Rapid soil micro-plastic pre-estimation method based on mobile phone super-microspur imaging

Abstract

The invention relates to the technical field of soil pollutant evaluation, and discloses a soil micro-plastic rapid estimation method based on mobile phone super-macro imaging, which comprises the following steps: pre-treating a soil sample and enriching the soil sample in a filter membrane; under a fixed condition, shooting a filter membrane image containing a scale mark by using a super-macro function of the mobile phone; scale correction and geometric distortion correction are carried out based on scale marks through an image processing technology, and then the number and size distribution of suspected micro-plastic particles are identified and counted through background suppression, threshold segmentation and form and optical feature screening; and calculating an abundance index in combination with the sample amount, and carrying out risk grade matching and output in combination with particle size distribution characteristics. According to the method, rapid and low-cost screening and risk estimation of soil micro-plastic pollution are realized by utilizing popular smart phone equipment, the dependence on large professional instruments is reduced, and the method is suitable for a field or basic detection scene with a large sample size.

Description

Technical Field

[0001] This invention relates to the field of soil pollutant assessment technology, and more specifically, to a rapid prediction method for soil microplastics based on mobile phone super macro imaging. Background Technology

[0002] Microplastic particles in soil environments are characterized by complex origins, discrete spatial distribution, and a wide range of particle sizes. Common morphologies include fragments, fibers, and films. To characterize the content or abundance of microplastic particles in soil, current techniques typically rely on spectroscopic confirmatory methods to identify the chemical composition and count the particles after sample pretreatment. For example, microscopic infrared imaging detection based on Fourier transform infrared spectroscopy can scan and collect particles on the surface of filter media and match them with a spectral library, thereby distinguishing polymer particles from natural particles and further completing counting and size statistics.

[0003] However, the above-mentioned testing routes often require the configuration of specialized instruments, which have high purchase and maintenance costs and are used in laboratory settings, making it difficult to cover the routine monitoring needs of multiple locations and large sample sizes.

[0004] Meanwhile, existing methods often require particle-by-particle spectral acquisition or long-term spectral imaging scans at the single-sample level. When the sample size increases from tens to hundreds or more, the detection cycle is lengthened, making it difficult to meet the timeliness requirements for rapid assessment and screening.

[0005] In addition, for rapid screening scenarios, there are also semi-quantitative screening methods based on fluorescent staining, such as using fluorescent dyes to selectively label suspected plastic particles and using a fluorescent imaging system to statistically analyze particle characteristics. Although these methods can increase throughput, they are sensitive to matrix residues and staining conditions, and still rely on dedicated fluorescent imaging equipment, thus limiting their applicability.

[0006] In actual management and investigation, there are many application needs that only require completing pollution level classification or risk screening. They often focus more on the abundance characterization and classification results that can support decision-making. However, existing technologies are still insufficient in achieving the above objectives under the conditions of low cost, high throughput and scalability. Summary of the Invention

[0007] In view of this, the present invention proposes a rapid prediction method for soil microplastics based on mobile phone super macro imaging, aiming to solve the problems that existing soil microplastic detection methods generally have high dependence on special instruments, long processing and analysis time for single samples, and difficulty in achieving rapid abundance prediction and risk classification output under large sample size and basic conditions.

[0008] In one aspect, this invention proposes a rapid prediction method for soil microplastics based on mobile phone super macro imaging, comprising: The collected soil samples were air-dried and sieved; the sieved soil samples were then subjected to density separation and organic matter removal treatments; the suspended phase obtained from the density separation treatment was filtered to obtain an enrichment filter membrane loaded with suspected microplastic particles, and the soil sample mass and volume were recorded. The enrichment filter membrane is placed on the carrier support surface, and scale marks are set on the carrier support surface or in the imaging field of view adjacent to the carrier support surface; the original image is acquired under fixed shooting distance and consistent illumination conditions, and the original image includes the enrichment filter membrane image and the scale mark image; The original image is imported into the image processing software, the scale conversion parameters are calculated based on the scale-marked image, and the scale conversion parameters are used to perform scale correction on the enrichment filter image to obtain the scale-corrected image. The scale-corrected image is preprocessed, thresholded, and identified sequentially to obtain a candidate particle set. The number of particles in the candidate particle set is counted, and the particle size distribution is obtained by size conversion and segmented statistics based on the scale conversion parameters. Microplastic abundance index was calculated based on particle number, particle size distribution, soil sample mass, and soil sample volume. Obtain the risk classification boundary conditions corresponding to the microplastic abundance index, match the microplastic abundance index with the risk classification boundary conditions, and output the microplastic pollution risk level.

[0009] Furthermore, an imaging condition locking step is included before acquiring the original image. The imaging condition locking step includes: After the enrichment filter membrane is fixed on the carrier support surface, the focus position, exposure parameters and white balance parameters of the mobile phone camera are locked. Multiple frames of raw images were continuously acquired under the locked imaging conditions, and sharpness characterization parameters were calculated for each frame of raw images. Sharpness characterization parameters include edge intensity characterization parameters and high-frequency energy characterization parameters. The original image whose sharpness representation meets the screening criteria is selected from the original images of multiple frames as the target original image. The screening criteria include: the edge intensity representation of the target original image is located in the upper quantile interval of the corresponding edge intensity representation of the original images of multiple frames, and the high frequency energy representation of the target original image does not show a monotonically decreasing trend relative to the high frequency energy representation of the original images of multiple frames.

[0010] Furthermore, when calculating the scale conversion parameters based on the scale-marked image, the following steps are included: Locate the geometric primitives of the scale markers in the scale-marked image. The geometric primitives are pairs of line segment endpoints or pairs of adjacent points in a lattice. Calculate the pixel spacing for any set of geometric primitives to form a pixel spacing set; Perform a consistency check on the pixel spacing set and remove pixel spacings that deviate from the central tendency of the pixel spacing set; Based on the pixel spacing set after consistency verification and the calibration length of the scale marker, a conversion relationship from pixel to actual length is established to obtain the scale conversion parameters.

[0011] Furthermore, when performing scale correction on the enrichment filter image using scale conversion parameters, the following steps are included: Locate the boundary of the enriched filter region in the original image and crop the image of the enriched filter region; extract the boundary direction set based on the boundary of the enriched filter region; The boundary direction sets are grouped into two intersecting direction groups based on their similarity in direction. The two intersecting direction groups correspond to two adjacent boundary directions of the enrichment filter membrane region, respectively. Determine whether the angle between two intersecting direction groups satisfies the orthogonality consistency condition; if the angle does not satisfy the orthogonality consistency condition, determine that the image of the enriched filter region has geometric distortion, and perform geometric correction on the image of the enriched filter region based on the two intersecting direction groups; The scale-corrected image is obtained by performing scale normalization on the geometrically corrected enriched filter region image based on the scale conversion parameters.

[0012] Furthermore, image preprocessing of the scale-corrected image includes: Image preprocessing for scale-corrected images includes: Background candidate regions are determined in the scale-corrected image. The background candidate regions are continuous regions with uniform filter texture and local contrast that meet a preset contrast threshold. The local contrast is calculated as follows: the scale-corrected image is divided into sliding windows of a preset size, and the ratio of the gray range to the gray mean in each sliding window is calculated to obtain the window contrast. When the window contrast is less than or equal to the preset contrast threshold, the region corresponding to the sliding window is determined to be a low-contrast region. Continuous low-contrast regions that meet the preset area threshold are determined as background candidate regions. Background brightness surfaces are extracted based on background candidate regions, and the background brightness surfaces are used as background compensation benchmarks to perform background compensation on scale-corrected images in order to suppress uneven illumination. Noise suppression and contrast enhancement are performed on the image after background compensation to obtain a preprocessed image for threshold segmentation.

[0013] Furthermore, when performing thresholding segmentation on the scale-corrected image, the following steps are included: Calculate the grayscale histogram of the preprocessed image and extract the peak-valley structure of the histogram; The segmentation threshold is determined based on the peak-valley structure. The segmentation threshold is taken as the gray value corresponding to the valley position between the background peak and the particle peak or the representative gray value in the neighborhood of the valley position. A binary segmentation map is generated based on the segmentation threshold, and connectivity correction processing is performed on the binary segmentation map, including hole filling processing and boundary smoothing processing, to obtain a particle mask map.

[0014] Furthermore, particle identification yields a candidate particle set, including: Perform connected component labeling on the particle mask graph to obtain a set of connected components; For each connected component in the set of connected components, extract the contour and calculate the morphological feature set, which includes area features, perimeter features, aspect ratio features and roundness features. Morphological consistency determination is performed based on morphological feature groups. Morphological consistency determination includes comparing the morphological features corresponding to connected components with the morphological feature distribution of other connected components in the same image, and removing connected components that fall into discrete edge regions. The retained connected components are determined as the candidate particle set.

