Cotton transparent foreign fiber recognition method combining polarization detection and image processing

Abstract

This invention relates to the field of image recognition technology, specifically a method for identifying transparent foreign fibers in cotton using polarization detection combined with image processing. The method includes the following steps: acquiring images using a polarization sensor and extracting temporal and angular information; calculating pixel phase delay; performing frequency domain analysis to determine the principal direction and generating a birefringence response; screening skeleton points based on grayscale and constructing fiber paths; combining phase gradients for extreme value analysis and intensity reconstruction to obtain global coordinates; and then weighted filtering based on directional differences to eliminate redundancy and obtain fiber position coordinates. In this invention, by combining image acquisition and feature expression paths, polarization response and texture information are fused to form multi-dimensional discriminative features, reducing noise and uneven illumination interference. Combining fiber continuity and optical response analysis enhances the ability to identify weak responses, making boundaries clearer and more stable. Furthermore, redundant points are compressed under directional classification and spatial constraints, improving positioning consistency and recognition accuracy, and achieving stable identification of transparent foreign fibers in complex scenarios.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method for identifying transparent heterogeneous fibers in cotton by combining polarization detection with image processing. Background Technology

[0002] Image recognition technology mainly involves the technical system of target detection, classification and discrimination by acquiring image information and analyzing and processing it. Its core contents include the selection of image acquisition methods, image enhancement and preprocessing methods, feature information extraction and expression, and judgment rules for target differentiation based on features. This field is widely used in industrial inspection quality control and material sorting scenarios, emphasizing the use of information such as gray-scale distribution, texture structure, edge morphology and optical response differences in images to distinguish and identify target objects.

[0003] Among them, the traditional method of identifying transparent foreign fibers in cotton by combining polarization detection with image processing refers to obtaining transmission or reflection images of cotton fibers in multiple polarization directions by setting up a polarized light source and a polarizer, and then performing grayscale conversion, contrast stretching and binarization processing on the acquired images. The connected area and boundary contour shape parameters of the region are calculated by combining the pixel brightness distribution, and the transparent foreign fibers and cotton fibers are classified and distinguished according to the differences in light intensity changes, texture details and morphological size.

[0004] Existing technologies rely on frame-by-frame phase delay calculation and frequency domain analysis to construct directional features during polarization image processing. However, they are sensitive to temporal consistency and noise disturbances during pixel-level arrangement and phase extraction, which can easily lead to insufficient stability of the phase sequence and thus affect the accuracy of main direction determination. At the same time, when screening skeleton points and connecting paths based on grayscale thresholds, they have limited adaptability to complex backgrounds or areas with uneven brightness, which can easily lead to fiber breakage or misconnection, resulting in deviations in spatial gradient calculation. In addition, the extreme value search and intensity reconstruction process relies on local mutation features, and the recognition boundary becomes blurred when facing weak response scenes of transparent and heterogeneous fibers, ultimately resulting in decreased positioning accuracy and unstable redundancy point removal. Summary of the Invention

[0005] To achieve the above objectives, the present invention employs the following technical solution: a method for identifying transparent heterogeneous fibers in cotton using polarization detection combined with image processing, comprising the following steps: S1: Obtain polarized image frames of the cotton detection scene through a polarization sensor and extract time markers and polarization angles. Arrange the pixels of the polarized image frames according to the time markers and calculate the phase delay of the corresponding pixels based on the polarization angle to construct a phase delay sequence. S2: Extract the corresponding phase delay set in pixel space from the phase delay sequence and perform frequency domain analysis. Select the angle corresponding to the main peak of the spectral amplitude as the main direction of the cotton phase, calculate the extreme difference in the phase delay set, and generate the birefringence response amplitude. S3: Extract the brightness of the polarized image frame and select the pixels that meet the preset grayscale threshold as candidate skeleton points. Connect the candidate skeleton points and analyze the fiber continuous path. Combine the birefringence response amplitude to perform differential analysis and generate a spatial phase gradient. S4: Based on the continuous fiber path, call the spatial phase gradient to perform extreme value search, divide the abrupt segment in the continuous fiber path and extract the maximum phase delay for intensity reconstruction calculation, extract the pixel points corresponding to the reconstruction value that meet the preset intensity threshold for spatial connectivity judgment, and generate the global coordinates of the transparent heterogeneous fiber. S5: Calculate the angle difference of neighboring pixels by calling the cotton phase principal direction for the global coordinates of the transparent foreign fiber and classify the attributes. Construct a direction group set and perform a weighted summation based on the reciprocal of the angle deviation. Extract the coordinate point corresponding to the largest weighted summation value and perform redundancy removal on the global coordinates of the transparent foreign fiber to generate the position coordinates of the transparent foreign fiber.

[0006] As a further embodiment of the present invention, the phase delay sequence includes phase delay, time series index, and pixel position mapping; the birefringence response amplitude includes extreme difference amplitude, amplitude distribution characteristics, and response intensity; the spatial phase gradient includes gradient amplitude, gradient direction, and spatial rate of change; the global coordinates of the transparent anisotropic fiber include a set of spatial position coordinates, connected region identifiers, and structural distribution range; and the position coordinates of the transparent anisotropic fiber include a set of filtered coordinate points, direction consistency markers, and a redundant spatial index.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: The polarization image frame of the cotton detection scene is acquired by the polarization sensor and the associated time marker and polarization angle of the polarization image frame are extracted. The pixel coordinates within each polarization image frame are encoded. The indexes of the same coordinates are sorted according to the time marker and rearranged into a pixel vector sequence to obtain pixel time sequence arrangement data. S102: Based on the pixel time sequence arrangement data, call the polarization angle of the corresponding pixel position at multiple time points, perform angle difference on the polarization angle at multiple time points, perform periodic normalization on the difference results, and arrange the normalized angle sequence in order to obtain a polarization angle difference sequence set. S103: Extract the angle difference sequence corresponding to each pixel according to the polarization angle difference sequence set, call the adjacent angle differences to perform trigonometric function conversion and combine them with the preset optical path difference reference value for numerical mapping, and combine and arrange the mapping results in time order to obtain the phase delay sequence.

