Single-exposure hyperspectral imaging system and detection method based on intensity-polarization joint modulation mask

Abstract

The present invention proposes a single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask and its detection method, comprising an excitation light source group, a compression acquisition module, a reconstruction module, and a polarization noise reduction module. The excitation light source group generates linearly polarized light to illuminate the object under test, causing it to produce photoluminescence or Raman scattering. The compression acquisition module receives the signal and completes the compression acquisition. The reconstruction module uses a deep learning algorithm to reconstruct the full polarization spectrum image data from the compressed observation image. The polarization noise reduction module removes the mirror reflection noise component and outputs the noise-reduced photoluminescence or Raman scattering hyperspectral image data. The present invention can quickly and stably complete photoluminescence and Raman scattering spectral imaging, improve material detection efficiency and signal-to-noise ratio, and is suitable for fields such as semiconductor detection.

Description

Technical Field

[0001] The present invention relates to the technical fields of spectral imaging and spectral detection, and in particular to a single-exposure hyperspectral imaging system based on intensity polarization combined modulation mask and a detection method thereof. Background Art

[0002] In the field of material property analysis, especially in the semiconductor testing industry, photoluminescence and Raman scattering spectroscopy analysis technology plays a vital role. With the continuous development of new energy vehicles, the application prospects of this technology in the future preparation of new energy circuits and components are also very broad.

[0003] However, the existing detection instruments have many limitations. Although the traditional single-point spectrometer solution can analyze the spectral components of each wavelength with high sensitivity and high resolution, it can only analyze a very small area of the object to be measured at a time due to the limitations of the probe's light collection distance and field of view. If you want to complete large-area measurements, you must rely on a high-precision displacement scanning mechanism and spend a lot of time to scan the entire area. This not only makes the measurement efficiency low and makes it difficult to meet the real-time detection needs of the online production line, but also increases the mechanical vibration interference of the system, further pushing up the system cost. Although the detection solution based on traditional cameras can capture images of a large area to be measured at one time, it can only obtain the total integrated intensity of all wavelength components of the excitation light or Raman scattered light of the object to be measured, and does not have the ability to analyze the response of each wavelength. Therefore, it is impossible to deeply analyze the wavelength differences of the material, and it is difficult to accurately measure the key characteristics of the material such as the band gap width and composition based on the characteristic peak information of the excitation scattered light.

[0004] In addition, existing hyperspectral imaging technology also has some problems. For example, when measuring high-reflectivity objects, there is no targeted structural and algorithm optimization. The objects being measured in photoluminescence and Raman scattering measurement scenarios are likely to be high-reflectivity objects, such as semiconductor materials such as silicon carbide. This makes it difficult for traditional single-exposure hyperspectral imaging solutions to obtain high-quality measurement results in this scenario.

[0005] There is also the interferometer-based Raman hyperspectral imaging solution described in CN112268618B. Although it has the characteristics of high light flux and high spatial resolution, the large number of lenses in the system, especially the increase in the number of reflectors, significantly increases the complexity of its optical design and the difficulty of assembly and adjustment, resulting in a significant increase in production costs.

[0006] In summary, existing detection technologies still have much room for improvement in acquiring global spatial and spectral information of materials quickly, efficiently, stably and at low cost. Summary of the Invention

[0007] The embodiments of the present invention provide a single-exposure hyperspectral imaging system and detection method based on intensity polarization combined modulation mask, which addresses the significant deficiencies in current technologies in efficiently and accurately detecting material photoluminescence and Raman scattering spectral images, and the difficulty in simultaneously meeting the measurement requirements of fast, stable, low cost and high signal-to-noise ratio.

[0008] The core technology of this invention is mainly to complete the photoluminescence / Raman scattering spectral data acquisition of large-field-of-view objects in a single exposure through single-exposure hyperspectral imaging technology using intensity polarization combined with modulation mask, combined with deep learning spectral reconstruction and polarization noise reduction algorithm, and use polarization information to suppress specular reflection noise.

[0009] In a first aspect, the present invention provides a single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask, comprising the steps of:

[0010] The excitation light source module is used to generate linearly polarized light and illuminate the object to be measured, causing the object to produce photoluminescence or Raman scattering. The excitation light source module includes a laser light source system, a filter set and a polarizer;

[0011] A compression acquisition module is used to receive the photoluminescence signal or Raman scattering signal and the mirror reflection noise signal generated by the object under test, and complete the compression acquisition of the full polarization spectrum image signal. The compression acquisition module includes a first filter group, a first lens group, an intensity polarization joint modulation mask, a second lens group, a dispersion device and a detector array;

[0012] A reconstruction module is used to reconstruct the full polarization spectrum image data to be measured from the compressed observation image;

[0013] The polarization noise reduction module is used to remove the specular reflection noise component and output the noise-reduced photoluminescence or Raman scattering hyperspectral image data.

[0014] Furthermore, the laser light source generated by the laser light source system includes a collimation system, and each light source is centrally symmetrically distributed about the line connecting the center of the object to be measured and the center of the compression acquisition module lens. The angle between the optical axis of the light source and the plane of the object to be measured is maintained at a fixed specific angle so that the laser spot evenly covers the object to be measured and maximizes energy utilization efficiency.

[0015] Furthermore, the polarizer adopts a wire grid polarizer to rotate or filter the polarization state of the laser light source to a specific linear polarization state, ensuring that the light source system emits or reflects only the specified linear polarization state component for subsequent noise filtering processing.

