Blue glass detection method and system

Abstract

The invention discloses a blue glass detection method and system, and the method comprises the steps: configuring dynamic adaptive illumination based on the defect type and stress distribution of blue glass; generating a bright field light source, a dark field light source and a polarization field light source based on dynamic adaptive illumination, and performing exposure to obtain three light source images; synchronously acquiring bright-field light source, dark-field light source and polarization-field light source images to obtain a fused image; separating the bright field image, the dark field image and the polarization field image based on the fused image, extracting defects in the bright field and the dark field, and extracting stress distribution of the polarization field; and taking the bright field image, the dark field image and the polarization field image as input, constructing a defect-stress model, and outputting a defect type and a stress feature level. Through multi-light-source fusion (bright field, dark field and polarization) and model construction, efficient and accurate detection of defect types and stress distribution is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of new material detection, and in particular to a blue glass detection method and a system thereof. Background Art

[0002] As an absorptive filter material, blue glass's core characteristic lies in its ability to selectively absorb specific wavelengths of light through absorbing substances within the glass matrix. This makes it crucial for applications in optical imaging, laser systems, and precision instrumentation (e.g., high-pixel camera filters and infrared light suppression devices). However, its production process is prone to various defects, including but not limited to uneven surface coating, internal bubbles / inclusions, thickness variations, and structural integrity deficiencies (e.g., cracks or stress concentrations). These defects can directly impact the blue glass's spectral stability, light transmission, and ultimately the reliability of the final product. Summary of the Invention

[0003] In response to the above-mentioned technical deficiencies, the purpose of the present invention is to provide a blue glass detection method and system, which can achieve efficient and accurate detection of defect types and stress distribution through multi-light source fusion (bright field, dark field, polarization) and model construction.

[0004] A first aspect of the present invention provides a blue glass detection method, comprising: Configure dynamic adaptive lighting based on blue glass defect types and stress distribution; Generate bright field light source, dark field light source and polarized field light source based on dynamic adaptive illumination, and perform exposure to obtain three light source images; Synchronously collect images of bright field light source, dark field light source and polarized field light source to obtain fused images; Based on the fused image, the bright field image, dark field image, and polarization field image are separated to extract defects in the bright field and dark field, as well as the stress distribution in the polarization field; Taking bright field images, dark field images and polarized field images as input, a defect-stress model is constructed to output defect types and stress characteristic levels.

[0005] Preferably, configuring dynamic adaptive lighting includes: Bright field illumination; Dark field illumination; Polarized field light source illumination; Through synchronous control, the three light sources can be time-division multiplexed.

[0006] Preferably, generating a bright field light source, a dark field light source, and a polarized field light source based on dynamic adaptive illumination, and performing exposure to acquire an image includes: Based on the time stamp mark or frame ID encoding, the three acquired light source images are aligned on the time axis.

[0007] Preferably, synchronously collecting images of a bright field light source, a dark field light source, and a polarized field light source to obtain a fused image includes: Bright field-dark field fusion image; Dark field-polarization field fusion image; Bright field-dark field-polarization field fusion image; Among them, by setting the fusion weight, the bright area of ​​the bright field light source image is compressed, and the dark area of ​​the dark field and polarization field is stretched.

[0008] Preferably, separating the bright field image, the dark field image, and the polarization field image based on the fused image includes: Bright field image: Edge enhancement is performed on the bright field image and edges are converted into single-pixel width skeletons; Dark field image: correct image background, segment defects, and perform feature screening; Polarization field image: Enhances image contrast and highlights defects.

[0009] Preferably, the bright field image, dark field image and polarization field image are used as input to construct a defect-stress model, and the output defect types include: Set labels based on bright field defect type, dark field defect type, and polarization field stress characteristic level; Splice the bright field image, dark field image, and polarization field image into multi-channel input; Extract defect features and stress features; Construct a joint feature library, including bright field defect types, dark field defect types, and polarization field stress characteristics, and annotate them with labels; Pre-training of bright field defect types and dark field defect types; Polarization field stress features are added for joint training, loss weights are dynamically adjusted, and defect types and stress feature levels are output.

[0010] A second aspect of the present invention provides a blue glass detection system, the system comprising: Lighting module, used to configure dynamic adaptive lighting; An image acquisition module is used to generate a bright field light source, a dark field light source, and a polarized field light source based on dynamic adaptive illumination, and perform exposure to acquire three light source images; An acquisition module is used to synchronously acquire images of a bright field light source, a dark field light source, and a polarized field light source to obtain a fused image; A separation module is used to separate the bright field image, dark field image, and polarization field image based on the fused image, extract defects in the bright field and dark field, and extract stress distribution in the polarization field; The model building module builds a defect-stress model based on bright field images, dark field images, and polarized field images as input; Output module, used to output defect types and stress characteristic levels.

