Intelligent identification method and device for micro-particle motion trajectory

Abstract

This invention provides an intelligent method and apparatus for recognizing the motion trajectories of microparticles. The method includes: acquiring consecutive frames of target images; for each frame of the consecutive target images, dynamically calculating an adaptive threshold corresponding to each local region based on the grayscale distribution characteristics of each local region in the target image, and segmenting the target image based on the adaptive threshold to obtain a binary image; determining particle information in the target image based on the binary image; constructing a tree-shaped search space based on the particle information; and generating the target motion trajectory of the particles based on the tree-shaped search space. This invention enables high-precision trajectory tracking of high-density microparticles.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to an intelligent method and apparatus for recognizing the motion trajectory of tiny particles. Background Technology

[0002] In fields such as fluid mechanics, particle dynamics, and biomedicine, recording the motion of multiple target microparticles using high-speed photography equipment and accurately extracting particle positions and tracking their trajectories from continuous image sequences are key technologies for flow field analysis, cell migration research, and industrial visual inspection. However, existing methods typically employ fixed or global threshold segmentation techniques in the image segmentation stage, which are ill-suited to complex imaging conditions such as uneven illumination and varying contrast. This leads to inaccurate separation of particles from the background, severely impacting the overall accuracy of microparticle trajectory recognition. Summary of the Invention

[0003] This invention provides an intelligent method and apparatus for recognizing the trajectory of microparticles, which solves the problem that the use of fixed threshold segmentation technology in the prior art affects the overall recognition accuracy of the trajectory of microparticles.

[0004] This invention provides an intelligent method for recognizing the trajectory of tiny particles, comprising the following steps: Acquire consecutive frames of target images; For each frame of the continuous target image, based on the gray-scale distribution characteristics of each local region in the target image, an adaptive threshold corresponding to each local region is dynamically calculated, and the target image is segmented based on the adaptive threshold to obtain a binary image; Based on the binary image, determine the particle information in the target image; Based on the particle information, a tree-shaped search space is constructed; Based on the tree-shaped search space, the target motion trajectory of the particle is generated.

[0005] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, the acquisition of continuous frame target images includes: Obtain the raw images of consecutive frames; The original images of the consecutive frames are preprocessed to obtain the target images of the consecutive frames. The preprocessing includes at least one of highlight suppression, shadow compensation, grid removal, and multi-level smoothing filtering.

[0006] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, the step of preprocessing the continuous frame original images to obtain continuous frame target images includes: For each frame of the continuous original image, the original image is divided into several local blocks, and the local threshold of each local block is calculated. Pixels with gray values ​​higher than the local threshold in each local block are identified as highlight pixels, and a highlight mask is generated based on the highlight pixels. Based on the highlighted pixels, the proportion of highlighted area in the original image is calculated. When the proportion of the highlight area is less than a preset threshold, a morphological dilation operation is performed on the highlight mask to obtain the target highlight mask. Based on the target specular mask, specular inpainting is performed on the original image to obtain the target image.

[0007] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, the step of preprocessing the continuous frame original images to obtain continuous frame target images includes: For each frame of the continuous frame original image, obtain the luminance component of the original image; Perform a morphological closing operation on the brightness components to obtain a background brightness distribution map; Calculate the difference between the luminance component and the background luminance distribution map, and perform shadow compensation on the luminance component based on the difference to obtain the target luminance component; The target image is determined based on the original image and the target brightness component.

[0008] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, the step of dynamically calculating the adaptive threshold corresponding to each local region based on the gray-level distribution characteristics of each local region in the target image includes: Based on the gray-level distribution characteristics of each local region in the target image, the local image features corresponding to each local region are determined, wherein the local image features include local standard deviation, gradient mean, and information entropy. Based on the local image features, a comprehensive feature value is determined; The comprehensive feature values ​​are mapped to a preset response value range by a nonlinear mapping function to obtain the dynamic response value corresponding to each local region. Based on the local standard deviation, the dynamic response value is constrained and adjusted to obtain the target dynamic response value, and an adaptive threshold corresponding to each local region is calculated based on the target dynamic response value.

[0009] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, the step of determining the particle information in the target image based on the binary image includes: The connected component contours are extracted from the binary image, and valid particle contours are selected from the connected component contours based on geometric properties. Calculate the average perimeter based on the effective particle profile; The number of particles in the effective particle profile is determined based on the ratio of the perimeter of the effective particle profile to the average perimeter. Based on the number of particles, equally divided contour points are determined, and particle information in the target image is determined based on the equally divided contour points.

[0010] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, the step of determining the particle information in the target image based on the binary image includes: Based on the binary image, determine the initial particle information in the target image; In response to a user's manual addition or deletion operation, the initial particle information is corrected to obtain the particle information in the target image.

[0011] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, the step of constructing a tree-shaped search space based on the particle information includes: Based on the particle information, each particle in the first frame of the target image is used as the root node of each trajectory tree. Candidate particles associated with the particles in the current frame in each of the other frames of the target image (excluding the first frame) are added as child nodes to the corresponding trajectory trees to construct a tree-shaped search space.

[0012] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, generating the target motion trajectory of the particle based on the tree-shaped search space includes: Define constraints, including motion speed constraints, motion smoothness constraints, directional consistency constraints, and distance constraints. The tree-shaped search space is traversed using a depth-first search strategy, and the motion trajectories that satisfy the constraints are selected as candidate motion trajectories. Based on the candidate motion trajectories, the target motion trajectory of the particle is determined.

[0013] According to the intelligent recognition method for the motion trajectory of microparticles provided by the present invention, determining the target motion trajectory of the particle based on the candidate motion trajectories includes: The number of path points of the candidate motion path is determined, and the comprehensive evaluation index of the candidate motion path is calculated; wherein, the comprehensive evaluation index includes a smoothness cost-effectiveness index and a linearity index; Based on the number of path points and the comprehensive evaluation index, the optimal motion trajectory is selected from the candidate motion paths as the target motion trajectory of the particle.