[0016] Furthermore, the intensity-polarization joint modulation mask is used to perform transverse spatial-correlated intensity modulation on the full polarization spectrum image signal to be measured, and the four components of the Stokes vector representing different polarization information are subjected to different transverse shifts to complete polarization modulation.

[0017] Furthermore, the reconstruction module includes an image preprocessing algorithm, a reconstruction deep neural network and an image post-processing algorithm, which are used to process and reconstruct the collected compressed observation signal image to obtain the full polarization spectrum image data to be measured.

[0018] Furthermore, the polarization noise reduction module removes specific polarization signals, namely the mirror reflection noise component, based on the orthogonal decomposition model. The mirror reflection noise component is the light of the excitation light source that is not filtered out by the filter or the light that is secondary illuminated by photoluminescence or Raman scattered light and secondary reflected to the detector.

[0019] In a second aspect, the present invention provides a detection method for a single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask, comprising the following steps:

[0020] S1. Excitation and signal generation: The excitation light source module emits linearly polarized light to illuminate the object being measured, stimulating photoluminescence or Raman scattering signals.

[0021] S2. Compressed Acquisition: The compressed acquisition module performs intensity and polarization joint modulation, dispersion spectroscopy, and detector array acquisition on the signal and noise light to obtain a compressed observation image.

[0022] S3. Spectral Reconstruction: Reconstructing a fully polarized hyperspectral image from the compressed observation image using a deep neural reconstruction network based on the reconstruction module.

[0023] S4. Polarization noise reduction: The polarization noise reduction module uses a polarization degree separation algorithm to remove specular reflection noise, extract depolarized light as signal light, and improve the system's signal-to-noise ratio. The resulting spectral image is then output after noise reduction.

[0024] Furthermore, the specific method for the excitation light source module to generate linearly polarized light is: the laser light source system generates a plurality of laser light sources, which are filtered by a filter group, and then the polarization state is rotated or filtered to a specific linear polarization state by a polarizer.

[0025] Furthermore, the compression acquisition module completes the process of full polarization spectrum image signal compression acquisition, including: the full polarization spectrum image signal to be measured is imaged to the first image plane through the first filter group and the first lens group, the intensity polarization joint modulation mask modulates the signal and then images it to the second image plane, the second image plane image is collimated and scaled by the second lens group and then enters the dispersion device for dispersion modulation, and finally the full polarization spectrum compression signal acquisition is completed on the detector array surface.

[0026] Furthermore, the reconstruction module uses a deep learning algorithm to obtain a fully polarized three-dimensional spectral image from the compressed observation image, including the steps of training a deep neural network and using the trained deep neural network to reconstruct the image.

[0027] The main contributions and innovations of the present invention are as follows:

[0028] 1. Efficient detection: Using a directional global excitation light source with a specific spatial arrangement, it can simultaneously excite a large area of the object being tested, achieving large-scale synchronous occurrence of photoluminescence and Raman scattering phenomena, greatly improving detection efficiency and meeting the real-time detection needs of online production lines.

[0029] 2. No motion mechanism required: No motion mechanism is required for the probe or the object being measured, which effectively improves the overall structural stability of the system, reduces the cost of optical and mechanical design, reduces mechanical vibration interference, and enhances the reliability and stability of the system.

[0030] 3. High-precision analysis: With the help of hyperspectral imaging technology, spectral data with higher wavelength resolution can be obtained, thereby achieving a more comprehensive and accurate analysis of material properties, and providing strong support for the accurate measurement of key properties such as the band gap width and composition of the material.

[0031] 4. Advantages of joint feature analysis: Combining the advantages of single-point spectrometer solutions and traditional camera solutions, the collected data has both high-precision structural information characteristics and high-resolution spectral information characteristics, which is conducive to improving the accuracy of material analysis through joint feature analysis. For example, the defect type of semiconductor material can be accurately determined by combining spectral characteristic peak information and impurity area morphology.

[0032] 5. Advantages of single-exposure hyperspectral imaging: It avoids the high input signal strength requirements of traditional line-scan hyperspectral cameras and the high cost of interferometric hyperspectral cameras. It can quickly and stably complete photoluminescence and Raman scattering spectral imaging, greatly improving the speed and stability of detection, reducing the signal strength requirements, and broadening the application range of the technology.

[0033] 6. Polarization Information Utilization: By introducing an intensity-polarization joint modulation mask, polarization modulation and spatial intensity coded modulation are simultaneously performed, acquiring polarization information while simultaneously acquiring hyperspectral data. This polarization information is used to suppress specular reflection noise, further improving the system's signal-to-noise ratio in photoluminescence and Raman scattering measurement scenarios, thereby enhancing the accuracy and reliability of the detection results.

[0034] The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below so that other features, objects, and advantages of the invention are more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0036] Figure 1is a flow chart of a single-exposure hyperspectral imaging detection method based on intensity-polarization joint modulation mask according to an embodiment of the present invention;

[0037] Figure 2 is a principle block diagram of a detection method according to an embodiment of the present invention;

[0038] Figure 3 is a structural diagram of an implementation method according to an embodiment of the present invention;

[0039] Figure 4 is a side view schematic diagram of the relative spatial positions of the excitation light sources according to an embodiment of the present invention;

[0040] Figure 5 is a side view schematic diagram of the position constraint of the excitation light source according to an embodiment of the present invention;

[0041] Figure 6 2. It is a schematic diagram of the principle of single-exposure compression acquisition with intensity and polarization joint modulation according to an embodiment of the present invention;

[0042] Figure 7 It is a deep neural reconstruction network algorithm framework according to an embodiment of the present invention;

[0043] Figure 8 2 is a diagram showing the effect of a specular reflection noise removal algorithm based on polarization degree according to an embodiment of the present invention;

[0044] Figure 9 This is a post-processing diagram of a single-wavelength result of a silicon carbide photoluminescence signal collected according to an embodiment of the present invention.

