Four-channel polarisation state measurement system for in-vivo corneal reflection

Abstract

This invention discloses a four-channel polarization state measurement system for live corneal reflection, specifically relating to the fields of ophthalmic polarization imaging and metasurface optics. The system includes performing near-infrared illumination on the live cornea and imaging the corneal reflected light, outputting the corneal reflected light acquired in the same exposure; performing four-state spectral analysis and spatial imaging on the corneal reflected light, enabling the imaging sensor to form a four-channel polarimetric map under the same exposure, and outputting a four-channel intensity map arranged in a fixed channel order; performing state processing on the four-channel corneal reflection intensity map formed in the same exposure, and modeling the current reflection state obtained from the processing against the pre-calibrated source state item by item; after determining that the reflection state corresponding to the current acquisition is within the applicable range of the calibrated instrument matrix, calling the corresponding instrument matrix to complete polarization inversion and result verification.

Description

Technical Field

[0001] This invention relates to the fields of ophthalmic polarization imaging and metasurface optics, and more specifically, to a four-channel polarization state measurement system for living corneal reflection. Background Technology

[0002] In the field of live corneal reflectance polarization measurement, existing technologies mainly focus on obtaining multi-channel polarization observations sufficient to support Stokes inversion within a short acquisition time. In engineering, the corneal reflected light is generally decomposed into multiple analysis channels using polarization beam splitting structures or polarization analysis metasurfaces. The system calibration is first completed with the help of a model eye, a standard reflective target, or a polarization state generator to form a corresponding instrument matrix. Then, during actual testing, the instrument matrix is ​​directly called to reconstruct the polarization of the multi-channel intensity obtained from a single acquisition. Taking the live examination in dry eye screening and ocular surface condition assessment as an example, the subject can usually only complete the collection in a short period of natural eye opening. It is not suitable to rely on multiple retakes to repeatedly correct the results, nor can a new calibration basis be established during on-site testing. At the same time, the subject must also bear the changes in reflection state caused by corneal curvature differences, tear film spreading changes, local high reflectivity compression, and slight changes in eye position. Under these conditions, a recurring problem arises: the instrument matrix obtained during the calibration phase clearly meets the reconstruction requirements, but during live detection, the local energy distribution, reflection core morphology, and spatial separation relationship in the four-channel sub-image deviate from the corresponding calibration state. If the original instrument matrix is ​​still used for direct inversion, although polarization results can still be output, there will be significant fluctuations between the two results, distortion in the interpretation of local areas, or inconsistent reconstruction results under the same ocular surface condition. The root cause is that the existing solution focuses on whether the instrument matrix obtained from calibration is valid, but does not solve the key problem of whether the current live corneal reflection is still within the applicable range of the instrument matrix. The technical problem to be solved by this application is: how to determine whether the reflection state corresponding to the current acquisition is within the applicable range of the calibrated instrument matrix during the four-channel polarization state measurement of live corneal reflection, and on this basis, call the corresponding instrument matrix to perform polarization inversion, thereby improving the stability and consistency of single live measurement results. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a four-channel polarization state measurement system for live corneal reflection. The system performs state processing on the four-channel corneal reflection intensity map formed by the same exposure, and models the current reflection state obtained by processing against the pre-calibrated source state item by item. After determining that the reflection state corresponding to the current acquisition is within the applicable range of the calibrated instrument matrix, the corresponding instrument matrix is ​​called to complete the polarization inversion and result verification, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a four-channel polarization state measurement system for live corneal reflection, comprising: The illumination and imaging module is used to perform near-infrared illumination on the living cornea and to perform imaging and acquisition of corneal reflected light, and output the corneal reflected light acquired in the same session; The polarization imaging module is used to perform four-state spectral splitting and spatial imaging on corneal reflected light, enabling the imaging sensor to form a four-channel polariton map under the same exposure and output a four-channel intensity map arranged in a fixed channel order. The state organization module is used to extract the reflection kernel position, reflection kernel width, edge unfolding direction and sub-image spacing from the four-channel intensity map, and combine them into current state groups according to pixel position or pixel block position, and output the current state table; The dynamic modeling module is used to read the pre-stored matrix table, compare the current state table with the source state group in the matrix table item by item, construct the correspondence model between the current acquisition state and the calibrated matrix state according to the reflection kernel position, reflection kernel width, edge unfolding direction and sub-image spacing, select the target instrument matrix when the correspondence is consistent, and output the target matrix table. The polarization calculation module is used to call the target instrument matrix in the target matrix table to perform Stokes inversion on the four-channel intensity map at the corresponding position, and substitute the inversion result back to the corresponding target instrument matrix to perform channel reconstruction. When the reconstructed four-channel intensity arrangement is consistent with the original four-channel intensity arrangement, the corresponding Stokes result is retained, and the effective polarization result table is output.

[0005] In a preferred embodiment, it further includes: The results output module is used to write the Stokes results from the valid polarization results table back to generate the live corneal reflection polarization results in the original position order, and to mark the positions where the target instrument matrix was not obtained or where the Stokes results were not retained as restricted positions, and output the polarization measurement results.

[0006] In a preferred embodiment, the illumination imaging module includes: The near-infrared light source is subjected to ring-shaped zonal emission control, and the beams of each zone are projected onto the living cornea in a predetermined incident direction. The corneal reflection brightness distribution in the corresponding imaging sensor is read, and the target illumination zone that makes the corneal reflection nucleus fall completely into the imaging range and the corneal reflection edge is continuously closed is selected. The target illumination state corresponding to the current acquisition is output. Under target illumination, the corneal reflected light is acquired in the same exposure, the position of the corneal reflection nucleus, the outward expansion direction of the reflection nucleus, and the closed range of the reflection edge are extracted, and the position of the corneal reflection nucleus, the outward expansion direction of the reflection nucleus, and the closed range of the reflection edge are written into the same acquisition record corresponding to the target illumination state, and the corneal reflected light acquired in the same exposure is output. For the same acquisition record, an illumination consistency check is performed. If the corneal reflection nucleus position corresponding to the current exposure is within the range of the target illumination zone, the outward expansion direction of the reflection nucleus is consistent with the target incident direction, and the closed range of the reflection edge is not broken, the corneal reflection light acquired in the same acquisition is retained as the input for subsequent polarization image separation.

[0007] In a preferred embodiment, the polarization separation module includes: The corneal reflected light is split into four analytical states, so that the reflected light corresponding to the four analytical states is projected onto the four image regions of the imaging sensor respectively. The center position, region spacing and region boundary of the four image regions are recorded in a predetermined arrangement order, and a four-channel image position table is output. Based on the four-channel image splitting position table, intensity reading and region cropping are performed on the four image splitting regions formed under the same exposure. The intensity range of the reflection nucleus, the edge transition range and the effective coverage range of each image splitting region are extracted according to the predetermined arrangement order. The intensity data of the corresponding four analysis states are combined according to the fixed channel order to output a four-channel intensity map. Perform a pixel consistency check on the four-channel intensity map. If the predetermined spacing relationship is maintained at the center of the four pixel regions, the region boundaries do not overlap, and the corresponding reflection nuclei are all within the effective coverage of each pixel region, the four-channel intensity map is retained as the input for subsequent state processing.

[0008] In a preferred embodiment, the state management module includes: Perform co-location expansion on the four-channel intensity map according to a fixed channel order, extract the center position, left and right boundary positions and outer intensity decreasing direction of each channel reflection nucleus, and solve the sub-map spacing based on the distance between the center positions of the corresponding channel reflection nucleus, and output the four-channel state component table. At the same pixel position or the same pixel block position, perform corresponding sorting on the four-channel state component table, combine the center position of the reflection kernel of the corresponding four channels into a reflection kernel position group, combine the difference between the left and right boundary positions of the corresponding four channels into a reflection kernel width group, combine the outer intensity decreasing direction of the corresponding four channels into an edge expansion direction group, and combine the center position distance of the corresponding four channels into a sub-image spacing group, and output the current state group. Write the current state group into the state table in the order of pixel position or pixel block position. When the four-channel reflection kernel position group, reflection kernel width group, edge unfolding direction group and sub-image spacing group all come from the same exposure and maintain a fixed channel correspondence, retain the corresponding current state group and output the current state table.

[0009] In a preferred embodiment, the dynamic modeling module includes: Perform corresponding expansion on each current state group in the current state table and each source state group in the matrix table according to a fixed field order. The fixed field order is: reflection kernel position group, reflection kernel width group, edge expansion direction group, and subgraph spacing group. Solve the reflection kernel position difference group, reflection kernel width difference group, edge expansion direction corresponding value group, and subgraph spacing difference group respectively. When all items in the edge expansion direction corresponding value group are in the same direction and all items in the subgraph spacing difference group have the same sign, establish candidate corresponding edges between each current state group and each source state group, and output the candidate corresponding graph. Following the order of the current state group in the current state table, perform inheritance expansion on the candidate corresponding edges in the candidate corresponding graph. Use the source state group position of the candidate corresponding edge of the previous current state group as the inheritance benchmark of the candidate corresponding edge of the next current state group. When the position of the next source state group is after the position of the previous source state group, the items of the reflection kernel position difference group have the same sign as the corresponding items of the previous candidate corresponding edge, the items of the reflection kernel width difference group have the same sign as the corresponding items of the previous candidate corresponding edge, and the items of the subgraph spacing difference group have the same sign as the corresponding items of the previous candidate corresponding edge, establish the previous and next inheritance relationship and output the candidate inheritance chain group.