[0014] The present invention also provides an intelligent recognition device for the trajectory of tiny particles, comprising the following modules: The acquisition module is used to acquire consecutive frames of target images; The calculation module is used to dynamically calculate the adaptive threshold corresponding to each local region for each frame of the target image in the continuous frame target image, based on the gray-level distribution characteristics of each local region in the target image, and to segment the target image based on the adaptive threshold to obtain a binary image; An extraction module is used to determine particle information in the target image based on the binary image; A construction module is used to construct a tree-shaped search space based on the particle information; The generation module is used to generate the target motion trajectory of the particles based on the tree-shaped search space.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent recognition method for the motion trajectory of microparticles as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent recognition method for the motion trajectory of microparticles as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements an intelligent recognition method for the motion trajectory of microparticles as described above.

[0018] This invention provides an intelligent method and apparatus for recognizing the motion trajectory of tiny particles. The method involves acquiring consecutive frames of target images; for each frame of the consecutive target images, dynamically calculating an adaptive threshold corresponding to each local region based on the grayscale distribution characteristics of each local region in the target image, and segmenting the target image based on the adaptive threshold to obtain a binary image; determining particle information in the target image based on the binary image; constructing a tree-shaped search space based on the particle information; and generating the target motion trajectory of the particles based on the tree-shaped search space. This invention solves the technical problem of the overall recognition accuracy of tiny particle motion trajectories being affected by fixed threshold segmentation techniques. Compared with existing technologies, this invention dynamically calculates adaptive thresholds based on the grayscale distribution characteristics of local regions for segmentation, automatically adjusting segmentation parameters according to the contrast and lighting conditions of different regions of the image. This effectively overcomes the influence of uneven lighting and complex backgrounds on the segmentation effect, significantly improving the accuracy of particle-background separation, thus ensuring the reliability of particle information extraction at the source, and ultimately achieving high-precision recognition and robust tracking of tiny particle motion trajectories under complex imaging conditions. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts illustrating the intelligent recognition method for the motion trajectory of microparticles provided by this invention.

[0021] Figure 2 This is a system overall framework diagram of the intelligent recognition method for the motion trajectory of microparticles provided by the present invention.

[0022] Figure 3 This is the second flowchart of the intelligent recognition method for the motion trajectory of microparticles provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the structure of the intelligent recognition device for the motion trajectory of microparticles provided by the present invention.

[0024] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] The following is combined with Figure 1 and Figure 3 The present invention describes an intelligent identification method for the motion trajectory of microparticles. This method is applicable to the intelligent identification of the motion trajectory of any microparticle. The subject executing this method can be an electronic device or an intelligent identification device for the motion trajectory of microparticles installed in the electronic device. The intelligent identification device for the motion trajectory of microparticles can be implemented by software, hardware, or a combination of both.

[0027] Figure 1 This is one of the flowcharts illustrating the intelligent recognition method for the motion trajectory of microparticles provided by this invention, such as... Figure 1 As shown, the method includes the following: Step 101: Obtain consecutive frame images of the target; It should be noted that the continuous frame target images are a sequence of images captured by high-speed photography equipment, used to capture the motion trajectory of particles in the time dimension. The acquired image sequence supports multiple formats such as BMP, JPG, PNG, and TIFF, and can be automatically converted to grayscale images to ensure the stability of subsequent processing; it is also compatible with resolutions of 1024×768 and above, meeting the requirements of microparticle motion analysis for image detail and spatiotemporal resolution.

[0028] Understandably, after acquiring consecutive frames of target images, preprocessing can be performed to further improve the accuracy and robustness of subsequent particle detection and trajectory recognition. For example, median filtering or Gaussian filtering can be used to remove noise interference introduced during image acquisition; histogram equalization or contrast stretching can enhance image contrast to make particle edges clearer; and for images with uneven illumination, top-hat transform or homomorphic filtering can be used for background correction to eliminate the influence of illumination.

[0029] Step 102: For each frame of the target image in the continuous frame target image, based on the gray-scale distribution characteristics of each local region in the target image, dynamically calculate the adaptive threshold corresponding to each local region, and segment the target image based on the adaptive threshold to obtain a binary image; In practical implementation, the gray-level distribution characteristics (such as contrast and texture) of each local region can be analyzed first, and the local standard deviation, gradient mean, and information entropy can be extracted as local image features to comprehensively characterize the gray-level fluctuations, edge strength, and texture complexity of the local region. Then, the local image features are fused into comprehensive feature values, and mapped to a preset response value range through a nonlinear mapping function to obtain dynamic response values, thereby enhancing the adaptability to complex imaging conditions. Next, the adaptive threshold of each local region is calculated based on the dynamic response values. Finally, based on the adaptive threshold of each local region, each pixel in the local region can be binarized: for example, if the pixel gray-level value is greater than or equal to the adaptive threshold of the local region, the pixel is judged as a foreground pixel; otherwise, it is judged as a background pixel, thus obtaining the binary image of the entire image.

[0030] In the specific implementation, to further improve the quality of the binary image and ensure the integrity and clarity of the particle contours, post-processing optimization can be performed on the binary image obtained from the initial segmentation: First, a slight morphological opening operation is performed to remove scattered noise, and the size of the structuring element is adaptively selected according to the estimated particle size, while filtering out connected regions with too small an area, thus protecting the real particle contours while denoising; Second, a sharpening template such as the Laplacian operator is used to perform convolution operations on the image to highlight the particle edge information; Next, the overall contrast of the image is improved through linear transformation and other methods to enhance the distinction between particles and the background; Finally, a morphological dilation operation is performed using a small-sized structuring element to connect adjacent broken edges, ensuring the connectivity of each particle region, thereby outputting a high-quality final binary image.

[0031] Step 103: Based on the binary image, determine the particle information in the target image; It should be noted that particle information typically includes particle position and particle identifier. In binary images, particles are usually represented as connected regions composed of foreground pixels. Therefore, it is necessary to extract all connected region contours in the binary image and filter out the valid particle contours based on geometric properties (such as area, roundness, etc.). Finally, the centroid of the filtered valid particle contours is calculated to determine the precise position coordinates of each particle.