[0011] The beneficial effects of the present invention are: Dynamic adaptive lighting, through synchronous control, enables time-division multiplexing of three light sources, improving imaging contrast and effectively enhancing the visibility of different types of defects (such as bubbles, scratches) and stress characteristics.

[0012] Multi-light source images are collected and fused simultaneously, and three types of images, bright field, dark field, and polarization, are obtained and fused, which significantly improves defect recognition.

[0013] A defect-stress model was constructed, and three types of images were input into the model. The defect type and stress level were jointly analyzed using a multi-channel input method, effectively achieving accurate joint determination of the defect type and stress level. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0015] Figure 1 The present invention provides a flow chart of a blue glass detection method.

[0016] Figure 2 This is a schematic diagram of the process of a blue glass detection system provided by the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0018] Example 1: like Figure 1 As shown, the present invention provides a blue glass detection method, the method comprising: S1: Configure dynamic adaptive lighting based on blue glass defect type and stress distribution.

[0019] Specifically, configuring dynamic adaptive lighting involves: Bright field illumination, using a ring-shaped LED (wavelength matching the light transmission characteristics of blue glass, such as 450nm blue light); Dark field illumination, using a ring fiber coupler (angle offset 5°~15° to avoid direct reflection interference); Polarized field light source illumination, linear polarizer (P polarization or S polarization, alternating switching); Through synchronous control, the three light sources can be time-division multiplexed, and high-speed switches (such as FPGA control) can be used to achieve time-division multiplexing of the three light sources (such as 1ms / channel).

[0020] S2: Generate a bright field light source, a dark field light source, and a polarized field light source based on dynamic adaptive illumination, and perform exposure to obtain three light source images.

[0021] Specifically, based on the time stamp mark or frame ID encoding, the three acquired light source images are aligned on the time axis.

[0022] The specific steps for time-division multiplexing acquisition are as follows: Bright field acquisition: Turn on the ring LED, turn off the dark field and polarized field light sources, and capture surface macro defects (such as scratches and cracks); Dark field acquisition: turn off the bright field and turn on the dark field light source to capture the scattered light from sub-surface defects (such as bubbles and impurities); Polarization field acquisition: Turn off bright field / dark field, turn on the polarized light source and rotate the polarizer (e.g. 0°→45°→90°) to capture the stress distribution of the coating layer.

[0023] Synchronization control: Ensure that the three sets of images are aligned on the time axis through timestamp marking or frame ID encoding.

[0024] S3: Synchronously collect images of bright field light source, dark field light source and polarized field light source to obtain a fused image.

[0025] Among them, including: Bright field-dark field fusion image; Dark field-polarization field fusion image; Bright field-dark field-polarization field fusion image; Specifically, by setting fusion weights, the brightfield light source image is compressed to highlight areas, while the darkfield and polarization fields are stretched to dark areas. Traditionally, separate captures require at least two exposures or two light source switches, and may require moving the camera or light source, slowing down detection. By processing the fused image, brightfield, darkfield, and polarization signals can be simultaneously captured in a single exposure, eliminating the need for multiple captures or mechanical adjustments, significantly improving detection efficiency.

[0026] The fusion weight follows the setting method of bright field as the main and dark field and polarization as the auxiliary: Bright field is used for surface defect detection and requires strong lighting, with a recommended proportion of 50% to 60%; Dark field is used for sub-surface defects and requires medium lighting, with a recommended proportion of 20% to 30%; Polarization fields are used for stress analysis. The signal is weak and excessive interference should be avoided. (10%~20%)

[0027] Specifically, through time division multiplexing, the bright field (1ms), dark field (0.5ms), and polarization field (0.5ms) are triggered in sequence, and the high-speed camera is used to collect frame by frame (total period 2ms). The total energy is distributed according to time, for example, 60% of the time is bright field, 25% of the time is dark field, and 15% of the time is polarization.

[0028] Furthermore, after the images are fused, each channel signal is normalized (for example, the bright field is scaled to [0, 1], the dark field is scaled to [0, 0.5], and the polarization field is scaled to [0, 0.3]), as follows:

[0029] Where, is the normalized channel data, is the original channel data to be processed; is the minimum and maximum value of the channel data; To set the upper limit of the preset range, set bright field = 1.0, dark field = 0.5, and polarized field = 0.3; Indicates the channel type, such as 1 for bright field, 2 for dark field, and 3 for polarized field.