[0010] In a preferred embodiment, the dynamic modeling module further includes: For each candidate connection chain group, reverse verification and unique retention are performed. The reflection kernel position difference groups corresponding to the corresponding edges of each candidate connection chain are concatenated end-to-end in field order to form a position difference string. The reflection kernel width difference groups are concatenated end-to-end in field order to form a width difference string. The edge expansion direction corresponding value groups are concatenated end-to-end in field order to form a direction corresponding string. The subgraph spacing difference groups are concatenated end-to-end in field order to form a spacing difference string. Then, a comparison is performed item by item in the following order: position difference string, width difference string, direction corresponding string, and spacing difference string. When different items appear at the same comparison position, retain the candidate succession chain with the lower absolute value corresponding to that position; when no retention result is obtained at any of the comparison positions of the position difference string, width difference string, direction correspondence string and spacing difference string, retain the first candidate succession chain with the lower code value after comparing the source state group sequence string item by item according to the sequence encoding order, delete the remaining candidate succession chains and re-execute succession expansion until each current state group corresponds to one candidate succession chain, or no new succession relationship is formed in the candidate correspondence graph, and output a stable correspondence chain; Based on the stable correspondence chain, each current state group and its corresponding source state group are written into the corresponding model, and the instrument matrix bound to each corresponding source state group is retrieved to form the target matrix table. Then, each instrument matrix is ​​used to perform inversion and back substitution on the four-channel intensity of the corresponding current state group to obtain the back-substituted four-channel intensity group. When the back-substituted four-channel intensity group corresponds to the original four-channel intensity group in a fixed channel order and the position of each channel reflection nucleus is kept in the same arrangement order, the corresponding instrument matrix is ​​retained as the target instrument matrix, and the target matrix table is output.

[0011] In a preferred embodiment, the polarization calculation module includes: For each target instrument matrix in the target matrix table, the four-channel intensity group at the corresponding position is arranged in a fixed channel order to form a quaternary solution group. The initial Stokes group is solved, and the initial Stokes group is substituted back to the corresponding target instrument matrix to generate the back-substituted four-channel intensity group. Then, the original four-channel intensity group and the back-substituted four-channel intensity group are subtracted channel by channel to form the channel residual group. The residual sign group is formed by judging channel by channel, and the initial solution table is output. Perform residual error correction expansion on the initial solution table. When there is a channel symbol in the residual symbol group that is opposite to the other three channel symbols, replace the corresponding intensity in the original four-channel intensity group with the back-substitution intensity of that channel in the back-substitution four-channel intensity group to form a corrected four-channel intensity group. Then, use the corrected four-channel intensity group and the original target instrument matrix to solve the corrected Stokes group again. Finally, use the corrected Stokes group to back-substitute and generate the corrected back-substitution four-channel intensity group and the corrected residual symbol group, and output the corrected solution table.

[0012] In a preferred embodiment, the polarization calculation module further includes: Perform a continuity check on the correction solution table according to the arrangement order of the corresponding positions in the four-channel intensity map. When the correction residual sign group of the previous position and the next position are consistent, the correction back substitution four-channel intensity group maintains a fixed channel order, and the positive and negative directions of each component in the correction Stokes group are consistent, a solution continuity relationship is established between the previous position and the next position, and the solution continuity chain group is output. Perform reverse verification and effective retention on the solution chain group. Concatenate the correction residual symbol groups in each solution chain in order of position to form a residual symbol string. Concatenate the four-channel intensity groups of each correction back-substitution in a fixed channel order to form a back-substitution intensity string. When the residual symbol string is consistent, the back-substitution intensity string maintains a fixed channel order, and the component direction obtained after the corresponding correction Stokes group is back-substituted and re-solved remains unchanged, retain the corresponding correction Stokes group and write it into the effective polarization result table. Output the effective polarization result table.

[0013] In a preferred embodiment, the result output module includes: Read the Stokes results corresponding to each position in the effective polarization result table according to the position order, and write back each Stokes result according to the original pixel position or the original pixel block position. Write the polarization result value at the position where the Stokes result is obtained, and write the restricted mark at the position where the target instrument matrix is ​​not obtained and the position where the Stokes result is not retained. Output the position result table. Perform adjacent position consolidation on the position result table, combine adjacent positions that are both written with polarization result values ​​into valid position segments, combine adjacent positions that are both written with restricted marks into restricted position segments, and mark the positions located at the intersection of valid position segments and restricted position segments as boundary positions, and output the result segment table; Based on the result segmentation table, generate the live corneal reflection polarization results in the original position order. Write the corresponding Stokes result sequence in the effective position segment and the corresponding restricted position sequence in the restricted position segment. Write the preceding and following segmentation relationship corresponding to the boundary position into the polarization measurement results and output the polarization measurement results.

[0014] The technical effects and advantages of this invention are as follows: 1. This scheme, by matching the current state group with the source state group item by item and only retrieving the target instrument matrix after forming a stable correspondence chain, can first determine whether the current acquisition reflection state is within the applicable range of the calibrated instrument matrix before performing polarization inversion, thereby relatively improving the stability and consistency of single live measurement results. 2. By performing target illumination zone selection, same-exposure acquisition, and illumination consistency verification on corneal reflected light, the input entering the four-analysis-state beam splitting link can maintain consistent illumination source and reflection kernel shape, thereby relatively suppressing the interference of illumination state shift on subsequent polarization calculation; 3. By generating a four-channel polariton map under the same exposure and verifying the consistency of the center position, region boundary and effective coverage of the image separation area, the spatial correspondence of the four-channel observation can be maintained, thereby relatively reducing the channel mismatch caused by image separation overlap and reflection nucleus overstepping. 4. Organizing the four-channel intensity maps into groups of reflection kernel position, reflection kernel width, edge expansion direction, and sub-map spacing can unify the scattered channel-level observations into comparable state objects, thereby providing a unified input for subsequent matrix matching and relatively reducing the corresponding deviation caused by the mixing of position scales; 5. Performing back substitution, residual sign determination, single-channel replacement error correction, and re-resolve on the initial Stokes group can identify and correct the solution anomalies caused by local channel mismatch, thereby relatively improving the usability of polarization results and mitigating the amplification of the overall inversion results by single-channel bias. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the system module structure of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Refer to the instruction manual appendix Figure 1 The four-channel polarization state measurement system for live corneal reflection of the present invention includes: The illumination and imaging module is used to perform near-infrared illumination on the living cornea and to perform imaging and acquisition of corneal reflected light, and output the corneal reflected light acquired in the same session; In this embodiment, the illumination imaging module first determines the target illumination state that matches the current live corneal reflection, then completes the same exposure acquisition under the target illumination state, and performs illumination consistency verification on the acquisition results to ensure that subsequent polarization image splitting inputs correspond to the same exposure, the same illumination state, and the same reflection nucleus source; this implementation process includes the following steps: First, the annular partition table of the near-infrared light source, the predetermined incident direction table corresponding to each partition, and the imaging range of the imaging sensor are read. Each annular partition is illuminated sequentially according to its partition number, and the corneal reflectance brightness map output by the imaging sensor is read while each partition is illuminated. Connectivity extraction is performed on each corneal reflectance brightness map. Regions with continuous high brightness and closed boundaries are recorded as candidate reflectance kernel regions. The center position and outer closed boundary of each candidate reflectance kernel region are determined, and it is determined whether the entire candidate reflectance kernel region is within the imaging range. When all candidate reflectance kernel regions are within the imaging range and the outer closed boundaries are connected end-to-end, the corresponding partition is recorded as a candidate illumination partition. The first candidate illumination partition is retained as the target illumination partition according to its partition number. The partition number, corresponding predetermined incident direction code, and current acquisition identifier of the target illumination partition are combined to form the target illumination state and output. If no candidate illumination partition is formed in any of the annular partitions, an invalid acquisition record is written, and subsequent exposure acquisition is stopped. Under target illumination conditions, the target illumination zone is kept continuously lit, and a formal exposure acquisition is performed to obtain the corneal reflected light image corresponding to the same exposure. The position of the reflection nucleus, the outward expansion direction of the reflection nucleus, and the closed range of the reflection edge are extracted from this corneal reflected light image. The position of the reflection nucleus is formed by weighting the coordinates of each brightness pixel within the reflection nucleus region to obtain the center coordinates. The outward expansion direction of the reflection nucleus is formed by statistically calculating the continuous length of decreasing brightness along a predetermined coordinate axis starting from the reflection nucleus position and encoding the corresponding direction. The closed range of the reflection edge is formed by extracting the outermost closed boundary of the reflection nucleus region. Subsequently, the position of the reflection nucleus, the outward expansion direction of the reflection nucleus, and the closed range of the reflection edge are written into the acquisition record corresponding to the zone number, predetermined incident direction code, current acquisition identifier, and formal exposure identifier in the target illumination conditions, and the corneal reflected light acquired in the same exposure is output. If no closed reflection nucleus region is extracted in the formal exposure image, a reflection nucleus missing marker is written into the acquisition record, and the corneal reflected light from that exposure is recorded as a restricted input. Read the same acquisition record and perform illumination consistency verification on the reflection kernel position, reflection kernel expansion direction, and reflection edge closure range corresponding to the current exposure. First, solve the target illumination zone range according to the target illumination zone boundary and the predetermined incident direction code in the target illumination state, and determine whether the reflection kernel position is within the corresponding range of the target illumination zone. Then, perform a correspondence comparison between the reflection kernel expansion direction code and the predetermined incident direction code, and perform a head-to-tail connectivity verification on the boundary point sequence in the reflection edge closure range. Only when the reflection kernel position is within the corresponding range of the target illumination zone, the reflection kernel expansion direction corresponds to the predetermined incident direction, and there are no breaks in the boundary point sequence in the reflection edge closure range, is the corneal reflected light acquired in the same acquisition retained as the subsequent polarization image input, and the retention mark is written into the same acquisition record. If any of the above conditions are not met, the corneal reflected light is written into the restricted input mark, and the corresponding reason field is written into the same acquisition record. Through the above processing, the illumination imaging module completes the target illumination state selection, same-exposure acquisition and illumination consistency verification, so that the corneal reflected light sent to the subsequent polarization imaging module simultaneously meets the requirements of consistent illumination source, consistent exposure source and consistent reflection nucleus shape, and writes the target illumination state, reflection nucleus position, reflection nucleus outward expansion direction and reflection edge closure range into the same-exposure record for direct reading in subsequent steps; In practical applications: When a subject undergoes a live corneal reflectance measurement, the illumination imaging module sequentially illuminates the annular zones and selects the first zone that satisfies the condition that the reflective nucleus is completely within the imaging range and the reflective edge is closed as the target illumination zone. The formal exposure is then completed while the zone remains illuminated. If the position of the reflective nucleus corresponding to the formal exposure is still within the range of the target illumination zone, and the outward expansion direction of the reflective nucleus is consistent with the corresponding incident direction, and there is no break in the closed range of the reflective edge, then the corneal reflectance light is retained as the polarization image input. If any verification condition is not met, then the corneal reflectance light is marked as a restricted input.

