A method and system for detecting and monitoring and evaluating the ecological dynamics of a sea cucumber population

Abstract

The present application relates to sea cucumber breeding technical field, provide a kind of sea cucumber detection and population ecological dynamic monitoring evaluation method and system, comprising: based on the preset survey trajectory vertical collection target sea area includes the underwater calibration image of sea cucumber;Using pre-trained SCYOLO-seg model extracts sea cucumber pixel boundary;Adopt polygon area formula to calculate sea cucumber pixel area, and calculate sea cucumber real physical area based on pre-calibrated unit pixel real physical area;Using the sea cucumber weight trait prediction model based on the mapping relationship between sea cucumber area and weight estimates sea cucumber real weight.In actual detection and monitoring, sea cucumber does not need to be caught, off-water operation, realizes the individual specification measurement of completely non-contact type, both avoids the stress damage caused by traditional contact type measurement to sea cucumber living body, guarantees the normal growth of sea cucumber, while greatly improves the operation efficiency of sea cucumber measurement.

Description

Technical Field

[0001] This invention relates to the field of sea cucumber farming technology, and in particular to a method and system for sea cucumber detection and population ecological dynamic monitoring and evaluation. Background Technology

[0002] As an important marine economic species, sea cucumbers' growth traits and population spatial ecological distribution patterns are core indicators for assessing the ecological value of marine ranches and formulating scientific harvesting strategies. Traditional sea cucumber trait measurement relies on manual diving for harvesting and then weighing using calipers and electronic scales. However, once sea cucumbers are removed from the water or subjected to physical contact, they are prone to violent body contractions and twisting, even triggering a "evisceration reaction," leading to severe distortion of weight and size data and causing irreversible mechanical damage. Furthermore, traditional underwater transect or quadrat survey methods suffer from inherent drawbacks such as low efficiency, high cost, and significant subjective errors, making it difficult to meet the practical needs of modern marine ranches for high-frequency, high-throughput, non-contact, continuous ecological monitoring of benthic organisms.

[0003] In recent years, computer vision technology has been gradually applied to marine organism identification, but its mature application in sea cucumber instance segmentation and morphology extraction tasks still lacks maturity. Sea cucumbers mostly inhabit muddy areas on the seabed, have strong camouflage capabilities, and their bodies are covered with irregular parapodia. The downsampling operation of conventional segmentation models (such as YOLOv8-seg) easily erases high-frequency edge details, and the fixed square convolution receptive field is difficult to adapt to the dynamic deformation of sea cucumbers underwater, such as their slenderness and curvature. This results in low edge fit of the generated segmentation mask or even breakage, which seriously affects the baseline accuracy of area measurement.

[0004] Furthermore, existing visual weight estimation models mostly employ simple linear regression of length and width, which is insufficient to meet the stable measurement requirements in scenarios with highly variable sea cucumber morphology. Real sea cucumber data often contains anomalous data such as eviscerated individuals, extremely shrunken individuals, or individuals with attached mud and sand. Existing methods lack a mechanism for cleaning anomalous samples based on physical laws, which can easily lead to machine learning models collapsing or overfitting, resulting in extremely poor generalization ability in complex sea areas.

[0005] Currently, individual visual recognition and population ecological spatial pattern assessment are still conducted in isolation. Existing computer vision systems often stop at outputting the size box or two-dimensional area of ​​a single target, failing to effectively couple data with multidimensional habitat environmental factors and macro-ecological spatial models. Therefore, there is an urgent need for a method that can overcome interference from complex underwater backgrounds, accurately fit the deformed contours of sea cucumbers, and realize sea cucumber weight measurement and ecological assessment. Summary of the Invention

[0006] To address the problem that existing technologies struggle to measure sea cucumber traits using computer vision, the first objective of this invention is to provide a method for sea cucumber detection and population ecological dynamic monitoring and assessment, comprising:

[0007] Underwater calibration images of sea cucumbers in the target sea area were vertically acquired based on a preset survey trajectory.

[0008] Use a pre-trained SCYOLO-seg model to extract the pixel boundaries of sea cucumbers;

[0009] The pixel area of ​​the sea cucumber is calculated using the polygon area formula, and the real physical area of ​​the sea cucumber is calculated based on the pre-calibrated real physical area of ​​the unit pixel.

[0010] The true weight of sea cucumbers was estimated using a sea cucumber weight trait prediction model based on the mapping relationship between sea cucumber area and weight.

[0011] Specifically, the SCYOLO-seg model, based on the original YOLO11-seg visual recognition model, replaces the conventional downsampling layer Conv in the backbone and neck networks with Haar wavelet downsampling modules, replaces the feature extraction module C3k2 in the backbone and neck networks with deformable convolutional C2f_DCNv4 modules, replaces the C2PSA module in the feature extraction and fusion stage with a heterogeneous multi-scale attention network, and replaces the original segmentation head with a dynamic snake-shaped convolutional segmentation head.

[0012] Specifically, the Haar wavelet downsampling module uses a two-dimensional Haar discrete wavelet transform to transform the input feature map. Decomposed into low-frequency components Horizontal detail components Vertical detail components and diagonal detail components ;

[0013] The four components are concatenated along the channel dimension to generate an intermediate feature map. ;

[0014] The output features are downsampled by mapping to the target number of output channels C' through 1×1 pointwise convolution, followed by batch normalization and ReLU activation. .