[0008] As a further aspect of the present invention, the optical path difference reference value is determined by collecting the center wavelength value of the incident light source in the detection scene, multiplying it with the birefringence coefficient of a preset standard medium, extracting the initial phase offset of the stress-free environment and converting it into an equivalent optical path difference as a compensation factor, and then adding the product result with the compensation factor to determine the value.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Extract the corresponding phase delay set in pixel space from the phase delay sequence and perform discrete Fourier transform to map the phase delay value sequence to the frequency component sequence. At the same time, perform amplitude square accumulation processing on multiple frequency components and rearrange them according to the frequency index order to obtain the frequency amplitude distribution sequence. S202: Based on the frequency amplitude distribution sequence, retrieve the amplitude data corresponding to multiple frequency positions, calculate the difference between the amplitudes of adjacent frequency positions, record the frequency positions where the current amplitude exceeds both the amplitude of the previous position and the amplitude of the next position as peak positions and convert them into angle data to obtain the main peak angle index; S203: Locate the direction data corresponding to the phase delay set according to the main peak angle index, and perform maximum and minimum value difference operation on all phase delay values ​​downward, and perform numerical mapping on the obtained difference value to generate birefringence response amplitude.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Extract the brightness of the polarized image frame and retrieve the grayscale values ​​of multiple pixels. Compare the grayscale values ​​with the preset grayscale filtering threshold point by point. Record the coordinate index of the pixel whose grayscale value exceeds the grayscale filtering threshold. Arrange all the recorded coordinate indices according to their spatial positions to obtain the candidate skeleton coordinate sequence. S302: Based on the candidate skeleton coordinate sequence, retrieve the spatial coordinate index of multi-pixel points and perform spatial eight-neighbor traversal to extract adjacent spatial points, fit the connection between adjacent spatial points, sequentially splice the fitted line segments and rearrange the index of associated points along the direction of the connection to construct the fiber path topology sequence. S303: According to the fiber path topology sequence, call the birefringence response amplitude value corresponding to the multi-pixel position in the path, perform difference calculation on the amplitude of adjacent path positions and accumulate and arrange them in the path order, and at the same time perform gradient mapping to generate spatial phase gradient.

[0011] As a further aspect of the present invention, the grayscale threshold is determined by acquiring the global pixel grayscale values ​​of the polarized image frame, summing all pixel grayscale values ​​and dividing by the total number of pixels to obtain the global grayscale average value, calculating the sum of squares of the differences between the multi-pixel grayscale values ​​and the global grayscale average value to obtain the grayscale standard deviation, and combining the global grayscale average value and the grayscale standard deviation with a preset weighting coefficient to perform a weighted summation operation.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the continuous fiber path, the spatial phase gradient is called to perform extreme value search. The phase gradient magnitudes of multiple pixels in the path are arranged. The center point corresponding to the current pixel gradient magnitude exceeding the gradient magnitudes of the adjacent pixels on both sides is extracted as the extreme point. The continuous fiber path is then split to obtain the abrupt segment index sequence. S402: Extract the phase delay value set within the corresponding interval according to the abrupt segment index sequence, compare the phase delay values ​​of multiple pixels in the set point by point, map the numerical size relationship into an ordered sequence, and select the phase delay value corresponding to the position of the largest value in the sequence as a representative parameter to generate the maximum phase delay value set. S403: Perform intensity reconstruction calculation based on the maximum phase delay value set, map multiple phase delay values ​​to corresponding intensity values ​​and filter the pixel index set that meets the preset intensity threshold, determine the spatial neighborhood connectivity of the pixel index set and aggregate the coordinates of adjacent pixels to generate the global coordinates of the transparent heterogeneous fiber.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Calculate the neighboring pixel angle difference by calling the cotton phase main direction for the global coordinates of the transparent heterogeneous fiber. Extract the neighboring pixel angle for each coordinate point and calculate the difference with the corresponding main direction angle. Determine the difference result and classify the attributes according to the preset angle segmentation interval threshold to obtain the angle difference grouping index set. S502: Based on the angle difference group index set, extract the angle deviation corresponding to the pixel in the multi-group, call the angle deviation reciprocal operation rule to transform the angle deviation of the multi-pixel inverse, use the reciprocal as the weight to pair with the pixel position index and accumulate at the corresponding position, and superimpose the weights of the multi-group to obtain the weight aggregate coordinate mapping table. S503: Retrieve the cumulative weight values ​​according to the weight aggregation coordinate mapping table and compare their magnitudes. Extract the set of coordinate points corresponding to the largest weight values. Deduplicate and remove duplicate coordinate points that appear repeatedly in the global coordinates of the transparent foreign fiber, retain the unique coordinate combination, and generate the position coordinates of the transparent foreign fiber.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by comprehensively reconstructing the image information acquisition method and feature expression path, and integrating polarization response differences and texture structure information, a more discriminative expression form is formed in a multi-dimensional feature space. This makes target recognition no longer solely dependent on local brightness or single phase changes, thereby reducing the interference of noise and uneven illumination on the recognition results. At the same time, by jointly analyzing the continuity of fiber structure and optical response characteristics, the recognition ability of weak response areas is enhanced, making fiber boundaries clearer and more stable. Furthermore, under the constraints of directional attribute classification and spatial relationship, redundant points are effectively compressed, improving positioning consistency and overall recognition accuracy. This enables reliable discrimination and stable output of transparent heterogeneous fibers in complex scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a method for identifying transparent heterogeneous fibers in cotton using polarization detection combined with image processing, comprising the following steps: S1: Obtain polarized image frames of the cotton detection scene through a polarization sensor and extract time markers and polarization angles. Arrange the pixels of the polarized image frames according to the time markers and calculate the phase delay of the corresponding pixels based on the polarization angle to construct a phase delay sequence. S2: Extract the corresponding phase delay set in pixel space from the phase delay sequence and perform frequency domain analysis. Select the angle corresponding to the main peak of the spectral amplitude as the main direction of the cotton phase, calculate the extreme difference in the phase delay set, and generate the birefringence response amplitude. S3: Extract the brightness of the polarized image frame and select the pixels that meet the preset grayscale threshold as candidate skeleton points. Connect the candidate skeleton points and analyze the continuous fiber path. Combine the birefringence response amplitude to perform differential analysis and generate spatial phase gradient. S4: Based on the continuous fiber path, call the spatial phase gradient to perform extreme value search, divide the abrupt segment in the continuous fiber path and extract the maximum phase delay for intensity reconstruction calculation, extract the pixels corresponding to the reconstruction value that meet the preset intensity threshold for spatial connectivity judgment, and generate the global coordinates of the transparent heterogeneous fiber. S5: Calculate the angle difference of neighboring pixels by calling the cotton phase principal direction for the global coordinates of the transparent foreign fiber and classify the attributes. Construct a direction group set and perform a weighted sum based on the reciprocal of the angle deviation. Extract the coordinate point corresponding to the maximum weighted sum value and remove redundancy from the global coordinates of the transparent foreign fiber to generate the position coordinates of the transparent foreign fiber.

