An optical imaging method and system for intelligent analysis of plankton

Abstract

The present application relates to a kind of optical imaging method, system for phytoplankton intelligent analysis, comprising the following steps: S1, according to the characteristic size of phytoplankton, the depth of field of microscope system, objective numerical aperture, total magnification are determined;S2, phytoplankton in sample liquid is extruded to form liquid flow, and the phytoplankton pre-image of phytoplankton is photographed by front camera;S3, the phytoplankton pre-image is processed, and the growth cycle characteristics, morphological characteristics, activity characteristics of corresponding phytoplankton are extracted;S4, according to the growth cycle characteristics, the morphological characteristics, the activity characteristics, dynamically determine the liquid flow velocity of corresponding phytoplankton, drive corresponding phytoplankton to enter the field of view range of the microscope system, improve the probability that the phytoplankton passes through the field of view range along the optical axis direction of microscope with its minimum geometric dimension;S5, the main image of phytoplankton is collected by the microscope system.

Description

Technical Field

[0001] This invention relates to the field of optical imaging of phytoplankton, specifically to an optical imaging method and system for intelligent analysis of phytoplankton. Background Technology

[0002] The fundamental cause of red tides lies in eutrophication of seawater. Excessive nitrogen, phosphorus, and other nutrients from industrial wastewater, domestic sewage, and agricultural discharges enter the ocean, providing a material basis for the outbreak of phytoplankton (such as diatoms and dinoflagellates). When environmental conditions such as water temperature and light are suitable, red tides are easily induced. Given the sudden onset, complex causes, and significant harm of red tides, timely and accurate monitoring is crucial. These algae are tiny, with particle sizes mostly ranging from 5 to 50 µm. Therefore, achieving precise identification of specific algae and quantitative analysis of their population dynamics is a key prerequisite for effective monitoring, early warning, and subsequent research.

[0003] Current technologies typically employ electron microscopy to image phytoplankton, then use the images to determine species, growth cycle, and other characteristics based on algal cell morphology. However, electron microscopes are optimized for observing thin, two-dimensional samples, with a depth of field of only about 0.5µm, allowing observation of details only on a single horizontal cross-section of the algal cell. Algal cells, however, are three-dimensional structures, varying in size, shape, and orientation. After exiting the microscope, algal cells move along Jeffrey tracks in the liquid, causing them to flip and resulting in random placement once inside the microscope's field of view. Current techniques require multiple focusing adjustments to observe algal cells in multiple longitudinal dimensions, potentially losing crucial details and affecting detection accuracy.

[0004] The purpose of this invention is to design an optical imaging method and system for intelligent analysis of plankton, addressing the problems existing in the prior art. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides an optical imaging method and system for intelligent analysis of plankton, which can effectively solve at least one of the problems existing in the prior art.

[0006] The technical solution of this invention is: An optical imaging method for intelligent analysis of plankton includes the following steps: S1. Determine the depth of field, numerical aperture of the objective lens, and total magnification of the microscope system based on the characteristic size of phytoplankton. S2, the phytoplankton in the sample solution is squeezed to form a liquid flow, and a pre-image of the phytoplankton is captured by the front camera; S3, process the phytoplankton pre-image and extract the corresponding phytoplankton growth cycle characteristics, morphological characteristics and activity characteristics; S4. Based on the growth cycle characteristics, morphological characteristics, and activity characteristics, dynamically determine the liquid flow velocity corresponding to the phytoplankton, drive the corresponding phytoplankton into the field of view of the microscope system, and increase the probability that the phytoplankton passes through the field of view along the optical axis of the microscope with its minimum geometric size. S5, acquire the main image of the phytoplankton using the microscope system.

[0007] Further, step S1 includes: S1.1, the median of the size range corresponding to the growth cycle in which the phytoplankton is most abundant is selected as the feature size; S1.2, Set the depth of field of the microscope system to be equal to the feature size, select the objective magnification of the microscope system, calculate the total magnification of the microscope system based on the objective magnification, and calculate the numerical aperture of the objective lens using the microscope depth of field calculation formula.

[0008] Furthermore, in step S2, The sample liquid containing phytoplankton is output through a sample injection pump, and a sheath flow is output through at least two sheath liquid injection pumps, wherein the output end of the sheath liquid injection pump surrounds the output end of the sample injection pump, thereby encapsulating and squeezing the sample liquid to form a liquid flow. The liquid flow is output to the sample cell within the field of view of the microscope system. The liquid outlet of the sample cell is equipped with a micro valve. When the micro valve is closed, the sample injection pump and the sheath injection pump stop outputting liquid.

[0009] Further, step S3 includes: S3.1, an adaptive threshold segmentation technique is adopted, which calculates the Gaussian weighted average value of the pixel neighborhood in the phytoplankton pre-image and dynamically sets the local threshold to segment algal cells from the background; S3.2, use morphological opening operation to remove noise points and use morphological closing operation to fill the pores inside the cell to obtain the complete outline of the algal cell; S3.3, Determine the growth cycle characteristics of phytoplankton based on their size range, wherein the growth cycle characteristics include sporangium stage, vegetative cell stage, fission and reproduction stage, and extinction stage; The aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio of phytoplankton are calculated as the morphological characteristics described above. Cell outline integrity, internal structural regularity, and boundary clarity are calculated as the activity characteristics.