[0015] Furthermore, a pseudo-particle removal step is included before statistically analyzing the candidate particle set. The pseudo-particle removal step includes: For each candidate particle in the candidate particle set, an optical consistency feature group is extracted. The optical consistency feature group includes at least color channel distribution features, edge gradient stability features, and local texture uniformity features. The pseudo-particle determination is performed based on the optical consistency feature group. The pseudo-particle determination includes comparing the color channel distribution features of the candidate particles with the color channel distribution features of the background candidate region, and combining the edge gradient stability features to determine whether the boundary of the candidate particles exhibits a unilateral abrupt change characteristic caused by reflection or shadow. Candidate particles that are identified as pseudo particles are removed from the candidate particle set, and the remaining candidate particles are determined as the candidate particle set for particle number statistics and particle size distribution calculation.

[0016] Furthermore, calculating the particle size distribution includes: For each candidate particle in the candidate particle set, calculate the equivalent particle size and principal axis length at the pixel scale; The equivalent grain size and principal axis length at the pixel scale are converted to the equivalent grain size and principal axis length at the actual scale based on the scale conversion parameters. The particle size distribution is obtained by segmenting the equivalent particle size at the actual scale according to the particle size segmentation rule. The particle size segmentation rule is generated based on the distribution pattern of the particle size distribution. The particle size segmentation rule includes identifying the aggregation intervals according to the distribution pattern and using the boundaries of the aggregation intervals as the segmentation boundaries.

[0017] Furthermore, the risk levels for exported microplastic pollution include: Obtain the risk classification boundary conditions, which are simultaneously associated with the microplastic abundance index range and the structural feature range of particle size distribution. The structural feature range of particle size distribution includes the range of fine particle proportion and the range of coarse particle proportion. A risk level set is constructed based on the risk classification boundary conditions. The risk level set includes at least the first risk level, the second risk level, and the third risk level. The risk level is determined by jointly matching the microplastic abundance index with the structural feature intervals of particle size distribution according to a progressive judgment order, where: When the microplastic abundance index falls into the first abundance range and the proportion of fine particles falls into the first fine particle proportion range, it is determined to be the first risk level. When the microplastic abundance index falls into the second abundance range, or when the microplastic abundance index falls into the first abundance range and the fine particle proportion falls into the second fine particle proportion range, it is determined to be the second risk level. When the microplastic abundance index falls into the third abundance range, or the microplastic abundance index falls into the second abundance range and the proportion of fine particles falls into the second fine particle proportion range, or the proportion of coarse particles falls into the third coarse particle proportion range, it is determined to be the third risk level. Output the microplastic pollution risk level and write the microplastic abundance index, particle number, particle size distribution and scale conversion parameters into the detection result record to obtain a traceable record.

[0018] Compared with existing technologies, the advantages of this invention are as follows: By uniformly loading suspected microplastic particles from soil samples after separation and enrichment onto an enrichment filter membrane, and acquiring original images containing the enrichment filter membrane and scale markers under fixed shooting distance and consistent illumination conditions, the imaging inputs of different samples have comparable geometric scales and illumination responses; furthermore, by calculating scale conversion parameters based on the scale marker images and performing scale correction on the enrichment filter membrane images, the size conversion and segmented statistics of candidate particles are established on a unified pixel-to-actual length mapping relationship, thereby reducing the impact of slight deviations in shooting distance and differences in imaging magnification on particle size statistics and abundance estimation. To mitigate biases, image preprocessing, threshold segmentation, and particle identification are sequentially performed on scale-corrected images to form a candidate particle set and obtain particle quantity and size distribution. This allows the microplastic abundance index to be calculated by combining particle quantity structure and size structure, reducing the loss of structural information caused by relying solely on single quantity statistics. Finally, the microplastic abundance index is matched with risk classification boundary conditions to output the microplastic pollution risk level. This enables the rapid prediction results to be directly converted into risk outputs that can be used for screening and graded management. Furthermore, the consistency and repeatability of the prediction process are improved through a continuous processing chain of data collection, scale correction, identification statistics, abundance calculation, and risk matching. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 The flowchart illustrates a method for rapid prediction of soil microplastics based on mobile phone super macro imaging, as provided in an embodiment of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] See Figure 1 As shown, this application proposes a rapid prediction method for soil microplastics based on mobile phone super macro imaging, including: S1: The collected soil samples were air-dried and sieved; the sieved soil samples were subjected to density separation and organic matter removal treatments in sequence; the suspended phase obtained from the density separation treatment was filtered to obtain an enrichment filter membrane loaded with suspected microplastic particles, and the soil sample mass and volume were recorded. S2: Place the enrichment filter membrane on the carrier support surface and set scale marks on the carrier support surface or in the imaging field of view adjacent to the carrier support surface; acquire the original image under fixed shooting distance and consistent illumination conditions, the original image includes the enrichment filter membrane image and the scale mark image; S3: Import the original image into the image processing software, calculate the scale conversion parameters based on the scale-marked image, and use the scale conversion parameters to perform scale correction on the enrichment filter image to obtain the scale-corrected image. S4: Perform image preprocessing, threshold segmentation, and particle recognition on the scale-corrected image in sequence to obtain a candidate particle set; count the number of particles in the candidate particle set, and perform size conversion and segmented statistics on the candidate particle set based on the scale conversion parameters to obtain the particle size distribution. S5: Calculate microplastic abundance index based on particle number, particle size distribution, soil sample mass, and soil sample volume; S6: Obtain the risk classification boundary conditions corresponding to the microplastic abundance index, match the microplastic abundance index with the risk classification boundary conditions, and output the microplastic pollution risk level.

[0022] Specifically, air drying and sieving treatment refers to removing free moisture and breaking up agglomerations from the collected soil samples under natural ventilation or constant temperature conditions, transforming the soil samples from a moist and sticky state to a dispersed state. Then, stones, plant debris, and large-diameter impurities are separated from the soil samples using a sieve. The sieve aperture can be selected based on the sample particle size background, with a 5mm sieve being a preferred embodiment. Density separation treatment involves mixing the sieved soil sample with a density separation medium, thoroughly shaking and mixing, and then allowing it to stand or centrifuge. This allows high-density mineral particles to settle, while low-density suspected microplastic particles enter the suspended phase, forming a collectable supernatant suspension. In a preferred embodiment, the separation medium can be a saturated sodium chloride solution, and the density separation process can be repeated and the suspended phases can be combined to improve enrichment consistency. Organic matter removal treatment refers to performing oxidation or chemical digestion on the system during the density separation process to weaken the adhesion interference of humic substances, colloidal organic matter, and fibrous residues on the particle boundaries. In a preferred embodiment, 30% H2O2 or Fenton's reagent can be used, and density separation and filtration can be performed again after digestion. Filtration of the suspended phase obtained from the density separation treatment refers to passing the suspended phase through a filter membrane to retain particles and form an enrichment filter membrane loaded with particles. In a preferred embodiment, the pore size of the filter membrane can be 0.The 45µm enrichment filter membrane, with particles on its surface, serves as the target for subsequent imaging and identification. Simultaneously, soil sample mass and volume are recorded to establish a conversion relationship between particle statistics and sample baseline quantities. During the imaging stage, the enrichment filter membrane is placed on a carrier support surface, and scale markers are set on the carrier support surface or within the imaging field of view adjacent to it. These scale markers are reference graphics with known calibration lengths or known dot distances, used to convert image pixel scales to actual scales. The fixed shooting distance and consistent illumination conditions in the original image acquisition refer to maintaining a constant shooting distance between the mobile phone camera and the enrichment filter membrane while keeping the illumination direction and illuminance stable. In a preferred embodiment, the fixed shooting distance can be 4cm, ensuring comparable geometric relationships and brightness responses between the enrichment filter membrane image and the scale marker image. After importing the original image into image processing software, scale conversion parameters are calculated based on the scale marker image and used to perform scale correction on the enrichment filter membrane image, resulting in a scale-corrected image with a unified scale baseline. In a preferred embodiment, ImageJ can be used to perform image preprocessing, threshold segmentation, and... Particle identification; Image preprocessing refers to suppressing background texture and uneven illumination in scale-corrected images and enhancing particle boundary contrast, enabling threshold segmentation to distinguish particle regions from background regions; Threshold segmentation refers to generating binarized results of particle and background regions based on grayscale or brightness differences, obtaining a particle mask map for particle identification; Particle identification refers to locating connected particle regions based on the particle mask map and establishing a candidate particle set, then counting the number of particles in the candidate particle set, and completing size conversion and segmented statistics based on scale conversion parameters to obtain particle size distribution; In the abundance calculation stage, the microplastic abundance index is jointly determined by particle number, particle size distribution, soil sample mass, and soil sample volume, used to characterize the microplastic particle abundance level under mass or volume benchmarks; In the risk assessment stage, the risk classification boundary conditions are the classification criteria for establishing a correspondence between the corresponding intervals of microplastic abundance index and the microplastic pollution risk level. By matching the microplastic abundance index with the risk classification boundary conditions, the microplastic pollution risk level is output, thereby achieving rapid prediction and classification of soil microplastic pollution levels.