[0020] The phase delay sequence includes phase delay, time series index, and pixel position mapping; the birefringence response amplitude includes extreme difference amplitude, amplitude distribution characteristics, and response intensity; the spatial phase gradient includes gradient amplitude, gradient direction, and spatial rate of change; the global coordinates of the transparent anisotropic fiber include a set of spatial position coordinates, connected region identifiers, and structural distribution range; and the position coordinates of the transparent anisotropic fiber include a set of filtered coordinate points, orientation consistency markers, and a redundant spatial index.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: The polarization image frame of the cotton detection scene is acquired by the polarization sensor and the associated time marker and polarization angle of the polarization image frame are extracted. The pixel coordinates within each polarization image frame are encoded. The indexes of the same coordinates are sorted according to the time marker and rearranged into a pixel vector sequence to obtain pixel time sequence arrangement data. A multispectral polarization camera positioned above the cotton detection area continuously acquires polarization image frames of the detection scene. The acquisition process sets the sampling frequency to 60 frames per second and the image resolution to 1920 pixels horizontally and 1080 pixels vertically. During image processing, the timestamp recorded in the metadata encapsulation area of ​​the polarization image frame header is extracted as a correlation time sequence marker. Simultaneously, the angle value returned by the camera sensor is extracted as the polarization angle, which is then forcibly calibrated to 0 degrees, 45 degrees, 90 degrees, or 135 degrees. For each pixel within a polarization image frame, linear single-dimensional encoding operations are performed on its horizontal and vertical coordinates. The vertical coordinate value of the currently processed pixel is multiplied by the fixed horizontal resolution of 1920 to obtain an initial intermediate product. This intermediate product is then multiplied by the horizontal coordinate of the current pixel. The algorithm performs algebraic addition to derive a globally unique pixel coordinate index. For example, if a pixel's vertical coordinate is 10, multiplying it by the horizontal resolution of 1920 yields 19200. Adding the horizontal coordinate of 5 gives the current pixel coordinate index as 19205. After calculating the full-image index, based on the temporal order of associated time stamps, the polarization angles at the same coordinate index position across multiple consecutive frames are extracted from the entire image queue. By comparing the timestamp values, the 45-degree polarization angles obtained in the first time frame and the 90-degree polarization angles obtained in the second time frame, belonging to the specific coordinate index 19205, are strictly shifted and rearranged according to the earliest to latest timestamp order. This generates a pixel vector sequence composed of continuous polarization angles in memory, which is then iteratively applied to all pixel positions within the current detection scene to obtain the pixel temporal arrangement data. The advantage of this operational logic is that it transforms spatial coordinates into linear indices and sorts them through multiplication and addition operations.

[0022] S102: Based on the pixel time sequence arrangement data, call the polarization angle of the corresponding pixel position at multiple time points, perform angle difference on the polarization angle at multiple time points, perform periodic normalization on the difference results, and arrange the normalized angle sequence in order to obtain the polarization angle difference sequence set. Based on the received pixel time-series data, the polarization angle values ​​recorded at multiple consecutive time sampling points corresponding to the specific coordinate index 19205 of the pixel position are extracted. The polarization angle data recorded at the first time node (45 degrees), the second time node (90 degrees), and the third time node (135 degrees) are retrieved. The extracted multi-time polarization angle values ​​are then subjected to angle difference subtraction operations strictly according to the chronological order of the physical time sampling points. The initial angle difference data is obtained by subtracting the polarization angle value of the immediately preceding time sampling point from the polarization angle value of the later time sampling point. Specifically, the first initial difference of 45 degrees is calculated by subtracting the 90 degrees of the second time node from the 45 degrees of the first time node, and the second initial difference of 45 degrees is calculated by subtracting the 135 degrees of the third time node from the 90 degrees of the second time node. Subsequently, all calculated... The initial angle difference data are processed one by one using periodic normalization and correction operations. It is determined whether each initial difference falls precisely within the standard polarization period closed interval of 0 to 180 degrees. If an initial difference is determined to be strictly less than 0 degrees, a correction accumulation operation of adding a 180-degree constant term is performed. If an initial difference is determined to be greater than or equal to 180 degrees, a correction cancellation operation of subtracting a 180-degree constant term is performed. The first and second initial differences of 45 degrees are both within the standard closed interval and are directly identified as the first and second normalized angle values. If a difference operation produces a fourth initial difference of 190 degrees, a correction cancellation operation of subtracting 180 degrees is performed to obtain a normalized angle value of 10 degrees. All processed normalized angle values ​​are continuously spliced ​​and queued according to the time sequence of the difference subtraction operation to construct a polarization angle difference sequence set. The advantage of this operation logic is that redundant periodic data is eliminated by combining difference subtraction with period determination and addition / subtraction correction logic.

[0023] S103: Extract the angle difference sequence corresponding to each pixel based on the polarization angle difference sequence set, call the adjacent angle differences to perform trigonometric function conversion and combine them with the preset optical path difference reference value for numerical mapping, and combine and arrange the mapping results in time order to obtain the phase delay sequence. Based on the output polarization angle difference sequence set, the angle difference sequence corresponding to each pixel is extracted. The angle difference sequence at coordinate index 19205 is retrieved, and the first difference of 45 degrees and the second difference of 10 degrees are extracted. Trigonometric function positive switching operation is performed on adjacent angle differences one by one to calculate the tangent function value corresponding to the first difference of 45 degrees and obtain the conversion result 1. Then, a mapping multiplication operation is performed with the preset optical path difference reference value and the conversion result. The optical path difference reference value is obtained by real-world calibration using 100 sets of standard cotton samples with known optical path characteristics. In the testing phase, the grid search method is used to iterate in increments of 0.1 micrometers within the range of 1.0 to 2.0 micrometers. In the test, when the parameter was set to 1.5 micrometers, the mapping error of each cotton sample was at its minimum of 0.02 micrometers. Therefore, the optical path difference reference value was fixed at 1.5 micrometers. The first difference conversion result 1 was multiplied by the reference value of 1.5 micrometers to obtain the first time node mapping value of 1.5 micrometers. For the second difference of 10 degrees, the tangent function value was calculated to obtain 0.176. Multiplying 0.176 by the reference value of 1.5 micrometers yielded the second node mapping value of 0.264 micrometers. The 1.5 micrometer and 0.264 micrometer values ​​were then strictly filled into a one-dimensional array in chronological order and combined to obtain the phase delay sequence of the current pixel. The calibration data is compiled into Table 1, the cotton optical path difference reference table.

[0024] Table 1. Comparison Table of Optical Path Difference for Cotton

[0025] Table 1 shows a detailed record of the error data comparison results under different parameter settings. The innovation of this operation logic lies in completing the mapping operation of real physical features through the multiplication operation of the tangent function and the reference value.

[0026] Please see Figure 3 The specific steps of S2 are as follows: S201: Extract the corresponding phase delay set in pixel space from the phase delay sequence and perform discrete Fourier transform to map the phase delay value sequence to the frequency component sequence. At the same time, perform amplitude square accumulation on multiple frequency components and rearrange them according to the frequency index order to obtain the frequency amplitude distribution sequence. Based on the output phase delay sequence, for the spatial position corresponding to the vertical coordinate 10 and the horizontal coordinate 5 in the cotton detection scene, the phase delay set corresponding to the pixel space formed by all continuous acquisition time nodes is extracted. The phase delay values ​​of the first four time nodes are retrieved as 1.5 μm, 0.264 μm, 1.2 μm, and 0.5 μm, respectively. Frequency domain transformation is performed on each extracted phase delay set. Each phase delay value is multiplied by the sine basis function value and cosine basis function value of the corresponding frequency. Then, the product results of each time node are accumulated and summed to separate the real part value and imaginary part value of the specific frequency position. This completes the operation of transforming the time domain phase delay value sequence into a frequency component sequence containing multiple frequency components. For the base frequency position with frequency index 1, the real part value is calculated. The real part is 2.0 and the imaginary part is 1.5. Then, the amplitude squared summation is performed on each frequency component. For a single frequency position, the real part is multiplied by itself to obtain the square of the real part, and the imaginary part is multiplied by itself to obtain the square of the imaginary part. The two squared values ​​are added to derive the comprehensive amplitude data for the corresponding frequency position. Substituting the real part 2.0 and the imaginary part 1.5, the square of the real part is calculated to be 4.0 and the square of the imaginary part to be 2.25. Addition yields a comprehensive amplitude data of 6.25 for this base frequency position. For the frequency position at twice the frequency index 2, the comprehensive amplitude data is calculated to be 12.4. After completing the calculations for all frequencies, all comprehensive amplitude data are rearranged and combined in ascending order of frequency index from 1 to the end, finally outputting the frequency amplitude distribution sequence. The innovation of this operation logic lies in achieving energy focusing extraction through basis function product and squared summation.