[0010] Further, step S4 includes: S4.1 Calculate the growth cycle score, morphological score, and activity score corresponding to the growth cycle characteristic, the morphological characteristic, and the activity characteristic, respectively. S4.2, Match the corresponding basic sheath flow rate according to the growth cycle score, adjust the basic sheath flow rate according to the morphology score, and limit the upper limit of the flow rate according to the activity score to obtain the sheath fluid flow rate; S4.3, Calculate the sample liquid flow rate based on the preset linear relationship between the sheath fluid flow rate and the sample liquid flow rate, and drive the corresponding phytoplankton into the field of view of the microscope system.

[0011] Further, step S4.1 includes: S4.1.1 Extract standard images corresponding to the sporangium stage, vegetative cell stage, division and reproduction stage, and extinction stage; find the standard image that is closest to the phytoplankton pre-image to determine the corresponding stage and the corresponding scoring interval; calculate the similarity between the phytoplankton pre-image and the corresponding standard image; and calculate the growth cycle score based on the higher the similarity. S4.1.2, after normalizing the aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio of phytoplankton, the morphological score is obtained by weighted summation. S4.1.3, calculate the ratio of actual contour pixels to theoretical contour pixels as cell contour integrity, calculate phytoplankton energy through gray-level co-occurrence matrix as internal structural regularity, calculate the standard deviation of phytoplankton edge intensity as boundary sharpness, and then normalize and weighted sum to obtain the activity score.

[0012] Further, step S4.2 includes: S4.2.1, define the growth cycle score as G, the base flow rate as Qbase, and the unit of flow rate as μL / min; If 80≤G≤100, ; If 50 ≤ G < 80, ; If 20 ≤ G < 50, ; If 0 ≤ G < 20, ; S4.2.2, defining the morphological score as M and the adjusted flow rate as Q, then, ; S4.2.3, define the activity score as V. If 0.9 ≤ V ≤ 1.0, then the upper limit of the flow velocity is 90%. If 0.5 ≤ V < 0.9, then the upper limit of the flow velocity is 50%. If V < 0.5, then the upper limit of the flow rate is 20%.

[0013] Further, after step S5, the following steps are performed: inputting the phytoplankton master image into a pre-trained lightweight large model to output phytoplankton species identification and counting.

[0014] Furthermore, the phytoplankton master image is input into a pre-trained lightweight large model for phytoplankton species identification and counting measurement, including: The collected phytoplankton main images were divided into training set, validation set and test set in a 6:2:2 ratio; The lightweight backbone network MobileNetV4 was selected and the CBAM attention mechanism was incorporated to enhance the ability to extract key features of algae. The MobileNetV4 weights pre-trained on ImageNet were used as initial parameters. The AdamW optimizer and cosine annealing learning rate scheduler were used to train on the training set. The SoftTargetCrossEntropy loss function was used, and mixed precision training and exponential moving average model updates were enabled. The loss was monitored through the validation set and early stopping was triggered to prevent overfitting. Finally, the phytoplankton identification model was obtained after validation on the test set.

[0015] Furthermore, an optical imaging system for intelligent analysis of plankton is provided, which implements the aforementioned optical imaging method for intelligent analysis of plankton during operation, and includes the following modules: The microscope system parameter determination module is used to determine the depth of field, objective lens numerical aperture, and total magnification of the microscope system based on the characteristic dimensions of phytoplankton. The sample liquid output module is used to squeeze the phytoplankton in the sample liquid to form a liquid stream, and to capture a pre-image of the phytoplankton through the front camera; The feature processing module is used to process the phytoplankton pre-image and extract the corresponding phytoplankton growth cycle features, morphological features, and activity features. The flow rate calculation module is used to dynamically determine the flow rate of the liquid flow corresponding to the phytoplankton based on the growth cycle characteristics, morphological characteristics, and activity characteristics, and drive the corresponding phytoplankton into the field of view of the microscope system, thereby increasing the probability that the phytoplankton passes through the field of view along the optical axis of the microscope with its minimum geometric size. The image acquisition module is used to acquire main images of phytoplankton through the microscope system.

[0016] Therefore, the present invention provides the following effects and / or advantages: This application first combines sheath fluid flow with pre-imaging, and then uses feature extraction to pre-observe the phytoplankton's growth cycle characteristics, morphological characteristics, and activity characteristics. This allows for matching and appropriately adjusting the sheath fluid flow rate, ultimately increasing the probability that the phytoplankton, with its minimum geometric size, passes through the microscope's optical axis within the field of view, thus obtaining clear optical measurement images and preventing random tumbling of phytoplankton in the fluid. By using a single fixed depth of field, it is possible to photograph cells under various conditions, obtaining optimally focused and clear images for more accurate algae identification.

[0017] This application enables automated identification of phytoplankton of specific species or within a specific size range. Fixed depth-of-field imaging means that structures at all levels can be captured without frequent refocusing, thus improving detection efficiency. Since the depth-of-field parameters of the optical system are known, the optimal model input size that best matches the target object can be directly calculated.