[0045] In the figure, 1, compression acquisition module; 2, laser light source system; 3, object to be measured; 21, light shield; 22, laser light source; 23, first filter group; 24, second filter group. DETAILED DESCRIPTION

[0046] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The implementations described in the following exemplary embodiments are not intended to represent all implementations consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with certain aspects of one or more embodiments of this specification, as detailed in the appended claims.

[0047] It should be noted that in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than those described in this specification. In addition, a single step described in this specification may be broken down into multiple steps for description in other embodiments, and multiple steps described in this specification may be combined into a single step for description in other embodiments.

[0048] Existing technologies are unable to simultaneously achieve rapid acquisition, high-resolution analysis, and high-reflection noise suppression of photoluminescence and Raman scattering spectra within a large field of view, resulting in low detection efficiency, insufficient accuracy, and complex systems.

[0049] Based on this, the present invention solves the problems existing in the prior art by using a single-exposure hyperspectral imaging technology based on intensity polarization combined with modulation mask.

[0050] Example 1

[0051] The present invention aims to propose a single-exposure hyperspectral imaging system based on intensity polarization joint modulation mask, specifically, referring to Figure 2 , the system comprising:

[0052] 1. Excitation light source module: This module is used to emit linearly polarized light of a specific wavelength to excite the object 3 to produce photoluminescence or Raman scattering signals. It includes a laser light source system 2, a first filter set 23, and a polarizer. The laser light source system 2 is centrally symmetrically distributed with respect to the center of the object 3 and the optical axis of the compression acquisition module 1. The angle between the optical axis and the plane of the object 3 is fixed, so that the light spot on the surface of the object 3 is distributed in a perfect circle.

[0053] In this embodiment, the laser light source system 2 is used to generate a plurality of laser light sources 22 that are irradiated onto the object under test 3. Preferably, the laser light source 22 should only contain the target excitation wavelength component to avoid overlapping of the characteristic peaks of photoluminescence and Raman scattering. At the same time, the light source system may include a collimation system, and the light sources are centrally symmetrically distributed about the line connecting the center of the object under test 3 and the center of the lens of the compression acquisition module 1. The angle between the optical axis of the light source and the plane of the object under test 3 should be maintained at a fixed specific angle so that the Gaussian beam emitted by the laser light source 22 is projected onto the plane of the object under test 3 in the shape of a perfect circle. Ultimately, this ensures that the laser spot evenly covers the object under test 3 and maximizes energy utilization efficiency.

[0054] The first filter group 23 includes several filters for filtering out the light components of the target excitation wavelength component to further improve the signal-to-noise ratio of the excitation light source and avoid the introduction of additional noise by the secondary illumination of the excitation light source module by the photoluminescence signal and the Raman scattering signal.

[0055] The polarizer is used to rotate or filter the polarization state of the laser light source to a specific linear polarization state. At the same time, the polarizer needs to adopt a structure such as a wire grid polarizer to ensure that all light reflected by it has only a specific polarization component, ultimately ensuring that the light emitted or reflected by the light source system has only a specified linear polarization state component for subsequent noise filtering processing.

[0056] 2. Compression acquisition module 1: This module receives signal light and noise light from the object under test 3 and performs compression acquisition of the full polarization spectrum signal. It includes a second filter group 24, a first lens group, a combined intensity and polarization modulation mask, a second lens group, a dispersion device, and a detector array. The combined intensity and polarization modulation mask is deployed on the first image plane and performs lateral spatial intensity modulation and polarization state offset modulation on the light field.

[0057] In this embodiment, the compression acquisition module 1 includes a second filter group 24, a first lens group, an intensity-polarization joint modulation mask, a second lens group, a dispersion device, and a detector array. The specific process is as follows:

[0058] The full polarization spectrum image signal to be measured will first be imaged to the first image plane through the filter group and the first lens group. The intensity polarization joint modulation mask will be deployed on the first image plane. The signal will be imaged to the second image plane after completing the intensity polarization joint modulation on the first image plane. Then, the image on the second image plane will pass through the second lens group to complete secondary collimation and scaling to ensure that the light entering the dispersion device is parallel light and the final image size matches the detector array size. Finally, the optical signal will enter the dispersion device to complete the dispersion modulation and complete the full polarization spectrum compression signal acquisition on the third image plane, that is, the detector array surface.

[0059] The functions of each component are as follows: the filter group is used to filter out noise light components outside the signal light wavelength range, so that the acquisition module receives only photoluminescence and Raman scattering spectrum image signals as much as possible, avoiding the impact of other wavelength light components on the received signal-to-noise ratio;

[0060] The first lens group is used to image the full polarization spectrum image signal to be measured from the object plane to the first image plane, completing the first image plane matching;

[0061] The intensity-polarization joint modulation mask is used to perform transverse spatially correlated intensity modulation on the full polarization spectrum image signal to be measured, and to perform different transverse shifts on the four components of the Stokes vector representing different polarization information to achieve polarization modulation. Preferably, the above modulation should be dispersion-managed as much as possible to ensure that only intensity- and polarization-dependent modulation is achieved, rather than wavelength-dependent modulation. The modulated signal is imaged onto the second image plane.