[0032] Step 104: Based on the particle information, construct a tree-shaped search space; It should be noted that each particle in the first frame of the target image is used as the root node of the trajectory tree. For each subsequent frame of the target image, the malicious program adds candidate particles that match the particles in the current frame as child nodes to the corresponding trajectory tree based on the correlation between particle information (such as distance, motion continuity, etc.). The tree search space is expanded frame by frame to enumerate all possible trajectory path assumptions.

[0033] In the specific implementation, the steps for constructing the tree-structured search space are as follows: (1) Initialize root node: Assign a unique trajectory ID to each particle in each frame of the target image and create the corresponding root node; (2) Associating candidate child nodes: For a certain node (particle) in the i-th frame, search for all possible candidate particles in the i+1-th frame and several subsequent frames, and add them to the tree as its child nodes; (3) Tree structure features: The whole tree represents all possible trajectory development paths starting from the initial frame, and the branches developed from each root node represent a potential complete trajectory.

[0034] Step 105: Based on the tree-shaped search space, generate the target motion trajectory of the particle.

[0035] In a specific implementation, based on the particle information, each particle in the first frame target image is used as the root node of each trajectory tree, and candidate particles associated with the particles in the current frame in each of the other frames of target images (excluding the first frame) are added as child nodes to the corresponding trajectory trees to construct a tree-shaped search space.

[0036] In the specific implementation, a depth-first search strategy can be used to traverse the entire tree structure and collect all paths that meet the preset constraints (such as motion speed, smoothness, and direction consistency) as candidate motion trajectories. Then, a comprehensive evaluation index (used to quantify trajectory quality) is calculated for each candidate motion trajectory. Finally, a set of mutually non-conflicting optimal motion trajectories is determined from the candidate trajectories according to the selection strategy as the target motion trajectory of the particle.

[0037] In specific implementations, such as Figure 2 The system architecture diagram shown illustrates the flow of the target image from input to output through various core modules. First, the image input module reads a sequence of consecutive frames from high-speed photography, supporting multiple formats and automatically converting them to grayscale. Next, the adaptive image preprocessing module sequentially performs highlight suppression, shadow compensation, mesh filtering, multi-level smoothing filtering, adaptive threshold segmentation, and image sharpening on each frame, outputting a high-contrast, low-noise binary image. Then, the target recognition module performs contour detection and filtering on the binary image, separating adhered particles using an intelligent adhesion segmentation algorithm and extracting sub-pixel-level coordinates to generate a high-precision particle coordinate list. Finally, the trajectory tracking module constructs a tree-shaped search space based on the particle coordinates, generating complete particle motion trajectories through multi-constraint recursive search and global optimal path selection. Throughout this process, the human-computer collaborative interaction module provides a visual interface, supporting manual addition / deletion of particles, batch selection processing, and undo / redo operations to correct automatic recognition results and ensure the reliability of the final outcome. Finally, the data export module outputs particle coordinates, trajectory data, visual images, and statistical reports in formats such as CSV, Excel, PNG, and PDF for subsequent analysis and application. The entire system is managed by a parameter configuration module, which supports dynamic adjustment of parameters of each module through XML configuration files or interactive interfaces to adapt to different experimental scenarios and needs.

[0038] In practical implementation, the target motion trajectories of all particles can be integrated and output, providing multi-dimensional data and visualization. Specifically: (1) Coordinate data: Export the subpixel coordinates of all particles in all frames in CSV or Excel format, optionally including physical coordinates; (2) Trajectory data: Export the complete coordinate sequence, trajectory ID, length and quality evaluation index (KPI) of each trajectory in a structured format, so as to facilitate import into MATLAB, Python and other tools for subsequent analysis; (3) Visualized images: Image identification with labels: labeling particle positions and numbers; Trajectory display image: Particle trajectories are drawn in different colors and labeled with trajectory numbers; Preprocessing images: showing the results of each stage, such as highlight removal and mesh suppression; Statistical Report: Generates a PDF report summarizing key performance indicators such as particle recognition rate, adhesion segmentation accuracy, average trajectory length, integrity rate, and average processing time.

[0039] The intelligent recognition method for the motion trajectory of microparticles provided in this invention involves acquiring consecutive frames of target images; for each frame of the consecutive target images, dynamically calculating an adaptive threshold corresponding to each local region based on the gray-level distribution characteristics of each local region in the target image, and segmenting the target image based on the adaptive threshold to obtain a binary image; determining particle information in the target image based on the binary image; constructing a tree-shaped search space based on the particle information; and generating the target motion trajectory of the particle based on the tree-shaped search space. This method solves the technical problem that the use of fixed threshold segmentation techniques affects the overall recognition accuracy of microparticle motion trajectories. Compared with existing technologies, this invention dynamically calculates adaptive thresholds based on the gray-level distribution characteristics of local regions for segmentation, which can automatically adjust the segmentation parameters according to the contrast and lighting conditions of different regions of the image. This effectively overcomes the influence of uneven lighting and complex backgrounds on the segmentation effect, significantly improving the accuracy of particle-background separation, thereby ensuring the reliability of particle information extraction at the source, and ultimately achieving high-precision recognition and robust tracking of microparticle motion trajectories under complex imaging conditions.

[0040] Based on any of the above embodiments, obtaining consecutive frame target images includes: Obtain the raw images of consecutive frames; The original images of the consecutive frames are preprocessed to obtain the target images of the consecutive frames. The preprocessing includes at least one of highlight suppression, shadow compensation, grid removal, and multi-level smoothing filtering.

[0041] It should be noted that images of moving microparticles are often affected by factors such as lighting changes, equipment noise, and background textures during acquisition. Directly using the raw images may lead to false positives or false negatives. Therefore, preprocessing of the raw images is necessary to eliminate or suppress these interfering factors. Highlight suppression reduces areas of strong reflection or overexposure, preventing light spots from being misidentified as particles; shadow compensation corrects local underexposure caused by uneven lighting, ensuring that particles in low-contrast areas can be effectively identified; grid filtering targets periodic background textures (such as microscope scale grids) that may be introduced by the imaging equipment, preventing them from interfering with particle contour extraction; and multi-level smoothing filtering removes random noise through filtering operations at different scales while preserving particle edge details as much as possible.

