A method for monitoring planktonic larval communities of sea cucumber in nearshore waters

Abstract

This invention relates to the field of marine ecological monitoring technology, specifically to a method for monitoring planktonic larvae communities of *Sinocyclocheilus spp.* in nearshore waters. The method includes the following steps: S1, using a continuous plankton sampler with layered trawls to obtain samples containing *Sinocyclocheilus spp.* larvae and microscopically acquiring raw image data; S2, preserving the samples at a constant temperature and protected from light, and generating digital atlases containing texture features, edges, and confidence masks under microscopy; S3, inputting a multi-stage feature network to construct embedding vectors and outputting classification labels and instance masks; S4, performing morphological comparison and confidence backtracking correction based on the labels and masks; S5, combining filtered water volume to calculate abundance vectors and stage structure matrices to output ecological monitoring data. This invention achieves accurate identification, quantity statistics, and community abundance expression of *Sinocyclocheilus spp.* planktonic larvae, improving the intelligence and data assetization level of nearshore marine ecological monitoring.

Description

Technical Field

[0001] This invention relates to the field of marine ecological monitoring technology, and in particular to a method for monitoring planktonic larval communities of sea cucumbers in nearshore waters. Background Technology

[0002] As an important economic echinoderm in nearshore waters, the population structure changes and reproductive success of sea snails directly affect the stability of nearshore ecosystems and the sustainable development of fishery resources. The planktonic larval stage is a key link in its life cycle, which is not only related to the replenishment and continuation of the population, but also an important indicator for marine ecological monitoring and resource assessment. At present, ecological monitoring stations usually use planktonic continuous samplers to obtain water samples and use microscopic examination to identify the planktonic community composition in order to assess the ecological dynamics and abundance distribution of target species.

[0003] However, existing monitoring methods face numerous technical bottlenecks when dealing with planktonic larval communities in sea cucumbers. First, traditional microscopic examination relies on manual identification, which is inefficient, subjective, and makes it difficult to reliably distinguish the ambiguous morphological boundaries between developmental stages. Second, existing image recognition methods mostly focus on single-stage targets and lack the ability to systematically extract and continuously model the developmental morphologies of multiple stages, such as auricularia, jaromorphia, and late-stage larvae. Furthermore, in complex image backgrounds, instance overlap and pseudo-classification occur frequently, leading to inaccurate statistical results and making it difficult to accurately reflect the nearshore ecological structure. Summary of the Invention

[0004] This invention provides a method for monitoring planktonic larvae communities of *Sinocyclocheilus spp.* in nearshore waters. It integrates key steps such as automatic acquisition of microscopic images, digital atlas construction, multi-stage nested classification, and morphological comparison and correction. This method can accurately identify planktonic larvae of *Sinocyclocheilus spp.* at different developmental stages, and realize the construction of abundance vectors and stage structure matrices for ecological monitoring. It improves the accuracy, continuity, and processing efficiency of nearshore water ecological data assets, and provides high-quality support for ecological risk early warning and dynamic resource management.

[0005] A method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters includes the following steps:

[0006] S1. During the breeding season of sea cucumber, a plankton continuous sampler is used at night to perform a layered vertical trawling from the bottom to the surface at a preset monitoring station to obtain a mixed water sample including sea cucumber planktonic larvae. The mixed water sample is then collected by a microscope to form a raw image dataset with depth labels and sampling metadata.

[0007] S2, the water mixture sample is transferred to a constant temperature and light-proof device to maintain the same ambient water temperature at the sampling point, and then quantitatively placed under a microscope for microscopic analysis. Microscopic image frames covering planktonic larvae of sea cucumber at different developmental stages are collected, and a digital atlas is constructed based on the image frames. The digital atlas includes microscopic image frame number, regional texture features, edge contour features and pixel-level confidence mask.

[0008] S3, input the digital map into the multi-stage structural feature extraction network to construct multi-scale morphological embedding vectors for the ear-shaped larvae, jar-shaped larvae and late larvae of the sea cucumber, and introduce nested classification units based on developmental continuity modeling to identify and fuse the fuzzy morphological boundaries between adjacent developmental stages, and output the classification labels and instance masks of larvae at each stage.