[0023] In some embodiments of this application, an imaging condition locking step is included before acquiring the original image. The imaging condition locking step includes: After the enrichment filter membrane is fixed on the carrier support surface, the focus position, exposure parameters and white balance parameters of the mobile phone camera are locked. Multiple frames of raw images were continuously acquired under the locked imaging conditions, and sharpness characterization parameters were calculated for each frame of raw images. Sharpness characterization parameters include edge intensity characterization parameters and high-frequency energy characterization parameters. The original image whose sharpness representation meets the screening criteria is selected from the original images of multiple frames as the target original image. The screening criteria include: the edge intensity representation of the target original image is located in the upper quantile interval of the corresponding edge intensity representation of the original images of multiple frames, and the high frequency energy representation of the target original image does not show a monotonically decreasing trend relative to the high frequency energy representation of the original images of multiple frames.

[0024] Specifically, the purpose of the imaging condition locking step is to maintain comparable optical responses and geometric imaging relationships for the same enrichment filter during the imaging process, reducing inconsistent changes in particle boundary contrast caused by repeated autofocus jumps, auto exposure compensation drift, or auto white balance correction fluctuations. Locking the focus position of the phone camera means fixing the focus plane on the imaging plane where the enrichment filter surface is located, allowing the particle boundaries in the enrichment filter image to transition from a sharp state to a stable state. For example, after focusing the phone camera on the filter texture or representative particle boundaries in the central area of ​​the enrichment filter, maintaining focus and preventing further autofocus triggering. Locking exposure parameters means fixing the shutter speed and photosensitivity gain or equivalent exposure combination, ensuring that the same particle remains in focus across multiple frames. Maintaining consistent brightness levels in the original images is crucial. For example, after a single exposure metering, the exposure compensation is kept constant to prevent flickering brightness in multiple consecutively captured original images due to changes in automatic gain. Locking white balance parameters means fixing the color temperature and color gain to prevent inter-frame shifts in color channel response between the enrichment filter image and the scale marker image. For instance, under fixed lighting conditions, locking the white balance to the same color temperature mode prevents the color channel distribution of multiple original images from drifting with automatic white balance adjustments. Continuously acquiring multiple original images under locked imaging conditions helps to compensate for occasional blurred frames caused by minor hand tremors, slight filter warping, and instantaneous illumination fluctuations, and provides a reference set with the same field of view for subsequent sharpness screening. Multiple frames of original images were continuously acquired without moving the enrichment filter membrane on the same carrier support surface, ensuring that only minor jitter differences existed between the multiple frames without introducing changes in the field of view. The edge intensity parameter in the sharpness representation describes the degree of gradient abruptness at particle boundaries. It can be obtained by performing gradient operator calculations on the enrichment filter membrane image region and accumulating or statistically analyzing the edge gradient amplitudes. A higher edge intensity parameter indicates a sharper boundary and a more concentrated gradient amplitude. For example, after performing gradient calculations on the enrichment filter membrane image region, the accumulated gradient amplitudes show that frames with larger gradient accumulation results correspond to sharper particle boundaries. The high-frequency energy parameter describes the proportion of high-frequency components in image detail texture and particle contours. It can be obtained by performing gradient calculations on the enrichment filter membrane image region... The high-frequency components are obtained by performing frequency domain transformation or second-order difference operators, and the energy of the high-frequency components is summarized. This results in a decrease in the high-frequency energy characterization when defocusing or motion blur causes detail attenuation. For example, the high-frequency component energy is calculated for multiple original images and a sequence is formed. If the high-frequency energy of a certain frame is significantly lower, it indicates that the frame details are blurred and weakened. The original images that meet the sharpness characterization criteria in the multiple original images are selected as the target original images. The upper quantile interval represents a candidate set that ranks high in the edge intensity characterization ranking of the multiple original images. It is used to preferentially retain frames with clear boundaries. For example, after ranking the edge intensity characterization of the multiple original images, the candidate frame set that ranks high is selected as the preferred set of clear frames.The absence of a monotonically decreasing trend indicates that the high-frequency energy representation of the target original image does not show a continuous downward trend in the acquisition sequence. This is used to exclude frame-by-frame blurring caused by increased hand tremors, slight displacement, or focus drift. For example, comparing the temporal changes of high-frequency energy representation in the candidate frame set, frames with continuously decreasing high-frequency energy representation are discarded. For example, after acquiring multiple frames of original images, the edge intensity representation of each frame is calculated within the enrichment filter image region, and several candidate frames with high edge intensity representation are selected. Then, the sequence of high-frequency energy representation changes is compared among the candidate frames, and candidate frames with continuously decreasing high-frequency energy representation are discarded. Finally, the original image that simultaneously satisfies the condition of edge intensity representation being in the upper quantile range and high-frequency energy representation not showing a monotonically decreasing trend is selected as the target original image. This ensures that subsequent scale correction, threshold segmentation, and particle recognition processing objects have stable and clear particle boundaries and detailed structures. For example, after screening, the target original image is imported into the image processing software and enters the scale conversion parameter calculation step, ensuring that subsequent processing is always based on the same clear image.

[0025] In some embodiments of this application, calculating the scale conversion parameters based on the scale-marked image includes: Locate the geometric primitives of the scale markers in the scale-marked image. The geometric primitives are pairs of line segment endpoints or pairs of adjacent points in a lattice. Calculate the pixel spacing for any set of geometric primitives to form a pixel spacing set; Perform a consistency check on the pixel spacing set and remove pixel spacings that deviate from the central tendency of the pixel spacing set; Based on the pixel spacing set after consistency verification and the calibration length of the scale marker, a conversion relationship from pixel to actual length is established to obtain the scale conversion parameters.

[0026] Specifically, the process of calculating scale conversion parameters based on scale-marked images is a calibration process that converts scale marks from image pixel space to actual length space. Preferably, scale marks are set as line segments with a defined calibration length, or as dot matrix marks with a fixed dot spacing. For example, a scale line segment with a known calibration length can be printed on the carrier support surface as a line segment mark, or a regular dot matrix with a known spacing between adjacent dots can be printed as a dot matrix mark. Locating the geometric primitives of the scale marks refers to first determining the region where the scale marks are located in the scale-marked image, and then... Within the image, reproducible reference point pairs are extracted. In line segment marking scenarios, the endpoints of the line segments are identified as endpoint pairs. In raster marking scenarios, the center points of two adjacent points are identified as adjacent raster point pairs. For example, in scale-marked images, the location of scale line segments is first defined, and the two endpoints of the line segments are extracted as endpoint pairs. Alternatively, multiple adjacent point pairs can be selected within a raster region, and the center point coordinates of each point can be extracted. Calculating the pixel spacing for any set of geometric primitives involves calculating the pixel distances of endpoint pairs or adjacent point pairs in the image coordinate system and summarizing them to form a set of pixel spacings. This allows for the utilization of multiple sets of geometric primitives. The primitives offset single measurement errors. For example, the pixel distance is calculated for multiple adjacent point pairs and multiple pixel spacing samples are formed to reduce the conversion deviation caused by the positioning error of a single point pair. The consistency test is used to remove abnormal pixel spacing caused by edge blurring, reflection, occlusion, and false detection of noise points. Specifically, the central tendency is first calculated for the set of pixel spacings and the central tendency is used as the representative level of pixel spacing. The central tendency can be the median or the mean. Then, the deviation of each pixel spacing from the central tendency is calculated and a deviation sequence is formed. Pixel spacings whose deviation deviates from the main group are judged as abnormal pixel spacings and removed. The judgment can be achieved by at least one of the following methods: the pixel spacing exceeds the effective range determined by the quantile interval; or, the pixel spacing exceeds the discrete threshold set based on the interquartile range; or, the pixel spacing is a statistical outlier in the deviation sequence. For example, when the center point of a certain adjacent point pair of the dot matrix is ​​mispositioned due to reflection, making the pixel spacing of the point pair greater than that of other point pairs, the pixel spacing is identified as abnormal and removed from the pixel spacing set, thereby ensuring that the retained pixel spacing set corresponds to the stable measurement results of the same mark scale.After completing the consistency check, a conversion relationship from pixel to actual length is established based on the pixel spacing set after the consistency check and the calibrated length of the scale marker. This means establishing a proportional mapping between the calibrated length and the representative pixel spacing of the pixel spacing set, thereby obtaining the scale conversion parameter. The scale conversion parameter can be expressed as the actual length corresponding to a unit pixel or the number of pixels corresponding to a unit actual length. This parameter serves as a unified scale benchmark for subsequent scale correction of the enrichment filter image, size conversion of the candidate particle set, and segmented statistics. For example, in the dot matrix marker scenario, the representative pixel spacing of the pixel spacing set after the consistency check is used as the pixel representation of the dot matrix dot pitch, and combined with the calibrated length of the dot matrix dot pitch, the actual length corresponding to a unit pixel is obtained. Then, in the subsequent particle size conversion, the pixel scale equivalent particle size of the candidate particle is converted into the actual scale equivalent particle size.