[0027] S202: Based on the frequency amplitude distribution sequence, retrieve the amplitude data corresponding to multiple frequency positions, calculate the difference between the amplitudes of adjacent frequency positions, record the frequency positions where the current amplitude exceeds both the amplitudes of the previous and subsequent positions as peak positions and convert them into angle data to obtain the main peak angle index; Based on the generated frequency amplitude distribution sequence, the comprehensive amplitude data corresponding to multiple consecutive frequency positions is retrieved. The comprehensive amplitude data corresponding to the base frequency position (frequency index 1) is extracted as 6.25; the comprehensive amplitude data corresponding to the second frequency position (frequency index 2) is extracted as 12.4; and the comprehensive amplitude data corresponding to the third frequency position (frequency index 3) is extracted as 4.1. Difference comparison calculations are performed on the comprehensive amplitude data of adjacent frequency positions one by one. The currently traversed second frequency position is selected as the center node, and its comprehensive amplitude data 12.4 is subtracted from the comprehensive amplitude data 6.25 of the base frequency position preceding it, resulting in a forward difference of 6.15. Simultaneously, its comprehensive amplitude data 12.4 is subtracted from the comprehensive amplitude data 4.1 of the third frequency position following it, resulting in a backward difference of 8.3. The forward and backward differences are extracted and compared with a constant 0. If a judgment is made... If both the forward and backward differences are strictly greater than the constant 0, meaning the current position amplitude simultaneously exceeds both the previous and backward position amplitudes, then the current processed second frequency position is marked and recorded as the peak position. The frequency index value 2 corresponding to this peak position is extracted, and a multiplication mapping operation is performed using a preset angle conversion step size coefficient. The preset angle conversion step size coefficient is obtained through a traversal scanning experiment of 20 standard polarization calibration light sources in a darkroom environment, with iterative testing in a step size of 1 degree within the parameter range of 10 to 30 degrees. When the parameter is set to 22.5 degrees, the absolute error between the reconstructed main peak angle and the theoretical light source angle is at a minimum of 0.5 degrees. Therefore, the preset angle conversion step size coefficient is assigned to 22.5 degrees. The extracted frequency index value 2 is multiplied by the preset angle conversion step size coefficient of 22.5 degrees, and the calculated value of the main peak polarization angle pointed to by this peak position is 45.0 degrees. This final result is established as the main peak angle index. The innovation of this operation logic lies in achieving peak spatial positioning by combining forward and backward difference determination with step size multiplication operation.

[0028] S203: Locate the direction data corresponding to the phase delay set based on the main peak angle index, and perform maximum and minimum value difference operation on all phase delay values ​​downwards. Then, perform numerical mapping on the obtained difference values ​​to generate birefringence response amplitude. Based on the generated main peak angle index value of 45.0 degrees, a fixed-point search operation is performed in the original three-dimensional spatial polarization data matrix to locate the data slice corresponding to the 45.0 degree polarization direction channel. All continuous phase delay values ​​stored in this direction are extracted. Data from 100 sampling nodes in this direction channel is retrieved, and the maximum and minimum delay parameters are obtained through numerical comparison. The maximum delay parameter in the current direction is found to be 1.8 micrometers, and the minimum delay parameter is 0.3 micrometers. A difference subtraction operation is performed on these extreme parameters, subtracting the minimum delay parameter from the maximum of 1.8 micrometers to 0.3 micrometers, resulting in an extreme difference data of 1.5 micrometers. This difference is then introduced... A numerical multiplication mapping operation was performed between the preset material response refractive index and the extreme value difference data. The preset material response refractive index was obtained through a transmittance comparison experiment on 50 standard cotton slices with known birefringence characteristics. Grid cross-validation was performed within the range of 0.1 to 1.0 with a step size of 0.05. When the coefficient was set to 0.8, the calculated response amplitude showed a peak linear correlation of 0.98 with the actual slice thickness. Therefore, the preset material response refractive index was fixed at 0.8. The extreme value difference data of 1.5 micrometers was multiplied with the preset material response refractive index of 0.8, resulting in a final mapping result of 1.2 micrometers. This result was directly defined as the birefringence response amplitude characterizing the optical characteristics of cotton fibers. The birefringence response amplitude of 1.2 micrometers was compared with the preset effective reference range of 0.5 to 2.5 micrometers. The result showed that 1.2 micrometers was within the effective reference range, indicating that the calculated response amplitude data belonged to the normal physical response range, ensuring that the result could be directly used to reflect the actual material characteristics. The innovation of this calculation logic lies in extracting the true internal stress index of the material through the multiplication of extreme value difference and material response coefficient.

[0029] Please see Figure 4 The specific steps of S3 are as follows: S301: Extract the brightness of the polarized image frame and retrieve the grayscale values ​​of multiple pixels. Compare the grayscale values ​​with the preset grayscale filtering threshold point by point. Record the coordinate index of the pixel whose grayscale value exceeds the grayscale filtering threshold. Arrange all the recorded coordinate indices according to their spatial positions to obtain the candidate skeleton coordinate sequence. The system acquires polarized image frames in a two-dimensional matrix format. It extracts the red, green, and blue channel values ​​from the image data structure. Color weighting coefficients are introduced, and numerical multiplication and summation operations are performed. The weighting coefficients for the red channel are set to 0.30, the green channel to 0.59, and the blue channel to 0.11. For a given pixel, the red channel value 100 is multiplied by 0.30 to obtain 30.0; the green channel value 150 is multiplied by 0.59 to obtain 88.5; and the blue channel value 200 is multiplied by 0.11 to obtain 22.0. Adding 30.0, 88.5, and 22.0 yields the current pixel's grayscale value of 140.5. The system then retrieves grayscale values ​​from multiple pixel locations and compares each grayscale value with a preset grayscale filtering threshold point by point. The preset grayscale screening threshold was obtained through histogram statistical testing of 50 typical cotton sample images. Iterative testing was conducted within the grayscale value range of 100 to 200, with a step size of 10. When the parameter was set to 150, the combined efficiency of background noise filtering and target retention reached its maximum while ensuring the target loss percentage remained extremely low. Therefore, the preset grayscale screening threshold was determined to be 150. The pixel grayscale value 180 corresponding to the aforementioned specific coordinate index 19205 was extracted and compared with the threshold 150. It was determined that the grayscale value 180 was strictly greater than the grayscale screening threshold 150, and the pixel coordinate index 19205 of this grayscale value was recorded. All recorded coordinate indices meeting the condition were arranged in a continuous queue according to the spatial horizontal and vertical coordinate increment rule to obtain the candidate skeleton coordinate sequence. The test calibration data is compiled into Table 2, the cotton image grayscale threshold calibration comparison table.