[0009] S4. Based on the classification label and instance mask, the number of individuals is counted using the intra-frame pixel area-morphology comparison mechanism. A backtracking optimization function including image frame confidence, developmental span weight and sample boundary continuity is constructed to correct abnormally overlapping individuals or pseudo-classification results, and the corrected number statistics of tadala ear-shaped larvae, jar-shaped larvae and late larvae are output.

[0010] S5. Combining the statistical values ​​and the filtration volume parameters of the continuous planktonic sampler, a population abundance vector of sea cucumber planktonic larvae at the monitoring station is constructed, and a stage structure matrix reflecting the proportion of auricular larvae, jaromorph larvae and late-stage larvae is generated as the data asset output result of nearshore marine ecological monitoring.

[0011] Optionally, S1 includes:

[0012] S11, during the peak breeding season of sea cucumbers (nighttime at the transition from spring to summer), sampling windows were set based on tidal changes and ecological history data of the sea area, and several fixed monitoring stations were selected based on the nearshore hydrodynamic distribution. ;

[0013] S12, using a continuous plankton sampler with a depth sensor, at each monitoring station Perform a layered vertical trawl operation from the bottom to the surface, generating a unique number for each layer of mixed water sample. ;

[0014] S13. Quantitatively measure the mixed water sample from each layer and place it in a temperature-controlled transparent sampling dish. Then, place the dish in a microscope platform equipped with an automatic moving stage and focusing control to acquire sampling image frames. ;

[0015] S14, metadata of each sampled image frame and its corresponding sample The binding, including metadata such as sampling time, location, depth, temperature, salinity, and trawl speed, ultimately generates the original image dataset. .

[0016] Optionally, S2 includes:

[0017] S21, mixing water samples from different stations and depths. Immediately inject into the temperature-controlled, light-protected chamber, and monitor the ambient temperature in real time. With cavity temperature And maintain the ambient temperature and cavity temperature to meet the requirements. ,in, Maximum permissible temperature difference threshold ( );

[0018] S22, from water-mixed samples Extract a fixed volume ( The sample is injected into a miniature observation dish, placed into a microscopic imaging system for automatic focusing and image acquisition, and microscopic image frames are acquired. ;

[0019] S23, for each microscopic image frame Perform image segmentation and texture feature extraction, defining the image as divided into... Local sub-blocks And calculate the gray-level co-occurrence matrix texture vector of local sub-blocks. The Canny algorithm is used to extract image edges and output an edge contour map. ;

[0020] S24, a lightweight CNN model is used to process the microscopic image frames. Perform forward prediction to obtain a probability map of each pixel belonging to a valid larval structure. And generate a confidence mask. ;

[0021] S25 encapsulates the microscopic image frame number, gray-level co-occurrence matrix texture vector, edge contour map, and confidence mask into a unified spectral unit. Finally, monitoring station locations are generated. depth layer Digital map .

[0022] Optionally, S23 includes:

[0023] S231, each microscopic image frame Divided into Non-overlapping equal-sized sub-blocks Each sub-block is an independent texture analysis unit;

[0024] S232, each sub-block Construct its gray-level co-occurrence matrix ;

[0025] S233, based on gray-level co-occurrence matrix For each sub-block Extract texture feature triples to generate gray-level co-occurrence matrix texture vectors. ;

[0026] S234, for the entire microscopic image frame The Canny algorithm is applied for edge extraction to obtain the edge contour map. .

[0027] Optionally, S3 includes:

[0028] S31 will come from digital graphs Each map unit Input a multi-scale encoder network to extract structural semantic embedding vectors at different scales. It outputs the feature embeddings of each microscopic image frame at multiple scales. ;

[0029] S32, embedding features across all scales Channel fusion and position encoding are performed to form a unified morphological embedding tensor. ;

[0030] S33, to Pixel-by-pixel classification is performed using a nested developmental stage classifier. Output category probabilities, including the probability of auricularia. Probability of jar-shaped larvae and the probability of later larvae ;

[0031] S34, Introducing a developmental stage continuity regularization term. Optimize the smoothness of adjacency between class probabilities;

[0032] S35, the category corresponding to the highest category probability is used as the label of each pixel, and an instance mask is generated based on pixel clustering. .

