used to determine whether the optical glass has defects. When the system detects that the defect confidence of at least one second optical glass surface image exceeds the preset threshold of 0.6, the defect determination mechanism will be triggered immediately. The system marks the optical glass as a defective product and automatically records the optical glass defect location and defect category corresponding to the highest confidence, which will be regarded as the final detection result.

[0044] In addition, a test report containing detailed test results is automatically generated. This report not only includes the determination result of whether the optical glass has defects, but also lists in detail the defect location with the highest confidence, the defect category, and the corresponding image data link or thumbnail, so that relevant personnel can intuitively understand the test situation.

[0045] Embodiment 2: Figure 2 As shown, the present invention provides an optical glass detection system based on image enhancement, the system comprising: An image acquisition module 1 is used for deploying a linear array camera obliquely above the optical glass inspection platform and a light source deployed at a symmetrical position of the linear array camera to capture a first optical glass surface image in real time; The rotation control module 2 is used to control the optical glass inspection platform to rotate clockwise by a preset degree after each image is taken, until the optical glass inspection platform rotates more than or equal to one circle; An image enhancement module 3 is used to enhance the first optical glass surface image using a contrast-limited adaptive histogram equalization algorithm to obtain a second optical glass surface image; The defect detection module 4 is used to input the second optical glass surface image into the optical glass detection model, and output the confidence level, defect category and defect location of the defect on the optical glass surface; The result determination module 5 is used to determine, based on the output result of the optical glass detection model, that the optical glass has defects when the confidence level of defects in at least one second optical glass surface image is greater than a preset threshold.

[0046] 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.

Claims

1. An optical glass detection method based on image enhancement, characterized in that: The method comprises: Step S100: using a linear array camera disposed obliquely above the optical glass inspection platform and a light source disposed at a symmetrical position to the linear array camera to capture an image of a surface of a first optical glass located on the optical glass inspection platform in real time; Step S200: after each image is taken, the optical glass inspection platform is rotated clockwise by a preset degree, and the next image is taken, until the optical glass inspection platform rotates for one circle or more; Step S300: performing image enhancement on the first optical glass surface image using a contrast-limited adaptive histogram equalization algorithm to obtain a second optical glass surface image; Step S400: inputting the second optical glass surface image into the optical glass detection model to output the confidence level, defect type and defect location of defects on the optical glass surface; Step S500: According to the output result of the optical glass detection algorithm, when the confidence level of at least one second optical glass surface image defect is greater than a preset threshold, it is determined that the optical glass has a defect.

2. The optical glass detection method according to claim 1, characterized in that: The shooting angle of the linear array camera is at a preset angle with the optical glass to be inspected, and the light emitted by the light source is at the same preset angle with the optical glass to be inspected.

3. The optical glass detection method according to claim 1, characterized in that: The contrast-limited adaptive histogram equalization algorithm specifically includes: The first optical glass surface image is divided into a plurality of non-overlapping rectangular blocks, and a histogram is calculated for each rectangular block area; The histogram is cropped using a preset threshold, and the cropped part is evenly distributed to other parts of the histogram for histogram reconstruction. Based on the reconstructed histogram, the cumulative distribution function is calculated; A transformation function is calculated based on the cumulative distribution function, and the grayscale value of each pixel is mapped to a new grayscale value using the transformation function to obtain a second optical glass surface image.

4. The optical glass detection method according to claim 1, characterized in that: The optical glass detection model is trained by an improved YOLO v8 algorithm and uses an optical glass surface grayscale dataset.

5. The optical glass detection method according to claim 4, characterized in that: The improved yolo v8 algorithm specifically includes: The backbone network uses a convolutional additive self-attention mechanism to perform global context modeling on the input second optical glass surface image and obtain a feature map of the second optical glass surface image; The neck network uses a scale-compensated feature pyramid network to resize feature maps of different scales. It uses an image feature alignment mechanism to align the features of optical glass images at different rotation angles. Combined with variable convolution, it performs adaptive sampling and image feature extraction for each position in the feature map. The detection head uses the Anchor-Free detection method to predict the defect bounding box and combines the convolution layer to predict the defect category and confidence level. The output layer outputs the location, confidence and defect category of optical glass defects.

6. The optical glass detection method according to claim 4, characterized in that: The optical glass surface grayscale data set is formed by using a historical second optical glass surface image.

7. The optical glass detection method according to claim 1, characterized in that: The step S500 specifically includes: Summarize the optical glass defect positions, optical glass defect confidences and optical glass defect categories of all second optical glass surface images of the optical glass; When there is at least one image of the second optical glass surface with a defect confidence greater than a preset threshold, the optical glass is marked as a defective product, and the optical glass defect position and optical glass defect category corresponding to the highest confidence are recorded as the detection result; Generate a test report containing the test results, and store the test report in association with the corresponding image data.

8. An optical glass detection system based on image enhancement, characterized in that: The system comprises: An image acquisition module is used to deploy a linear array camera obliquely above the optical glass inspection platform and a light source deployed at a symmetrical position of the linear array camera to capture a surface image of the first optical glass in real time; A rotation control module, used to control the optical glass inspection platform to rotate clockwise by a preset degree after each image is taken, until the optical glass inspection platform rotates more than or equal to one circle; An image enhancement module, used for performing image enhancement on the first optical glass surface image using a contrast-limited adaptive histogram equalization algorithm to obtain a second optical glass surface image; A defect detection module, used for inputting the second optical glass surface image into the optical glass detection model, and outputting the confidence level, defect category and defect location of the defects on the optical glass surface; The result judgment module is used to judge that the optical glass has defects according to the output result of the optical glass detection model when the confidence of at least one second optical glass surface image defect is greater than a preset threshold.

Images