[0027] In some embodiments of this application, when performing scale correction on enrichment filter images using scale conversion parameters, the following steps are included: Locate the boundary of the enriched filter region in the original image and crop the image of the enriched filter region; extract the boundary direction set based on the boundary of the enriched filter region; The boundary direction sets are grouped into two intersecting direction groups based on their similarity in direction. The two intersecting direction groups correspond to two adjacent boundary directions of the enrichment filter membrane region, respectively. Determine whether the angle between two intersecting direction groups satisfies the orthogonality consistency condition; if the angle does not satisfy the orthogonality consistency condition, determine that the image of the enriched filter region has geometric distortion, and perform geometric correction on the image of the enriched filter region based on the two intersecting direction groups; The scale-corrected image is obtained by performing scale normalization on the geometrically corrected enriched filter region image based on the scale conversion parameters.

[0028] Specifically, the process of scale correction of the enrichment filter image using scale conversion parameters involves extracting the imaging region related to the enrichment filter in the original image before threshold segmentation and particle recognition. This process eliminates geometric distortions caused by slight tilting of the shooting angle or uneven placement of the enrichment filter, ensuring that subsequent particle size conversion is based on a unified scale and geometric shape. Locating the boundary of the enrichment filter region in the original image and cropping the image of the enrichment filter region refers to identifying the boundary transition position between the enrichment filter and the carrier support surface in the original image, obtaining the closed boundary surrounding the enrichment filter, and using this closed boundary as the basis for further processing. The enriched filter region image is cropped out from the boundary, thereby reducing the inclusion of scale markers or background textures in the subsequent particle recognition range. For example, when the original image contains both scale markers and carrier textures, cropping only the region inside the closed boundary ensures that subsequent processing only affects the enriched filter region. Extracting the boundary direction set based on the enriched filter region boundary involves statistically analyzing the direction of the boundary contours of the enriched filter region image. The orientation of each boundary segment on the boundary contour is summarized as a direction angle or direction vector to form a boundary direction set, which is used to characterize the overall orientation of the enriched filter region in the image coordinate system. For example, for closed boundaries... The boundary is segmented and the direction angle of each segment is counted to obtain a set of directions reflecting the main orientation of the boundary. The boundary direction set is then grouped into two intersecting direction groups based on their similarity. This means grouping boundary segments with similar orientations into the same direction group, so that the two direction groups correspond to two adjacent boundary directions of the enriched filter membrane region. The two direction groups intersecting indicates that the two adjacent boundary directions form an intersecting directional relationship in the image. For example, approximately horizontal boundary segments are grouped into one direction group, and approximately vertical boundary segments are grouped into another direction group, thus obtaining two sets of intersecting main directions. The determination of the two intersecting directions... Whether the included angle of the two directions satisfies the orthogonality consistency condition refers to calculating the included angle of the principal directions of the two directions and judging whether the included angle exhibits a stable relationship close to a right angle. The orthogonality consistency condition is used to express that the adjacent boundary directions of the enrichment filter should be orthogonal under ideal top-down shooting conditions. When the included angle deviates from the orthogonality relationship, it indicates that there is geometric distortion in the image of the enrichment filter area caused by the shooting tilt angle or perspective effect. For example, when a slight tilt occurs during mobile phone shooting, the enrichment filter area appears tilted or trapezoidal in the image. In this case, the included angle of the two principal directions will deviate from the right angle relationship and trigger the judgment that the orthogonality consistency condition is not satisfied.When geometric distortion is detected, geometric correction is performed on the enriched filter region image based on two intersecting direction groups. This means using the principal directions of the two direction groups as directional constraints for geometric correction, rotating, shearing, or performing perspective inversion transformations on the enriched filter region image in the image coordinate system. This restores the two sets of adjacent boundary directions of the enriched filter region to a geometric relationship that satisfies the orthogonality consistency condition, thus making the enriched filter region image geometrically correspond to the corrected top-view shape. For example, rotation aligns one set of principal directions with the image coordinate axes, and then perspective inversion restores the trapezoidal boundary to an orthogonal boundary shape. After geometric correction, scale normalization is performed based on scale conversion parameters. This means using scale conversion parameters to uniformly map the pixel scale of the corrected enriched filter region image to the actual scale, making the pixel scale consistent with the actual scale. The actual length corresponding to a pixel length remains consistent across the entire image, resulting in a scale-corrected image. For example, the distance of any pixel in the corrected image is converted to its actual length using scale conversion parameters, ensuring a unified pixel-to-actual-length mapping between the equivalent particle size and principal axis length of subsequent candidate particle sets. For instance, if a slight angled shot during mobile phone photography causes the enrichment filter to appear trapezoidal in the image, and the angle between the two directional groups no longer exhibits a stable right-angle relationship, geometric correction restores the trapezoidal appearance to a regular shape. Scale normalization then ensures that the size conversion and segmentation statistics of subsequent candidate particle sets are not affected by systematic deviations caused by geometric distortion. For example, before correction, the pixel scale correspondence between the edge and center of the same particle is inconsistent, while after correction and scale normalization, the correspondence tends to be consistent across the entire image.

[0029] In some embodiments of this application, image preprocessing of scale-corrected images includes: Image preprocessing for scale-corrected images includes: Background candidate regions are determined in the scale-corrected image. The background candidate regions are continuous regions with uniform filter texture and local contrast that meet a preset contrast threshold. The local contrast is calculated as follows: the scale-corrected image is divided into sliding windows of a preset size, and the ratio of the gray range to the gray mean in each sliding window is calculated to obtain the window contrast. When the window contrast is less than or equal to the preset contrast threshold, the region corresponding to the sliding window is determined to be a low-contrast region. If a continuous low-contrast region appears and meets the preset area threshold, it is determined to be a background candidate region. Background brightness surfaces are extracted based on background candidate regions, and the background brightness surfaces are used as background compensation benchmarks to perform background compensation on scale-corrected images in order to suppress uneven illumination. Noise suppression and contrast enhancement are performed on the image after background compensation to obtain a preprocessed image for threshold segmentation.

[0030] Specifically, the process of determining candidate background regions in a scale-corrected image involves first calculating the window contrast based on a sliding window and marking low-contrast regions. Then, based on spatial continuity and area constraints, candidate background regions suitable for fitting the background brightness surface are selected from the low-contrast regions. The determination that a region is considered low-contrast when its window contrast is less than or equal to a preset contrast threshold means comparing the window contrast calculated for each sliding window with the preset contrast threshold. If the window contrast is not greater than the preset contrast threshold, the grayscale fluctuations within the sliding window are under control relative to the grayscale mean, indicating that the region better matches the background characteristics of a uniform filter membrane texture. This allows the sliding window to be considered a low-contrast region. The area corresponding to the window is marked as a low-contrast region. Conversely, when the window contrast is greater than a preset contrast threshold, it indicates that there is a strong grayscale jump or obvious boundary structure within the sliding window. The region may contain grain boundaries, bright reflective stripes, shadow boundaries, or local wrinkles and textures, and therefore is not included in the low-contrast region. After marking the low-contrast regions of the sliding windows in the entire image, if a continuous low-contrast region appears and meets the preset area threshold, it is determined as a candidate background region. Here, continuous means that multiple low-contrast sliding windows are adjacent in spatial position and form a connected area. The formation of continuous low-contrast regions indicates that the region is not an occasional single window with low contrast, but rather a continuous low-contrast region within a certain spatial range. The background exhibits uniform texture and controlled grayscale fluctuations. When low-contrast areas are discontinuous and scattered, even if the contrast of a single window meets the threshold, it may be an occasional result caused by local flatness inside the particles, occasional noise suppression, or local occlusion of the filter membrane pores. Directly using it for background brightness surface fitting can easily introduce local biases, so it is not identified as a candidate background region. When low-contrast areas are continuous but their area does not reach the preset area threshold, it indicates that the coverage of the continuous area is insufficient, making it difficult to provide enough background sampling points to support the stable extrapolation of the background brightness surface in space. Therefore, the continuous area is considered as a candidate but not as the final candidate background region. When low-contrast areas are continuous... When the area reaches or exceeds the preset area threshold, it indicates that the contiguous area has sufficient spatial coverage and can provide a continuous set of sampling points that can represent the fluctuations in background brightness, thus being identified as a candidate background area. For example, in one implementation, the scale-corrected image is divided into N sliding windows and the window contrast sequence is calculated. After determining the preset contrast threshold based on the window contrast sequence, windows with a contrast less than or equal to the preset contrast threshold are marked as low-contrast windows. Then, the low-contrast windows are connected and aggregated to obtain several low-contrast contiguous areas. If a contiguous area consists of only a small number of low-contrast windows or presents a broken and scattered state, it is not considered as a candidate background area.If a contiguous region is continuously composed of adjacent low-contrast windows and the area covered by the contiguous region reaches a preset area threshold, then the contiguous region is determined as a background candidate region. A background brightness surface is then fitted based on the background candidate region, and background compensation is performed to ensure that the background compensation benchmark comes from a spatially continuous and textured background region.