[0033] Optionally, S4 includes:

[0034] S41, For each microscopic image frame, based on pixel-level classification labels and instance masks... Extract candidate instance regions for each category. ,in, For the first Candidate instance regions, For the first Frame belongs to category The number of candidate instances;

[0035] S42, for each candidate instance region Extract its morphological features, such as area and aspect ratio, and compare them with the prior morphological model of this category. The comparison is performed to obtain the matching confidence of the candidate instances. ;

[0036] S43, If the microscopic image frames are temporally continuous frames, the confidence score is adjusted using the spatial span consistency of developmental changes to obtain the developmental span-weighted confidence score. ;

[0037] S44, calculate the consistency score using the Sobel gradient graph and neighborhood consistency. , representing pixels Boundary stability at each candidate instance region Calculate the average boundary continuity ,like If it is, then it is considered a pseudo-instance or an abnormally overlapping region, where, This is the morphological difference threshold used for image morphological consistency determination;

[0038] S45, Output the... Each type of larva in the frame Corrected statistics .

[0039] Optionally, S41 includes:

[0040] S411, For each microscopic image frame, based on pixel-level classification labels and instance masks... , build categories joint mask , is represented as:

[0041] ;

[0042] S412, for combined mask Perform 8-neighborhood connectivity analysis to extract the set of all connected pixels as candidate instance regions.

[0043] Optionally, S42 includes:

[0044] S421, for each microscopic image frame Categories The Candidate instance regions Extract its morphological feature vector ;

[0045] S422, Pre-statistics for each category Average morphological vector And calculate the current candidate region and the prior morphological model. The square of the Euclidean distance between them;

[0046] S423, Calculate the matching confidence level based on the morphological difference comparison results. .

[0047] Optionally, S5 includes:

[0048] S51, for each monitoring station and microscopic image frames Combined with filtration volume parameters Calculate the abundance vector per unit volume of various larvae of *Solanum lycopersicum*. ;

[0049] S52, based on abundance vector Construct a stage structure matrix that reflects the proportions and interaction structures of each stage. ;

[0050] S53 outputs the stage structure matrix and station information as standardized ecological monitoring data assets, forming a community abundance distribution model for nearshore sea areas.

[0051] The beneficial effects of this invention are:

[0052] This invention achieves automated acquisition and structured annotation of nearshore water samples and microscopic image data by using a continuous plankton sampler to perform layered vertical trawls during the breeding season of sea cucumbers, combined with constant-temperature, light-protected sample transfer and microscopic imaging processing. This not only ensures the temperature consistency and spatiotemporal representativeness of the samples, but also establishes a raw image dataset containing multi-dimensional information such as sampling station location, depth, time, temperature and salinity parameters, and trawl speed, significantly improving the stability and traceability of sea cucumber planktonic larvae sampling.

[0053] This invention constructs a digital atlas processing framework for microscopic images based on a lightweight convolutional neural network and introduces a multi-stage structural feature extraction network and a developmental continuity nested classification unit. This framework can distinguish auricular larvae, puffin larvae, and late-stage larvae at the pixel level, accurately capturing the blurred boundaries between different developmental stages. Furthermore, by combining instance segmentation and area-morphology comparison mechanisms, a developmental span weight and boundary continuity backtracking optimization function is established, effectively suppressing the interference of overlapping individuals and pseudo-classifications, improving the robustness and quantitative accuracy of larval identification, and thus significantly improving the identification rate and stage classification accuracy of planktonic larvae compared to traditional manual statistics.

[0054] This invention integrates morphologically corrected quantitative statistics with the filtration volume parameters of a planktonic sampler to form a monitoring station-level abundance vector and stage structure matrix of sea snail planktonic larvae communities. This enables a quantitative expression of the proportional relationships between different developmental stages. This matrix data structure not only supports automatic analysis of the spatiotemporal dynamic changes of the community but also has the function of outputting standardized ecological monitoring data assets. It can provide a unified data interface and algorithm support for marine ecosystem health assessment, dynamic monitoring during the breeding season, and correction of marine ecological models. Attached Figure Description

[0055] 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 only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a schematic diagram of the monitoring method according to an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the stage structure data asset construction process according to an embodiment of the present invention. Detailed Implementation

[0058] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0059] like Figures 1-2 As shown, a method for monitoring planktonic larval communities of *Sinocyclocheilus hainanensis* in nearshore waters includes the following steps:

[0060] S1. During the breeding season of sea cucumber, a plankton continuous sampler is used at night to conduct a layered vertical trawling from the bottom to the surface at a preset monitoring station to obtain a mixed water sample including sea cucumber planktonic larvae. The mixed water sample is then collected by microscope to form a raw image dataset with depth labels and sampling metadata.