[0030] Table 2. Comparison Table of Gray-Scale Threshold Calibration for Cotton Images

[0031] Table 2 shows a detailed comparison of the filtering features under different grayscale threshold settings. The innovation of this operation logic lies in accurately removing redundant background pixels by comparing the weighted multiplication sum with a constant.

[0032] S302: Based on the candidate skeleton coordinate sequence, retrieve the spatial coordinate index of multi-pixel points and perform spatial eight-neighbor traversal to extract adjacent spatial points. Fit the connection between adjacent spatial points, sequentially splice the fitted line segments, and rearrange the index of associated points along the direction of the connection to construct the fiber path topology sequence. Based on the output candidate skeleton coordinate sequence, the spatial coordinate index data corresponding to multiple pixels within the sequence is retrieved. Substituting this into the previously extracted specific pixel coordinate index 19205, it is restored to a spatial center node with an x-coordinate of 5 and a y-coordinate of 10. An eight-neighbor traversal operation is performed on this node to extract adjacent spatial points. The x-coordinate and y-coordinate are respectively added by 1, subtracted by 1, and kept unchanged. After removing the center node itself, the coordinates of its eight surrounding adjacent positions are obtained. These adjacent position coordinates are compared with the recorded data items in the candidate skeleton coordinate sequence to extract adjacent spatial points existing within the sequence as associated nodes. The adjacent node with an x-coordinate of 6 and a y-coordinate of 11 is located. A line segment fitting operation is performed on the coordinate association between the center node and the adjacent spatial points to extract... The vertical difference between two nodes is calculated by subtracting their ordinates. Simultaneously, the horizontal difference is calculated by subtracting their abscissas. Subtracting the central node's ordinate (10) from the adjacent node's ordinate (11) yields a vertical difference of 1. Subtracting the central node's abscissa (5) from the adjacent node's abscissa (6) yields a horizontal difference of 1. Dividing these vertical and horizontal differences derives the slope of the line connecting the two points. Based on this slope, the fitted line segments are sequentially joined end-to-end. The coordinate indices of associated nodes are reordered along the line's direction, placing coordinate index 19205 adjacent to the coordinate index of the adjacent node. This process completes the node sorting of the entire connected region, constructing a fiber path topology sequence reflecting the physical orientation of the fiber. The innovation of this operational logic lies in reconstructing the topology of spatially discrete pixels through coordinate addition / subtraction traversal and slope division operations.

[0033] S303: Based on the fiber path topology sequence, call the birefringence response amplitude values ​​corresponding to the multi-pixel positions within the path, perform difference calculations on the amplitudes of adjacent path positions and accumulate and arrange them according to the path order, and simultaneously perform gradient mapping to generate a spatial phase gradient. Based on the generated fiber path topology sequence, and according to the order of the nodes in the sequence, the birefringence response amplitude values ​​extracted from the multi-pixel positions within the path are called one by one. The birefringence response amplitude value of 1.2 micrometers at the coordinate index 19205 calculated in the previous sequence is substituted as the preceding position amplitude parameter. Simultaneously, the birefringence response amplitude value of 1.5 micrometers corresponding to the immediately adjacent subsequent position coordinates is retrieved. A difference subtraction operation is performed on the amplitude values ​​of adjacent path positions, using the logic of subtracting the preceding position amplitude parameter from the subsequent position amplitude parameter. Subtracting 1.2 micrometers from 1.5 micrometers yields an amplitude difference of 0.3 micrometers. All calculated amplitude difference data are then continuously spliced ​​and cumulatively arranged according to the path connectivity order to form a sequence data structure. Subsequently, a multiplication operation was performed between a preset spatial gradient mapping coefficient and the amplitude difference data to perform gradient mapping. The preset spatial gradient mapping coefficient was obtained by tensile testing on 20 cotton fiber samples with known tensile strain distributions. Iterative testing was conducted in increments of 0.1 within the parameter range of 0.1 to 1.0. When the coefficient was set to 0.5, the absolute error between the virtual stress change rate and the data measured by the actual sensor reached a minimum of 2%. Therefore, the preset spatial gradient mapping coefficient was fixed at 0.5. The previously calculated amplitude difference data of 0.3 micrometers was multiplied with the gradient mapping coefficient of 0.5, and the comprehensive mapping result was calculated to be 0.15 micrometers per pixel. The mapping results corresponding to all nodes were arranged in sequence to generate a spatial phase gradient. The spatial phase gradient result of 0.15 micrometers per pixel was compared with the preset reasonable response range of 0.05 micrometers per pixel to 0.3 micrometers per pixel. It was determined that 0.15 micrometers per pixel was within this range, meaning that the currently extracted fiber spatial stress change characteristics were within the range of normal material deformation physical laws. The innovation of this operational logic lies in locating the strain concentration region inside the material by using path direction difference subtraction and mapping coefficient multiplication.

[0034] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the continuous fiber path, call the spatial phase gradient to perform extreme value search, arrange the phase gradient magnitudes of multiple pixels in the path, extract the center point corresponding to the current pixel gradient magnitude that exceeds the gradient magnitudes of the adjacent pixels on both sides as the extreme point, and split the continuous fiber path to obtain the abrupt segment index sequence. Based on the output fiber path topology sequence, the spatial phase gradient values ​​corresponding to each pixel position within the path are called one by one. The spatial phase gradient value of 0.15 micrometers per pixel corresponding to the coordinate index 19205 of the previous calculation is substituted into this value and used as the current central node data. Simultaneously, the spatial phase gradient value of 0.08 micrometers per pixel corresponding to the coordinate index 19204 of the preceding adjacent node is extracted by moving forward one unit step, and the spatial phase gradient value of 0.09 micrometers per pixel corresponding to the coordinate index 19206 of the following adjacent node is extracted by moving backward one unit step. The numerical relationship between the current pixel gradient magnitude and the gradient magnitudes of the adjacent pixels on both sides is then determined. It is determined that the central node value of 0.15 micrometers per pixel is strictly greater than the preceding node value of 0.08 micrometers per pixel and simultaneously strictly greater than the following node value of 0.09 micrometers per pixel. Therefore, coordinate index 19205 was established as a local extremum and marked. For each extremum point generated by the marking, a path segmentation buffer was introduced to perform continuous path splitting operations. The path segmentation buffer was obtained through cutting experiments on 50 sets of cotton fiber datasets with known fracture characteristics. Iterative testing was conducted with a step size of 1 pixel within a range of 1 to 10 pixels. When the parameter was set to 2 pixels, the breakpoint feature retention rate reached a maximum of 99%. Therefore, the path segmentation buffer was fixed at 2 pixels. Using the extremum point 19205 as the cutting center, a buffer distance of 2 pixels was extended forward and backward along the path to perform the breakage operation. The original long path was decomposed into multiple independent sub-path segments consisting of the preceding starting point to coordinate 19203 and the following starting point to the following ending point. The start and end coordinates of each segment after splitting were recorded sequentially to form a mutation segment index sequence. The gradient magnitude of the extremum point (0.15 micrometers per pixel) was compared with the preset gradient judgment benchmark (0.10 micrometers per pixel). Exceeding the benchmark indicated a real structural mutation at that location. The advantage of this operational logic is that it achieves precise stripping of fiber stress concentration areas by comparing forward and backward gradient magnitudes in combination with buffer splitting logic.