[0018] This application innovatively integrates a CBAM attention mechanism module into the AI ​​training model. This module effectively compensates for the accuracy loss that may be caused by preprocessing, ensuring the accuracy of algae detection. This synergistic solution, combining lightweight preprocessing and enhanced AI, achieves efficient, automated, and high-precision algae monitoring. Compared to relying solely on high-computing AI or a single preprocessing technology, it achieves a better balance between processing efficiency and accuracy.

[0019] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0020] It should be understood that the above summary and the following detailed description of the invention are exemplary and explanatory, and are intended to provide further explanation of the invention as claimed. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating one embodiment of the present invention. Detailed Implementation

[0022] To facilitate understanding by those skilled in the art, the present invention will now be described in further detail with reference to the embodiments: refer to Figure 1 , An optical imaging method for intelligent analysis of plankton, characterized by the following steps: S1. Determine the depth of field, numerical aperture of the objective lens, and total magnification of the microscope system based on the characteristic size of phytoplankton. S2, the phytoplankton in the sample solution is squeezed to form a liquid flow, and a pre-image of the phytoplankton is captured by the front camera; S3, process the phytoplankton pre-image and extract the corresponding phytoplankton growth cycle characteristics, morphological characteristics and activity characteristics; S4. Based on the growth cycle characteristics, morphological characteristics, and activity characteristics, dynamically determine the liquid flow velocity corresponding to the phytoplankton, drive the corresponding phytoplankton into the field of view of the microscope system, and increase the probability that the phytoplankton passes through the field of view along the optical axis of the microscope with its minimum geometric size. S5, acquire the main image of the phytoplankton using the microscope system.

[0023] The specific process is as follows.

[0024] Further, step S1 includes: S1.1, the median of the size range corresponding to the growth cycle in which the phytoplankton is most abundant is selected as the feature size; S1.2, Set the depth of field of the microscope system to be equal to the feature size, select the objective magnification of the microscope system, calculate the total magnification of the microscope system based on the objective magnification, and calculate the numerical aperture of the objective lens using the microscope depth of field calculation formula.

[0025] In this step, samples can be taken from the target water area, and then preliminary images of the phytoplankton can be used to determine the stage of most of the phytoplankton cells. The phytoplankton cell growth cycle includes: In the vegetative cell stage, algal cells are spherical or oval, plump, with clear internal structure and distinct boundaries. In the sporangium stage, the cells lose water and shrink, the cell walls thicken, and the interior is filled with storage substances. Sometimes there are bright red or yellow granules, which are spherical or oval in shape, dark overall, and have a rough texture. During the cell division and reproduction stage, the cells undergo binary division, resulting in transverse and longitudinal grooves, significant elongation of the cells, obvious depressions or division grooves in the middle of the cells, a significant increase in overall size, irregular outline, and separation of internal structures. During the death phase, the cells become dull in color, distorted in shape, have a disordered internal structure, and blurred boundaries.

[0026] By manually identifying or using image recognition, based on the image characteristics of each growth cycle of algal cells, it can be determined which growth cycle the most phytoplankton cells are currently in. Then, by consulting the data on the algae or measuring the size range of cells in that growth cycle from the image, the median value is selected as the feature size.

[0027] Then, taking a feature size of 25µm as an example, according to the depth of field formula... λ is the wavelength of light; we generally use white light, taking 0.55µm (550 nanometers). The refractive index n used for depth-of-field calculation is determined by the medium directly in contact with the sample point. Assuming our sample cell medium is water (n≈1.33), we take n=1.33. NA is the numerical aperture of the objective lens. M is the total magnification of the microscope. M = objective lens magnification × adapter lens magnification. e is the minimum resolvable distance of the detector. e = 2 × camera pixel size; here, we assume the camera pixel size is 3.45µm, so the value of e in the formula is e=6.9µm. Choosing a 50x objective lens and an adapter lens magnification of 1×, the magnification M=50.

[0028] Based on the above calculations, to obtain a depth of field of 25µm, a microscope with a 50x objective lens (numerical aperture NA = 0.1747) and a CCD camera with a pixel size of 3.45µm are required. The CCD camera is mainly used for imaging, preliminary analysis, and connecting to a computer to establish a data sample library.

[0029] Given a 50x objective lens and a 3.45µm CCD camera, by photographing algal cells occupying 20 micrometers along their long axis, the pixel count of the target object in the captured image can be calculated. This allows for the selection of appropriate equipment to capture images that retain the desired detail without obtaining excessively large images. For example, the target object could occupy approximately 290 pixels along the long axis of the original image.

[0030] Furthermore, In step S2, a sample liquid containing phytoplankton is output through a sample injection pump, and a sheath flow is output through at least two sheath liquid injection pumps. The output end of the sheath liquid injection pump surrounds the output end of the sample injection pump, thereby encapsulating and squeezing the sample liquid to form a liquid flow. The liquid flow is output to the sample cell within the field of view of the microscope system. The liquid outlet of the sample cell is equipped with a micro valve. When the micro valve is closed, the sample injection pump and the sheath injection pump stop outputting liquid.