[0061] S2. The water mixture sample is transferred to a constant temperature and light-proof device to maintain a consistent ambient water temperature at the sampling point. Then, it is quantitatively placed under a microscope for microscopic analysis. Microscopic image frames covering planktonic larvae of sea cucumber at different developmental stages are collected, and a digital atlas is constructed based on the image frames. The digital atlas includes microscopic image frame number, regional texture features, edge contour features, and pixel-level confidence mask.

[0062] S3 inputs the digital map into the multi-stage structural feature extraction network to construct multi-scale morphological embedding vectors for the ear-shaped larvae, jar-shaped larvae and late larvae of the sea cucumber, and introduces nested classification units based on developmental continuity modeling to identify and fuse the fuzzy morphological boundaries between adjacent developmental stages, and outputs the classification labels and instance masks of larvae at each stage.

[0063] S4, based on classification labels and instance masks, uses the intra-frame pixel area-morphology comparison mechanism to perform quantity statistics, and constructs a backtracking optimization function including image frame confidence, developmental span weight and sample boundary continuity to correct abnormally overlapping individuals or pseudo-classification results, and outputs the corrected quantity statistics of sea cucumber ear-shaped larvae, jar-shaped larvae and late larvae.

[0064] S5 combines the quantitative statistics with the filtration volume parameters of the planktonic continuous sampler to construct the abundance vector of the planktonic larvae community of the monitoring station, and generates a stage structure matrix reflecting the total proportion of auricular larvae, jaromorph larvae and late larvae of the planktonic larvae, as the data asset output result of nearshore marine ecological monitoring.

[0065] S1 includes:

[0066] S11, during the peak breeding season of sea cucumbers (nighttime at the transition from spring to summer), sampling windows were set based on tidal changes and ecological history data of the sea area, and several fixed monitoring stations were selected based on the nearshore hydrodynamic distribution. , is represented as:

[0067] ;

[0068] in, To monitor the longitude of the monitoring stations, To monitor the latitude of the stations, The target sampling depth range;

[0069] S12, using a continuous plankton sampler with a depth sensor, at each monitoring station Perform a layered vertical trawl operation from the bottom to the surface, generating a unique number for each layer of mixed water sample. , is represented as:

[0070] ;

[0071] in, For the first One monitoring station, For the first Layer depth labels This is the timestamp of the start of the trawling operation. Generate a unique sample identifier for the hash function;

[0072] S13. Quantitatively measure the mixed water sample from each layer and place it in a temperature-controlled transparent sampling dish. Then, place the dish in a microscope platform equipped with an automatic moving stage and focusing control to acquire sampling image frames. ;

[0073] S14, metadata of each sampled image frame and its corresponding sample The binding, including metadata such as sampling time, location, depth, temperature, salinity, and trawl speed, ultimately generates the original image dataset. , is represented as:

[0074] ;

[0075] ;

[0076] in, For the first The monitoring station number The first depth layer was collected from the first depth layer A data unit composed of a sampled image frame. This represents the total number of monitoring stations. The number of sampling depth layers per station. The number of sampled image frames captured for each layer.

[0077] S2 includes:

[0078] S21, mixing water samples from different stations and depths. Immediately inject into the temperature-controlled, light-protected chamber, and monitor the ambient temperature in real time. With cavity temperature And maintain the ambient temperature and cavity temperature to meet the requirements. ,in, Maximum permissible temperature difference threshold ( );

[0079] S22, from water-mixed samples Extract a fixed volume ( The sample is injected into a miniature observation dish, placed into a microscopic imaging system for automatic focusing and image acquisition, and microscopic image frames are acquired. ;

[0080] S23, for each microscopic image frame Perform image segmentation and texture feature extraction, defining the image as divided into... Local sub-blocks And calculate the gray-level co-occurrence matrix texture vector of local sub-blocks. The Canny algorithm is used to extract image edges and output an edge contour map. ;

[0081] S24, a lightweight CNN model is used to process the microscopic image frames. Perform forward prediction to obtain a probability map of each pixel belonging to a valid larval structure. And generate a confidence mask. Specifically, it includes:

[0082] S241, for microscopic image frames After normalization, it is represented as:

[0083] ;

[0084] in, These are normalized microscopic image frames;

[0085] S242 uses a lightweight CNN model for forward propagation, outputting a probability map of each pixel belonging to a valid juvenile structure. , is represented as:

[0086] ;

[0087] in, It is the Sigmoid activation function. , These are the convolution kernel and bias of the output layer, respectively. For upsampling function, For encoder feature extraction function, For jump join functions;

[0088] S243, Set confidence threshold for probability plot Binarization is performed to obtain a pixel-level confidence mask. , is represented as:

[0089] ;

[0090] in, For pixels The predicted probability of belonging to a juvenile target. The confidence threshold;

[0091] S25 encapsulates the microscopic image frame number, gray-level co-occurrence matrix texture vector, edge contour map, and confidence mask into a unified spectral unit. Finally, monitoring station locations are generated. depth layer Digital map , is represented as:

[0092] ;

[0093] ;

[0094] in, Assign frame numbers to the microscopic images. The gray-level co-occurrence matrix texture vector, This represents the total number of frames in the microscopic images.

[0095] S23 includes:

[0096] S231, each microscopic image frame Divided into Non-overlapping equal-sized sub-blocks Each sub-block is an independent texture analysis unit, represented as:

[0097] ;

[0098] S232, each sub-block Construct its gray-level co-occurrence matrix , is represented as:

[0099] ;

[0100] in, For pixel coordinates, , Grayscale level , These represent the offset in the specified direction. Count the number of pixel pairs that meet the conditions;

[0101] S233, based on gray-level co-occurrence matrix For each sub-block Extract texture feature triples to generate gray-level co-occurrence matrix texture vectors. , is represented as:

[0102] ;

[0103] in, , , , , These are the average gray values ​​for the rows and columns, respectively. , These are the corresponding standard deviations. To prevent the small constant of logarithm 0;

[0104] S234, for the entire microscopic image frame The Canny algorithm is applied for edge extraction to obtain the edge contour map. Specifically, it includes:

[0105] S2341, Microscopic image frame First, convolve with a Gaussian kernel to remove noise, as shown below:

[0106] ;

[0107] in, These are the pixel values ​​after Gaussian smoothing. It is a two-dimensional Gaussian kernel. The standard deviation of the Gaussian kernel. The radius of the core;

[0108] S2342, using the Sobel operator to approximate the gradients of the image in the horizontal and vertical directions, is expressed as:

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] in, , These are horizontal and vertical gradients, respectively. , These are Sobel convolution kernels, This represents the edge intensity (gradient magnitude). Edge direction (angle);

[0114] S2343, along the gradient direction, the gradient magnitude... Perform nonmaximum suppression to refine the edges, as shown below:

[0115] ;

[0116] in, The gradient magnitude after non-maximum suppression;

[0117] S2344, using a low threshold With high threshold Edge classification is represented as:

[0118] ;

[0119] in, High threshold, controls strong edge preservation. A low threshold is used to control the potential for weak edge connections. For pixels in the image Edge category labels;

[0120] S2345 connects weak edges with adjacent strong edges using hysteresis thresholding to generate an edge contour map. , represented as:

[0121] .

[0122] S3 includes:

[0123] S31 will come from digital graphs Each map unit Input a multi-scale encoder network to extract structural semantic embedding vectors at different scales. It outputs the feature embeddings of each microscopic image frame at multiple scales. , represented as:

[0124] ;

[0125] in, It is the Sigmoid activation function. , These are the weight matrices for the first and second level convolutional layers, respectively. , These are the bias terms for the first and second level convolutional layers, respectively. It is the ReLU activation function;

[0126] ;

[0127] in, The number of scale layers;

[0128] S32, embedding features across all scales Channel fusion and position encoding are performed to form a unified morphological embedding tensor. , represented as:

[0129] ;

[0130] ;

[0131] in, For the first Attention weights for layer features For the first Learnable attention vectors of the layer;

[0132] S33, to Pixel-by-pixel classification is performed using a nested developmental stage classifier. Output category probabilities, including the probability of auricularia. Probability of jar-shaped larvae and the probability of later larvae , is represented as:

[0133] ;