[0042] The intelligent recognition method for the motion trajectory of tiny particles provided in this invention preprocesses the original images of consecutive frames by performing highlight suppression, shadow compensation, grid filtering and multi-level smoothing filtering. This eliminates the obscuring and misjudgment of particle outlines by the highlight areas formed by particle reflections or device light spots, alleviates local over-darkness caused by uneven lighting to enhance the contrast between particles and background, filters out the interference of periodic background textures on particle extraction, and maintains the clarity of particle edges while suppressing random noise. This makes the target image sequence more realistically restore the particle distribution state and effectively improves the recognizability of tiny particles and image quality.

[0043] Based on any of the above embodiments, the preprocessing of the continuous frame original images to obtain the continuous frame target images includes: For each frame of the continuous original image, the original image is divided into several local blocks, and the local threshold of each local block is calculated. Pixels with gray values ​​higher than the local threshold in each local block are identified as highlight pixels, and a highlight mask is generated based on the highlight pixels. Based on the highlighted pixels, the proportion of highlighted area in the original image is calculated. When the proportion of the highlight area is less than a preset threshold, a morphological dilation operation is performed on the highlight mask to obtain the target highlight mask. Based on the target specular mask, specular inpainting is performed on the original image to obtain the target image.

[0044] It should be noted that specular suppression is used to eliminate specular areas formed by particle reflections or device spots, preventing them from obscuring the true particle outlines or being misidentified as particles; specular mask is used to locate the range of specular areas that need to be repaired in the original image; and target specular mask is a mask obtained by optimizing the initial specular mask through morphological dilation, which can more completely cover the edges and surrounding transition areas of the specular area.

[0045] In practice, the steps for specular highlight suppression are as follows: (1) Local statistical analysis: Divide the original image into several local blocks (such as 8×8 pixel blocks), and calculate the mean gray value μ and standard deviation σ of each local block; (2) Highlight pixel detection: Pixels with gray values ​​higher than (μ+3σ) in local blocks are identified as highlight pixels, and the area ratio of the highlight region is calculated (i.e., the highlight area ratio). (3) Adaptive processing decision: If the proportion of the highlight area exceeds the preset threshold (e.g., 5%), then no highlight repair is required to avoid over-processing of atypical highlight images; if it does not exceed the preset threshold, then highlight repair is performed. (4) Mask optimization: After generating a specular mask based on the specular pixels, perform a morphological dilation operation on the specular pixel mask (e.g., using a 3×3 rectangular structuring element) to obtain the target specular mask, so as to ensure complete coverage of the edge of the specular area; (5) Highlight restoration: Based on the target highlight mask, an image restoration algorithm (such as the Telea algorithm) is used to fill the highlight area, and gamma transformation (such as γ=2.2) is combined to perform intensity compression on the restored area and its surroundings in order to retain more image details.

[0046] The intelligent particle motion trajectory recognition method provided in this invention effectively eliminates highlight areas formed by particle reflections or device light spots by performing highlight suppression preprocessing on consecutive frames of original images, preventing them from obscuring the true particle outlines or being misidentified as particles. This method uses local adaptive thresholding to accurately detect highlight pixels, accurately generates a highlight mask, and intelligently decides whether to perform repair by statistically analyzing the highlight area ratio, effectively avoiding overprocessing of atypical highlight images. For images requiring repair, morphological dilation is used to optimize the mask to completely cover the highlight area edges, and then image inpainting algorithms and gamma transform are combined for filling and intensity compression, preserving rich image details while removing highlights. The target image obtained after highlight suppression processing can more realistically restore the particle distribution state, significantly improving the recognizability of microparticles and image quality.

[0047] Based on any of the above embodiments, the preprocessing of the continuous frame original images to obtain the continuous frame target images includes: For each frame of the continuous frame original image, obtain the luminance component of the original image; Perform a morphological closing operation on the brightness components to obtain a background brightness distribution map; Calculate the difference between the luminance component and the background luminance distribution map, and perform shadow compensation on the luminance component based on the difference to obtain the target luminance component; The target image is determined based on the original image and the target brightness component.

[0048] It should be noted that shadow compensation processing is used to alleviate shadow areas caused by uneven lighting, improve the contrast between particles and the background, and avoid missed or false detections of particles in low-contrast areas. Specifically, firstly, the original image is converted from RGB space to YCbCr space, and the luminance component Y is extracted from it; then, morphological closing operations (such as using a 7×7 rectangular structuring element) are performed on the luminance component, and small dark spots in the luminance component are eliminated by dilation followed by erosion, while maintaining the overall luminance distribution characteristics, thus obtaining a background luminance distribution map, which reflects the overall luminance change trend of the unevenly lit background; next, the difference between the luminance component and the background luminance distribution map is calculated. The pixel values ​​in the shadow areas of this difference image are low. After inverting and normalizing to the 0-255 grayscale dynamic range, the luminance of the shadow areas can be effectively enhanced, making them more consistent with the surrounding background; finally, based on the target luminance component obtained after compensation, combined with the chromaticity information of the original image, the target image after shadow compensation is obtained.

[0049] The intelligent recognition method for the motion trajectory of tiny particles provided in this invention extracts the brightness component and performs morphological closing operation to accurately estimate the background brightness distribution. Then, it calculates the difference between the brightness component and the background brightness distribution map to perform shadow compensation, which effectively alleviates the local dark areas caused by uneven lighting, significantly improves the contrast between particles and the background, and makes the corrected target brightness component more realistically reflect the particle distribution. The final target image sequence has a more uniform brightness and clearer particle outlines.

[0050] Based on any of the above embodiments, the preprocessing of the continuous frame original images to obtain the continuous frame target images includes: For each frame of the continuous original image, perform a Fourier transform on the original image to obtain a frequency domain image; The target frequency domain image is obtained by filtering the frequency domain image using a band-stop filter, wherein the band-stop filter is designed based on the frequency components corresponding to the periodic grid lines in the frequency domain. Perform an inverse Fourier transform on the target frequency domain image to obtain the target image after filtering out the grid lines.