[0031] In this embodiment, the output ends of multiple sheath fluid injection pumps are spaced at equal angles around the output end of the sample injection pump. When both the internal sample flow and the external sheath fluid are flowing stably in a laminar state, the high-speed sheath fluid exerts a uniform compression effect on the sample flow at the output end, and the two do not mix. At the outlet, the sample flow is confined within the sheath fluid. By adjusting the flow rate of the external sheath fluid, the degree of compression on the internal sample flow can be precisely controlled, thereby changing the width of the sample flow. When the sample liquid contains analyte particles, the outer sheath fluid can compress and focus them into an extremely narrow straight line. In this way, the particles will be arranged in a single file and pass through the detection area sequentially, creating conditions for subsequent accurate measurement. The total flow rate of the sheath fluid on both sides is much higher than the flow rate of the sample liquid in the middle (for example, the sheath fluid flow rate is 10-50 times higher than the sample flow rate) and is injected from the inlet at the top of the cavity, forming a ring-shaped rapid liquid flow around the sample flow. In the focusing area, the high-speed flowing sheath fluid, through laminar flow, uniformly compresses the low-speed sample flow in the center from all sides, causing its diameter to decrease rapidly. Cells in the sample stream are physically confined, forced to align in a straight single file at the center of the flow axis. This causes phytoplankton cells to flow one by one into a transparent, flat, diamond-shaped, hollow, sealed glass sample cell. When the phytoplankton enters the sample cell and is detected by a photodetector located at the entrance of the observation chamber, the outlet valve downstream of the observation chamber is closed, and the flow rate of the sample injection pump is shut off or significantly reduced, confining the phytoplankton within the observation chamber and allowing them to be photographed under a microscope. After imaging, the phytoplankton is carried by the water flow into the wastewater tank.

[0032] Meanwhile, in order to accurately control the output speed of the sheath fluid injection pump in the subsequent process, a front-mounted camera is set at the output end of the sample injection pump in this embodiment. The front-mounted camera can take a preliminary image of the phytoplankton through the front-mounted microscope system. At this time, the pre-image of the phytoplankton may be a preliminary and blurry image. Its core function is to quickly acquire key information sufficient to infer the cell state with low resolution and high speed, thereby guiding the subsequent flow rate regulation step S4 and precise imaging.

[0033] Further, step S3 includes: S3.1, an adaptive threshold segmentation technique is adopted, which calculates the Gaussian weighted average value of the pixel neighborhood in the phytoplankton pre-image and dynamically sets the local threshold to segment algal cells from the background; In this step, the phytoplankton pre-image can be converted from RGB to Lab color space, and the luminance channel can be enhanced using CLAHE. Then, an adaptive thresholding segmentation technique is used. By calculating the Gaussian weighted average of the pixel neighborhood and dynamically setting the local threshold, the problem of uneven illumination can be effectively solved, and the microorganisms can be accurately segmented from the complex background. S3.2, use morphological opening operation to remove noise points and use morphological closing operation to fill the pores inside the cell to obtain the complete outline of the algal cell; S3.3, Determine the growth cycle characteristics of phytoplankton based on their size range, wherein the growth cycle characteristics include sporangium stage, vegetative cell stage, fission and reproduction stage, and extinction stage; The aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio of phytoplankton are calculated as the morphological characteristics described above. Cell outline integrity, internal structural regularity, and boundary clarity are calculated as the activity characteristics.

[0034] In this step, the growth cycle of phytoplankton can be determined based on its size range. The characteristics of the sporangium stage, vegetative cell stage, division and reproduction stage, and extinction stage have been described above. Therefore, the size of the long axis of phytoplankton can be used to preliminarily determine whether the cell is proliferating, shrinking, or in a normal state, and then match it with the corresponding growth cycle.

[0035] Then, the morphological characteristics of phytoplankton are calculated. Among them, the nuclear eccentricity is the distance from the nuclear centroid to the cell's geometric centroid, and the degree of nuclear deviation reflects the cell cycle stage; the chloroplast distribution uniformity is the statistical distribution uniformity of chloroplast regions. The more uniform the distribution, the higher the activity, which can be used to identify whether it is an abnormal cell; the vacuolar area ratio is the ratio of the total vacuolar area to the total cell area. Vacuole expansion indicates senescence or death, which can identify cells in the death phase and avoid ineffective flow rate regulation.

[0036] Further, step S4 includes: S4.1 Calculate the growth cycle score, morphological score, and activity score corresponding to the growth cycle characteristic, the morphological characteristic, and the activity characteristic, respectively. In step S3, sub-features corresponding to different growth cycle characteristics, morphological characteristics, and activity characteristics are obtained, thereby enabling the calculation of corresponding scores. These growth cycle scores, morphological scores, and activity scores can be used for subsequent calculations of sheath fluid flow rate.