[0134] in, Represents pixels The class probability, It is a nested classifier structure;

[0135] Nested classifier structure Represented as:

[0136] ;

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] ;

[0142] in, For the first Pixel position in frame image The probability of belonging to the first category (ear-like larvae), Given that it already belongs to the first-level category, the pixel The conditional probability of belonging to the second-level category (gourd larvae), Given that the pixel already belongs to the second level category The conditional probability of belonging to the third-level category (late-stage larvae), For joint probability, pixel The probability of belonging to the first-level category. For joint probability, pixel The probability of belonging to the second-level category. For joint probability, pixel The probability of belonging to the third category. , , These are linear weight vectors used for first-level, second-level, and third-level classification prediction, respectively. , , These are the corresponding bias terms. Use the Sigmoid activation function;

[0143] S34, Introducing a developmental stage continuity regularization term. The optimization of the adjacency smoothness between class probabilities is expressed as:

[0144] ;

[0145] S35, the category corresponding to the highest category probability is used as the label of each pixel, and an instance mask is generated based on pixel clustering. , is represented as:

[0146] ;

[0147] InstanceSegment , Ear-shaped, bottle-shaped, later stage ;

[0148] in, For each pixel, the final classification label is... For the first The first frame The instance mask corresponding to the class juvenile, and InstanceSegment is the instance segmentation.

[0149] S4 includes:

[0150] S41, For each microscopic image frame, based on pixel-level classification labels and instance masks... Extract candidate instance regions for each category. ,in, For the first Candidate instance regions, For the first Frame belongs to category The number of candidate instances;

[0151] S42, for each candidate instance region Extract its morphological features, such as area and aspect ratio, and compare them with the prior morphological model of this category. The comparison is performed to obtain the matching confidence of the candidate instances. ;

[0152] S43, If the microscopic image frames are temporally continuous frames, the confidence score is adjusted using the spatial span consistency of developmental changes to obtain the developmental span-weighted confidence score. , is represented as:

[0153] ;

[0154] in, As a developmental uniformity enhancement factor, Score the trajectory continuity of this instance in adjacent frames;

[0155] S44, calculate the consistency score using the Sobel gradient graph and neighborhood consistency. , representing pixels Boundary stability at each candidate instance region Calculate the average boundary continuity ,like If it is, then it is considered a pseudo-instance or an abnormally overlapping region, where, The morphological difference threshold used for image morphological consistency determination is expressed as:

[0156] ;

[0157] in, , The first Pixels in a frame image , The final classification label results, For pixels The set of neighboring pixels;

[0158] Morphological difference threshold Represented as:

[0159] ;

[0160] in, , Categories Mean and standard deviation of the area-morphology product of individuals. The confidence adjustment coefficient;

[0161] S45, Output the... Each type of larva in the frame Corrected statistics , is represented as:

[0162] ;

[0163] in, This is an indicator function; it is 1 if the condition is met, and 0 otherwise.

[0164] S41 includes:

[0165] S411, For each microscopic image frame, based on pixel-level classification labels and instance masks... , build categories joint mask , is represented as:

[0166] ;

[0167] S412, for combined mask Perform 8-neighborhood connectivity analysis to extract the set of all connected pixels as candidate instance regions, specifically including:

[0168] (1) Define the 8-neighborhood pixel set: for any pixel Its 8-neighborhood is defined as:

[0169] ;

[0170] (2) Constructing a pixel connected graph ,in, For a set of nodes, Given the set of edges, each connected subgraph of this graph... Corresponding to a candidate instance region;

[0171] (3) Candidate region extraction: The final set of candidate regions is:

[0172] ;

[0173] in, For the first The set of pixel coordinates corresponding to each region This represents the total number of connected regions. To extract the coordinates of all pixels in a connected graph.

[0174] S42 includes:

[0175] S421, for each microscopic image frame Categories The Candidate instance regions Extract its morphological feature vector , is represented as:

[0176] ;

[0177] in, Candidate instance region The pixel area, that is, the total number of pixels in that area. The aspect ratio of the region. , These are the primary and secondary axis lengths of the region, or the length and width of the smallest enclosing rectangle, respectively.