[0018] The polarization imaging module is used to perform four-state spectral splitting and spatial imaging on corneal reflected light, enabling the imaging sensor to form a four-channel polariton map under the same exposure and output a four-channel intensity map arranged in a fixed channel order. In this embodiment, the polarization imaging module is used to synchronously separate the corneal reflected light obtained from the same exposure into four imaging regions of the imaging sensor according to four analytical states. Then, based on a fixed channel order, a four-channel intensity map that can be directly processed is formed, and the imaging consistency check eliminates imaging position mismatch, region overlap, and reflection kernel out-of-bounds input. This implementation process includes the following steps: First, the corneal reflected light acquired in the same batch is input into a four-analysis-state beam splitter. Simultaneously, the four-analysis-state and channel number correspondence table, the image splitting region pre-arrangement table, and the image plane coordinate table of the imaging sensor are retrieved. Four-analysis-state beam splitting is then performed on the corneal reflected light, so that the reflected light corresponding to each of the four analysis states is projected onto the four image splitting regions of the imaging sensor. The image splitting region pre-arrangement table includes at least the image splitting region number, fixed channel order, initial value of the region center position, and initial value of the region boundary. The four-analysis-state and channel number correspondence table is used to lock the one-to-one correspondence between the analysis state number and the channel number. Subsequently, the… The luminance connectivity regions of the four image regions formed under the same exposure are extracted. The center position of each image region is determined by the center coordinates of the luminance connectivity regions in each image region. The region spacing is determined by the difference between the center coordinates of adjacent image regions. The region boundary is determined by the outermost luminance coverage boundary of each image region. The image region number, fixed channel order, image region center position, region spacing, and region boundary are written into the four-channel image position table in a predetermined arrangement order. If the number of image regions formed under the same exposure is not four, the current exposure identifier is written into the image missing record and subsequent intensity reading is stopped. After obtaining the four-channel image segmentation location table, intensity reading and region cropping are performed on the four image segmentation regions formed under the same exposure according to the four-channel image segmentation location table to obtain the reflection kernel intensity range, edge transition range, and effective region coverage range consistent with the subsequent state processing. Specifically, firstly, region cropping is performed on the same exposure image according to the region boundaries in the four-channel image segmentation location table to obtain the local intensity matrix of the four image segmentation regions. Then, continuous bright regions are extracted from each local intensity matrix. The set of pixel intensity within the continuous bright region is recorded as the reflection kernel intensity range, the set of pixel intensity outside the continuous bright region along the brightness decreasing direction to the background intensity range is recorded as the edge transition range, and the remaining region within the region boundary after deducting the edge transition range is recorded as the effective region coverage range. The background intensity range is obtained by statistically analyzing the non-reflective areas outside the image segmentation region in the same exposure image. Continuous bright areas are determined by areas whose brightness is greater than the average background intensity and remain connected. Subsequently, the local intensity matrices corresponding to the four analytical states are combined into a four-channel intensity map in a fixed channel order, and the reflection kernel intensity range, edge transition range, and effective coverage range of each channel are written into the association field of the four-channel intensity map. If no continuous bright areas are extracted in a certain image segmentation region, a reflection kernel missing marker is written into the corresponding channel field, and the channel is recorded as a restricted channel. Subsequently, a consistency check was performed on the four-channel intensity map and the four-channel image splitting position table to confirm that the four image splitting regions formed after four analytical states still maintain a unified image splitting structure and correspond to the same corneal reflection source. Specifically, the center positions of the four image splitting regions were first compared according to a predetermined arrangement order to determine whether the region spacing sequence formed by the difference in the center positions of adjacent image splitting regions maintained the same arrangement order as the region spacing sequence in the four-channel image splitting position table. Then, boundary intersection calculations were performed on the region boundaries of each image splitting region, and region overlap was recorded when there was a common pixel position at the boundary of any two image splitting regions. After that, the reflection nuclei corresponding to each channel were read. The intensity range and effective coverage area are determined to ensure that the center and boundary positions of the reflection kernels for each channel are all within the effective coverage area of ​​their respective regions. Only when the region spacing sequence maintains the same arrangement order, there are no common pixel positions between region boundaries, and the reflection kernels for all four channels are within their respective effective coverage areas, is the four-channel intensity map retained as input for subsequent state processing, and a valid image segmentation flag is written. If any condition is not met, the corresponding reason field in the image segmentation mismatch flag, region overlap flag, or reflection kernel boundary crossing flag is written, and the four-channel intensity map is recorded as a restricted input and is no longer sent to the subsequent state processing module. Through the above processing, the polarization imaging module completes four-state beam splitting, four-image region positioning, four-channel intensity map generation, and image splitting consistency verification. This ensures that the four-channel intensity map entering the subsequent state processing module simultaneously meets the requirements of clear fixed channel order, clear image region positional relationship, non-overlapping region boundaries, and reflection nuclei within the effective coverage range. It also establishes a one-to-one correspondence between the image region center position, region spacing, and region boundary in the four-channel image splitting position table and the reflection nuclei intensity range, edge transition range, and effective region coverage range in the four-channel intensity map, allowing for direct reading in the subsequent state processing. In practical applications: After the illumination imaging module outputs the corneal reflected light acquired in the same session, the polarization imaging module synchronously projects the corneal reflected light to four imaging regions according to the correspondence table of four analytical states and channel numbers. If all four imaging regions form independent luminance connected regions, the center position, region spacing, and region boundary of the four imaging regions are solved, and a four-channel imaging position table is generated. Subsequently, the reflection kernel intensity range, edge transition range, and effective coverage range of the region are extracted from the four imaging regions respectively, and a four-channel intensity map is generated according to a fixed channel order. If the verification finds that the region boundaries of the second and third imaging regions have common pixel positions, or that the center position of the reflection kernel of a certain channel falls outside the effective coverage range of the corresponding region, the four-channel intensity map is marked as a restricted input and does not enter the state sorting module. If the four-channel imaging relationship and the reflection kernel coverage relationship are both valid, the four-channel intensity map is retained as the state sorting input.