[0051] It should be noted that mesh filtering is used to eliminate interference from periodic mesh lines introduced by microscopic imaging or specific acquisition equipment, preventing these background textures from being misidentified as particle outlines or affecting segmentation accuracy during subsequent particle extraction. Specifically, two optional mesh filtering strategies are provided: The first is a frequency domain filtering method. First, a Fourier transform is performed on the original image to obtain a frequency domain image. Then, a band-stop filter is designed based on the frequency components corresponding to the periodic mesh lines in the frequency domain for filtering. Finally, an inverse Fourier transform is used to recover the spatial domain image, thereby effectively filtering out mesh lines. The second is a morphological top-hat transform method. A morphological opening operation is performed on the original image using linear structuring elements that match the mesh line width. Then, the opening operation result is subtracted from the original image to obtain the top-hat transformed image, thereby suppressing mesh noise.

[0052] The intelligent identification method for the motion trajectory of tiny particles provided in this invention transforms the original image into the frequency domain by performing a Fourier transform, then uses a band-stop filter designed for the frequency components of periodic grid lines to filter out the corresponding frequency components, and finally restores the spatial domain image by inverse Fourier transform. This effectively eliminates the interference of periodic grid lines introduced by microscopic imaging or acquisition equipment, avoids the background texture being misjudged as particle outlines in subsequent processing, and improves the accuracy of particle extraction and image quality.

[0053] Based on any of the above embodiments, the preprocessing of the continuous frame original images to obtain the continuous frame target images includes: For each frame of the continuous original image, a small-scale Gaussian kernel is used to perform multiple iterative filterings on the original image, and a reflection boundary mode is used to process the image edges in each filtering process to obtain a smoothed target image.

[0054] It should be noted that multi-level smoothing filtering is used to suppress random noise in the image while maintaining the edge sharpness of tiny particles as much as possible.

[0055] Understandably, since adaptive thresholding and particle localization are sensitive to edge details, using a large-scale Gaussian kernel for single-pass filtering, while effective in denoising, can easily lead to edge blurring; conversely, using only a small-scale Gaussian kernel for single-pass filtering has limited denoising capabilities. Therefore, a small-scale Gaussian kernel (e.g., 3×3, σ=0.5) can be used for multiple iterative filtering steps. This gradually reduces noise interference through successive smoothing while preserving particle edge details. Furthermore, a reflection boundary mode is used to process image edges during each filtering process, filling the convolution window with the mirror image values ​​of pixels outside the edges, effectively preventing image edge distortion or artifacts.

[0056] The intelligent recognition method for the motion trajectory of tiny particles provided in this invention uses a small-scale Gaussian kernel for multiple iterative filtering to gradually suppress random noise during the smoothing process, while avoiding particle edge blurring caused by large-scale filtering. Combined with reflection boundary mode to process image edges, it effectively prevents distortion or artifacts in edge areas, thereby achieving a good balance between noise reduction and edge preservation, obtaining a smooth target image with clear edges and a clean background, and significantly improving image quality.

[0057] Based on any of the above embodiments, the step of dynamically calculating the adaptive threshold corresponding to each local region based on the gray-level distribution characteristics of each local region in the target image includes: Based on the gray-level distribution characteristics of each local region in the target image, the local image features corresponding to each local region are determined, wherein the local image features include local standard deviation, gradient mean, and information entropy. Based on the local image features, a comprehensive feature value is determined; The comprehensive feature values ​​are mapped to a preset response value range by a nonlinear mapping function to obtain the dynamic response value corresponding to each local region. Based on the local standard deviation, the dynamic response value is constrained and adjusted to obtain the target dynamic response value, and an adaptive threshold corresponding to each local region is calculated based on the target dynamic response value.

[0058] It should be noted that local standard deviation Used to characterize the contrast strength of a local region; gradient mean Used to reflect the edge strength of a local region; information entropy Used to describe the randomness and complexity of grayscale distribution in a local area.

[0059] In practical implementation, the dynamic C value (i.e., dynamic response value) of each local region can be calculated using a nonlinear mapping function. Specifically, it can be... , , The three eigenvalues ​​are normalized to the range [0,1], resulting in... , , Then, the normalized features are weighted and summed to obtain a comprehensive feature value F: in, , , Weighting coefficients (can be taken as follows) , , ),satisfy .

[0060] Next, the composite eigenvalue F is mapped to the C value range [3.0, 6.0] using a nonlinear mapping function, such as a piecewise linear function or a sigmoid function. For example, a piecewise function can be used: Alternatively, a sigmoid function can be used: Where k is an adjustment parameter (typical value k=6).

[0061] Finally, to further enhance adaptability, strategic constraints can be imposed on the dynamic C value. For example: when When <0.3 (low contrast region), forced constraint To improve sensitivity; when When >0.7 (high contrast region), forced constraint To suppress noise.

[0062] The intelligent recognition method for the motion trajectory of tiny particles provided in this invention analyzes the gray-level distribution characteristics of local regions, extracts local standard deviation, gradient mean, and information entropy as local image features, and comprehensively characterizes the gray-level fluctuations, edge intensity, and texture complexity of each region. Then, it fuses these multi-dimensional features into a comprehensive feature value, transforms it to a preset response value range through nonlinear mapping, and obtains a dynamic response value to sensitively reflect local imaging differences. Finally, it uses the local standard deviation to constrain and adjust the dynamic response value, avoiding threshold instability caused by noise or sudden gray-level changes, thereby generating an accurate adaptive threshold for each local region. This method can dynamically adjust the segmentation threshold according to the local characteristics of the image, effectively overcoming the problem that global thresholds cannot adapt to uneven local illumination or complex backgrounds, and significantly improving the accuracy and adaptability of particle segmentation.

[0063] Based on any of the above embodiments, determining the particle information in the target image based on the binary image includes: The connected component contours are extracted from the binary image, and valid particle contours are selected from the connected component contours based on geometric properties. Calculate the average perimeter based on the effective particle profile; The number of particles in the effective particle profile is determined based on the ratio of the perimeter of the effective particle profile to the average perimeter. Based on the number of particles, equally divided contour points are determined, and particle information in the target image is determined based on the equally divided contour points.

[0064] It should be noted that, in a binary image, the contour of a connected region refers to the boundary of a region formed by connecting adjacent foreground pixels (usually white pixels, representing particle regions); in a binary image, each connected region corresponds to a pixel block composed of foreground pixels, while the contour is the set of pixels at the edge of the region, describing the shape and boundary position of the region.