[0178] S422, Pre-statistics for each category Average morphological vector And calculate the current candidate region and the prior morphological model. The square of the Euclidean distance between them is expressed as:

[0179] ;

[0180] ;

[0181] in, , Categories Average area and aspect ratio, The squared Euclidean distance between the current candidate region and the prior model of this type;

[0182] S423, Calculate the matching confidence level based on the morphological difference comparison results. , is represented as:

[0183] ;

[0184] in, For category The morphological sensitivity parameter.

[0185] S5 includes:

[0186] S51, for each monitoring station and microscopic image frames Combined with filtration volume parameters Calculate the abundance vector per unit volume of various larvae of *Solanum lycopersicum*. , is represented as:

[0187] ;

[0188] in, , , These represent the individual abundance per unit volume for the three developmental stages. This represents the total number of microscopic image frames acquired at this station. , , These are the morphologically corrected statistical values ​​for the number of ear-shaped, barrel-shaped, and late-stage larvae;

[0189] S52, based on abundance vector Construct a stage structure matrix that reflects the proportions and interaction structures of each stage. , is represented as:

[0190] ;

[0191] in, Indicates the stage and stage The individual proportion relationship between them;

[0192] S53 outputs the stage structure matrix and station location information as standardized ecological monitoring data assets, forming a community abundance distribution model for nearshore sea areas, represented as:

[0193] ;

[0194] in, For ecological monitoring data asset collection, This refers to the number of monitoring stations.

[0195] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0196] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters, characterized in that, Includes the following steps: S1. During the breeding season of sea cucumber, a plankton continuous sampler is used at night to perform a layered vertical trawling from the bottom to the surface at a preset monitoring station to obtain a mixed water sample including sea cucumber planktonic larvae. The mixed water sample is then collected by a microscope to form a raw image dataset with depth labels and sampling metadata. S2, the water mixture sample is transferred to a constant temperature and light-proof device to maintain the same ambient water temperature at the sampling point, and then quantitatively placed under a microscope for microscopic analysis. Microscopic image frames covering planktonic larvae of sea cucumber at different developmental stages are collected, and a digital atlas is constructed based on the image frames. The digital atlas includes microscopic image frame number, regional texture features, edge contour features and pixel-level confidence mask. S3, input the digital map into the multi-stage structural feature extraction network to construct multi-scale morphological embedding vectors for the ear-shaped larvae, jar-shaped larvae and late larvae of the sea cucumber, and introduce nested classification units based on developmental continuity modeling to identify and fuse the fuzzy morphological boundaries between adjacent developmental stages, and output the classification labels and instance masks of larvae at each stage. S4. Based on the classification label and instance mask, the number of individuals is counted using the intra-frame pixel area-morphology comparison mechanism. A backtracking optimization function including image frame confidence, developmental span weight and sample boundary continuity is constructed to correct abnormally overlapping individuals or pseudo-classification results, and the corrected number statistics of tadala ear-shaped larvae, jar-shaped larvae and late larvae are output. S5. Combining the statistical values ​​and the filtration volume parameters of the continuous planktonic sampler, a population abundance vector of sea cucumber planktonic larvae at the monitoring station is constructed, and a stage structure matrix reflecting the proportion of auricular larvae, jaromorph larvae and late-stage larvae is generated as the data asset output result of nearshore marine ecological monitoring.

2. The method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 1, characterized in that, S1 includes: S11, during the peak breeding season of sea cucumbers, sampling windows were set based on tidal changes and ecological history data of the sea area, and several fixed monitoring stations were selected based on the nearshore hydrodynamic distribution. ; S12, using a continuous plankton sampler with a depth sensor, at each monitoring station Perform a layered vertical trawl operation from the bottom to the surface, generating a unique number for each layer of mixed water sample. ; S13. Quantitatively measure the mixed water sample from each layer and place it in a temperature-controlled transparent sampling dish. Then, place the dish in a microscope platform equipped with an automatic moving stage and focusing control to acquire sampling image frames. ; S14, metadata of each sampled image frame and its corresponding sample The binding, including metadata such as sampling time, location, depth, temperature, salinity, and trawl speed, ultimately generates the original image dataset. .