[0035] S402: Extract the phase delay value set within the corresponding interval based on the abrupt segment index sequence, compare the multi-pixel phase delay values ​​within the set point by point, map the numerical size relationship into an ordered sequence, and select the phase delay value corresponding to the position with the largest value in the sequence as a representative parameter to generate the maximum phase delay value set. Based on the obtained mutation segment index sequence, all pixel coordinates within each independent interval are extracted. Substituting these coordinates into a specific mutation segment with a starting coordinate of 19207 and an ending coordinate of 19215, the phase delay values ​​corresponding to multiple pixels within this interval are extracted point-by-point in the global storage space to construct a phase delay value set. The values ​​contained in this set are 1.5 μm, 1.8 μm, 1.2 μm, and 1.6 μm. For each multi-pixel phase delay value in the set, a value comparison operation is performed. Any two values ​​are compared by difference to determine their relative size. A bubble sort logic is used to map the value size relationship into a descending ordered sequence, resulting in the sorted values ​​of 1.8 μm, 1.6 μm, 1.5 μm, and 1.2 μm. Subsequently, the phase delay value of 1.8 μm, corresponding to the largest value at the first position in the ordered sequence, is selected as the... To compensate for the numerical attenuation caused by local light scattering, a preset delay compensation amplification factor is introduced and multiplied with this representative parameter. This preset delay compensation amplification factor is obtained through optical projection comparison experiments on 100 anisotropic fibers that have undergone rigorous physical measurements. The factor is increased in increments of 0.05 within a parameter range of 1.0 to 1.5. When the factor is set to 1.15, the mean square error between the virtual compensation result and the actual physical thickness reaches a minimum of 0.03 micrometers. Therefore, the preset delay compensation amplification factor is set to 1.15. The extracted maximum phase delay value of 1.8 micrometers is multiplied with the preset delay compensation amplification factor of 1.15, resulting in a final corrected representative parameter of 2.07 micrometers. This extreme value extraction and multiplication correction operation is performed iteratively across all abrupt changes, and all the obtained corrected representative parameters are summarized to generate a maximum phase delay value set. The calculated corrected representative parameter of 2.07 micrometers is compared with the preset effective thickness lower limit of 1.0 micrometers. The result is confirmed to be higher than the lower limit, indicating that the extracted data has reliable material characteristic discrimination. The advantage of this operational logic is that it effectively filters out local minor fluctuation noise by combining interval extreme value extraction with multiplication compensation operation.

[0036] S403: Perform intensity reconstruction calculation based on the maximum phase delay value set, map multiple phase delay values ​​to corresponding intensity values ​​and filter the pixel index set that meets the preset intensity threshold, determine the spatial neighborhood connectivity of the pixel index set and aggregate the coordinates of adjacent pixels to generate the global coordinates of the transparent heterogeneous fiber. The physical intensity reconstruction calculation is performed based on the maximum phase delay value set of the output. A specific correction representative parameter of 2.07 micrometers is extracted from the set. The optical path intensity mapping constant is introduced and directly multiplied with this correction representative parameter to map it to the corresponding intensity value. The optical path intensity mapping constant is obtained from the transmitted light intensity measurement experiment of 30 standard transparent nylon asymmetrical fiber sheets in a standard darkroom. The grid search iteration test is performed in the range of 10 candela per square meter per micrometer to 100 candela per square meter per micrometer with a step size of 5. When the constant is set to 40 candela per square meter per micrometer, the reconstruction intensity error rate is at a minimum of 3%. The optical path intensity mapping constant is determined to be set to 40 candela per square meter per micrometer. The correction representative parameter of 2.07 micrometers is multiplied with 40 candela per square meter per micrometer, and the reconstruction intensity value of the target pixel is derived to be 82.8 candela. The reconstructed intensity value of 82.8 candela per square meter was compared with a preset intensity screening threshold of 75.0 candela per square meter. The pixel coordinate index corresponding to this value was recorded and stored in a pixel index set. For all isolated coordinate points stored in this pixel index set, a spatial neighborhood connectivity determination operation was performed. Starting with any retained coordinate point, the spatial coordinates of its four adjacent positions (up, down, left, and right) were checked to see if they existed in the current pixel index set. If they existed, they were considered to be in the same connected region, and adjacent pixel coordinate aggregation was performed. This process was repeated to merge all coordinate points that met the connectivity conditions, ultimately generating the global coordinates of the transparent anisotropic fiber, representing its physical boundary. The reconstructed intensity value of 82.8 candela per square meter was compared with the theoretically high impurity background value of 60.0 candela per square meter, confirming a clear distinction from the background. This provided direct data support for the accurate delineation of the final coordinates.

[0037] Please see Figure 6 The specific steps of S5 are as follows: S501: Calculate the angle difference of neighboring pixels by calling the cotton phase main direction for the global coordinates of the transparent heterogeneous fiber. Extract the angle of neighboring pixels for each coordinate point and calculate the difference with the corresponding main direction angle. Determine the difference result and classify the attributes according to the preset angle segmentation interval threshold to obtain the angle difference group index set. The generated transparent heterogeneous fiber global coordinate data is used to extract specific coordinate positions, namely x-coordinate 6 and y-coordinate 11. The cotton phase principal direction in the current environment is then used to determine the global fiber position coordinates. This cotton phase principal direction is obtained by substituting the pre-sequence phase delay sequence into a frequency domain analysis and selecting the angle corresponding to the main peak of the spectral amplitude, which is directly obtained as 45.0 degrees. Subsequently, a neighboring pixel within the spatial neighborhood of the center pixel is extracted, and its corresponding neighboring pixel angle is obtained as 52.0 degrees. For each coordinate point, the neighboring pixel angle is extracted and the difference is calculated with the corresponding principal direction angle, i.e., the neighboring pixel angle 52.0 degrees is subtracted from the cotton phase principal direction angle 45.0 degrees. The calculated angle difference is 7.0 degrees. A preset angle segmentation interval threshold is introduced to determine the difference result and classify its attributes. This preset angle segmentation interval threshold is obtained through clustering and annotation experiments on a dataset of 200 cotton samples, using a sliding window with a step size of 5.0 degrees within the range of 5.0 degrees to 30.0 degrees. In iterative testing, the classification error rate reached a minimum of 3% when the upper limit of the first segment interval was set to 10.0 degrees and the upper limit of the second segment interval was set to 25.0 degrees. The first segment interval was determined to be 0.0 degrees to 10.0 degrees, and the second segment interval was determined to be 10.0 degrees to 25.0 degrees. The angle difference result of 7.0 degrees was compared with the upper and lower boundaries of each interval. It was determined that 7.0 degrees was within the first segment interval of 0.0 degrees to 10.0 degrees, and it was classified into the first group. The corresponding coordinate position and group label were recorded. All coordinate points were processed iteratively to obtain the angle difference group index set. The calculated angle difference result of 7.0 degrees was compared with the upper limit of the interval of 10.0 degrees. It was confirmed that it was less than the upper limit, indicating that the physical arrangement direction of the current pixel point was consistent with the main direction. This provided a direct classification basis for subsequent allocation calculations. The advantage of this operation logic is that it completes the fine determination of the orientation attribute of the fiber microstructure by combining the angle mean division and difference operation with dynamic boundary comparison.