[0065] In practical implementation, a contour discovery algorithm (such as the Suzuki algorithm) can be used to extract all connected component contours from the binary image, and geometric features such as area, perimeter, and aspect ratio of the circumscribed rectangle are calculated for each contour. Then, noise and non-particle contours are filtered out according to a preset range of particle geometric parameters (such as minimum area, maximum area, aspect ratio threshold, etc.) to obtain a list of effective particle contours. Next, the average perimeter of all effective particle contours is calculated and used as the reference size of a single particle. For each effective particle contour, the ratio of its perimeter to the average perimeter is calculated, and the number of particles N contained in the contour is determined according to a preset ratio range (e.g., 1.5~2.5 times indicates 2 particles are stuck together, 2.5~3.5 times indicates 3 particles are stuck together, and so on). If N=1, the centroid coordinates of the contour are directly calculated as particle information; if N>1, the total length is calculated along the discrete point sequence of the contour, and (N-1) equally spaced division points are determined on the contour according to N, and the coordinates of these division points are used as the center positions of each particle after segmentation. To further ensure the rationality of the segmentation, the distance between adjacent equally spaced points can be verified. If the distance is too close or too far, it can be fine-tuned to make the segmentation result more consistent with the actual particle distribution.

[0066] In the specific implementation, to further improve particle positioning accuracy and meet the high-precision requirements of scientific research, sub-pixel level coordinate extraction is performed on each particle: First, the zero-order moment m00 and first-order moments m10 and m01 are calculated for each particle contour, and the sub-pixel level centroid coordinates are obtained using the centroid coordinate formula; then, the coordinate accuracy is optimized to multiple decimal places using methods such as linear interpolation; finally, the pixel coordinates are converted into actual physical coordinates according to preset mapping parameters to obtain the final high-precision particle information. Specifically, the calculation formula for calculating the particle centroid coordinates based on moments is as follows: In the formula, m00 is the zeroth moment, representing the area of ​​the particle region (i.e., the total number of pixels in the region); m10 and m01 are the first moments of the image in the x and y directions, respectively, representing the sum of the x and y coordinates of all pixels in the particle region.

[0067] It is understandable that for binary images, pixel grayscale values ​​are only 0 or 1, therefore the above moment calculation is based solely on pixel coordinates and does not require consideration of grayscale weighting. The (x...) calculated using this formula... c,y c This refers to the sub-pixel level centroid coordinates of the particle, whose accuracy can break through the integer pixel limit and reach multiple decimal places, providing accurate particle information data for subsequent high-precision trajectory analysis.

[0068] The intelligent particle motion trajectory recognition method provided in this invention extracts connected component contours from a binary image and filters valid particle contours based on geometric characteristics, effectively eliminating noise and false target interference. Then, it calculates the average perimeter of all valid contours as a benchmark, and adaptively determines the number of particles contained within each contour based on the ratio of its perimeter to the average perimeter, achieving intelligent segmentation of particle clumps with different degrees of adhesion. Finally, it determines equally divided contour points based on the determined number of particles, and uses the coordinates of these points as the center positions of each segmented particle, completing particle extraction. This method does not require pre-specified seed points or reliance on complex models, and can accurately separate adhered particles and precisely locate each particle, significantly improving the accuracy and robustness of particle extraction.

[0069] Based on any of the above embodiments, determining the particle information in the target image based on the binary image includes: Based on the binary image, determine the initial particle information in the target image; In response to a user's manual addition or deletion operation, the initial particle information is corrected to obtain the particle information in the target image.

[0070] In practical implementation, to further improve the accuracy of particle extraction, during the automatic recognition process, some particles may be missed or falsely detected due to factors such as uneven lighting, particle adhesion, or background interference. Therefore, a human-machine collaborative correction function is provided: users can view the initial particle positions of the current frame through a graphical interface, manually click to add missed particles, and the sub-pixel coordinates of the added particles are automatically calculated and included in the initial particle information; at the same time, users can also choose to delete falsely detected particles, removing them from the initial particle information. All add and delete operations support multi-step undo and redo, facilitating user adjustments.

[0071] The intelligent identification method for the motion trajectory of microparticles provided in this invention adds a human-machine collaborative correction step on the basis of automatic particle information identification. By responding to the user's manual addition operation, missed particles are added, and by responding to the user's manual deletion operation, falsely detected particles are removed. This effectively compensates for missed and false detections caused by uneven lighting, particle adhesion or background interference during the automatic identification process, and significantly improves the integrity and accuracy of particle information.

[0072] Figure 3 This is the second flowchart illustrating the intelligent recognition method for the motion trajectory of microparticles provided by this invention, as shown below. Figure 3As shown, step 105 also includes steps 1051 to 1053: Step 1051: Set constraints, wherein the constraints include motion speed constraints, motion smoothness constraints, direction consistency constraints, and distance constraints; It should be noted that the constraints are used to establish physical plausibility screening criteria during the inter-frame correlation process of particles, ensuring that candidate particles added to the trajectory tree conform to real-world motion laws. Specifically, the motion velocity constraint limits the particle's motion velocity between consecutive frames to a preset minimum velocity v. min With maximum speed v max Within the specified range, false associations caused by noise or mismatches that are too fast or too slow are eliminated; motion smoothness constraints require that the speed variation of adjacent segments on the trajectory does not exceed the allowable range, ensuring the continuity and stability of particle motion; direction consistency constraints are for longer trajectories, measuring the continuity of motion direction by the linearity of the fitted straight line at multiple points, avoiding frequent abnormal turns or jitters in the trajectory; distance constraints limit the Euclidean distance between particles in frames to a reasonable range [dmin, dmax], directly eliminating false matches at long distances from a spatial position perspective.

[0073] In practical implementation, appearance constraints can also be introduced as auxiliary screening conditions, requiring candidate particles to be similar to the current particle in appearance features such as grayscale and size, thereby further improving the accuracy of inter-frame association.