[0062] The second lens group is used to optically align the second image plane image with the detector array size and form parallel light to input the dispersion device;

[0063] The dispersion device is used to split the wide-band light beam that has been intensity- and polarization-modulated and to perform dispersion modulation on light beams of different wavelengths separately, so that the image signals of different wavelengths that have been intensity- and polarization-modulated can be imaged in different lateral areas of the detector array, realizing compressed observation; the detector array is used to collect compressed observation images and complete the photoelectric signal conversion.

[0064] 3. Reconstruction module: This module includes a deep neural reconstruction network, which is used to reconstruct the full polarization hyperspectral image based on the observation data of the compressed acquisition module 1;

[0065] In this embodiment, the reconstruction module includes an image preprocessing algorithm, a reconstruction deep neural network, and an image post-processing algorithm. The image preprocessing algorithm mainly performs noise reduction, grayscale equalization, cropping, and image alignment on the collected compressed observation image to match the input data requirements of the deep neural reconstruction network. The deep neural reconstruction network mainly forward-propagates the preprocessed compressed observation image through a pre-trained deep neural network to reconstruct the compressed observation image into the full polarization spectrum image data to be measured. The image post-processing algorithm is mainly used to perform preliminary noise reduction, cropping, and labeling on the full polarization spectrum image to be measured, so as to further extract the photoluminescence and Raman scattering components.

[0066] 4. Polarization noise reduction module: Used to remove specular reflection noise based on the polarization degree separation algorithm and output a denoised spectral image.

[0067] In this embodiment, the specular reflection noise component may be composed of light from the excitation light source that is not filtered out by the filter, or light that is secondary illuminated by photoluminescence or Raman scattered light and secondary reflected to the detector. The removal principle is based on the fact that the light after specular reflection still maintains a specific polarization state that is the same as or similar to its reflection source, while the Raman scattered light and photoluminescence signals are mostly weakly polarized or completely depolarized. Based on this, an orthogonal decomposition model can be established to remove specific polarization signals, that is, to remove noise.

[0068] All optoelectronic devices in the above method are not restricted in their operating wavelength range or specific parameters. The operating wavelength and other parameters should be selected to match the application scenario. For example, for typical UV-excited photoluminescence measurements of silicon carbide, the operating center wavelength of laser light source 22 can be 280-360nm. The illumination power depends on the surface power receivable by the object being measured and the excitation threshold, typically in the watt range. In this specific embodiment, the maximum wavelength that can be transmitted by the filter of the excitation light source module should be less than 360nm, while the minimum wavelength that can be transmitted by the filter of the compression acquisition module 1 should be greater than 360nm.

[0069] The system design diagram of some embodiments of the excitation light source module and compression reconstruction module of the present invention is as follows: Figure 3As shown, the object under test is in the xoy zero plane, and is illuminated by an excitation light source distributed in a regular polygon on the xoy plane, causing photoluminescence or Raman scattering. Then, the photoluminescence or Raman scattering signal light and other noise light form a signal to be measured and are transmitted to the compression acquisition module 1 through the light shield 21 to complete the compression observation.

[0070] For the excitation light source module, in order to ensure that the excitation light source illuminates the object as evenly as possible and is as close to the object 3 as possible without entering the field of view of the compression acquisition module 1 to increase the intensity of the photoluminescence or Raman scattering signal, preferably, the position of the excitation light source should be arranged according to the specified spatial position: the side view schematic diagram of the relative spatial position of the excitation light source is as follows Figure 4 As shown in , when the excitation beam waist position is much larger than the distance between the light source and the object and the excitation light spatial transverse mode only has the fundamental mode, the beam path can be simplified to a conical cylinder region, as shown in Figure 4 The midpoints ABEF form a quadrilateral, where point O represents the center of the light source. represents the radius of the excitation light source aperture, The height of the representative point O from the plane of the object being measured 3, The distance between the representative point O and the optical axis of the compression acquisition module 1, Represents the light source divergence half angle, represents the angle between the normal vector of the excitation light source (i.e., OC) and the plane of the object 3 under test, BE represents the range of the light spot on the plane of the object 3 under test, point C represents the intersection of the light emitted from point O and the plane of the object 3 under test, points A and F represent the maximum boundary points of the light that can be emitted by the excitation light source, and point D represents the intersection of the optical axis of the compression acquisition module 1 and the plane of the object 3 under test. In order to ensure that the object 3 under test is excited and illuminated as evenly as possible within the field of view, the light spot on the plane of the object 3 under test must be as symmetrical as possible about the center of the field of view of the final captured image, that is, symmetrical about the center of the optical axis of the compression acquisition module 1. Therefore, the midpoint of BE must be as close to point D as possible. According to geometric relationships and trigonometric functions, it is easy to obtain:

[0071] (1)

[0072] in are the distances between point A\F and point O in the Y and Z directions, respectively, so we can get As follows:

[0073] (2)

[0074] At the same time, in order to prevent the excitation light source from being imaged into the field of view of the compression acquisition module 1 to eliminate interference, the distance between the innermost boundary of the excitation light source and the optical axis is Certain constraints must be met, such as Figure 5 As shown, Represents the half angle of the field of view of the compressed acquisition module 1, Represents the equivalent aperture radius collected by compression acquisition module 1, Represents the distance between the equivalent aperture plane of the compressed acquisition module 1 and the plane of the measured object 3. According to the geometric relationship, the maximum field of view radius of the compressed acquisition module 1 on the XOY plane passing through point F and The constraint relationship can be expressed as follows, which is subsequently referred to as the first constraint relationship:

[0075] (3)

[0076] Preferably, in order to maximize the intensity of the excitation light source to improve the final desired scattering signal or luminescence signal signal-to-noise ratio, the Euclidean distance (i.e., OD length) between the excitation light source and the object 3 should be as short as possible. This constraint relationship is called the second constraint relationship and is expressed as follows:

[0077] (4)

[0078] In summary, by substituting formulas (1) and (2) into (3) and (4), the joint constraint relationship that satisfies both the first constraint relationship and the second constraint relationship can be expressed as follows:

[0079] (5)

[0080] In the actual design process, the indicators related to the compression acquisition module 1 and the light source divergence angle and aperture indicators are usually fixed at the time of selection. Only some parameters need to be optimized later, such as That's it.