[0031] For example, in one implementation, the scale-corrected image is scanned using sliding windows of a preset size, assuming a total of N=100 sliding windows are obtained. The window contrast is calculated for each sliding window as (maximum grayscale value - minimum grayscale value) / (average grayscale value within the sliding window), resulting in 100 window contrast values, denoted as C1, C2, ..., C100. C1 to C100 are sorted from smallest to largest, resulting in the sorted sequence S1≤S2≤...≤S100. Then, the 25th value S25 in the sorted sequence is taken as the contrast value corresponding to the 25th percentile, and the 75th value S75 is taken as the contrast value corresponding to the 75th percentile, thus obtaining the concentrated distribution interval of the window contrast sequence as [S25, S75]. For example, if after sorting, S25=0.12 and S75=0.22, it indicates that the scale-corrected image is within a certain range. In all sliding windows of the corrected image, at least 25 sliding windows have a window contrast of no more than 0.12, and at least 25 sliding windows have a window contrast of no less than 0.22, with the window contrast of most sliding windows concentrated between 0.12 and 0.22. Based on this, a preset contrast threshold of 0.22 can be selected as the upper boundary value of the concentrated distribution interval, and sliding windows with a window contrast of less than or equal to 0.22 are marked as low contrast regions. Adjacent low contrast regions are then aggregated. If the number of low contrast windows contained in a certain aggregated region reaches a preset area threshold (e.g., 20 or 25 windows), the aggregated region is determined as a background candidate region for subsequent background brightness surface fitting and background compensation processing.

[0032] In some embodiments of this application, threshold segmentation of a scale-corrected image includes: Calculate the grayscale histogram of the preprocessed image and extract the peak-valley structure of the histogram; The segmentation threshold is determined based on the peak-valley structure. The segmentation threshold is taken as the gray value corresponding to the valley position between the background peak and the particle peak or the representative gray value in the neighborhood of the valley position. A binary segmentation map is generated based on the segmentation threshold, and connectivity correction processing is performed on the binary segmentation map, including hole filling processing and boundary smoothing processing, to obtain a particle mask map.

[0033] Specifically, threshold segmentation serves to binary divide the suspected particle region and the filter membrane background region in the preprocessed image at the pixel level, thus providing clear regional input for subsequent connected component labeling and particle recognition. Calculating the grayscale histogram of the preprocessed image and extracting its peak-valley structure involves statistically analyzing the frequency of each grayscale value in the preprocessed image and forming a distribution curve. The peak-valley structure reflects the aggregation state of the main pixel groups in the image. Typically, background pixels form background peaks due to their relatively concentrated brightness, while particle pixels form particle peaks due to differences in boundary and internal texture. The valley between the two peaks corresponds to the transition zone where the background and particles are most easily separated at the grayscale level. For example, when preprocessing… When the background compensation effect of the image is good, the background peak appears as a concentrated and sharp main peak, while the particle peak appears as a relatively dispersed secondary peak. The valley range between the two corresponds to the grayscale range of the transition pixels between the particle boundary and the background. Determining the segmentation threshold based on the peak-valley structure means selecting a representative grayscale value at the valley position between the background peak and the particle peak as the segmentation threshold, or selecting a representative grayscale value in the neighborhood of the valley position that can make the separation of the two types of pixels more stable. The neighborhood of the valley position is used to accommodate the instability of the valley value caused by subtle changes in the filter membrane texture or residual noise, making the threshold selection robust. For example, when the valley position causes a slight fluctuation in adjacent grayscale segments due to a small amount of residual texture, a representative grayscale value can be selected in the neighborhood of the valley position. The process involves generating corresponding binary segmentation maps for several candidate grayscale values ​​and selecting the candidate grayscale value that results in less fragmentation of the particle region and less missegmentation of the background as the representative grayscale value. Generating a binary segmentation map based on a segmentation threshold involves assigning pixels in the preprocessed image that meet the threshold criteria as particle pixels and the remaining pixels as background pixels, thus forming a binarized result containing only particle and background regions. For example, pixels below the threshold are uniformly labeled as background, and pixels above the threshold are uniformly labeled as particles, or labeled according to the opposite criteria, ensuring that particle regions form connected pixel blocks in the binary segmentation map. Performing connectivity correction processing on the binary segmentation map refers to... The results show that structural repair is performed on fractures and voids caused by filter membrane pore texture, particle surface reflection, or threshold cutting boundaries. Among them, the hole filling process is used to fill in the blank holes formed by uneven brightness within the same particle area, so that the particle area presents a continuous solid shape. For example, when a particle forms a ring-shaped particle because the central area is judged as background by the threshold due to reflection, the blank inside is filled into particle pixels by hole filling. The boundary smoothing process is used to eliminate burrs and jagged edges of binary boundaries, so that the particle outline changes from broken edges to smooth edges. For example, when jagged burrs or small sharp corners appear on the particle boundary after thresholding, the sharp corners are flattened by boundary smoothing and the outline is closer to the real particle boundary.The final particle mask image is obtained. This image can be understood as a region labeling result for suspected microplastic particle areas in the scale-corrected image. In subsequent particle identification steps, the particle mask image is used to locate each connected particle region and establish a candidate particle set. This ensures that particle count and size distribution calculations are based on regions with complete connectivity and stable boundaries. For example, each connected particle pixel region in the particle mask image corresponds to a connected component object in the subsequent connected component labeling process.

[0034] In some embodiments of this application, particle identification yields a candidate particle set, including: Perform connected component labeling on the particle mask graph to obtain a set of connected components; For each connected component in the set of connected components, extract the contour and calculate the morphological feature set, which includes area features, perimeter features, aspect ratio features and roundness features. Morphological consistency determination is performed based on morphological feature groups. Morphological consistency determination includes comparing the morphological features corresponding to connected components with the morphological feature distribution of other connected components in the same image, and removing connected components that fall into discrete edge regions. The retained connected components are determined as the candidate particle set.

[0035] Specifically, the process of obtaining a candidate particle set through particle identification involves transforming the pixel blocks in the particle mask image that are binarized and labeled as particle regions into a set of particle objects that can be statistically analyzed and measured. This provides object-level input for subsequent calculations of particle quantity and size distribution. Specifically, performing connected component labeling on the particle mask image involves clustering particle pixels in the particle mask image according to their pixel connectivity, merging interconnected particle pixels into the same connected component, so that each connected component corresponds to an independent candidate particle region, thus obtaining a set of connected components. For example, when multiple separate particle pixel blocks appear in the particle mask image, connected component labeling will number each pixel block and form multiple connected component pairs. The concept of "contour" refers to extracting the contour and calculating a set of morphological features for each connected component in a set of connected components. This involves contour tracing of the outer boundary of the connected components and calculating feature quantities describing the geometric shape of the particles based on the contour and region pixels. Area features characterize the pixel area covered by the connected component or its converted actual area; perimeter features characterize the contour length and boundary complexity; aspect ratio features characterize the elongation along the principal axis of the connected component; and roundness features characterize the degree to which the contour of the connected component approximates a circle and are used to distinguish approximate particle shapes from strip-like or fragmented regions. For example, when a connected component is a long, thin strip, its aspect ratio feature increases and its roundness feature decreases, thus differentiating it from most approximate particle shapes. Differences in connected component formation; morphological consistency determination based on morphological feature groups refers to comparing the morphological feature group of each connected component with the morphological feature distribution of all connected components in the same image. The morphological feature distribution can be understood as the overall range of candidate particle morphologies formed by the same enriched filter under the current imaging conditions. When the area, aspect ratio, or roundness features of a connected component significantly deviate from the overall range, the connected component is more likely to be caused by filter wrinkles and shadows, local dirt clusters, edge truncation, or threshold segmentation errors. Therefore, connected components falling into discrete edge regions are removed. Discrete edge regions refer to sparse areas far from the main clustering area in the morphological feature distribution, which do not possess the morphological features used to express the connected component's characteristics as multiple The statistical properties of connected components are consistent. For example, when the roundness of most connected components is concentrated in a stable range, while the roundness of some connected components is low and the aspect ratio is high, the connected components are located at the sparse edge in the distribution and can be determined to fall into the discrete edge region and be eliminated. After the elimination is completed, the remaining connected components are determined as the candidate particle set, so that the candidate particle set is mainly composed of particle objects with consistent morphological features and overall distribution, and provides a stable particle object basis for subsequent pseudo particle elimination, particle number statistics and particle size distribution calculation. For example, the connected components retained by the morphological consistency judgment are more likely to correspond to the real particle outline rather than connected component objects formed by missegmentation of background texture or shadow.