[0030] Example: Inputting Polarization Field Raw Data (weak signal, range [−0.2, 0.1]); After normalization, we get ∈[0,0.3], the defect detection rate is significantly improved in the fused image.

[0031] S4: Based on the fused image, the brightfield, darkfield, and polarization field images are separated to extract defects in the brightfield and darkfield images, as well as the stress distribution in the polarization field. After separation, the brightfield and darkfield images can be processed separately to accurately identify different types of defects. For example, brightfield images are more sensitive to macroscopic surface defects (such as scratches and cracks) but are less sensitive to subsurface defects. Darkfield images are more sensitive to subsurface defects (such as bubbles and impurities) detected through scattered light, but may lack detail on surface defects.

[0032] Specifically, image separation includes: (1) Bright field image: Edge enhancement is performed on the bright field image to extract macro defects of different widths (such as scratches and cracks), and the edges are converted into single-pixel wide skeletons (through morphological skeleton extraction); (2) Dark field image: Collect defect-free sample images as background templates to correct the image background, segment defects (such as dust, bubbles, and micropores), and perform feature screening and elimination based on defect area, circularity, and grayscale contrast, such as: Area: Remove the circularity of noise points <3 pixels; Distinguish between bubbles (roundness > 0.85) and irregular dust; Grayscale contrast: excludes non-defective bright spots.

[0033] (3) Polarization field image: using S The 0 component suppresses specular reflections and highlights surface geometric defects; polarization field images are collected at four angles: 0°, 45°, 90°, and 135°, as follows: S 0= I 0∘ + I 90∘ = I 45∘ + I 135∘ ; in, S 0 represents the total light intensity; I Indicates the light intensity at the four corresponding angles mentioned above; Extraction is performed by correction formula: S 0 校正 = S 0 原始 − k ⋅ S 1 Where, k is the scattering coefficient; .

[0034] S5: Take the bright field image, dark field image, and polarization field image as input, build a defect-stress model, and output the defect type and stress characteristic level.

[0035] Specifically: (1) Set labels based on bright field defect type, dark field defect type, and polarization field stress characteristic level. Specifically: Defect type: scratches, bubbles, cracks, etc.; Stress levels: low, medium, high.

[0036] (2) The bright field image, dark field image, and polarization field image are stitched together into a multi-channel input. For example, if 8 channels are used, the details are as follows: 1) Bright field: RGB channels, 3 channels (R, G, B), directly acquired by a polarization camera; 2) Dark field: grayscale channel, 1 channel (0-255); total light intensity = ; 3) Polarization field: Degree of polarization: DoP channel, 1 channel (0~1); ; Polarization angle decomposition: AoP channel, 3 channels (trigonometric function value + original angle). and Decomposition (when θ changes continuously, 2θ changes continuously, and (\cos) and (\sin) are continuous functions, so their values ​​also change continuously without sudden changes), eliminating the angle sudden change problem caused by the discontinuity between 0° and 180° caused by the 180° periodicity of the polarization angle; in:

[0037]

[0038]

[0039] in, , at 0°-180°.

[0040] To further illustrate, the following example is given: Assume that a pixel measures the light intensity in four polarization directions (unit: grayscale value): = 100, 0° filter; = 40, 45° filter; = 20, 90° filter; = 60, 135° filter; S 0= I 0∘ + I 90∘ = 100+20=120; S 1 = I 0∘ - I 90∘ =100-20=120 =80; S 2= − =−20; Calculate the degree of polarization (DoP): DoP= ; ; ; The continuous value is (0.98, -0.25). Calculate the original angle (with mutation problem):

[0041] pass and : =

[0042] At this time, it is at -90°-90°, avoiding the mutation point of 0°-180°.

[0043] (4) Extract defect features and stress features; use the YOLOv8 (You Only Look Once version 8) real-time target detection model to detect defects (such as scratch location) and output defects.

[0044] (5) Construct a joint feature library, including bright field defect types, dark field defect types, and polarization field stress features, and annotate them with labels; label annotation includes defect type labels, stress level labels, and defect location labels, where the defect location labels are obtained through YOLOv8 bounding box coordinates.

[0045] (6) Pre-training of bright field defect types and dark field defect types; (7) Add polarization field stress features, conduct joint training, dynamically adjust loss weights, and output defect types and stress feature levels.