[0038] S502: Extract the angle deviation corresponding to the pixel in multiple groups based on the angle difference group index set, call the angle deviation reciprocal operation rule to transform the angle deviation of multiple pixels by reciprocal, use the reciprocal as the weight to pair with the pixel position index and accumulate at the corresponding position, and superimpose the weights of multiple groups to obtain the weight aggregate coordinate mapping table. Based on the output angle difference grouping index set, the angle deviations corresponding to multiple pixel groups within the first group are extracted from the system memory structure. Substituting this into the previously calculated angle deviation of 7.0 degrees corresponding to the x-coordinate 6 and y-coordinate 11 positions, a preset angle deviation reciprocal operation rule is invoked to perform a reciprocal transformation on the multi-pixel angle deviations. To mitigate the overflow risk of a zero denominator, a fixed smoothing constant factor of 1.0 degrees is introduced. The extracted angle deviation of 7.0 degrees is added to the constant factor of 1.0 degrees to obtain a smoothed deviation value of 8.0 degrees. Then, the constant value of 1.0 is divided by the smoothed deviation value of 8.0 degrees, deriving a reciprocal angle deviation of 0.125 degrees per degree. This calculated reciprocal value is then compared with the pixel position... The index is set to 6 x and 11 y. Data is paired with columns. The global hash map is used to check if the index at this location contains historical cumulative data. If the location already has a historical inverse value of 0.100 per degree in other neighborhood association calculations, the new inverse value of 0.125 per degree is added to the historical inverse value of 0.100 per degree. The combined total cumulative weight is 0.225 per degree. This inverse division and historical weight addition operation is repeated for all pixel locations within each group. All pixel coordinates and their cumulative results are arranged according to memory address order, resulting in a weighted aggregated coordinate mapping table. The data obtained from the test analysis are arranged as shown in Table 3, the weighted aggregated coordinate mapping table.

[0039] Table 3 Weight Aggregate Coordinate Mapping Table

[0040] As shown in Table 3, the final weight distribution of multiple spatial coordinate nodes after undergoing nonlinear transformation and neighborhood iterative accumulation is listed in detail. The calculated accumulated weight of 0.225 per degree is compared with the preset effective response lower limit of 0.150 per degree. It is determined that 0.225 per degree is strictly greater than the lower limit parameter, which means that the pixel position has extremely prominent structural variation response features in the multi-directional feature fusion process. The advantage of this operation logic is that it highlights the feature weight of small angle variation regions through nonlinear amplification by reciprocal division and superposition of multi-dimensional historical data.

[0041] S503: Retrieve the cumulative weight values ​​according to the weight aggregation coordinate mapping table and compare their magnitudes. Extract the set of coordinate points corresponding to the largest weight values. Deduplicate and remove duplicate coordinate points in the global coordinates of the transparent foreign fiber, retain the unique coordinate combination, and generate the position coordinates of the transparent foreign fiber. Based on the generated weighted aggregated coordinate mapping table, the cumulative weight value corresponding to each coordinate node is retrieved cyclically in the data storage queue. This is substituted into the cumulative weight of 0.225 per degree corresponding to the x-coordinate 6 and y-coordinate 11 in the preceding table entry. Simultaneously, the cumulative weight of 0.315 per degree corresponding to the x-coordinate 7 and y-coordinate 12 of the subsequent adjacent node is extracted. These two values ​​are then compared in a comparator. If 0.315 per degree is strictly greater than 0.225 per degree, a quicksort mechanism is introduced to swap the positions of all coordinate nodes in the mapping table in descending order based on their cumulative weight. The top 50 coordinate points with the highest weight values ​​in the sorted result queue are extracted. For each data item in this high-weight point set, duplicate coordinate points in the global fiber position coordinates are extracted for deduplication and removal. The x-coordinate 7 and y-coordinate 12 of the currently extracted node are read and converted into plain text. This character format is concatenated to generate a unique verification code. A string matching search is performed in the confirmed historical record database. If a matching verification code with an x-coordinate of 7 and y-coordinate of 12 already exists in the historical record database, the current node is considered a redundant duplicate, and memory destruction is immediately performed to remove the node. If no match is found, it is loaded into the final retention database. After traversing all 50 nodes, a unique coordinate combination without any duplicates is retained. This is then encapsulated in a data structure to generate transparent anisotropic fiber position coordinates. The extracted high-weight data of 0.315 per degree is compared across segments with the extremely high background noise weight value of 0.200 per degree calibrated using a clean blank sample. The comparison confirms that 0.315 per degree significantly exceeds the distribution range of clean background noise, indicating that the final selected spatial coordinates directly eliminate false feature interference caused by random refraction of stray light from the environment.

[0042] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying transparent foreign fibers in cotton using polarization detection in combination with image processing, characterized by, Includes the following steps: S1: Obtain polarized image frames of the cotton detection scene through a polarization sensor and extract time markers and polarization angles. Arrange the pixels of the polarized image frames according to the time markers and calculate the phase delay of the corresponding pixels based on the polarization angle to construct a phase delay sequence. S2: Extract the corresponding phase delay set in pixel space from the phase delay sequence and perform frequency domain analysis. Select the angle corresponding to the main peak of the spectral amplitude as the main direction of the cotton phase, calculate the extreme difference in the phase delay set, and generate the birefringence response amplitude. S3: Extract the brightness of the polarized image frame and select the pixels that meet the preset grayscale threshold as candidate skeleton points. Connect the candidate skeleton points and analyze the fiber continuous path. Combine the birefringence response amplitude to perform differential analysis and generate a spatial phase gradient. S4: Based on the continuous fiber path, call the spatial phase gradient to perform extreme value search, divide the abrupt segment within the continuous fiber path and extract the maximum phase delay for intensity reconstruction calculation, extract the pixel points corresponding to the reconstruction values ​​that meet the preset intensity threshold for spatial connectivity judgment, and generate the global coordinates of the transparent heterogeneous fiber.

2. The cotton transparent foreign fiber recognition method of claim 1, wherein The phase delay sequence includes phase delay, time series index, and pixel position mapping; the birefringence response amplitude includes extreme difference amplitude, amplitude distribution characteristics, and response intensity; the spatial phase gradient includes gradient amplitude, gradient direction, and spatial rate of change; and the global coordinates of the transparent anisotropic fiber include a set of spatial position coordinates, connected region identifiers, and structural distribution range.