[0037] S4.2, Match the corresponding basic sheath flow rate according to the growth cycle score, adjust the basic sheath flow rate according to the morphology score, and limit the upper limit of the flow rate according to the activity score to obtain the sheath fluid flow rate; In this step, the growth cycle score is the core factor affecting the sheath fluid flow rate. This is because, for the cyst stage, the algal cells are spherical, and in this case, there is no orientation requirement for entering the imaging field of the microscope system. At this time, a high flow rate will only increase energy consumption and collision risk. For the vegetative cell stage, the algal cells have a moderate aspect ratio, and a moderate shear rate is sufficient to allow the cells to enter the high-probability orientation zone, avoiding violent flipping. For the division and reproduction stage, the algal cells have a large aspect ratio, generally 3-6. The algal cells move along the Jeffrey track in the liquid and have a high natural flipping frequency. A high shear rate is required to force the long axis to flow parallel to the optical axis, so that the smallest size is aligned with the optical axis as much as possible. For the extinction stage, the algal cell structure is fragile, and a high flow rate will cause it to break. Only basic flow needs to be maintained to prevent blockage. Therefore, a suitable basic sheath flow rate is first matched by its growth cycle score.

[0038] For example, the basic flow rates of sheath flow for the sporangium stage, the vegetative cell stage, the division and reproduction stage, and the extinction stage are 5~20 µL / min, 30~60 µL / min, 80~120 µL / min, and less than 5 µL / min, respectively.

[0039] Next, the base flow rate of the sheath fluid needs to be adjusted based on the morphological score and the activity score. For example, if the morphological score indicates that the algal cells are more slender, have more pronounced cell outlines, and have nuclei that are more off-center, it means that the algal cells are prone to overturning at low flow rates, so the base flow rate of the sheath fluid should be increased. Alternatively, if the activity score indicates fewer vacuoles, regular texture, and orderly arrangement of organelles, it means that the algal cells have a stronger tolerance to high flow rates, in which case a higher sheath fluid flow rate can be allowed.

[0040] S4.3, Calculate the sample liquid flow rate based on the preset linear relationship between the sheath fluid flow rate and the sample liquid flow rate, and drive the corresponding phytoplankton into the field of view of the microscope system.

[0041] In this step, the sheath fluid flow rate needs to be much greater than the sample fluid flow rate. Therefore, a linear relationship can be set in advance, for example, the sheath fluid flow rate is 10-50 times the sample fluid flow rate. At this time, the sample fluid flow rate can be obtained based on the sheath fluid flow rate.

[0042] Further, step S4.1 includes: S4.1.1 Extract standard images corresponding to the sporangium stage, vegetative cell stage, division and reproduction stage, and extinction stage; find the standard image that is closest to the phytoplankton pre-image to determine the corresponding stage and the corresponding scoring interval; calculate the similarity between the phytoplankton pre-image and the corresponding standard image; and calculate the growth cycle score based on the higher the similarity. In this step, we first find the image that is most similar to the preset standard image to determine the corresponding stage, and then determine the corresponding growth cycle score based on the image similarity.

[0043] The scoring ranges for the sporangium stage, vegetative cell stage, fission and reproduction stage, and extinction stage can be set to 20~40, 50~70, 80~100, and 0~10, respectively. Then, based on the higher the similarity, the higher the score, and the corresponding growth cycle score is obtained from the scoring range.

[0044] S4.1.2, after normalizing the aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio of phytoplankton, the morphological score is obtained by weighted summation. In this step, the aspect ratio is the ratio of the major axis to the minor axis of the cell's circumscribed rectangle. The aspect ratio is approximately 1.5–2.5 in the vegetative cell stage and approximately 3–6 in the division and reproduction stage. A larger aspect ratio results in a higher frequency of tumbling of ellipsoidal particles in the shear flow, requiring a higher shear rate to lock them in the flow direction. A roundness close to 1 indicates that the cell is approximately spherical. Spherical particles experience symmetrical forces in all directions, have no preferred orientation, and generate stronger torque on the cell, necessitating active flow velocity control. A convexity less than 1 indicates the presence of concavity on the cell surface. Concave regions generate local eddies and unbalanced pressure distribution in the shear flow, causing the cell to tend to tumble. Forcing its orientation requires greater hydrodynamic force. Nuclear eccentricity is a morphological precursor to the cell entering mitosis. When the cell enters the division phase or is under stress, chloroplasts aggregate or degrade towards the poles, resulting in poor chloroplast distribution uniformity. This internal mass unevenness further exacerbates fluid disturbance; the more asymmetrical the internal mass distribution, the higher the flow velocity required to force orientation.

[0045] Specifically, the morphological score can be obtained by weighted summation according to the following formula. ; Where M represents the morphological score. These represent the weights of aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio, respectively. These represent the normalized values ​​for aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio, respectively. The entropy weighting method can be used to analyze the data distribution of each feature value from the massive amount of collected data containing labeled feature values ​​and corresponding known optimal flow velocities. If a feature varies greatly among different samples, it indicates that its discriminative ability is strong, and the entropy weighting method will automatically assign it a higher weight. Conversely, if the feature values ​​do not differ much, the weight will be reduced accordingly. The sum is 1.