[0046] Among them, the loss weight (classification loss: positioning loss) is specifically set as follows in YOLOv8: Set the value based on the actual detection situation. For example, when detecting a target, the precise location (positioning) of the bubble is more important than the category judgment (classification). Therefore, the positioning loss weight needs to be increased to focus on key points.

[0047] If high-precision positioning is required, such as micron-level scratches, the loss weight = classification loss: positioning loss = 4:6; If fine-grained classification is required, such as distinguishing between bubbles and dust, the loss weight = classification loss: positioning loss = 8:2.

[0048] Example 2: like Figure 2 As shown, the present invention provides a blue glass detection system, which includes: Lighting module, used to configure dynamic adaptive lighting; An image acquisition module is used to generate a bright field light source, a dark field light source, and a polarized field light source based on dynamic adaptive illumination, and perform exposure to acquire three light source images; An acquisition module is used to synchronously acquire images of a bright field light source, a dark field light source, and a polarized field light source to obtain a fused image; A separation module is used to separate the bright field image, dark field image, and polarization field image based on the fused image, extract defects in the bright field and dark field, and extract stress distribution in the polarization field; The model building module builds a defect-stress model based on bright field images, dark field images, and polarized field images as input; Output module, used to output defect types and stress characteristic levels.

[0049] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A blue glass detection method, characterized in that: include: Configure dynamic adaptive lighting based on blue glass defect types and stress distribution; Generate bright field light source, dark field light source and polarized field light source based on dynamic adaptive illumination, and perform exposure to obtain three light source images; Synchronously collect images of bright field light source, dark field light source and polarized field light source to obtain fused images; Based on the fused image, the bright field image, dark field image, and polarization field image are separated to extract defects in the bright field and dark field, as well as the stress distribution in the polarization field; Taking bright field images, dark field images and polarized field images as input, a defect-stress model is constructed to output defect types and stress characteristic levels.

2. A blue glass detection method according to claim 1, characterized in that: Configuring dynamic adaptive lighting includes: Bright field illumination; Dark field illumination; Polarized field light source illumination; Through synchronous control, the three light sources can be time-division multiplexed.

3. A blue glass detection method according to claim 2, characterized in that: Based on dynamic adaptive illumination, bright field light source, dark field light source and polarized field light source are generated and exposed to obtain images including: Based on the time stamp mark or frame ID encoding, the three acquired light source images are aligned on the time axis.

4. A blue glass detection method according to claim 2, characterized in that: Synchronously collect bright field light source, dark field light source and polarized field light source images to obtain fused images including: Bright field-dark field fusion image; Dark field-polarization field fusion image; Bright field-dark field-polarization field fusion image; Among them, by setting the fusion weight, the bright area of ​​the bright field light source image is compressed, and the dark area of ​​the dark field and polarization field is stretched.

5. The blue glass detection method according to claim 1, wherein: Separate the bright field image, dark field image, and polarization field image based on the fused image, including: Bright field image: Edge enhancement is performed on the bright field image and edges are converted into single-pixel width skeletons; Dark field image: correct image background, segment defects, and perform feature screening; Polarization field image: Enhances image contrast and highlights defects.

6. The blue glass detection method according to claim 1, wherein: The bright field image, dark field image, and polarization field image are used as input to construct a defect-stress model. The output defect types include: Set labels based on bright field defect type, dark field defect type, and polarization field stress characteristic level; Splice the bright field image, dark field image, and polarization field image into multi-channel input; Extract defect features and stress features; Construct a joint feature library, including bright field defect types, dark field defect types, and polarization field stress characteristics, and annotate them with labels; Pre-training of bright field defect types and dark field defect types; Polarization field stress features are added for joint training, loss weights are dynamically adjusted, and defect types and stress feature levels are output.

7. A blue glass detection system, characterized in that: For executing the method according to any one of claims 1 to 6, the system comprises: Lighting module, used to configure dynamic adaptive lighting; An image acquisition module is used to generate a bright field light source, a dark field light source, and a polarized field light source based on dynamic adaptive illumination, and perform exposure to acquire three light source images; An acquisition module is used to synchronously acquire images of a bright field light source, a dark field light source, and a polarized field light source to obtain a fused image; A separation module is used to separate the bright field image, dark field image, and polarization field image based on the fused image, extract defects in the bright field and dark field, and extract stress distribution in the polarization field; The model building module builds a defect-stress model based on bright field images, dark field images, and polarized field images as input; Output module, used to output defect types and stress characteristic levels.

Images