[0081] After the relative position of the light source is optimized, the excitation light source will pass through a bandpass filter that can only pass the working wavelength range of the light source and then pass through a polarizer. The Stokes vector can be expressed as follows:

[0082] (6)

[0083] in is the angle between the polarizing direction of the polarizer and the reference axis (such as the x-axis), are the minimum and maximum working wavelengths of the excitation light source, is the original light intensity distribution of the excitation light source, and the wavelength is , the spatial position is In photoluminescence detection applications, the typical minimum operating wavelength and maximum operating wavelength can be 355nm and 365nm.

[0084] In this embodiment, for the photoluminescence signal and Raman scattering signal acquisition part, since this system only focuses on the photoluminescence and Raman scattering signal components of the object under test 3 itself, and the photoluminescence signal and Raman scattering signal intensity are relatively weak, in order to further improve the system signal-to-noise ratio, the remaining signal components are regarded as noise. Preferably, a wire grid polarizer structure is processed on all surfaces that can be directly seen by the object under test (that is, all points on the surface of the object under test are taken as source points, full-angle rays are made, and there is an intersection with these rays and the intersection is the intersection closest to the source point on the ray, then the surface where the intersection is located is considered to be a directly visible surface) to make all light reflected by these surfaces become linearly polarized light. If only one reflection noise is considered, the three-dimensional light field Stokes vector that will enter the compression acquisition module 1 after being excited by the excitation light source without any filtering It can be expressed as:

[0085] (7)

[0086] (8)

[0087] (9)

[0088] in The target acquisition component is the photoluminescence signal or Raman scattering signal of the object being measured that is excited and not reflected or scattered by the surface of other objects; represents the noise component of the photoluminescence or Raman scattering signal that enters the acquisition module after being reflected or scattered by the environment. Indicates that when the line grid period direction is parallel to the x-axis The wire grid polarizer reflects the Mueller matrix at the angle, and the specific expression of the Mueller matrix is shown in formula (8): is the actual processed wire grid period direction angle; represents the noise component of the excitation light source that enters the acquisition module after reflection or scattering. In formula (7), It means that when the incident angle is The surface reflection Mueller matrix at the time of , the specific expression of the Mueller matrix is shown in formula (9), It represents the Fresnel reflection coefficient of the polarization s component and the polarization p component, is the incident angle of the tangent plane when the actual light irradiates the object 3.

[0089] in Since the component is homologous to the target collection component and has the same wavelength range, it cannot be filtered out by a filter alone. Therefore, it is preferred to implement the device by the aforementioned surface treatment of the device, while adding a wire grid polarizer and reducing the stray light shielding device to directly weaken the reflection, scattering, and ambient light intensity; Due to the wavelength shift characteristics of the photoluminescence signal and the Raman scattering signal, Some components can be filtered by combining filter sets without affecting the signal light, such as Figure 3 The transmission wavelength range of the second filter group 24 in the medium compression acquisition module 1 is and That is, it is only necessary to ensure that the transmission wavelength range of the acquisition module and the transmission wavelength range of the light source module do not overlap in order to filter out most of the noise components introduced by the light source. The present invention does not limit the form of the filter set and the wavelength transmission characteristics. The remaining light source noise components will be filtered out through subsequent polarization analysis. In a typical photoluminescence detection scenario, the typical minimum transmittance wavelength of the second filter set 24 can be set to 400nm. In a Raman scattering scenario based on 532nm green light excitation, this value is typically 534nm (considering only the Stokes component).

[0090] Preferably, if the object under test 3 is a non-metallic dielectric and the signal light It is approximately depolarized light or weakly polarized light because For linearly polarized light, For linearly polarized light, only the polarization angle may change, so formulas (7)-(9) can be degenerated into:

[0091] (10)

[0092] in represents the linear polarization state rotation angle introduced by the reflection of non-metallic dielectric materials, Represents the intensity distribution of the signal light. It is easy to see from the above formula that under this simplified model, the signal received by the final compression acquisition module 1 will contain depolarized light components and linearly polarized light components; by extension, if the object under test 3 is a metal dielectric or semiconductor medium, since the light absorption coefficient in the complex refractive index of the material is non-zero, then Will be due to By introducing a phase difference between the s- and p-polarization directions, the signal is converted to elliptically polarized light. The signal received by the compression acquisition module 1 will contain depolarized light components, linearly polarized light components, and elliptically polarized light components. Once the polarization components of the received signal are clearly identified, subsequent noise filtering algorithms can be designed based on this information.