[0019] The state organization module is used to extract the reflection kernel position, reflection kernel width, edge unfolding direction and sub-image spacing from the four-channel intensity map, and combine them into current state groups according to pixel position or pixel block position, and output the current state table; In this embodiment, the state organization module is used to organize the scattered channel-level reflection information in the four-channel intensity map into a unified state object that can directly participate in matrix matching. Its main processing steps are: first, to solve for the basic position and direction of each channel's reflection kernel under a fixed channel order; then, to complete the four-channel co-position combination at the same pixel position or the same pixel block position; and finally, to write it into the current state table in the original position order to ensure that the reflection kernel position group, reflection kernel width group, edge unfolding direction group, and sub-image spacing group read by subsequent dynamic modeling have the same exposure source, the same position reference, and a fixed field order. This implementation process includes the following steps: First, to form composable channel-level state components, the four-channel intensity map is expanded in a fixed channel order. Specifically, the channel number, corresponding image region boundary, reflection kernel intensity range, and edge transition range in the four-channel intensity map are read first. The pixel intensity within the reflection kernel intensity range of each channel is used as the central solution input. The coordinates of each pixel within the reflection kernel intensity range are weighted and summed according to the pixel intensity, and then divided by the sum of the corresponding pixel intensities to obtain the center position of the reflection kernel of that channel. Then, based on the center position of the reflection kernel, the boundary pixels of the reflection kernel intensity range are searched to the left and right sides along a predetermined horizontal coordinate direction. The position of the first boundary pixel on the left is recorded as the left boundary position, and the position of the first boundary pixel on the right is recorded as the right boundary position. Then, using the pixel intensity sequence of the edge transition range outside the left and right boundary positions as input, the intensity difference between adjacent pixels is calculated in the increasing direction of pixel position. The side with consecutive negative values ​​is recorded as the outer intensity decreasing direction. If consecutive negative value sequences are formed on both the left and right sides, the side with the longer consecutive negative value sequence is recorded as the outer intensity decreasing direction. After that, the center position of each channel reflection kernel is read in a fixed channel order, and the coordinate difference is calculated for the center positions of adjacent channel reflection kernels to obtain the center position distance sequence as the sub-image spacing. Finally, the channel number, reflection kernel center position, left boundary position, right boundary position, outer intensity decreasing direction, and corresponding sub-image spacing are written into the four-channel state component table in a fixed channel order. If the reflection kernel center position is not solved in any channel, or there is a missing position in the left and right boundary positions, a component missing mark is written in the corresponding channel field, and the corresponding position processing at that position is stopped. Next, to form comparable current position state objects, the four-channel state component table is processed for corresponding positions at the same pixel position or the same pixel block position. Here, the pixel position is the coordinate position of a single pixel of the imaging sensor, and the pixel block position is the center position of a block composed of multiple adjacent pixels divided according to a preset configuration. The preset configuration is determined by the sensor resolution and processing load and remains unchanged within the same acquisition. When pixel position processing is used, the four-channel state components corresponding to the same pixel position are directly read. When pixel block position processing is used, the four-channel state components corresponding to each pixel in the same pixel block are first extracted according to the fixed block division rules. Then, the coordinate average of the center position of the reflection nucleus in the same channel within the block, the position average of the left boundary position and the right boundary position are taken respectively, the distance average of the sub-image spacing is taken, and the direction encoding with the same frequency of occurrence is taken for the outer intensity decreasing direction to form the four-channel state components corresponding to the pixel block position. Then, the center positions of the corresponding four channels' reflection kernels are arranged in a fixed channel order to form a reflection kernel position group. The difference between the right and left boundary positions of the corresponding four channels is arranged in a fixed channel order to form a reflection kernel width group. The outer intensity decreasing direction codes of the corresponding four channels are arranged in a fixed channel order to form an edge spreading direction group. The center position distances of the corresponding four channels are arranged in a fixed channel order to form a sub-image spacing group. Then, the reflection kernel position group, reflection kernel width group, edge spreading direction group, and sub-image spacing group are combined with the current position identifier to form the current state group for output. If the four channel state components are incomplete at the same pixel position or the same pixel block position, or if the direction code within the block does not have a unique result, then the position is written with a missing position marker, and the current state group is not generated. To form a unified input for subsequent dynamic modeling, the current state group is written into the state table in the order of pixel position or pixel block position. Specifically, the current position identifiers are first read in their original position order. When using pixel position organization, the original position order is the fixed order of sensor coordinates in row priority; when using pixel block position organization, it is the fixed order of pixel block center coordinates in row priority. Then, the reflection kernel position group, reflection kernel width group, edge unfolding direction group, and sub-image spacing group corresponding to each current position identifier are written into the current state table in sequence, and the same exposure identifier, fixed channel order identifier, and position are written simultaneously. The scale identifier, where the location scale identifier is used to distinguish between pixel location and pixel block location; then, a source check is performed on each current state group to determine whether its four field groups all come from the same exposure, whether they are all formed in the same fixed channel order, and whether they correspond one-to-one with the current location identifier. The corresponding current state group is retained only when all three checks are true; if any one check is false, the current location identifier is written to the state restricted flag, and the field groups that have been written but failed the source check are deleted; finally, the retained current state groups constitute the current state table, which can be directly read by the subsequent dynamic modeling module; Through the above processing, the state sorting module converts the channel-level brightness information in the four-channel intensity map into a current state table with ordered positions, fixed fields, and consistent sources. This allows the subsequent dynamic modeling module to directly read the reflection kernel position group, reflection kernel width group, edge expansion direction group, and sub-image spacing group to perform corresponding expansion and continuation modeling without having to backtrack to the original four-channel intensity map again. At the same time, it avoids state mismatch caused by mixing different exposure sources, different position scales, and different channel orders. In practical applications: When the four-channel intensity map enters the state processing module, the system first solves the center position of the reflection kernel, the left and right boundary positions, the outer intensity decrease direction, and the distance between the center positions of adjacent channels for each channel in a fixed channel order from the first to the fourth channel. Then, at the same pixel block position, the four-channel state components are combined to form the reflection kernel position group, reflection kernel width group, edge unfolding direction group, and sub-image spacing group corresponding to that pixel block position. If the four-channel state components corresponding to the pixel block position are complete and all come from the same exposure, the current state group corresponding to the pixel block position is written into the current state table. If the third channel does not extract the right boundary position at the pixel block position, the pixel block position is marked with a missing position and no corresponding current state group is generated.