[0074] It should be noted that during the recursive traversal of the tree-shaped search space, if a particle in the current frame does not find a candidate particle that meets the constraints in the next frame, it is allowed to skip one or two frames and attempt to continue matching in frame i+2 or i+3, thereby re-associating the temporarily lost particle back with the current trajectory tree. This cross-frame search mechanism can be used to cope with the interruption of inter-frame association caused by particle occlusion, detection failure, or temporary disappearance, thereby effectively improving the continuity and integrity of the trajectory and avoiding trajectory breakage caused by the absence of individual frames.

[0075] Step 1052: Use a depth-first search strategy to traverse the tree search space and select the motion trajectories that satisfy the constraints as candidate motion trajectories; It should be noted that a depth-first search strategy is used to traverse the entire tree structure of the tree search space. Starting from each root node, child nodes are recursively visited along the inter-frame relationships. During the traversal, the current motion trajectory is continuously verified to ensure that it meets the constraints. Only trajectories that meet all constraints are retained as candidate motion trajectories. After the traversal is completed, each valid path extending from the root node to the leaf node represents a candidate motion trajectory.

[0076] Step 1053: Based on the candidate motion trajectories, determine the target motion trajectory of the particle.

[0077] In the specific implementation, the trajectory with the most frames can be selected first according to the principle of prioritizing the number of path points to ensure the integrity of the trajectory. When multiple candidate trajectories have the same number of path points, their comprehensive evaluation index (KPI) is compared, and the trajectory with the smallest KPI (i.e., the best combination of smoothness and linearity) is selected. At the same time, a minimum quality threshold is set to remove low-quality trajectory segments with poor KPI (such as linearity below 0.7). For multiple conflicting trajectories sharing the same particle node, only the one with the smallest KPI is retained to ensure that each particle has a unique trajectory identifier.

[0078] The intelligent recognition method for the motion trajectory of tiny particles provided in this invention constructs a trajectory search tree with the particle detected in the first frame as the root node. During the recursive search process, multiple constraints such as motion speed, motion smoothness, trajectory linearity, spatial distance, and apparent similarity are comprehensively introduced. It also has cross-frame search capability, enabling continuous association and intelligent recovery of the trajectory after the target is briefly lost. Furthermore, it outputs complete, smooth, and non-conflicting motion trajectories through global optimal path selection, effectively solving the problem of insufficient robustness of trajectory tracking in complex scenarios.

[0079] Based on any of the above embodiments, determining the target motion trajectory of the particle based on the candidate motion trajectory includes: The number of path points of the candidate motion path is determined, and the comprehensive evaluation index of the candidate motion path is calculated; wherein, the comprehensive evaluation index includes a smoothness cost-effectiveness index and a linearity index; Based on the number of path points and the comprehensive evaluation index, the optimal motion trajectory is selected from the candidate motion paths as the target motion trajectory of the particle.

[0080] In practice, the comprehensive evaluation index is calculated as follows: In the formula, This represents a smoothness index (0~1, the closer to 1 the smoother the surface). This represents the linearity index (0~1, the closer to 1, the more linear). Indicates the linearity weight (0~1, configurable, default 0.5).

[0081] In the specific implementation, the number of path points (i.e., the number of particles contained in the trajectory) of each candidate motion trajectory is first determined, and its comprehensive evaluation index KPI (which can be determined based on the smoothness index and linearity index) is calculated to quantify the overall quality of the trajectory. Then, the trajectory with the most path points is selected first to ensure the integrity and continuity of the trajectory. When multiple trajectories have the same number of path points, the trajectory with the lowest KPI is selected, i.e., the trajectory with the best quality. At the same time, a minimum quality threshold is set (e.g., trajectories with a linearity index < 0.7 are eliminated) to exclude trajectory segments with poor quality. For multiple conflicting trajectories sharing the same particle node, only the one with the lowest KPI is retained to ensure that each particle has a unique trajectory ID. Through the above strategies, a complete, accurate, and non-conflicting set of particle target motion trajectories is finally output.

[0082] The intelligent recognition method for the motion trajectory of microparticles provided in this invention determines the number of path points for each candidate motion trajectory and calculates a comprehensive evaluation index weighted by a smoothness evaluation index and a linearity evaluation index to quantitatively evaluate the integrity and motion quality of the trajectory. Then, based on the principle of prioritizing the number of path points and selecting the optimal comprehensive evaluation index, the target motion trajectory of the particle is selected to ensure that the final output trajectory has the longest number of consecutive frames and the best smoothness and directional consistency.

[0083] The intelligent recognition device for the motion trajectory of microparticles provided by this invention will be described below. The intelligent recognition device for the motion trajectory of microparticles described below can be referred to in correspondence with the intelligent recognition method for the motion trajectory of microparticles described above. Figure 4 As shown, the intelligent recognition device for the trajectory of the microparticles includes: Acquisition module 10 is used to acquire consecutive frame target images; The calculation module 20 is used to dynamically calculate the adaptive threshold corresponding to each local region for each frame of the target image in the continuous frame target image, based on the gray-level distribution characteristics of each local region in the target image, and to segment the target image based on the adaptive threshold to obtain a binary image. Extraction module 30 is used to determine particle information in the target image based on the binary image; Construction module 40 is used to construct a tree-shaped search space based on the particle information; The generation module 50 is used to generate the target motion trajectory of the particle based on the tree-shaped search space.

[0084] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute an intelligent recognition method for the motion trajectory of tiny particles. This method includes: acquiring consecutive frames of target images; for each frame of the consecutive frames of target images, dynamically calculating an adaptive threshold corresponding to each local region based on the grayscale distribution characteristics of each local region in the target image, and segmenting the target image based on the adaptive threshold to obtain a binary image; determining particle information in the target image based on the binary image; constructing a tree-shaped search space based on the particle information; and generating the target motion trajectory of the particles based on the tree-shaped search space.

[0085] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the intelligent recognition method for the motion trajectory of tiny particles provided by the above methods. The method includes: acquiring consecutive frame target images; for each frame of the consecutive frame target images, dynamically calculating an adaptive threshold corresponding to each local region based on the gray-scale distribution characteristics of each local region in the target image, and segmenting the target image based on the adaptive threshold to obtain a binary image; determining particle information in the target image based on the binary image; constructing a tree-shaped search space based on the particle information; and generating the target motion trajectory of the particles based on the tree-shaped search space.