[0093] 3D Light Field Stokes Vector It will be sent to the compression acquisition module 1 to complete the compression observation acquisition. First, the signal will be filtered by the aforementioned filter group 2 to only transmit the signal light wavelength component, and then the first lens group will focus on the intensity polarization joint modulation mask on the first image plane, and then start compression modulation and acquisition. The principle diagram of the compression acquisition module 1 is shown in the figure. Figure 6 As shown, the three-dimensional light field Stokes vector After the transfer function is The light field after the intensity and polarization joint modulation mask transmission It can be expressed as:

[0094] (11)

[0095] The light field is the encoded spectral image, where the key points are Design, such as Figure 6 As shown, the intensity-polarization joint modulation is mainly used to encode different polarization states and different transverse spatial positions with different intensities to meet the limited isometry constraints in single-exposure compressed sensing reconstruction, so as to facilitate the subsequent reconstruction of hyperspectral images with different polarization states. The main difference between the intensity-polarization joint modulation mask of the present invention and the traditional single-exposure intensity mask is that for a mask unit aligned with the sensor pixel, it will be further split into four sub-units. , each subunit has a different transmittance to complete intensity modulation. At the same time, each subunit has a different lateral offset on the xoy plane for different components of the signal Stokes vector to complete polarization modulation, ultimately realizing an intensity-polarization joint modulation mask. Preferably, the mask needs to perform dispersion management at the same time to ensure that the intensity and polarization modulation of each wavelength are as similar as possible, so as to improve the accuracy of the subsequent reconstruction network forward model. The mathematical model of the above modulation process can be expressed as follows:

[0096] (12)

[0097] in It represents the modulated Mueller matrix of the original input signal of the subunit numbered n, which contains the intensity transmittance modulation term and the polarization shift modulation term Furthermore, if we consider one of the embodiments of preparing an intensity-polarization joint modulation mask by using a metasurface, since the metasurface can independently design phase gradients in the x and y directions of the lateral space, in order to simplify the theoretical model, we only consider the modulation transfer function in the x direction and the incident light medium is air. At the same time, we consider that the metasurface is composed of nanopillars with adjustable rotation angle, duty cycle, and height. In the above formula, It can be further expanded into:

[0098] (13)

[0099] In the formula, all variables with subscripts n indicate that they belong to the subunit numbered n, where represents the spatial distribution of the angle (rotation angle) between the long axis of the metasurface nanopillar and the x-axis; The phase space distribution introduced for the metasurface; is the spatial phase gradient of the metasurface; is the polarization conversion efficiency coefficient of the subunit; is the spatial distribution of the equivalent refractive index of the metasurface; represents the spatial distribution of nanopillar duty cycle; represents the intrinsic material absorption coefficient; represents the nanopillar height; represents the periodic spatial distribution of nanopillars; represents the handedness of incident light; Indicates the refractive index of the substrate material used for the metasurface. In the scenario where the measured signal light is visible light, the typical substrate material can be ; Represents the spatial distribution of nanocolumn diameters. From the above, it can be seen that if different sub-units are required to achieve different intensity modulations, this can be achieved by changing the duty cycle distribution of the metasurface nanocolumns and the height of the nanocolumns. Furthermore, according to the above formula, the Mueller matrix of polarization state modulation of different sub-units depends on the spatial distribution of the rotation angles of the metasurface nanocolumns in each sub-unit, the duty cycle distribution, and the height of the nanocolumns, and the offset of the signal modulated by different sub-units in the lateral space can be further calculated based on the phase gradient change. If the second image plane is considered to be the equivalent focal plane introduced by the metasurface, the lateral spatial offset of the modulated signal on the second image plane can be approximately expressed by the following formula:

[0100] (14)

[0101] In summary, the intensity polarization joint modulation mask can be realized by preparation technology such as metasurface to The optical field is independently intensity modulated and polarized, and each component is separated. For scenarios where the photoluminescence and Raman scattering signals are green light, the typical offset achievable on the second image plane is typically tens of microns, considering a dispersion compensation range of 550nm±75nm.

[0102] After that, the light field after coding and modulation on the second image plane will pass through the second lens group to complete zooming and collimation and enter the dispersion element for dispersion modulation, so that the light fields of different wavelengths will have different offsets along a specific axis. Assuming that the axis here is the y-axis, the light field after dispersion shift, that is, the spectral image after dispersion, on the detector array plane can be expressed as:

[0103] (15)

[0104] in It represents the dispersion offset produced by the dispersive element in the detector array plane.

[0105] Afterwards, the pixel space coordinate system is introduced , respectively represent the rows and columns of pixels, so on the third image plane, i.e., the detector array plane, the photocurrent intensity that can be collected by each pixel can be expressed as:

[0106] (16)

[0107] in Indicates the responsivity of different regions of the detector array to different wavelengths. Further, the above formula is quantified to define the spatial dimension The number of pixels in the direction is , spectral range Contains channels, and includes The polarization components to be solved, such as 4 for the Stokes vector, are expressed as vectors. , then in the above formula Can be represented as a vector , expanded to:

[0108] (17)

[0109] in , is the transmission matrix of the compressed observation module. The difference between the present invention and the traditional compressed single-exposure hyperspectral imaging technology is whether the transmission matrix of the compressed observation module will encode the polarization state. The matrix is obtained by the calibration process in actual use. Preferably, when calibrating the forward model, it is necessary to use fully polarized light with different polarization states to calibrate the system to clarify the system's modulation of each polarization component; y can be called a compressed observation; the above formula can be called the system forward model. In the actual system, due to the system noise With the existence of , the system forward model will degenerate into:

[0110] (18)

[0111] As shown in the above formula, the method of the present invention acquires two-dimensional compressed observations in a single exposure mode without the need for a scanning mechanism, and thus can be acquired at a video frame rate. Furthermore, in order to recover the three-dimensional spectral image of each Stokes component of the object under test from the compressed observation y, that is, the full polarization three-dimensional spectral image, it is necessary to inversely solve the ill-posed inverse problem of formula (18). However, due to the sparse characteristics of natural spectral image signals, in N λ >>1, the signal can still be recovered from the compressed observation, so the compressed observation will be sent to the reconstruction module for spectral and polarization reconstruction.