[0036] In some embodiments of this application, a pseudo-particle removal step is included before counting the candidate particle set. The pseudo-particle removal step includes: For each candidate particle in the candidate particle set, an optical consistency feature group is extracted. The optical consistency feature group includes at least color channel distribution features, edge gradient stability features, and local texture uniformity features. The pseudo-particle determination is performed based on the optical consistency feature group. The pseudo-particle determination includes comparing the color channel distribution features of the candidate particles with the color channel distribution features of the background candidate region, and combining the edge gradient stability features to determine whether the boundary of the candidate particles exhibits a unilateral abrupt change characteristic caused by reflection or shadow. Candidate particles that are identified as pseudo particles are removed from the candidate particle set, and the remaining candidate particles are determined as the candidate particle set for particle number statistics and particle size distribution calculation.

[0037] Specifically, the pseudo-particle removal step aims to further exclude regions that resemble particles in morphology but whose optical response does not conform to particle characteristics, formed by filter membrane reflections, shadows, wrinkles, or residual impurities, from the candidate particle set. This makes the particle count and particle size distribution calculation closer to the true distribution of suspected microplastic particles. Extracting optical consistency feature groups for each candidate particle in the candidate particle set refers to extracting feature quantities that reflect imaging optical properties from the scale-corrected or pre-processed image, using the connected region corresponding to the candidate particle as the sampling range. Color channel distribution features are used to describe the brightness distribution and concentration of the candidate particle region in each color channel to distinguish color responses. Texture blocks close to the background and particle blocks with stable color responses that differ from the background can be analyzed. For example, the pixel value distribution of candidate particle regions in each color channel can be statistically analyzed, and the concentration, offset, or relative proportion between channels can be extracted as color channel distribution features. Edge gradient stability features are used to describe whether the gradient intensity of the candidate particle contour boundary is balanced and continuous in different orientations. Particle boundaries usually exhibit a continuous gradient distribution around the contour, while reflections or shadows often cause abnormal gradient enhancement or weakening on one side of the boundary. For example, the candidate particle contour can be divided into several boundary segments according to orientation, and the gradient intensity distribution of each boundary segment can be compared. When only one side of the boundary segment shows a concentrated and sudden increase in gradient while the gradient of the other boundary segments weakens, the gradient will be significantly reduced. At this time, the edge gradient stability feature will show obvious imbalance; the local texture uniformity feature is used to describe whether the texture undulation of pixels inside the candidate particle presents a relatively consistent particle surface texture. The filter membrane pore texture or wrinkle shadow usually shows a texture structure with stronger directionality or more obvious periodicity. For example, multiple local sub-regions can be divided inside the candidate particle region and the consistency of the texture undulation of the sub-regions can be compared. When the texture shows strip-like repetition in a certain direction or the periodic pore texture is dominant, the local texture uniformity feature will show enhanced directionality rather than uniform undulation; performing pseudo-particle determination based on optical consistency feature group refers to comparing the color channel distribution features of the candidate particle region with the color of the background candidate region. By comparing the channel distribution characteristics, if the color channel distribution of the candidate particle region is highly consistent with that of the background candidate region, the candidate particle is more likely to be a region that was mistakenly cut out by the threshold segmentation of the background texture. At the same time, the edge gradient stability characteristics are combined to determine whether the boundary of the candidate particle exhibits a unilateral abrupt change characteristic caused by reflection or shadow. The unilateral abrupt change characteristic indicates that the gradient intensity on one side of the candidate particle outline is concentrated or suddenly increased while the gradient on the other side is weakened. This characteristic is usually caused by reflective stripes or shadow edges rather than the closed boundary structure of the real particle. Furthermore, the local texture uniformity characteristics are combined to determine whether the internal texture of the candidate particle exhibits an obvious directional or periodic structure, thereby distinguishing the surface texture of the real particle from the pore texture of the filter membrane.For example, if the color channel distribution of a candidate particle is almost identical to that of the background candidate region, and the gradient intensity of the candidate particle boundary only increases sharply on the side closest to the light source while the gradient of the other boundary segments is weak, and the internal texture of the candidate particle exhibits a periodic texture consistent with the pores of the filter membrane, then the candidate particle can be determined to be a pseudo-particle formed by reflection or texture missegmentation and should be removed. Candidate particles identified as pseudo-particles are removed from the candidate particle set, and the remaining candidate particles are determined as the candidate particle set for particle quantity statistics and particle size distribution calculation. This ensures that the particle objects participating in the final statistics simultaneously meet the optical consistency requirements of being distinguishable from the background and having stable closed boundaries in terms of color response, boundary gradient, and internal texture, thereby reducing the probability of misjudgment into the microplastic abundance index calculation. For example, the candidate particles retained after pseudo-particle removal typically exhibit stable color differences from the background, continuous gradient distribution in all directions of the outline, and no obvious striped directionality in the internal texture.

[0038] In some embodiments of this application, calculating the particle size distribution includes: For each candidate particle in the candidate particle set, calculate the equivalent particle size and principal axis length at the pixel scale; The equivalent grain size and principal axis length at the pixel scale are converted to the equivalent grain size and principal axis length at the actual scale based on the scale conversion parameters. The particle size distribution is obtained by segmenting the equivalent particle size at the actual scale according to the particle size segmentation rule. The particle size segmentation rule is generated based on the distribution pattern of the particle size distribution. The particle size segmentation rule includes identifying the aggregation intervals according to the distribution pattern and using the boundaries of the aggregation intervals as the segmentation boundaries.