[0020] The dynamic modeling module is used to read the pre-stored matrix table, compare the current state table with the source state group in the matrix table item by item, construct the correspondence model between the current acquisition state and the calibrated matrix state according to the reflection kernel position, reflection kernel width, edge unfolding direction and sub-image spacing, select the target instrument matrix when the correspondence is consistent, and output the target matrix table. In this embodiment, the dynamic modeling module is used to establish a verifiable, referential, and uniquely retained correspondence between the current state group in the current state table and the source state group in the matrix table. Its main processing flow is as follows: First, expand the field correspondence between the current state group and the source state group under a fixed field order to form a candidate correspondence diagram; then, expand the candidate inheritance relationship along the original arrangement order of the current state table to form a candidate inheritance chain group; subsequently, perform reverse verification and unique retention on the candidate inheritance chain group to converge and obtain a stable correspondence chain; finally, retrieve the corresponding instrument matrix based on the stable correspondence chain and perform back-substitution verification to form the target matrix table, ensuring that the instrument matrix entering the polarization calculation module corresponds to both the current position state and the calibrated source state. This implementation process includes the following steps: First, to form a candidate relationship between the expandable current position and the source state, each current state group in the current state table and each source state group in the matrix table are expanded according to a fixed field order. The fixed field order is: reflection kernel position group, reflection kernel width group, edge expansion direction group, and sub-image spacing group. The matrix table includes at least the source state group, the source state group sequence, the bound instrument matrix, and the source position identifier. The field composition of the source state group is consistent with that of the current state group. Then, the corresponding results are calculated field by field for each current state group and each source state group. The reflection kernel position group is formed by subtracting coordinates according to the same channel sequence to form a reflection kernel position difference group. The reflection kernel width group is formed by subtracting width values ​​according to the same channel sequence to form a reflection kernel width difference group. The edge expansion direction group is formed by comparing direction codes according to the same channel sequence. When the direction codes are the same or belong to the same coordinate axis in the same direction, they are recorded as corresponding values ​​in the same direction. When the direction codes are opposite or the coordinate axes are different, they are recorded as corresponding values ​​in opposite directions to form an edge expansion direction value group. The sub-image spacing group is formed by subtracting distance values ​​according to the same channel sequence to form a sub-image spacing difference group. Next, the edge expansion direction corresponding value group and the subgraph spacing difference group are checked. When all items in the edge expansion direction corresponding value group are corresponding values ​​in the same direction, and all non-zero items in the subgraph spacing difference group have the same sign, a candidate corresponding edge is established between the current state group and the source state group. The current state group identifier, source state group identifier, source state group sequence, reflection kernel position difference group, reflection kernel width difference group, edge expansion direction corresponding value group, and subgraph spacing difference group are written into the candidate corresponding graph. If any field group is missing, or there are opposite corresponding values ​​in the edge expansion direction corresponding value group, or positive and negative values ​​appear simultaneously in the subgraph spacing difference group, no candidate corresponding edge is established, and the pair is written into the unestablished edge record. Next, in order to form a continuous state succession relationship across positions, the candidate corresponding edges in the candidate corresponding graph are expanded according to the order of the current state group in the current state table. Specifically, two adjacent current state groups are read first according to their original position order in the current state table. The candidate corresponding edge corresponding to the previous current state group is recorded as the previous candidate corresponding edge, and the candidate corresponding edge corresponding to the next current state group is recorded as the next candidate corresponding edge. Then, the source state group sequence of the previous candidate corresponding edge is used as the succession benchmark for the next candidate corresponding edge, and four types of succession checks are performed on the next candidate corresponding edge: sequence, position difference, width difference, and spacing difference. The sequence check requires that the sequence of the source state group of the next candidate edge is greater than that of the previous candidate edge; the position difference check requires that each item in the position difference group of the reflection kernel of the next candidate edge has the same sign as the corresponding item of the previous candidate edge or is simultaneously zero; the width difference check requires that each item in the width difference group of the reflection kernel of the next candidate edge has the same sign as the corresponding item of the previous candidate edge or is simultaneously zero; and the spacing difference check requires that each item in the spacing difference group of the subgraph of the next candidate edge has the same sign as the corresponding item of the previous candidate edge or is simultaneously zero. Only when all four types of continuity checks are met simultaneously, a continuity relationship is established between the previous and next candidate edges, and the continuity start point, continuity end point, and corresponding source state group sequence are concatenated to form a candidate continuity chain. Then, the chain continues to expand backward in the original position order until the end of the current state table, and the candidate continuity chain group is output. If a current state group does not have a candidate edge, the current state group is written into the continuity interruption record, and the expansion of the current candidate continuity chain is terminated at that position. Subsequently, in order to converge a unique and usable acceptance result from the candidate acceptance chain group, a reverse verification and unique retention are performed on the candidate acceptance chain group. Specifically, each candidate corresponding edge in each candidate acceptance chain is first concatenated end to end according to the field order to form a position difference string, a width difference string, a direction correspondence string, a spacing difference string, and a source state group sequence string. Among them, the position difference string is formed by concatenating the reflection kernel position difference groups of each candidate corresponding edge end to end according to the order within the chain and the fixed channel order within the group. The width difference string, the direction correspondence string, and the spacing difference string are formed in the same way. The source state group sequence string is formed by concatenating the sequence of each source state group within the chain end to end according to the acceptance order. Next, candidate succession chains with the same starting current state group and the same ending current state group are compared item by item. The comparison order is as follows: position difference string, width difference string, direction correspondence string, and spacing difference string. During the comparison, the strings are read item by item in the order of the fields. When different values ​​appear at the same comparison position, the candidate succession chain with the smaller absolute value at that position is retained. If no retention result is found at all comparison positions of the above four types of strings, the source state group sequence string is compared item by item in the sequence encoding order. The candidate succession chain with the smaller sequence encoding value is retained at the first different sequence position. Then, the candidate succession chains that are not retained are deleted, and the succession expansion and reverse verification are re-executed for the current state groups affected by the deletion. When each current state group corresponds to only one candidate succession chain, or when no new succession relationship is formed in the candidate correspondence graph, the re-expansion is stopped, and the stable correspondence chain is output. If a current state group no longer corresponds to any candidate succession chain after deletion, the current state group is written to the record of no stable correspondence and is not included in the subsequent correspondence model writing. Finally, to form a target matrix table that can be directly called, each current state group and its corresponding source state group are written into the corresponding model according to the stable correspondence chain, and the instrument matrix bound to each corresponding source state group is retrieved for back-substitution verification. Specifically, first, according to the sequential order in the stable correspondence chain, the current state group identifier, source state group identifier, source state group sequence, current position identifier, and bound instrument matrix identifier are written into the corresponding model; then, the instrument matrix bound to the corresponding source state group is retrieved item by item according to the corresponding model to form the target matrix table; then, for each instrument matrix in the target matrix table, the four-channel intensity group from which the corresponding current state group originates is read, the four-channel intensity group is inverted to obtain the Stokes intermediate result, and then the Stokes intermediate result is substituted back into the instrument matrix to generate the back-substituted four-channel intensity group; Then, the four-channel intensity groups are matched with the original four-channel intensity groups one by one according to the fixed channel order, and the reflection nucleus position group in the corresponding current state group is read. It is checked whether the arrangement order of the reflection nucleus positions of each channel in the original four-channel intensity group is consistent with the arrangement order in the four-channel intensity groups of the replacement. Only when the fixed channel order matches item by item and the reflection nucleus positions of each channel maintain the same arrangement order, the instrument matrix is ​​retained as the target instrument matrix and written into the target matrix table. If the channel order is misaligned or the arrangement order of the reflection nucleus positions changes, the corresponding record of the instrument matrix is ​​deleted and the current position identifier is written into the matrix replacement mismatch record. Through the above processing, the dynamic modeling module establishes a complete convergence link between the current position state in the current state table and the calibrated source state in the matrix table, from candidate edge construction, acceptance expansion, reverse verification, unique retention to back-substitution verification. This ensures that each target instrument matrix in the target matrix table simultaneously meets the four requirements of field correspondence, position acceptance, unique retention, and back-substitution verification. This avoids the mismatch problem caused by directly calling the instrument matrix based on a single field similarity and provides the subsequent polarization solution module with matrix input that is ordered in position, clear in source, and verified. In practical applications: After the state processing module outputs the current state table, the dynamic modeling module first expands each current state group with the source state group in the matrix table item by item to form a candidate correspondence graph; then, it performs a connection expansion on the candidate correspondence edges along the original position order of the current state table. If the position of a later source state group is not greater than the position of the previous source state group, or if the corresponding reflection nucleus position difference group, reflection nucleus width difference group, or subgraph spacing difference group has an item with a different sign from the previous candidate correspondence edge, then the candidate correspondence edge does not participate in the connection; subsequently, it performs item-by-item comparison of the position difference string, width difference string, direction correspondence string, spacing difference string, and source state group position string on the formed candidate connection chain group, retaining the unique candidate connection chain and deleting the rest of the candidate connection chains; finally, it retrieves the bound instrument matrix corresponding to the stable correspondence chain to perform inversion and back substitution. If the back substitution of the four-channel intensity group still corresponds to the original four-channel intensity group in a fixed channel order, and the arrangement order of the reflection nucleus positions of each channel remains unchanged, then the instrument matrix is ​​written into the target matrix table for the polarization solution module to read directly.

[0021] Furthermore, it should be noted that during the matrix table generation stage, the standard polarization state generator sequentially outputs four linearly independent standard polarization states. Each standard polarization state contains a total light intensity component, a horizontal and vertical linear polarization component, a ±45 degree linear polarization component, and a circular polarization component. After each standard polarization state enters the polarization imaging module, the polarization imaging module reads the intensity of the first channel, the second channel, the third channel, and the fourth channel according to a fixed channel order, forming a four-channel calibration intensity record. The instrument matrix is ​​obtained by combining the four standard polarization states and their corresponding four-channel calibration intensity records, and is bound to the source state group, the source state group sequence, the source position identifier, and the fixed channel order identifier and written into the matrix table. If the four standard polarization states do not meet the linear independence condition, or if any standard polarization state's corresponding four-channel calibration intensity record has missing channels, overlapping channels, or a reflection nucleus exceeding the boundary, then the calibration record for this round is deleted, and the standard polarization state input and four-channel calibration intensity acquisition are re-executed. Before writing the instrument matrix into the matrix table, pixel-level and channel-level corrections are performed on the four-channel calibration intensity records. Pixel-level corrections include dark field subtraction, flat field normalization, and pixel position offset correction. Dark field subtraction uses the dark field image acquired when the near-infrared light source is turned off as the subtraction reference. Flat field normalization uses the flat field image formed by the uniformly reflective target as the gain reference. Pixel position offset correction uses the row and column offsets of the center positions of the four image regions relative to the calibration center position as input to map the four-channel intensity to the same image plane coordinate reference. Channel-level corrections include channel gain correction and channel crosstalk correction. The positive sum channel sequence calibration includes channel gain calibration, which generates the channel gain coefficient based on the average value of each channel's flat field; channel crosstalk calibration, which forms the channel crosstalk subtraction relationship based on the responses of the remaining channels when a single analytical state is input; and channel sequence calibration, which is based on the correspondence table between the four analytical states and channel numbers, and writes the calibrated intensities of the first to fourth channels into the four-channel calibration intensity record in a fixed channel order. After completing pixel-level and channel-level calibration, the instrument matrix, source state group, calibration temperature, calibration image splitting region center position, calibration channel gain coefficient, and calibration channel crosstalk relationship are then written into the matrix table. Before and during in vivo corneal reflectance measurement, the system reads the current temperature, current image splitting region center position, current channel gain coefficient, and current channel crosstalk relationship, and compares them with the calibration temperature, calibration image splitting region center position, calibration channel gain coefficient, and calibration channel crosstalk relationship bound to the matrix table. When the temperature difference reaches the temperature drift trigger condition, the offset of any image splitting region center position reaches the assembly drift trigger condition, the gain difference of any channel reaches the channel drift trigger condition, or the crosstalk difference of any channel reaches the crosstalk drift trigger condition, a recalibration trigger flag is generated. After the marker is issued, the system pauses the original instrument matrix and re-executes standard polarization state input, pixel-level correction, channel-level correction, and instrument matrix generation. It then performs post-calibration error calculation on the recalibrated instrument matrix. The post-calibration error is obtained by statistically analyzing the residuals between the back-substituted four-channel intensities corresponding to the four standard polarization states and the calibrated four-channel intensities after correction. The recalibrated instrument matrix is ​​written into the matrix table and replaces the original instrument matrix only if the post-calibration error does not reach the upper limit of the calibration error. If the post-calibration error reaches the upper limit of the calibration error, a recalibration failure record is written, and the affected current position is marked in the restricted position.