[0112] In this embodiment, the reconstruction module of the present invention mainly uses a deep learning algorithm to obtain a full polarization three-dimensional spectral image from compressed observations, that is, the key component is a deep neural reconstruction network. The deep learning algorithm framework includes two steps: training and inference. Figure 7 As shown:

[0113] The first step is the deep neural network training phase: First, prepare a full polarization spectrum image dataset that matches the system channel. , where the subscripts represent the batches of training data; obtain the system forward model of the acquisition module ; Data is obtained through formula (17) Compressed observation ;Will and Input into the deep neural network for forward propagation operation to obtain the reconstructed full polarization spectrum image ; The data and Input into the loss function to calculate the loss value; perform backpropagation operation on the model based on the loss value and update the deep neural network parameters; after updating the neural network parameters, repeat the input of the next batch of data sets and compressed observations into the deep neural network operation and backpropagation until the loss value is less than the target loss value, stop training, save the current deep neural network model parameters, and end step one.

[0114] The second step is the deep neural network inference stage, that is, the actual application stage: the compressed observations collected by the acquisition module are and the system forward model Input into the trained deep neural network to obtain the reconstructed full polarization 3D spectral image The inference phase in step 2 typically takes milliseconds, depending primarily on the dispersion acquisition module parameters, neural network parameters, and computing resources. After obtaining the trained deep neural network model in step 1, the reconstruction module only needs to perform image reconstruction in step 2, inference, and no longer relies on step 1. As a result, the entire system can acquire a full polarization spectrum image in milliseconds, helping to improve inspection process efficiency in semiconductor applications.

[0115] The pre-processing algorithm in the reconstruction module mainly implements operations such as multiple exposures, cropping, and exposure matching to further improve the input neural network signal-to-noise ratio while matching the model input size requirements; the post-processing algorithm in the reconstruction module mainly completes operations such as filtering and cropping to further extract the photoluminescence or Raman scattering components of the signal.

[0116] Then, as shown in the block diagram Figure 2 As shown in the figure, the reconstructed full polarization hyperspectral image will be sent to the polarization noise reduction module to remove the specular reflection noise. The principle of specular reflection noise is as follows:

[0117] According to formulas (7)-(10), if we consider that most photoluminescence signals and Raman scattering signals are depolarized light or weakly polarized light, and due to the polarizer in the excitation light source module and the wire grid polarizer on the inner surface, the mirror reflection noise will mostly be fully polarized light such as linear polarized light or elliptically polarized light, and the final received signal here can be expressed as a linear superposition of noise and signal, then a polarization degree separation model can be established to distinguish the desired signal light from the noise light. The specific mathematical model can be expressed as follows:

[0118] (19)

[0119] in Hyperspectral image representing the depolarized light component, A hyperspectral image representing the fully polarized light component, Represents the reconstructed full-polarization hyperspectral image, that is, the hyperspectral image of the mth component in the hyperspectral Stokes vector. Based on this reconstruction result and the assumption that the signal light is depolarized or weakly polarized, the depolarized light in the reconstructed signal can be extracted by using the polarization degree separation algorithm, as shown in the following formula:

[0120] (20)

[0121] Where DoP is the degree of polarization. For depolarized light, DoP=0. Based on the above formula and the reconstruction results, the signal light extraction can be completed, and the signal-to-noise ratio of the system result can be further improved. The effect diagram of the above algorithm is as follows Figure 8 As shown in the figure, this is a single-wavelength image of the reconstructed first Stokes vector component. The red frame area on the left of the figure marks the significant reflection noise on the surface of the object 3 under test. The reflection noise is caused by the secondary illumination of the internal structure by the photoluminescence signal. The right image is the result after noise filtering.

[0122] Then, the noise-filtered results can be input into subsequent applications as a high signal-to-noise ratio photoluminescence or Raman scattering signal source for application-side analysis. Figure 9 It shows the small facet analysis of silicon carbide ingot based on the data source of the present invention, where the left picture is the signal source (original image), the middle picture is the small facet segmentation result based on the signal source, the dark purple part is the small facet of silicon carbide ingot, and the right picture shows the photoluminescence intensity distribution result of specific wavelength of silicon carbide ingot based on the signal source.

[0123] Example 2

[0124] like Figure 1 As shown, based on the same concept, the present invention also proposes a detection method for a single-exposure hyperspectral imaging system based on intensity polarization combined modulation mask, comprising:

[0125] S1. The excitation light source module emits linearly polarized light to illuminate the object 3, causing it to produce photoluminescence or Raman scattering. The excitation light source module emits linearly polarized light at a specific angle at a non-entry distance to illuminate the object 3, causing it to produce photoluminescence or Raman scattering, generating signal light.

[0126] S2. The photoluminescence signal / Raman scattering signal and the mirror reflection noise signal of the object 3 enter the compression acquisition module 1 and are imaged to the first image plane by the first lens group.

[0127] S3. The image on the first image plane is modulated by the intensity-polarization joint modulation mask and imaged onto the second image plane.

[0128] S4. The modulated second image plane image is collimated and scaled twice by the second lens group, and then sent to the dispersion device to complete dispersion modulation, and finally imaged to the third image plane where the image sensor is located, completing the full polarization spectrum compression signal acquisition.