[0046] S4.1.3, calculate the ratio of actual contour pixels to theoretical contour pixels as cell contour integrity, calculate phytoplankton energy through gray-level co-occurrence matrix as internal structural regularity, calculate the standard deviation of phytoplankton edge intensity as boundary sharpness, and then normalize and weighted sum to obtain the activity score.

[0047] In this embodiment, specific values ​​for cell outline integrity, internal structure regularity, and boundary clarity were obtained. Then, the sum of the weights corresponding to the specific values ​​of cell outline integrity, internal structure regularity, and boundary clarity was set to 1. Through manual empirical calibration, the weights corresponding to the specific values ​​of cell outline integrity, internal structure regularity, and boundary clarity were found to be 0.4, 0.35, and 0.25, respectively. After normalizing their respective values, they were weighted and summed according to the set weights.

[0048] Further, step S4.2 includes: S4.2.1, define the growth cycle score as G, the base flow rate as Qbase, and the unit of flow rate as μL / min; If 80≤G≤100, ; If 50 ≤ G < 80, ; If 20 ≤ G < 50, ; If 0 ≤ G < 20, ; In this step, growth cycle scoring G The value of G ranges from 0 to 100, and G determines the base flow rate level. The system uses a pre-defined piecewise linear mapping function to map the flow rate, thereby forcing or protecting the cells.

[0049] S4.2.2, defining the morphological score as M and the adjusted flow rate as Q, then, ; In this step, the higher the morphological score M, the stronger the shear force required for the cells to be oriented. Therefore, the base flow rate is neither reduced nor increased; the multiplier here is 0.5 + 0.5. M The maximum is 1.0, meaning that the morphology score can only decrease the flow rate and cannot be increased indefinitely, because the base flow rate has already been determined by the growth cycle score, and the morphology score can only be adjusted downwards.

[0050] S4.2.3, define the activity score as V, If 0.9 ≤ V ≤ 1.0, then the upper limit of the flow velocity is 90%. If 0.5 ≤ V < 0.9, then the upper limit of the flow velocity is 50%. If V < 0.5, then the upper limit of the flow rate is 20%.

[0051] In this step, a higher V indicates a healthier cell, and thus a higher upper limit for the flow rate that can be obtained.

[0052] Further, after step S5, the following steps are performed: inputting the phytoplankton master image into a pre-trained lightweight large model to output phytoplankton species identification and counting.

[0053] The phytoplankton master image is input into a pre-trained lightweight large model for phytoplankton species identification and counting measurement, including: S5.1, the collected phytoplankton main images are divided into training set, validation set and test set in a ratio of 6:2:2; S5.2 uses the lightweight backbone network MobileNetV4 and incorporates the CBAM attention mechanism to enhance the extraction of key algal features. It uses ImageNet pre-trained MobileNetV4 weights as initial parameters, and trains on the training set using the AdamW optimizer and cosine annealing learning rate scheduler. It uses the SoftTargetCrossEntropy loss function, enables mixed precision training and exponential moving average model updates, monitors the loss on the validation set and triggers early stopping to prevent overfitting, and finally obtains the phytoplankton identification model after validation on the test set.

[0054] In this step, a highly efficient image processing workflow is used to automatically locate microbial targets and calculate the optimal cropping box based on an intelligent decision-making algorithm. The method first converts the image to the Lab color space and applies contrast-adaptive histogram equalization (CLAHE) to the brightness channel to enhance the contrast between the microorganisms and the background. Then, an adaptive thresholding technique is employed, which calculates the Gaussian weighted average of the pixel neighborhood and dynamically sets a local threshold to effectively solve the problem of uneven illumination and accurately segment the microorganisms from the complex background. Next, morphological opening and closing operations are used to remove noise and fill holes to obtain a complete outline. In the target selection stage, area threshold and roundness (calculated as 4π × area / perimeter) are introduced as key criteria to effectively filter out impurities and select true microbial targets. Finally, based on the target area ratio and bounding box expansion ratio parameters, a cropping box is dynamically calculated centered on the target to ensure that the microorganism is prominent and the composition is reasonable. This method combines traditional image processing with parametric intelligent decision-making to automatically, accurately, and efficiently locate and crop microbial targets in microscope images. This step significantly reduces data redundancy and computational load in subsequent processing, ensuring high efficiency and real-time performance.

[0055] Set the target width and target height of the input training model image. By calculating the minimum scaling factor (scale=min(target width / original width, target height / original height)) between the target size and the aspect ratio of the original image, and combining it with the bicubic interpolation algorithm (Image.LANCZOS), the processed image maintains a relatively smooth outline and rich details. It also uses mean fill (calculating the mean of the image area as the background color) or custom fill (such as black) strategies to solve the problems of image distortion and visual disjointness in the filled area caused by traditional scaling.

[0056] Images that have undergone resizing are divided into training, validation, and test sets in a 6:2:2 ratio. To enhance the model's generalization ability, small-amplitude (±5°) rotations, Gaussian blurs, and color jitter are randomly applied to the input images during training. Image normalization is performed using independent mean and standard deviation for each channel. Model training employs the AdamW optimizer, coupled with a cosine annealing learning rate scheduler, and utilizes mixed-precision training and exponential moving average model update mechanisms. SoftTargetCrossEntropy is used as the loss function during training, while the standard cross-entropy loss function is used during validation, supplemented by gradient clipping and early stopping mechanisms to ensure training stability.