[0087] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements an intelligent method for recognizing the motion trajectory of tiny particles provided by the methods described above. The method includes: acquiring consecutive frame target images; for each frame of the consecutive frame target images, dynamically calculating an adaptive threshold corresponding to each local region based on the gray-scale distribution characteristics of each local region in the target image, and segmenting the target image based on the adaptive threshold to obtain a binary image; determining particle information in the target image based on the binary image; constructing a tree-shaped search space based on the particle information; and generating the target motion trajectory of the particles based on the tree-shaped search space.

[0088] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0090] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligently identifying the trajectory of a microparticle, characterized in that, include: Acquire consecutive frames of target images; For each frame of the continuous target image, based on the gray-scale distribution characteristics of each local region in the target image, an adaptive threshold corresponding to each local region is dynamically calculated, and the target image is segmented based on the adaptive threshold to obtain a binary image; Based on the binary image, determine the particle information in the target image; Based on the particle information, a tree-shaped search space is constructed; Based on the tree-shaped search space, the target motion trajectory of the particle is generated. 2.The intelligent identification method of micro-particle motion trajectory according to claim 1, characterized in that, The acquisition of consecutive frame target images includes: Obtain the raw images of consecutive frames; The original images of the consecutive frames are preprocessed to obtain the target images of the consecutive frames. The preprocessing includes at least one of highlight suppression, shadow compensation, grid removal, and multi-level smoothing filtering.

3. The intelligent recognition method for the motion trajectory of microparticles according to claim 2, characterized in that, The step of preprocessing the continuous frame original images to obtain continuous frame target images includes: For each frame of the continuous original image, the original image is divided into several local blocks, and the local threshold of each local block is calculated. Pixels with gray values ​​higher than the local threshold in each local block are identified as highlight pixels, and a highlight mask is generated based on the highlight pixels. Based on the highlighted pixels, the proportion of highlighted area in the original image is calculated. When the proportion of the highlight area is less than a preset threshold, a morphological dilation operation is performed on the highlight mask to obtain the target highlight mask. Based on the target specular mask, specular inpainting is performed on the original image to obtain the target image.

4. The intelligent recognition method for the motion trajectory of microparticles according to claim 2, characterized in that, The step of preprocessing the continuous frame original images to obtain continuous frame target images includes: For each frame of the continuous frame original image, obtain the luminance component of the original image; Perform a morphological closing operation on the brightness components to obtain a background brightness distribution map; Calculate the difference between the luminance component and the background luminance distribution map, and perform shadow compensation on the luminance component based on the difference to obtain the target luminance component; The target image is determined based on the original image and the target brightness component.

5. The intelligent recognition method for the motion trajectory of microparticles according to claim 1, characterized in that, The step of dynamically calculating the adaptive threshold corresponding to each local region based on the gray-level distribution characteristics of each local region in the target image includes: Based on the gray-level distribution characteristics of each local region in the target image, the local image features corresponding to each local region are determined, wherein the local image features include local standard deviation, gradient mean, and information entropy. Based on the local image features, a comprehensive feature value is determined; The comprehensive feature values ​​are mapped to a preset response value range by a nonlinear mapping function to obtain the dynamic response value corresponding to each local region. Based on the local standard deviation, the dynamic response value is constrained and adjusted to obtain the target dynamic response value, and an adaptive threshold corresponding to each local region is calculated based on the target dynamic response value.

6. The intelligent recognition method for the motion trajectory of microparticles according to claim 1, characterized in that, The step of determining particle information in the target image based on the binary image includes: The connected component contours are extracted from the binary image, and valid particle contours are selected from the connected component contours based on geometric properties. Calculate the average perimeter based on the effective particle profile; The number of particles in the effective particle profile is determined based on the ratio of the perimeter of the effective particle profile to the average perimeter. Based on the number of particles, equally divided contour points are determined, and particle information in the target image is determined based on the equally divided contour points.

7. The intelligent recognition method for the motion trajectory of microparticles according to claim 1, characterized in that, The step of determining particle information in the target image based on the binary image includes: Based on the binary image, determine the initial particle information in the target image; In response to a user's manual addition or deletion operation, the initial particle information is corrected to obtain the particle information in the target image.

8. The intelligent recognition method for the motion trajectory of microparticles according to claim 1, characterized in that, The construction of the tree-shaped search space based on the particle information includes: Based on the particle information, each particle in the first frame of the target image is used as the root node of each trajectory tree. Candidate particles associated with the particles in the current frame in each of the other frames of the target image (excluding the first frame) are added as child nodes to the corresponding trajectory trees to construct a tree-shaped search space.

9. The intelligent recognition method for the motion trajectory of microparticles according to claim 1, characterized in that, The generation of the target motion trajectory of the particle based on the tree-shaped search space includes: Define constraints, including motion speed constraints, motion smoothness constraints, directional consistency constraints, and distance constraints. The tree-shaped search space is traversed using a depth-first search strategy, and the motion trajectories that satisfy the constraints are selected as candidate motion trajectories. Based on the candidate motion trajectories, the target motion trajectory of the particle is determined.

10. The intelligent recognition method for the motion trajectory of microparticles according to claim 9, characterized in that, Determining the target trajectory of the particle based on the candidate trajectories includes: The number of path points of the candidate motion path is determined, and the comprehensive evaluation index of the candidate motion path is calculated; wherein, the comprehensive evaluation index includes a smoothness cost-effectiveness index and a linearity index; Based on the number of path points and the comprehensive evaluation index, the optimal motion trajectory is selected from the candidate motion paths as the target motion trajectory of the particle.

11. An intelligent recognition device for the trajectory of tiny particles, characterized in that, include: The acquisition module is used to acquire consecutive frames of target images; The calculation module is used to dynamically calculate the adaptive threshold corresponding to each local region for each frame of the target image in the continuous frame target image, based on the gray-level distribution characteristics of each local region in the target image, and to segment the target image based on the adaptive threshold to obtain a binary image; An extraction module is used to determine particle information in the target image based on the binary image; A construction module is used to construct a tree-shaped search space based on the particle information; The generation module is used to generate the target motion trajectory of the particles based on the tree-shaped search space.

Images