3. The cotton transparent foreign fiber recognition method of claim 1, wherein the polarized light detection is combined with image processing. The specific steps of S1 are as follows: S101: The polarization image frame of the cotton detection scene is acquired by the polarization sensor and the associated time marker and polarization angle of the polarization image frame are extracted. The pixel coordinates within each polarization image frame are encoded. The indexes of the same coordinates are sorted according to the time marker and rearranged into a pixel vector sequence to obtain pixel time sequence arrangement data. S102: Based on the pixel time sequence arrangement data, call the polarization angle of the corresponding pixel position at multiple time points, perform angle difference on the polarization angle at multiple time points, perform periodic normalization on the difference results, and arrange the normalized angle sequence in order to obtain a polarization angle difference sequence set. S103: Extract the angle difference sequence corresponding to each pixel according to the polarization angle difference sequence set, call the adjacent angle differences to perform trigonometric function conversion and combine them with the preset optical path difference reference value for numerical mapping, and combine and arrange the mapping results in time order to obtain the phase delay sequence.

4. The polarized detection combined with image processing method for identifying the transparent anisotropic fiber of cotton according to claim 3, characterized in that, The optical path difference reference value is determined by multiplying the center wavelength value of the incident light source in the detection scene with the birefringence coefficient of a preset standard medium, extracting the initial phase offset of the stress-free environment and converting it into an equivalent optical path difference as a compensation factor, and then adding the product result with the compensation factor.

5. The polarized detection combined with image processing method for identifying the transparent anisotropic fiber of cotton according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Extract the corresponding phase delay set in pixel space from the phase delay sequence and perform discrete Fourier transform to map the phase delay value sequence to the frequency component sequence. At the same time, perform amplitude square accumulation processing on multiple frequency components and rearrange them according to the frequency index order to obtain the frequency amplitude distribution sequence. S202: Based on the frequency amplitude distribution sequence, retrieve the amplitude data corresponding to multiple frequency positions, calculate the difference between the amplitudes of adjacent frequency positions, record the frequency positions where the current amplitude exceeds both the amplitude of the previous position and the amplitude of the next position as peak positions and convert them into angle data to obtain the main peak angle index; S203: Locate the direction data corresponding to the phase delay set according to the main peak angle index, and perform maximum and minimum value difference operation on all phase delay values ​​downward, and perform numerical mapping on the obtained difference value to generate birefringence response amplitude.

6. The polarized detection combined with image processing method for identifying the transparent anisotropic fiber of cotton according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Extract the brightness of the polarized image frame and retrieve the grayscale values ​​of multiple pixels. Compare the grayscale values ​​with the preset grayscale filtering threshold point by point. Record the coordinate index of the pixel whose grayscale value exceeds the grayscale filtering threshold. Arrange all the recorded coordinate indices according to their spatial positions to obtain the candidate skeleton coordinate sequence. S302: Based on the candidate skeleton coordinate sequence, retrieve the spatial coordinate index of multi-pixel points and perform spatial eight-neighbor traversal to extract adjacent spatial points, fit the connection between adjacent spatial points, sequentially splice the fitted line segments and rearrange the index of associated points along the direction of the connection to construct the fiber path topology sequence. S303: According to the fiber path topology sequence, call the birefringence response amplitude value corresponding to the multi-pixel position in the path, perform difference calculation on the amplitude of adjacent path positions and accumulate and arrange them in the path order, and at the same time perform gradient mapping to generate spatial phase gradient.

7. The polarized detection combined with image processing method for identifying the transparent anisotropic fiber of cotton according to claim 6, characterized in that, The grayscale threshold is determined by acquiring the global pixel grayscale values ​​of the polarized image frame, summing all pixel grayscale values ​​and dividing by the total number of pixels to obtain the global grayscale average value, calculating the sum of squares of the differences between the multi-pixel grayscale values ​​and the global grayscale average value to obtain the grayscale standard deviation, and then performing a weighted summation operation on the global grayscale average value and the grayscale standard deviation together with a preset weighting coefficient.

8. The polarized detection combined with image processing method for identifying the transparent anisotropic fiber of cotton according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the continuous fiber path, the spatial phase gradient is called to perform extreme value search. The phase gradient magnitudes of multiple pixels in the path are arranged. The center point corresponding to the current pixel gradient magnitude exceeding the gradient magnitudes of the adjacent pixels on both sides is extracted as the extreme point. The continuous fiber path is then split to obtain the abrupt segment index sequence. S402: Extract the phase delay value set within the corresponding interval according to the abrupt segment index sequence, compare the phase delay values ​​of multiple pixels in the set point by point, map the numerical size relationship into an ordered sequence, and select the phase delay value corresponding to the position of the largest value in the sequence as a representative parameter to generate the maximum phase delay value set. S403: Perform intensity reconstruction calculation based on the maximum phase delay value set, map multiple phase delay values ​​to corresponding intensity values ​​and filter the pixel index set that meets the preset intensity threshold, determine the spatial neighborhood connectivity of the pixel index set and aggregate the coordinates of adjacent pixels to generate the global coordinates of the transparent heterogeneous fiber.

9. The polarized detection combined with image processing method for identifying the transparent anisotropic fiber of cotton according to claim 1, characterized in that, The method further includes: S5: Calculate the angle difference of neighboring pixels by calling the cotton phase main direction for the global coordinates of the transparent foreign fiber and classify the attributes. Construct a direction group set and perform a weighted summation based on the reciprocal of the angle deviation. Extract the coordinate point corresponding to the maximum weighted summation value and perform redundancy removal on the global coordinates of the transparent foreign fiber to generate the position coordinates of the transparent foreign fiber. The location coordinates of the transparent heterogeneous fiber include a set of filtered coordinate points, a direction consistency marker, and a redundant space index.

10. The method for identifying transparent heterogeneous fibers in cotton by combining polarization detection and image processing according to claim 9, characterized in that, The specific steps of S5 are as follows: S501: Calculate the neighboring pixel angle difference by calling the cotton phase main direction for the global coordinates of the transparent heterogeneous fiber. Extract the neighboring pixel angle for each coordinate point and calculate the difference with the corresponding main direction angle. Determine the difference result and classify the attributes according to the preset angle segmentation interval threshold to obtain the angle difference grouping index set. S502: Based on the angle difference group index set, extract the angle deviation corresponding to the pixel in the multi-group, call the angle deviation reciprocal operation rule to transform the angle deviation of the multi-pixel inverse, use the reciprocal as the weight to pair with the pixel position index and accumulate at the corresponding position, and superimpose the weights of the multi-group to obtain the weight aggregate coordinate mapping table. S503: Retrieve the cumulative weight values ​​according to the weight aggregation coordinate mapping table and compare their magnitudes. Extract the set of coordinate points corresponding to the largest weight values. Deduplicate and remove duplicate coordinate points that appear repeatedly in the global coordinates of the transparent foreign fiber, retain the unique coordinate combination, and generate the position coordinates of the transparent foreign fiber.

Images