[0022] The polarization solution module is used to call the target instrument matrix in the target matrix table to perform Stokes inversion on the four-channel intensity map at the corresponding position, and substitute the inversion result back to the corresponding target instrument matrix to perform channel reconstruction. When the reconstructed four-channel intensity arrangement is consistent with the original four-channel intensity arrangement, the corresponding Stokes result is retained, and the effective polarization result table is output. In this embodiment, the polarization calculation module is used to perform inversion, back substitution, error correction, continuation, and verification on the four-channel intensity groups at the corresponding positions, based on the target instrument matrix already locked in the target matrix table. The output only contains a table of valid polarization results that have passed the consistency check. The main processing flow is as follows: first, the target instrument matrix and the original four-channel intensity groups form the initial calculation results; then, the back substitution residual is used to identify single-channel mismatch and complete the correction calculation; subsequently, the calculation continuation relationship is established along the original position sequence; finally, the valid corrected Stokes groups are retained through reverse verification and re-calculation. This implementation process includes the following steps: First, to obtain verifiable initial polarization results, a quaternary solution group is constructed by combining each target instrument matrix in the target matrix table with the corresponding four-channel intensity group in a fixed channel order. This quaternary solution group includes at least a position identifier, a target instrument matrix identifier, the first channel intensity, the second channel intensity, the third channel intensity, and the fourth channel intensity, with the fixed channel order consistent with that in the four-channel intensity map. Then, matrix inversion is performed on the corresponding four-channel intensity group using the target instrument matrix to obtain the initial Stokes group, which includes at least the Stokes component values ​​and their corresponding position identifiers. Finally, the initial Stokes group is substituted back into the corresponding target instrument matrix to generate the substituted four-channel intensity group. Next, the original four-channel intensity group and the back-substituted four-channel intensity group are subtracted one by one in a fixed channel order to form a channel residual group. The residual value of each channel is obtained by subtracting the back-substituted channel intensity from the original channel intensity. Then, the channel residual group is determined channel by channel to form a residual sign group, where a residual value greater than zero is recorded as positive, a residual value less than zero is recorded as negative, and a residual value equal to zero is recorded as zero. Finally, the position identifier, target instrument matrix identifier, initial Stokes group, back-substituted four-channel intensity group, channel residual group, and residual sign group are written into the initial solution table. If a target instrument matrix is ​​missing or the four-channel intensity group is incomplete at a certain position, the position is written into the solution missing record and will not proceed to the subsequent residual correction expansion. Next, to correct the polarization calculation deviation caused by single-channel mismatch, residual error correction expansion is performed on the initial calculation table. Specifically, the residual sign groups are read position by position, and the occurrence frequency of positive, negative, and zero signs is counted. When there is one and only one channel sign that is opposite to the other three channel signs, the channel is recorded as the channel to be corrected. Here, opposite means that one channel is positive and the other three channels are negative, or one channel is negative and the other three channels are positive. The residual sign group containing zero signs is not included in the determination of the channel to be corrected. Then, the intensity of the corresponding channel in the original four-channel intensity group is replaced by the intensity of the channel to be corrected in the back-substitution four-channel intensity group to form the corrected four-channel intensity group. The intensity of the other three channels remains unchanged in the original four-channel intensity group. Next, matrix inversion is performed again using the calibrated four-channel intensity group and the original target instrument matrix to solve for the calibrated Stokes group. The calibrated Stokes group is then substituted back into the original target instrument matrix to generate the calibrated back-substituted four-channel intensity group. Then, the calibrated four-channel intensity group is subtracted from the calibrated back-substituted four-channel intensity group channel by channel to form the calibrated channel residual group. The calibrated residual sign group is then formed according to the same sign determination rules as the initial solution table. Finally, the position identifier, target instrument matrix identifier, calibrated four-channel intensity group, calibrated Stokes group, calibrated back-substituted four-channel intensity group, and calibrated residual sign group are written into the calibration solution table. If the residual sign group at a certain position does not meet the single-channel opposite condition, the original initial Stokes group is retained without replacement, and the initial Stokes group, the original four-channel intensity group, and the residual sign group at that position are directly converted into the corresponding records in the calibration solution table. Subsequently, to identify the consecutively valid solution results along the original position sequence, a succession check was performed on the correction solution table according to the arrangement order of the corresponding positions in the four-channel intensity map. Specifically, the correction solution records of the previous and next positions were first read in the original position sequence, and the correction residual sign group, the correction back-substitution four-channel intensity group, and the correction Stokes group were extracted respectively. Then, three succession checks were performed on the previous and next positions. The first check was the consistency check of the correction residual sign group, requiring that the sign codes of the previous and next positions be identical in each of the four channels. The second check was the fixed channel order check, requiring that the correction back-substitution four-channel intensity group of the next position still follow the fixed channel order from the first to the fourth channel. The arrangement does not involve channel order swapping; the third check is the component direction check, which requires that the signs of the same Stokes components in the correction Stokes group of the previous and next positions be the same for each item, and zero-value components correspond only to zero-value components; a solution connection relationship is established between the previous and next positions only when all three connection checks are met, and the previous position identifier, the next position identifier, and their corresponding correction solution records are written into the solution connection chain group; then, it continues to expand backward in the original position order until the end of the four-channel intensity map; if there is no correction solution record at a certain position, or if any of the three connection checks is not met, the current solution connection chain is terminated at that position, and that position is written into the connection interruption record; Finally, to preserve the repeatable polarization calculation results, a reverse verification and effective retention are performed on the calculation chain group. Specifically, firstly, each correction residual symbol group in each calculation chain is concatenated end-to-end in positional order to form a residual symbol string. Then, each correction replacement four-channel intensity group is concatenated end-to-end in fixed channel order to form a replacement intensity string, and the correction Stokes group corresponding to each position in the chain is read simultaneously. Next, a consistency verification is performed on the residual symbol string, requiring that the residual symbol groups of adjacent positions from the first to the last position in the chain be consistent item by item. After that, a channel order verification is performed on the replacement intensity string, requiring that every four consecutive intensity values ​​after concatenation correspond to the first to the fourth channel in sequence, and there is no channel number jump. Subsequently, each corrected Stokes group is substituted back into the corresponding target instrument matrix and recalculated to obtain a recalculated Stokes group. The recalculated Stokes group is then compared component by component with the original corrected Stokes group. If the same component maintains the same positive or negative sign, it is recorded as if the component direction remains unchanged. Only when the residual sign string is consistent, the substitution intensity string maintains a fixed channel order, and the direction of each Stokes component obtained from the recalculation remains unchanged, is the corresponding corrected Stokes group retained and written into the valid polarization result table. The position identifier, corrected Stokes group, target instrument matrix identifier, and retention mark are also written. If any verification condition is not met, the corresponding record in the chain is deleted, not written into the valid polarization result table, and the corresponding position is written into the verification failure record. Through the above processing, the polarization solution module forms a complete solution link from initial inversion, residual error correction, acceptance verification to reverse verification, so that the correction Stokes group written into the effective polarization result table simultaneously meets four conditions: the target instrument matrix correspondence is valid, the single channel residual error correction is valid, the cross-position acceptance is valid, and the re-solution direction is consistent. This avoids directly outputting the solution results with local mismatch, channel misalignment, or unstable back substitution to the subsequent result output module. In practical applications: After the dynamic modeling module outputs the target matrix table, the polarization solution module performs an initial inversion on the four-channel intensity group and the target instrument matrix at a certain position. If the third channel in the resulting residual sign group is positive and the other three channels are negative, the original third channel intensity is replaced by the back-substitution intensity of the third channel, and the corrected Stokes group is solved again. Subsequently, a connection check is performed with the next position in the original position sequence. If the corrected residual sign groups of the two positions are consistent item by item, the corrected back-substitution four-channel intensity group maintains a fixed channel order, and the positive and negative signs of each component in the corrected Stokes group are consistent, a solution connection relationship is established. When the residual sign string is consistent after the reverse verification of the solution connection chain, the channel order of the back-substitution intensity string remains unchanged, and the direction of the component obtained by the re-solution does not change, the corrected Stokes group at each position in the chain is written into the effective polarization result table. If the positive and negative signs of a Stokes component after the re-solution at any position change, the corresponding record at that position is deleted and not written into the effective polarization result table.