[0129] S5. The compressed signal is transmitted from the compression acquisition module 1 to the reconstruction module to reconstruct the original full polarization spectrum image.

[0130] S6. The original full polarization spectrum image is sent to the polarization noise reduction module to filter the specular reflection component and finally output the denoised hyperspectral image.

[0131] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be repeated here.

[0132] In general, various embodiments may be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention may be implemented in hardware, while other aspects may be implemented in firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flow charts, or using some other graphical representation, it should be understood that, as non-limiting examples, the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0133] The embodiments of the present invention may be implemented by computer software that is executable by a data processor of a mobile device, such as in a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets and / or macros may be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer executable components that are configured to perform an embodiment when the program is run. One or more computer executable components may be at least one software code or a portion thereof. In addition, it should be noted at this point that, for example, Figure 1 Any block of the logic flow in the program may represent program steps, or interconnected logic circuits, blocks and functions, or a combination of program steps and logic circuits, blocks and functions. The software may be stored on physical media such as memory chips or memory blocks implemented within the processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs, etc. Physical media are non-transitory media.

[0134] Those skilled in the art should understand that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0135] The above embodiments merely illustrate several embodiments of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of the present invention. Therefore, the scope of the present invention shall be determined by the appended claims.

Claims

1. Single-exposure hyperspectral imaging system based on intensity-polarization joint modulation mask, characterized by: include: An excitation light source module is used to generate linearly polarized light and illuminate the object to be measured, causing the object to produce photoluminescence or Raman scattering. The excitation light source module includes a laser light source system, a filter set, and a polarizer; A compression acquisition module is used to receive the photoluminescence signal or Raman scattering signal and the specular reflection noise signal generated by the object under test, and complete the compression acquisition of the full polarization spectrum image signal. The compression acquisition module includes a first filter group, a first lens group, an intensity-polarization joint modulation mask, a second lens group, a dispersion device, and a detector array. The intensity-polarization joint modulation mask is used to perform transverse spatially correlated intensity modulation on the full polarization spectrum image signal under test, and to perform different transverse offsets on the four components of the Stokes vector representing different polarization information to complete polarization modulation. A reconstruction module is used to reconstruct the full polarization spectrum image data to be measured from the compressed observation image; The polarization noise reduction module is used to remove the specular reflection noise component and output the noise-reduced photoluminescence or Raman scattering hyperspectral image data.

2. The single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask according to claim 1, characterized in that: The laser light source generated by the laser light source system includes a collimation system, and each light source is centrally symmetrically distributed about the line connecting the center of the object to be measured and the center of the compression acquisition module lens. The angle between the optical axis of the light source and the plane of the object to be measured is maintained at a fixed specific angle so that the laser spot evenly covers the object to be measured and maximizes energy utilization efficiency.

3. The single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask according to claim 1, characterized in that: The polarizer is a wire grid polarizer, which is used to rotate or filter the polarization state of the laser light source to a specific linear polarization state, ensuring that the light source system emits or reflects only a specified linear polarization state component for subsequent noise filtering processing.

4. The single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask according to claim 1, characterized in that: The reconstruction module includes an image preprocessing algorithm, a reconstruction deep neural network and an image post-processing algorithm, which are used to process and reconstruct the collected compressed observation signal image to obtain the full polarization spectrum image data to be measured.

5. The single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask according to any one of claims 1 to 4, characterized in that: The polarization noise reduction module removes specific polarization signals, namely, the specular reflection noise component, based on an orthogonal decomposition model. The specular reflection noise component is light from the excitation light source that is not filtered out by the filter or light that is secondary illuminated by photoluminescence or Raman scattered light and secondary reflected to the detector.

6. The detection method of a single-exposure hyperspectral imaging system based on intensity-polarization combined modulation mask according to any one of claims 1 to 5, characterized in that: The following steps are involved: S1. Excitation and signal generation: The excitation light source module emits linearly polarized light to illuminate the object being measured, stimulating photoluminescence or Raman scattering signals. S2. Compressed Acquisition: The compressed acquisition module performs intensity and polarization joint modulation, dispersion spectroscopy, and detector array acquisition on the signal and noise light to obtain a compressed observation image. S3. Spectral Reconstruction: Reconstructing a fully polarized hyperspectral image from the compressed observation image using a deep neural reconstruction network based on the reconstruction module. S4. Polarization noise reduction: The polarization noise reduction module uses a polarization degree separation algorithm to remove specular reflection noise, extract depolarized light as signal light, and improve the system's signal-to-noise ratio. The resulting spectral image is then output after noise reduction.

7. The detection method according to claim 6, wherein The specific method of the excitation light source module to generate linearly polarized light is as follows: a laser light source system generates a plurality of laser light sources, which are filtered by a filter group, and then the polarization state is rotated or filtered to a specific linear polarization state by a polarizer.

8. The detection method according to claim 6, wherein The compression acquisition module completes the process of full polarization spectrum image signal compression acquisition, including: the full polarization spectrum image signal to be measured is imaged to the first image plane through the first filter group and the first lens group, the signal is modulated by the intensity polarization combined modulation mask and then imaged to the second image plane, the second image plane image is collimated and scaled by the second lens group and then enters the dispersion device for dispersion modulation, and finally the full polarization spectrum compression signal is acquired on the detector array surface.

9. The detection method according to claim 6, wherein The reconstruction module uses a deep learning algorithm to obtain a fully polarized three-dimensional spectral image from a compressed observation image, including the steps of training a deep neural network and using the trained deep neural network to reconstruct the image.

Images