[0039] Specifically, particle size distribution is used to characterize the quantity composition of candidate particle sets within different particle size ranges. This allows the microplastic abundance index to reflect not only the quantity of particles but also their size structure, providing input for subsequent risk classification. Specifically, calculating the equivalent particle size and principal axis length at the pixel scale for each candidate particle in the candidate particle set involves extracting geometrical parameters for each candidate particle in the particle mask image or candidate particle region. The equivalent particle size is used to equate the area of ​​irregular particles to the diameter of circular particles, enabling a unified characterization of fragmented or near-circular particles using a single size metric. The principal axis length characterizes the maximum extension of the candidate particle along its principal direction. The dimension is suitable for describing the length characteristics of fibrous or strip-shaped particles. Both are represented in pixels at the pixel scale. For example, when a candidate particle is fragmented and has an irregular outline, the pixel area of ​​the candidate particle can be calculated first, and the equivalent circular diameter can be obtained as the equivalent particle size at the pixel scale. When a candidate particle is a long strip, the principal direction can be determined on the outline of the candidate particle, and the maximum span along the principal direction can be calculated as the principal axis length at the pixel scale. Thus, the same candidate particle can simultaneously possess both area equivalent size and length extension size. The equivalent particle size and principal axis length at the pixel scale are converted into the equivalent particle size at the actual scale based on the scale conversion parameters. The diameter and principal axis length refer to mapping pixel units to actual length units using scale conversion parameters. This ensures the comparability of candidate particle sizes obtained under different shooting conditions and guarantees that subsequent segmented statistics are not affected by changes in image resolution. For example, when the scale conversion parameter represents the actual length corresponding to a unit pixel, the equivalent particle diameter and principal axis length at the pixel scale can be multiplied by the actual length corresponding to a unit pixel to obtain the equivalent particle diameter and principal axis length at the actual scale, thereby achieving a unified scale comparison across samples and across captured images. Segmenting and statistically analyzing the equivalent particle diameter at the actual scale according to particle diameter segmentation rules to obtain the particle size distribution refers to the actual scale of all candidate particles... The equivalent particle size is mapped to several particle size intervals, and the number or proportion of particles in each interval is counted to form a particle size distribution. This provides structural information on the proportion of different particle size segments at the statistical level. For example, all actual-scale equivalent particle sizes are first summarized into an equivalent particle size sequence. Then, the equivalent particle size sequence is divided into several segments according to the particle size segmentation rule and counted separately to obtain the number and proportion of particles in each segment. The particle size segmentation rule is generated based on the distribution pattern of particle size distribution. The distribution pattern refers to the aggregation and sparsity characteristics of the equivalent particle size in all candidate particles. Specifically, it can be represented by several relatively dense aggregation intervals and several relatively sparse transition intervals.Identifying clustering intervals based on distribution patterns and using the boundaries of these intervals as segmentation boundaries involves first sorting or performing histogram analysis on the equivalent particle size sequence to identify the size ranges where particle numbers are clearly concentrated as clustering intervals. Then, the boundaries between adjacent clustering intervals are determined as the clustering interval boundaries, which are then used as the boundaries for particle size segmentation. This ensures that the segmented intervals match the size distribution characteristics of the sample itself, reducing the risk of artificially splitting aggregated particles or excessively amplifying sparse intervals due to fixed segmentation. For example, histogram analysis of the equivalent particle size sequence can be performed, and the dense column clusters appearing in the histogram results can be used to determine the size range covered by the dense column clusters as the clustering intervals. The clustering intervals are then defined by identifying the size positions of low-frequency valleys between dense column clusters and adjacent dense column clusters. These clustering interval boundaries serve as segmentation boundaries, ensuring that candidate particles within the same clustering interval are grouped into the same particle size segment during segmented statistical analysis, resulting in a stable particle size distribution. For example, when an equivalent particle size sequence exhibits a distinct clustering interval within a smaller size range and another clustering interval within a larger size range, with a sparse transition interval between the two, the boundary position of the sparse transition interval can be used as the segmentation boundary. This ensures that subsequent calculations of the proportion of fine and coarse particles are based on particle size segmentation consistent with the actual distribution pattern.

[0040] In some embodiments of this application, the output microplastic pollution risk level includes: Obtain the risk classification boundary conditions, which are simultaneously associated with the microplastic abundance index range and the structural feature range of particle size distribution. The structural feature range of particle size distribution includes the range of fine particle proportion and the range of coarse particle proportion. A risk level set is constructed based on the risk classification boundary conditions. The risk level set includes at least the first risk level, the second risk level, and the third risk level. The risk level is determined by jointly matching the microplastic abundance index with the structural feature intervals of particle size distribution according to a progressive judgment order, where: When the microplastic abundance index falls into the first abundance range and the proportion of fine particles falls into the first fine particle proportion range, it is determined to be the first risk level. When the microplastic abundance index falls into the second abundance range, or when the microplastic abundance index falls into the first abundance range and the fine particle proportion falls into the second fine particle proportion range, it is determined to be the second risk level. When the microplastic abundance index falls into the third abundance range, or the microplastic abundance index falls into the second abundance range and the proportion of fine particles falls into the second fine particle proportion range, or the proportion of coarse particles falls into the third coarse particle proportion range, it is determined to be the third risk level. Output the microplastic pollution risk level and write the microplastic abundance index, particle number, particle size distribution and scale conversion parameters into the detection result record to obtain a traceable record.

[0041] Specifically, the process of outputting microplastic pollution risk levels involves converting the aforementioned microplastic abundance index and particle size distribution into discrete risk assessment results. This allows for a clear and intuitive representation of the microplastic pollution level in soil samples in a graded manner during rapid screening. The risk grading boundary conditions serve as a set of grading criteria to support the level determination. These conditions simultaneously correlate the microplastic abundance index range and the size distribution structural feature range. The microplastic abundance index range describes the range where the microplastic abundance index is at a low, medium, or high level. The size distribution structural feature range describes the structural morphology of the particle size distribution. In this application, the structural morphology is characterized by the fine particle proportion range and the coarse particle proportion range. The fine particle proportion range represents the range corresponding to the percentage of particles falling within the fine particle size range in the particle size distribution, while the coarse particle proportion range represents the range corresponding to the percentage of particles falling within the coarse particle size range. Constructing a risk level set based on the risk grading boundary conditions means limiting the risk assessment output to at least three discrete levels, including a first risk level, a second risk level, and a third risk level, thereby enabling the risk assessment output to be more comprehensive. It possesses a fixed and interpretable risk level system; the joint matching of microplastic abundance indices and size distribution structural feature intervals according to a progressive judgment order means first determining whether the combination conditions corresponding to the first risk level are met; if the first risk level conditions are not met, then the combination conditions corresponding to the second risk level are determined; if the second risk level conditions are still not met, then the combination conditions corresponding to the third risk level are determined, so that the risk judgment order has a progressive relationship from low to high; wherein, the combination conditions corresponding to the first risk level are that the microplastic abundance index falls into the first abundance interval and the proportion of fine particles falls into the first fine particle proportion interval. The combined conditions are used to characterize situations where both the microplastic abundance level and the proportion of fine particles are at relatively low levels. The combined conditions corresponding to the second risk level include two types of paths. The first type of path is that the microplastic abundance index falls into the second abundance range, indicating that the microplastic abundance level has entered a medium level and triggers the second risk level. The second type of path is that the microplastic abundance index falls into the first abundance range but the proportion of fine particles falls into the second fine particle proportion range, indicating that although the microplastic abundance level is still in the first abundance range, the particle size structure shows a trend of increasing fine particle proportion, thereby increasing the risk level through the size structure path.The combination conditions corresponding to the third risk level also include multiple paths. The first path is that the microplastic abundance index falls into the third abundance range, indicating that the microplastic abundance level has entered a high level and directly triggers the third risk level. The second path is that the microplastic abundance index falls into the second abundance range and the fine particle proportion falls into the second fine particle proportion range, indicating that the abundance level and the fine particle proportion are simultaneously in a high combination state and trigger the third risk level. The third path is that the coarse particle proportion falls into the third coarse particle proportion range, indicating that the coarse particle proportion in the particle size structure has reached the corresponding range and triggers the third risk level through the structural feature path. After outputting the microplastic pollution risk level, the microplastic abundance index, particle number, particle size distribution and scale conversion parameters are written into the detection result record. Among them, the microplastic abundance index is used to verify the abundance level of the sample, the particle number and particle size distribution are used to verify the statistical basis of the candidate particle set, and the scale conversion parameters are used to verify the scale benchmark of the size conversion process, thereby obtaining a traceable record, which is convenient for comparing and verifying the consistency of the risk level output of the same sample under different time or different batch collection conditions. Risk classification boundary conditions can be established by calibrating several soil samples obtained from the same region or the same monitoring task: First, calculate the microplastic abundance index for each sample and form an abundance index sequence; then, calculate the proportion of fine particles and coarse particles for each sample based on particle size distribution, and form fine particle proportion sequences and coarse particle proportion sequences respectively; based on this, sort the abundance index sequences from smallest to largest, and generate the first abundance interval, the second abundance interval, and the third abundance interval using quantiles. For example, divide the abundance index sequence into low, middle, and high segments using three equal quantiles, so that the first abundance interval corresponds to the low-segment samples, the second abundance interval corresponds to the middle-segment samples, and the third abundance interval corresponds to the high-segment samples; similarly, sort the fine particle proportion sequence from smallest to largest and generate the first and second fine particle proportion intervals using quantiles, and sort the coarse particle proportion sequence from smallest to largest and generate the third coarse particle proportion interval using quantiles. The third coarse particle proportion interval can be set as the high-end interval in the coarse particle proportion sequence to trigger the third risk level of the structural feature path. For example: Assume that the first abundance interval of the abundance index obtained through the above calibration is [0, A1), the second abundance interval is [A1, A2), and the third abundance interval is [A2, +∞). The first fine particle proportion interval is [0, F1], the second fine particle proportion interval is [F1, 1], and the third coarse particle proportion interval is [C2, 1] (A1, A2, F1, and C2 are all determined by the quantiles of the calibration sequence). When the calculated microplastic abundance index of a sample falls into [0, A1) and the fine particle proportion falls into [0, F1), the combination condition of the first abundance interval + the first fine particle proportion interval is satisfied, and the first risk level is output.When the microplastic abundance index of a sample falls within [A1, A2), the second risk level can be directly triggered regardless of the fine particle proportion. Alternatively, when the microplastic abundance index falls within [0, A1) but the fine particle proportion falls within [F1, 1], the second risk level is output through the structural feature path of the fine particle proportion shifting upwards. When the microplastic abundance index of a sample falls within [A2, +∞), the third risk level is directly output. Alternatively, when the microplastic abundance index falls within [A1, A2) and the fine particle proportion falls within [F1, 1], the third risk level is output. Alternatively, when the coarse particle proportion falls within [C2, 1], the third risk level is output through the structural feature path of the coarse particle proportion entering the high-end range.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A rapid prediction method for soil microplastics based on mobile phone super macro imaging, characterized in that, include: The collected soil samples were air-dried and sieved. The sieved soil samples were subjected to density separation and organic matter removal treatments in sequence. The suspended phase obtained from the density separation process was filtered to obtain an enriched filter membrane loaded with suspected microplastic particles, and the soil sample mass and volume were recorded. The enrichment filter membrane is placed on the carrier support surface, and scale marks are set on the carrier support surface or in the imaging field of view adjacent to the carrier support surface; the original image is acquired under fixed shooting distance and consistent illumination conditions, and the original image includes the enrichment filter membrane image and the scale mark image; The original image is imported into the image processing software, the scale conversion parameters are calculated based on the scale-marked image, and the scale conversion parameters are used to perform scale correction on the enrichment filter image to obtain the scale-corrected image. The scale-corrected image is preprocessed, thresholded, and identified sequentially to obtain a candidate particle set. The number of particles in the candidate particle set is counted, and the particle size distribution is obtained by size conversion and segmented statistics based on the scale conversion parameters. Microplastic abundance index was calculated based on particle number, particle size distribution, soil sample mass, and soil sample volume. Obtain the risk classification boundary conditions corresponding to the microplastic abundance index, match the microplastic abundance index with the risk classification boundary conditions, and output the microplastic pollution risk level.

2. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 1, characterized in that, Before acquiring the raw image, an imaging condition locking step is included, which includes: After the enrichment filter membrane is fixed on the carrier support surface, the focus position, exposure parameters and white balance parameters of the mobile phone camera are locked. Multiple frames of raw images were continuously acquired under the locked imaging conditions, and sharpness characterization parameters were calculated for each frame of raw images. Sharpness characterization parameters include edge intensity characterization parameters and high-frequency energy characterization parameters. The original image whose sharpness representation meets the screening criteria is selected from the original images of multiple frames as the target original image. The screening criteria include: the edge intensity representation of the target original image is located in the upper quantile interval of the corresponding edge intensity representation of the original images of multiple frames, and the high frequency energy representation of the target original image does not show a monotonically decreasing trend relative to the high frequency energy representation of the original images of multiple frames.

3. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 2, characterized in that, When calculating scale conversion parameters based on scale-labeled images, the following are included: Locate the geometric primitives of the scale markers in the scale-marked image. The geometric primitives are pairs of line segment endpoints or pairs of adjacent points in a lattice. Calculate the pixel spacing for any set of geometric primitives to form a pixel spacing set; Perform a consistency check on the pixel spacing set and remove pixel spacings that deviate from the central tendency of the pixel spacing set; Based on the pixel spacing set after consistency verification and the calibration length of the scale marker, a conversion relationship from pixel to actual length is established to obtain the scale conversion parameters.

4. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 3, characterized in that, When performing scale correction on enrichment filter images using scale conversion parameters, the following steps are included: Locate the boundary of the enriched filter region in the original image and crop the image of the enriched filter region; extract the boundary direction set based on the boundary of the enriched filter region; The boundary direction sets are grouped into two intersecting direction groups based on their similarity in direction. The two intersecting direction groups correspond to two adjacent boundary directions of the enrichment filter membrane region, respectively. Determine whether the angle between two intersecting direction groups satisfies the orthogonality consistency condition; if the angle does not satisfy the orthogonality consistency condition, determine that the image of the enriched filter region has geometric distortion, and perform geometric correction on the image of the enriched filter region based on the two intersecting direction groups; The scale-corrected image is obtained by performing scale normalization on the geometrically corrected enriched filter region image based on the scale conversion parameters.

5. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 4, characterized in that, Image preprocessing for scale-corrected images includes: Background candidate regions are determined in the scale-corrected image. The background candidate regions are continuous regions with uniform filter texture and local contrast that meet a preset contrast threshold. The local contrast is calculated as follows: the scale-corrected image is divided into sliding windows of a preset size, and the ratio of the gray range to the gray mean in each sliding window is calculated to obtain the window contrast. When the window contrast is less than or equal to the preset contrast threshold, the region corresponding to the sliding window is determined to be a low-contrast region. Continuous low-contrast regions that meet the preset area threshold are determined as background candidate regions. Background brightness surfaces are extracted based on background candidate regions, and the background brightness surfaces are used as background compensation benchmarks to perform background compensation on scale-corrected images in order to suppress uneven illumination. Noise suppression and contrast enhancement are performed on the image after background compensation to obtain a preprocessed image for threshold segmentation.

6. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 5, characterized in that, Thresholding segmentation of scale-corrected images includes: Calculate the grayscale histogram of the preprocessed image and extract the peak-valley structure of the histogram; The segmentation threshold is determined based on the peak-valley structure. The segmentation threshold is taken as the gray value corresponding to the valley position between the background peak and the particle peak or the representative gray value in the neighborhood of the valley position. A binary segmentation map is generated based on the segmentation threshold, and connectivity correction processing is performed on the binary segmentation map, including hole filling processing and boundary smoothing processing, to obtain a particle mask map.

7. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 6, characterized in that, Particle identification yields a candidate particle set, including: Perform connected component labeling on the particle mask graph to obtain a set of connected components; For each connected component in the set of connected components, extract the contour and calculate the morphological feature set, which includes area features, perimeter features, aspect ratio features and roundness features. The morphological consistency determination is performed based on the morphological feature group. The morphological consistency determination includes comparing the morphological features corresponding to the connected components with the morphological feature distribution of other connected components in the same image, and removing connected components that fall into discrete edge regions. The retained connected components are determined as the candidate particle set.

8. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 7, characterized in that, Before compiling the candidate particle set, a pseudo-particle removal step is included, which includes: For each candidate particle in the candidate particle set, an optical consistency feature group is extracted. The optical consistency feature group includes at least color channel distribution features, edge gradient stability features, and local texture uniformity features. The pseudo-particle determination is performed based on the optical consistency feature group. The pseudo-particle determination includes comparing the color channel distribution features of the candidate particles with the color channel distribution features of the background candidate region, and combining the edge gradient stability features to determine whether the candidate particle boundary exhibits a unilateral abrupt change characteristic caused by reflection or shadow. Candidate particles that are identified as pseudo particles are removed from the candidate particle set, and the remaining candidate particles are determined as the candidate particle set for particle number statistics and particle size distribution calculation.

9. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 8, characterized in that, The calculation of particle size distribution includes: For each candidate particle in the candidate particle set, calculate the equivalent particle size and principal axis length at the pixel scale; The equivalent grain size and principal axis length at the pixel scale are converted to the equivalent grain size and principal axis length at the actual scale based on the scale conversion parameters. The particle size distribution is obtained by segmenting the equivalent particle size at the actual scale according to the particle size segmentation rule. The particle size segmentation rule is generated based on the distribution pattern of the particle size distribution. The particle size segmentation rule includes identifying the aggregation intervals according to the distribution pattern and using the boundaries of the aggregation intervals as the segmentation boundaries.

10. The method for rapid prediction of soil microplastics based on mobile phone super macro imaging according to claim 9, characterized in that, The risk levels of microplastic pollution from output include: Obtain the risk classification boundary conditions, which are simultaneously associated with the microplastic abundance index range and the structural feature range of particle size distribution. The structural feature range of particle size distribution includes the range of fine particle proportion and the range of coarse particle proportion. A risk level set is constructed based on the risk classification boundary conditions. The risk level set includes at least the first risk level, the second risk level, and the third risk level. The risk level is determined by jointly matching the microplastic abundance index with the structural feature intervals of particle size distribution according to a progressive judgment order, where: When the microplastic abundance index falls into the first abundance range and the proportion of fine particles falls into the first fine particle proportion range, it is determined to be the first risk level. When the microplastic abundance index falls into the second abundance range, or when the microplastic abundance index falls into the first abundance range and the fine particle proportion falls into the second fine particle proportion range, it is determined to be the second risk level. When the microplastic abundance index falls into the third abundance range, or the microplastic abundance index falls into the second abundance range and the proportion of fine particles falls into the second fine particle proportion range, or the proportion of coarse particles falls into the third coarse particle proportion range, it is determined to be the third risk level. Output the microplastic pollution risk level and write the microplastic abundance index, particle number, particle size distribution and scale conversion parameters into the detection result record to obtain a traceable record.

Images