3. The method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 2, characterized in that, S2 includes: S21, mixing water samples from different stations and depths. Immediately inject into the temperature-controlled, light-protected chamber, and monitor the ambient temperature in real time. With cavity temperature And maintain the ambient temperature and cavity temperature to meet the requirements. ,in, This is the maximum permissible temperature difference threshold; S22, from water-mixed samples Extract a fixed volume The microscopic image frames are then captured by an autofocus and image acquisition system, which injects the microscopic observation dish into the microscopic imaging system. ; S23, for each microscopic image frame Perform image segmentation and texture feature extraction, defining the image as divided into... Local sub-blocks And calculate the gray-level co-occurrence matrix texture vector of local sub-blocks. The Canny algorithm is used to extract image edges and output an edge contour map. ; S24, a lightweight CNN model is used to process the microscopic image frames. Perform forward prediction to obtain a probability map of each pixel belonging to a valid larval structure. And generate a confidence mask. ; S25 encapsulates the microscopic image frame number, gray-level co-occurrence matrix texture vector, edge contour map, and confidence mask into a unified spectral unit. Finally, monitoring station locations are generated. depth layer Digital map .

4. The method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 3, characterized in that, S23 includes: S231, each microscopic image frame Divided into Non-overlapping equal-sized sub-blocks Each sub-block is an independent texture analysis unit; S232, each sub-block Construct its gray-level co-occurrence matrix ; S233, based on gray-level co-occurrence matrix For each sub-block Extract texture feature triples to generate gray-level co-occurrence matrix texture vectors. ; S234, for the entire microscopic image frame The Canny algorithm is applied for edge extraction to obtain the edge contour map. .

5. A method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 4, characterized in that, S3 includes: S31 will come from digital graphs Each map unit Input a multi-scale encoder network to extract structural semantic embedding vectors at different scales. It outputs the feature embeddings of each microscopic image frame at multiple scales. ; S32, embedding features across all scales Channel fusion and position encoding are performed to form a unified morphological embedding tensor. ; S33, to Pixel-by-pixel classification is performed using a nested developmental stage classifier. Output category probabilities, including the probability of auricularia. Probability of jar-shaped larvae and the probability of later larvae ; S34, Introducing a developmental stage continuity regularization term. Optimize the smoothness of adjacency between class probabilities; S35, the category corresponding to the highest category probability is used as the label of each pixel, and an instance mask is generated based on pixel clustering. .

6. A method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 5, characterized in that, S4 includes: S41, For each microscopic image frame, based on pixel-level classification labels and instance masks... Extract candidate instance regions for each category. ,in, For the first Candidate instance regions, For the first Frame belongs to category The number of candidate instances; S42, for each candidate instance region Extract its morphological features, such as area and aspect ratio, and compare them with the prior morphological model of this category. The comparison is performed to obtain the matching confidence of the candidate instances. ; S43, If the microscopic image frames are temporally continuous frames, the confidence score is adjusted using the spatial span consistency of developmental changes to obtain the developmental span-weighted confidence score. ; S44, calculate the consistency score using the Sobel gradient graph and neighborhood consistency. , representing pixels Boundary stability at each candidate instance region Calculate the average boundary continuity ,like If it is, then it is considered a pseudo-instance or an abnormally overlapping region, where, This is the morphological difference threshold used for image morphological consistency determination; S45, Output the... Each type of larva in the frame Corrected statistics .

7. A method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 6, characterized in that, S41 includes: S411, For each microscopic image frame, based on pixel-level classification labels and instance masks... , build categories joint mask , is represented as: ； S412, for combined mask Perform 8-neighborhood connectivity analysis to extract the set of all connected pixels as candidate instance regions.

8. A method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 7, characterized in that, S42 includes: S421, for each microscopic image frame Categories The Candidate instance regions Extract its morphological feature vector ; S422, Pre-statistics for each category Average morphological vector And calculate the current candidate region and the prior morphological model. The square of the Euclidean distance between them; S423, Calculate the matching confidence level based on the morphological difference comparison results. .

9. A method for monitoring planktonic larval communities of *Sinocyclocheilus spp.* in nearshore waters according to claim 8, characterized in that, S5 includes: S51, for each monitoring station and microscopic image frames Combined with filtration volume parameters Calculate the abundance vector per unit volume of various larvae of *Solanum lycopersicum*. ; S52, based on abundance vector Construct a stage structure matrix that reflects the proportions and interaction structures of each stage. ; S53 outputs the stage structure matrix and station information as standardized ecological monitoring data assets, forming a community abundance distribution model for nearshore sea areas.

Images