[0057] CBAM introduces attention in the channel and spatial dimensions, allowing the model to adaptively focus on more important regions and features in the image.

[0058] Position 1: Placed at the end of intermediate feature extraction, after the initial downsampling to medium resolution (96×96), specifically after the ExtraDW module (output channels 96) in the table (number 4). CBAM is introduced to filter the initially extracted intermediate features such as texture, providing cleaner input for deep networks.

[0059] Position 2: Placed at the beginning of deep feature extraction, after the feature map is downsampled to 48×48 and the number of channels is expanded to 192, specifically before the ConvNext module (number 5 in the table). This is the starting point for the network to perform complex semantic extraction. Adding CBAM can refine the features, allowing the model to focus more on key information in deep computation.

[0060] Position 3: Placed in the early stage of the advanced feature stage, specifically after the ExtraDW module of feature map number 5 in the table has been downsampled to 12×12 and the number of channels has been increased to 512. Adding CBAM to this advanced semantic feature layer can strengthen the most discriminative features before classification decisions are made.

[0061] The core of this step is to achieve high efficiency and lightweight design. This is achieved by constructing a collaborative optimization process from optical design to algorithm processing: First, the model input size that best matches the target size is directly derived from the depth-of-field parameters of the optical system, controlling the data volume from the source. For acquired images, a low-computing traditional image processing module is used for precise localization and cropping of algae targets. Then, intelligent scaling technology that maintains the aspect ratio is used to unify the image to a preset size, avoiding image distortion. For the model, the lightweight backbone network MobileNetV4 is selected, and the CBAM attention mechanism is incorporated to enhance the extraction of key algae features. This combined strategy significantly reduces system computational resource consumption while effectively ensuring and improving the final detection accuracy.

[0062] Furthermore, an optical imaging system for intelligent analysis of plankton is provided, which implements the aforementioned optical imaging method for intelligent analysis of plankton during operation, and includes the following modules: The microscope system parameter determination module is used to determine the depth of field, objective lens numerical aperture, and total magnification of the microscope system based on the characteristic dimensions of phytoplankton. The sample liquid output module is used to squeeze the phytoplankton in the sample liquid to form a liquid stream, and to capture a pre-image of the phytoplankton through the front camera; The feature processing module is used to process the phytoplankton pre-image and extract the corresponding phytoplankton growth cycle features, morphological features, and activity features. The flow rate calculation module is used to dynamically determine the flow rate of the liquid flow corresponding to the phytoplankton based on the growth cycle characteristics, morphological characteristics, and activity characteristics, and drive the corresponding phytoplankton into the field of view of the microscope system, thereby increasing the probability that the phytoplankton passes through the field of view along the optical axis of the microscope with its minimum geometric size. The image acquisition module is used to acquire main images of phytoplankton through the microscope system.

[0063] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0067] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

Claims

1. An optical imaging method for intelligent analysis of plankton, characterized in that: Includes the following steps: S1. Determine the depth of field, numerical aperture of the objective lens, and total magnification of the microscope system based on the characteristic size of phytoplankton. S2, the phytoplankton in the sample solution is squeezed to form a liquid flow, and a pre-image of the phytoplankton is captured by the front camera; S3, process the phytoplankton pre-image and extract the corresponding phytoplankton growth cycle characteristics, morphological characteristics and activity characteristics; S4, based on the growth cycle characteristics, morphological characteristics, and activity characteristics, dynamically determine the sap flow velocity corresponding to the phytoplankton, drive the corresponding phytoplankton into the field of view of the microscope system, and increase the probability that the phytoplankton passes through the field of view along the optical axis of the microscope with its minimum geometric size; including: S4.1, calculate the growth cycle score, morphological score, and activity score corresponding to the growth cycle characteristic, the morphological characteristic, and the activity characteristic, respectively; step S4.1 includes: S4.1.1 Extract standard images corresponding to the sporangium stage, vegetative cell stage, division and reproduction stage, and extinction stage; find the standard image that is closest to the phytoplankton pre-image to determine the corresponding stage and the corresponding scoring interval; calculate the similarity between the phytoplankton pre-image and the corresponding standard image; and calculate the growth cycle score based on the higher the similarity. S4.1.2, after normalizing the aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio of phytoplankton, the morphological score is obtained by weighted summation. S4.1.3, calculate the ratio of actual contour pixels to theoretical contour pixels as cell contour integrity, calculate phytoplankton energy through gray-level co-occurrence matrix as internal structural regularity, calculate the standard deviation of phytoplankton edge intensity as boundary sharpness, and then normalize and weighted sum to obtain the activity score; S4.2, the sample liquid containing phytoplankton is encapsulated and compressed by the sheath flow to form a liquid flow. The base flow rate of the sheath flow is matched according to the growth cycle score, the base flow rate of the sheath flow is adjusted according to the morphology score, and the upper limit of the flow rate is limited according to the activity score to obtain the sheath fluid flow rate; step S4.2 includes: S4.2.1, define the growth cycle score as G, the base flow rate as Qbase, and the unit of flow rate as μL / min; If 80≤G≤100, ; If 50 ≤ G < 80, ; If 20 ≤ G < 50, ; If 0 ≤ G < 20, ; S4.2.2, defining the morphological score as M and the adjusted flow rate as Q, then, ； S4.2.3, define the activity score as V. If 0.9 ≤ V ≤ 1.0, then the upper limit of the flow velocity is 90%. If 0.5 ≤ V < 0.9, then the upper limit of the flow velocity is 50%. If V < 0.5, then the upper limit of the flow rate is 20%. S4.3, Based on the preset linear relationship between the sheath fluid flow rate and the sample fluid flow rate, calculate the sample fluid flow rate and drive the corresponding phytoplankton into the field of view of the microscope system; S5, acquire the main image of the phytoplankton using the microscope system.