[0023] The results output module is used to write the Stokes results in the valid polarization results table back to generate the live corneal reflection polarization results in the original position order, and to mark the positions where the target instrument matrix was not obtained or where the Stokes results were not retained as restricted positions, and output the polarization measurement results. In this embodiment, the result output module is used to write back the valid solution results in the valid polarization result table to the corresponding positions in the original position order, and to unify the positions where the target instrument matrix was not obtained and the positions where Stokes results were not retained into restricted positions. Then, based on the position continuity relationship, valid position segments, restricted position segments, and boundary positions are formed, and finally, polarization measurement results that can be directly used for subsequent display, storage, and retrieval are generated. This implementation process includes the following steps: First, to form a result carrier that is traceable position by position, the Stokes results corresponding to each position are read according to the position order in the effective polarization result table, and position write-back is performed in conjunction with the target matrix table and the verification failure record. Specifically, the original position sequence table, position scale identifier, and effective polarization result table are read first. The original position sequence table is arranged according to the priority order of sensor coordinate rows when writing back using the original pixel position, and according to the priority order of pixel block center coordinate rows when writing back using the original pixel block position. Then, the Stokes result record in the effective polarization result table is checked item by item along the original position order. If it exists, the Stokes result is written into the polarization result value field corresponding to the current position. If it does not exist, continue searching for the existence of a target instrument matrix record in the target matrix table at that location. If no target instrument matrix record exists, write a matrix missing restriction flag at that location. If a target instrument matrix record exists but no valid polarization result record exists, write a solution failure restriction flag at that location. Then, write the location identifier, location scale identifier, polarization result value field, and restriction flag field into the location result table in the original position order. If both the polarization result value field and the restriction flag field appear at the same location, retain the current location record based on the polarization result value field and write the restriction flag into the conflict record, without entering the location result table. Next, in order to form a continuous and readable spatial result structure, the position result table is processed by merging adjacent positions. Specifically, the current position record and the next position record in the position result table are read item by item in the original position order. Positions whose serial numbers differ by one position in the original position order are recorded as adjacent positions. Then, the result type of the adjacent positions is determined. When both the current position and the next position are written to the polarization result value field, the two positions are merged into the same valid position segment. The start position identifier, end position identifier, and number of Stokes results in the segment are recorded in the valid position segment field. When both the current position and the next position are written to the restricted marker field, the two positions are merged into the same restricted position segment, and the start position identifier, end position identifier, and number of restricted positions within the segment are recorded in the restricted position segment field; when the current position is written to the polarization result value field and the next position is written to the restricted marker field, or when the current position is written to the restricted marker field and the next position is written to the polarization result value field, the boundary position between the current position and the next position is recorded as the boundary position. Specifically, the boundary position is taken as the two boundary positions when writing back from the original pixel position, and as the two boundary pixel block positions when writing back from the original pixel block position; then, the front segment identifier, back segment identifier, and boundary position identifier corresponding to the valid position segment, restricted position segment, and boundary position are written to the result segment table; if a position contains neither a polarization result value field nor a restricted marker field, then the position is written to an empty result record and marked separately as an empty result position in the result segment table, and does not participate in the combination of the valid position segment and the restricted position segment; Finally, to generate complete live corneal reflection polarization results, the results are spliced ​​and written back according to the original position order based on the result segmentation table. Specifically, each valid position segment and each restricted position segment is read according to the segment order in the result segmentation table. For each valid position segment, the corresponding Stokes results are extracted according to the position order within the segment to form a Stokes result sequence, and this Stokes result sequence is written into the valid result field of the polarization measurement results. Then, for each restricted position segment, the corresponding restricted markers are extracted according to the position order within the segment to form a restricted position sequence, and this restricted position sequence is written into the restricted result field of the polarization measurement results. Then, the boundary position records in the result segmentation table are read, and the front segment identifier, back segment identifier, boundary position identifier, and front and back segment result types corresponding to each boundary position are written into the boundary relationship field of the polarization measurement result. Then, the position scale identifier, original position sequence table identifier, valid result field, restricted result field, and boundary relationship field are combined to form the polarization measurement result output. If there is no valid position segment in the result segmentation table, the polarization measurement result only retains the restricted result field and the boundary relationship field. Through the above processing, the result output module writes the solution results in the effective polarization result table, the matrix missing information in the target matrix table, and the failure information in the solution stage back to the original position order, forming a position-by-position traceable, segmented continuous, and boundary-clear polarization measurement result. This allows the subsequent caller to directly distinguish between the effective result area, the restricted area, and the boundary relationship between the two types of areas, and to trace back whether the source of the restricted position is matrix missing or solution failure. In practical applications: when Stokes results exist at positions 1 to 10 in the effective polarization result table, but no target instrument matrix record exists at positions 11 to 13, and Stokes results exist at positions 14 to 20, the result output module first writes the polarization result values ​​for positions 1 to 10 and positions 14 to 20, and writes the matrix missing / restricted markers for positions 11 to 13, forming a position result table; then, positions 1 to 10 are combined into a first effective position segment, positions 11 to 13 are combined into a first restricted position segment, and positions 14 to 20 are combined into a second effective position segment, with positions 10 and 11, and 13 and 14 marked as boundary positions, forming a result segmentation table; finally, the Stokes result sequences for the first and second effective position segments, the restricted position sequence for the first restricted position segment, and the segmentation relationship between the two boundary positions are written into the polarization measurement results in the original position order.

[0024] Working Principle: This scheme first applies near-infrared illumination to the living cornea and selects a target illumination state from multiple annular illumination zones that allows the corneal reflective nucleus to fully enter the imaging range while maintaining a closed reflective edge. Then, under this illumination state, a single exposure acquisition is performed, and the corneal reflected light is simultaneously distributed to four imaging regions according to four analytical states, forming a four-channel intensity map with a fixed channel order. Next, the system extracts the reflective nucleus position, reflective nucleus width, edge unfolding direction, and sub-map spacing from the four-channel intensity map, forming a state group for the current position. These state groups are then matched item by item with pre-calibrated source state groups to screen out the truly matching target instrument matrix. Next, the system uses the target instrument matrix to perform Stokes inversion on the four-channel intensity group, and through back substitution, residual error correction, front-to-back position verification, and reverse verification, unstable or inconsistent results are removed, retaining only valid polarization results. Finally, the system writes back the valid results in the original position order, marking positions where no matrix was obtained or verification failed as restricted positions. The continuous valid regions, continuous restricted regions, and their boundaries are then organized into the final polarization measurement results. For example, in dry eye screening or ocular surface condition assessment scenarios, the subject only needs to gaze briefly in front of the device. The system automatically selects an appropriate ring illumination zone to form a stable and complete reflection nucleus on the cornea. Then, four polarization sub-images are obtained in a single exposure, eliminating the need for multiple repeated shots. The system then determines which calibrated instrument matrix is ​​more suitable for the current location based on these four sub-images and performs polarization calculations at each location. If the image separation relationship is abnormal at certain locations, the matrix is ​​mismatched, or the calculation results are inconsistent after substitution, these locations will not be considered valid results but will be marked as restricted locations. In this way, the doctor ultimately sees not an image that mixes reliable and unreliable results, but a polarization measurement result that has distinguished the valid area, restricted area, and boundary location, making it easier to judge the corneal reflection state and changes in the ocular surface.

[0025] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A four-channel polarization state measurement system for live corneal reflection, characterized in that, include: The illumination and imaging module is used to perform near-infrared illumination on the living cornea and to perform imaging and acquisition of corneal reflected light, and output the corneal reflected light acquired in the same session; The polarization imaging module is used to perform four-state spectral splitting and spatial imaging on corneal reflected light, enabling the imaging sensor to form a four-channel polariton map under the same exposure and output a four-channel intensity map arranged in a fixed channel order. The state organization module is used to extract the reflection kernel position, reflection kernel width, edge unfolding direction and sub-image spacing from the four-channel intensity map, and combine them into current state groups according to pixel position or pixel block position, and output the current state table; The dynamic modeling module is used to read the pre-stored matrix table, compare the current state table with the source state group in the matrix table item by item, construct the correspondence model between the current acquisition state and the calibrated matrix state according to the reflection kernel position, reflection kernel width, edge unfolding direction and sub-image spacing, select the target instrument matrix when the correspondence is consistent, and output the target matrix table. The polarization calculation module is used to call the target instrument matrix in the target matrix table to perform Stokes inversion on the four-channel intensity map at the corresponding position, and substitute the inversion result back to the corresponding target instrument matrix to perform channel reconstruction. When the reconstructed four-channel intensity arrangement is consistent with the original four-channel intensity arrangement, the corresponding Stokes result is retained, and the effective polarization result table is output.

2. The four-channel polarization state measurement system for live corneal reflection according to claim 1, characterized in that: Also includes: The results output module is used to write the Stokes results from the valid polarization results table back to generate the live corneal reflection polarization results in the original position order, and to mark the positions where the target instrument matrix was not obtained or where the Stokes results were not retained as restricted positions, and output the polarization measurement results.