2. The optical imaging method for intelligent analysis of plankton according to claim 1, characterized in that: Step S1 includes: S1.1, the median of the size range corresponding to the growth cycle in which the phytoplankton is most abundant is selected as the feature size; S1.2, Set the depth of field of the microscope system to be equal to the feature size, select the objective magnification of the microscope system, calculate the total magnification of the microscope system based on the objective magnification, and calculate the numerical aperture of the objective lens using the microscope depth of field calculation formula.

3. The optical imaging method for intelligent analysis of plankton according to claim 1, characterized in that: In step S2, The sample liquid containing phytoplankton is output through a sample injection pump, and a sheath flow is output through at least two sheath liquid injection pumps, wherein the output end of the sheath liquid injection pump surrounds the output end of the sample injection pump, thereby encapsulating and squeezing the sample liquid to form a liquid flow. The liquid flow is output to the sample cell within the field of view of the microscope system. The liquid outlet of the sample cell is equipped with a micro valve. When the micro valve is closed, the sample injection pump and the sheath injection pump stop outputting liquid.

4. The optical imaging method for intelligent analysis of plankton according to claim 3, characterized in that: Step S3 includes: S3.1, an adaptive threshold segmentation technique is adopted, which calculates the Gaussian weighted average value of the pixel neighborhood in the phytoplankton pre-image and dynamically sets the local threshold to segment algal cells from the background; S3.2, use morphological opening operation to remove noise points and use morphological closing operation to fill the pores inside the cell to obtain the complete outline of the algal cell; S3.3, Determine the growth cycle characteristics of phytoplankton based on their size range, wherein the growth cycle characteristics include sporangium stage, vegetative cell stage, fission and reproduction stage, and extinction stage; The aspect ratio, roundness, outline convexity, area-to-perimeter ratio, nucleus eccentricity, chloroplast distribution uniformity, and vacuolar area ratio of phytoplankton are calculated as the morphological characteristics described above. Cell outline integrity, internal structural regularity, and boundary clarity are calculated as the activity characteristics.

5. The optical imaging method for intelligent analysis of plankton according to claim 1, characterized in that: After step S5, the following steps are performed: input the main image of the phytoplankton into a pre-trained lightweight large model to output the identification and counting of phytoplankton species.

6. The optical imaging method for intelligent analysis of plankton according to claim 5, characterized in that: The phytoplankton master image is input into a pre-trained lightweight large model for phytoplankton species identification and counting measurement, including: The collected phytoplankton main images were divided into training set, validation set and test set in a 6:2:2 ratio; The lightweight backbone network MobileNetV4 was selected and the CBAM attention mechanism was incorporated to enhance the ability to extract key features of algae. The MobileNetV4 weights pre-trained on ImageNet were used as initial parameters. The AdamW optimizer and cosine annealing learning rate scheduler were used to train on the training set. The SoftTargetCrossEntropy loss function was used, and mixed precision training and exponential moving average model updates were enabled. The loss was monitored through the validation set and early stopping was triggered to prevent overfitting. Finally, the phytoplankton identification model was obtained after validation on the test set.

7. An optical imaging system for intelligent analysis of plankton, characterized in that: The optical imaging method for intelligent analysis of plankton, as described in any one of claims 1-6, comprises the following modules: The microscope system parameter determination module is used to determine the depth of field, objective lens numerical aperture, and total magnification of the microscope system based on the characteristic dimensions of phytoplankton. The sample liquid output module is used to squeeze the phytoplankton in the sample liquid to form a liquid stream, and to capture a pre-image of the phytoplankton through the front camera. The feature processing module is used to process the phytoplankton pre-image and extract the corresponding phytoplankton growth cycle features, morphological features, and activity features. The flow rate calculation module is used to dynamically determine the flow rate of the liquid flow corresponding to the phytoplankton based on the growth cycle characteristics, morphological characteristics, and activity characteristics, and drive the corresponding phytoplankton into the field of view of the microscope system, thereby increasing the probability that the phytoplankton passes through the field of view along the optical axis of the microscope with its minimum geometric size. The image acquisition module is used to acquire main images of phytoplankton through the microscope system.

Images