3. The four-channel polarization state measurement system for live corneal reflection according to claim 2, characterized in that: The illumination imaging module includes: The near-infrared light source is subjected to ring-shaped zonal emission control, and the beams of each zone are projected onto the living cornea in a predetermined incident direction. The corneal reflection brightness distribution in the corresponding imaging sensor is read, and the target illumination zone that makes the corneal reflection nucleus fall completely into the imaging range and the corneal reflection edge is continuously closed is selected. The target illumination state corresponding to the current acquisition is output. Under target illumination, the corneal reflected light is acquired in the same exposure, the position of the corneal reflection nucleus, the outward expansion direction of the reflection nucleus, and the closed range of the reflection edge are extracted, and the position of the corneal reflection nucleus, the outward expansion direction of the reflection nucleus, and the closed range of the reflection edge are written into the same acquisition record corresponding to the target illumination state, and the corneal reflected light acquired in the same exposure is output. For the same acquisition record, an illumination consistency check is performed. If the corneal reflection nucleus position corresponding to the current exposure is within the range of the target illumination zone, the outward expansion direction of the reflection nucleus is consistent with the target incident direction, and the closed range of the reflection edge is not broken, the corneal reflection light acquired in the same acquisition is retained as the input for subsequent polarization image separation.

4. The four-channel polarization state measurement system for live corneal reflection according to claim 3, characterized in that: The polarization image separation module includes: The corneal reflected light is split into four analytical states, so that the reflected light corresponding to the four analytical states is projected onto the four image regions of the imaging sensor respectively. The center position, region spacing and region boundary of the four image regions are recorded in a predetermined arrangement order, and a four-channel image position table is output. Based on the four-channel image splitting position table, intensity reading and region cropping are performed on the four image splitting regions formed under the same exposure. The intensity range of the reflection nucleus, the edge transition range and the effective coverage range of each image splitting region are extracted according to the predetermined arrangement order. The intensity data of the corresponding four analysis states are combined according to the fixed channel order to output a four-channel intensity map. Perform a pixel consistency check on the four-channel intensity map. If the predetermined spacing relationship is maintained at the center of the four pixel regions, the region boundaries do not overlap, and the corresponding reflection nuclei are all within the effective coverage of each pixel region, the four-channel intensity map is retained as the input for subsequent state processing.

5. The four-channel polarization state measurement system for live corneal reflection according to claim 4, characterized in that: The status processing module includes: Perform co-location expansion on the four-channel intensity map according to a fixed channel order, extract the center position, left and right boundary positions and outer intensity decreasing direction of each channel reflection nucleus, and solve the sub-map spacing based on the distance between the center positions of the corresponding channel reflection nucleus, and output the four-channel state component table. At the same pixel position or the same pixel block position, perform corresponding sorting on the four-channel state component table, combine the center position of the reflection kernel of the corresponding four channels into a reflection kernel position group, combine the difference between the left and right boundary positions of the corresponding four channels into a reflection kernel width group, combine the outer intensity decreasing direction of the corresponding four channels into an edge expansion direction group, and combine the center position distance of the corresponding four channels into a sub-image spacing group, and output the current state group. Write the current state group into the state table in the order of pixel position or pixel block position. When the four-channel reflection kernel position group, reflection kernel width group, edge unfolding direction group and sub-image spacing group all come from the same exposure and maintain a fixed channel correspondence, retain the corresponding current state group and output the current state table.

6. The four-channel polarization state measurement system for live corneal reflection according to claim 5, characterized in that: The dynamic modeling module includes: Perform corresponding expansion on each current state group in the current state table and each source state group in the matrix table according to a fixed field order. The fixed field order is: reflection kernel position group, reflection kernel width group, edge expansion direction group, and subgraph spacing group. Solve the reflection kernel position difference group, reflection kernel width difference group, edge expansion direction corresponding value group, and subgraph spacing difference group respectively. When all items in the edge expansion direction corresponding value group are in the same direction and all items in the subgraph spacing difference group have the same sign, establish candidate corresponding edges between each current state group and each source state group, and output the candidate corresponding graph. Following the order of the current state group in the current state table, perform inheritance expansion on the candidate corresponding edges in the candidate corresponding graph. Use the source state group position of the candidate corresponding edge of the previous current state group as the inheritance benchmark of the candidate corresponding edge of the next current state group. When the position of the next source state group is after the position of the previous source state group, the items of the reflection kernel position difference group have the same sign as the corresponding items of the previous candidate corresponding edge, the items of the reflection kernel width difference group have the same sign as the corresponding items of the previous candidate corresponding edge, and the items of the subgraph spacing difference group have the same sign as the corresponding items of the previous candidate corresponding edge, establish the previous and next inheritance relationship and output the candidate inheritance chain group.

7. The four-channel polarization state measurement system for live corneal reflection according to claim 6, characterized in that: The dynamic modeling module also includes: For each candidate connection chain group, reverse verification and unique retention are performed. The reflection kernel position difference groups corresponding to the corresponding edges of each candidate connection chain are concatenated end-to-end in field order to form a position difference string. The reflection kernel width difference groups are concatenated end-to-end in field order to form a width difference string. The edge expansion direction corresponding value groups are concatenated end-to-end in field order to form a direction corresponding string. The subgraph spacing difference groups are concatenated end-to-end in field order to form a spacing difference string. Then, a comparison is performed item by item in the following order: position difference string, width difference string, direction corresponding string, and spacing difference string. When different items appear at the same comparison position, retain the candidate succession chain with the lower absolute value corresponding to that position; when no retention result is obtained at any of the comparison positions of the position difference string, width difference string, direction correspondence string and spacing difference string, retain the first candidate succession chain with the lower code value after comparing the source state group sequence string item by item according to the sequence encoding order, delete the remaining candidate succession chains and re-execute succession expansion until each current state group corresponds to one candidate succession chain, or no new succession relationship is formed in the candidate correspondence graph, and output a stable correspondence chain; Based on the stable correspondence chain, each current state group and its corresponding source state group are written into the corresponding model, and the instrument matrix bound to each corresponding source state group is retrieved to form the target matrix table. Then, each instrument matrix is ​​used to perform inversion and back substitution on the four-channel intensity of the corresponding current state group to obtain the back-substituted four-channel intensity group. When the back-substituted four-channel intensity group corresponds to the original four-channel intensity group in a fixed channel order and the position of each channel reflection nucleus is kept in the same arrangement order, the corresponding instrument matrix is ​​retained as the target instrument matrix, and the target matrix table is output.

8. The four-channel polarization state measurement system for living corneal reflection according to claim 7, characterized in that: The polarization calculation module includes: For each target instrument matrix in the target matrix table, the four-channel intensity group at the corresponding position is arranged in a fixed channel order to form a quaternary solution group. The initial Stokes group is solved, and the initial Stokes group is substituted back to the corresponding target instrument matrix to generate the back-substituted four-channel intensity group. Then, the original four-channel intensity group and the back-substituted four-channel intensity group are subtracted channel by channel to form the channel residual group. The residual sign group is formed by judging channel by channel, and the initial solution table is output. Perform residual error correction expansion on the initial solution table. When there is a channel symbol in the residual symbol group that is opposite to the other three channel symbols, replace the corresponding intensity in the original four-channel intensity group with the back-substitution intensity of that channel in the back-substitution four-channel intensity group to form a corrected four-channel intensity group. Then, use the corrected four-channel intensity group and the original target instrument matrix to solve the corrected Stokes group again. Finally, use the corrected Stokes group to back-substitute and generate the corrected back-substitution four-channel intensity group and the corrected residual symbol group, and output the corrected solution table.

9. The four-channel polarization state measurement system for live corneal reflection according to claim 8, characterized in that: The polarization calculation module further includes: Perform a continuity check on the correction solution table according to the arrangement order of the corresponding positions in the four-channel intensity map. When the correction residual sign group of the previous position and the next position are consistent, the correction back substitution four-channel intensity group maintains a fixed channel order, and the positive and negative directions of each component in the correction Stokes group are consistent, a solution continuity relationship is established between the previous position and the next position, and the solution continuity chain group is output. Perform reverse verification and effective retention on the solution chain group. Concatenate the correction residual symbol groups in each solution chain in order of position to form a residual symbol string. Concatenate the four-channel intensity groups of each correction back-substitution in a fixed channel order to form a back-substitution intensity string. When the residual symbol string is consistent, the back-substitution intensity string maintains a fixed channel order, and the component direction obtained after the corresponding correction Stokes group is back-substituted and re-solved remains unchanged, retain the corresponding correction Stokes group and write it into the effective polarization result table. Output the effective polarization result table.

10. The four-channel polarization state measurement system for live corneal reflection according to claim 9, characterized in that: The result output module includes: Read the Stokes results corresponding to each position in the effective polarization result table according to the position order, and write back each Stokes result according to the original pixel position or the original pixel block position. Write the polarization result value at the position where the Stokes result is obtained, and write the restricted mark at the position where the target instrument matrix is ​​not obtained and the position where the Stokes result is not retained. Output the position result table. Perform adjacent position consolidation on the position result table, combine adjacent positions that are both written with polarization result values ​​into valid position segments, combine adjacent positions that are both written with restricted marks into restricted position segments, and mark the positions located at the intersection of valid position segments and restricted position segments as boundary positions, and output the result segment table; Based on the result segmentation table, generate the live corneal reflection polarization results in the original position order. Write the corresponding Stokes result sequence in the effective position segment and the corresponding restricted position sequence in the restricted position segment. Write the preceding and following segmentation relationship corresponding to the boundary position into the polarization measurement results and output the polarization measurement results.

Images