[0015] Specifically, the C2f_DCNv4 module will input the feature map After adjusting the number of channels using 1×1 convolution, the channels are evenly divided into reserved branches. and processing branches The processing branch There are n Bottleneck structures containing DCNv4;

[0016] The Bottleneck structure first extracts local features through a 3×3 standard convolution. Then, the offset is dynamically predicted using DCNv4. and modulation coefficient For the reference sampling point corresponding to the output position Adaptive sampling formula for:

[0017] ,

[0018] in, This represents the modulation coefficient of the k-th sampling point in group G. This indicates the reference sampling point at the current output position. This indicates a predefined sampling point offset. This represents the learned dynamic offset, and K represents the total number of sampling points;

[0019] Predicting offsets using additional convolutional branches and modulation scalar :

[0020] ;

[0021] Keep the branch Initial processing branch and the output of each Bottleneck The concatenation is performed along the channel dimension, followed by 1×1 convolutional fusion and dimensionality reduction to output the feature map. .

[0022] Specifically, the heterogeneous multi-scale attention network module will input the feature map First, the number of channels is adjusted using 1×1 convolution, and the features are then divided into g subgroups along the channel dimension. Each subgroup passes through a multi-scale convolutional module in parallel. This module contains at least three parallel branches, each extracting features from a different receptive field, and compressing the output channel count to [number missing]. The features output from each convolutional branch are concatenated along the channel dimension and then fused through a 1×1 convolution to obtain the multi-scale features corresponding to that subgroup. The features of each subgroup are input into the channel attention module and the spatial attention module respectively, and all subgroups are attention-enhanced features. The data is concatenated and connected to the input feature map via residual connections. Add them together to output the feature map. .

[0023] Specifically, the channel attention module focuses on the features of the Jth subgroup. Perform global average pooling and global max pooling respectively. Two were obtained. The two descriptors are fed into a shared multilayer perceptron (MLP), the outputs are summed, and then activated by a sigmoid function to obtain the channel weights. ;

[0024] The spatial attention module employs a directional decomposition mechanism, encoding features in both the height and width directions, and then processing the subgroup features. Perform max pooling and average pooling along the channel axis respectively to obtain two results. The feature maps are concatenated and then subjected to 7×7 convolution and sigmoid activation to obtain spatial weights. .

[0025] Specifically, the dynamic snake-shaped convolution segmentation head will input the feature map Predict the initial offset using a 3×3 convolution. Where K is the number of sampling points, and is scaled using the tanh function to... The offset prediction formula is:

[0026] ,

[0027] in, This represents the initial coordinate offset matrix for prediction, α is the scaling factor, and BN represents the batch normalization operation. This represents a standard convolution with a kernel size of 3×3;

[0028] Coordinates of the original sampling point Based on this, and according to the shape and orientation of the target to be segmented, the horizontal offset component of the initial prediction is adjusted along the X-axis or Y-axis. or vertical offset The process involves iterative accumulation, followed by bilinear interpolation to extract the final interpolated features corresponding to consecutive sampling points p. :

[0029] ,

[0030] In the formula, p represents continuous sampling points. Let p be the four integer pixels surrounding p, and q be the set. Integer pixels in the data. The original feature value corresponding to the integer pixel q. For distance-based interpolation weights, interpolation weights The calculation is as follows:

[0031] ;

[0032] The sampled features are weighted and summed according to the convolution kernel weights to obtain the final output features.

[0033] Specifically, the sea cucumber pixel area The calculation formula is:

[0034] ,

[0035] in, The total number of vertices. , Let a be the coordinates of the a-th vertex. , Let the coordinates be those of the (a+1)th vertex.

[0036] The sea cucumber weight trait prediction model includes:

[0037] Based on the growth patterns of sea cucumbers, a physical proportion characteristic variable P is constructed.

[0038] ,

[0039] In the formula, E is the allometric growth parameter for converting the two-dimensional projected area of ​​the sea cucumber into its three-dimensional equivalent volume;

[0040] The interquartile range algorithm is used to remove extreme outliers from the physical ratio P. Let the first quartile be Q1 and the third quartile be Q3, then:

[0041] ,

[0042] Samples that meet the following criteria are retained as high-quality training samples:

[0043] ,

[0044] Samples that do not meet the above conditions will be considered abnormal samples and removed.

[0045] Multiple machine learning regression models were built for fitting, and 2-3 mean absolute percentage error (MAPE) models were selected from them. And the coefficient of determination The regression model with a value greater than 0.95 was used to average and fuse the prediction results to obtain the final predicted value of sea cucumber weight.

[0046] Specifically, this also includes: obtaining the number of quadrats t in the underwater calibration image, the total number of sea cucumber individuals T in the quadrat set, and the number of sea cucumber individuals observed in the u-th quadrat. A log-likelihood function is constructed based on the symmetric Dirichlet-polynomial model.

[0047] ,

[0048] in, For gamma function, Let be the probability mass function.

[0049] For the observed dataset, construct its log-likelihood function. :

[0050] ,

[0051] The maximum likelihood estimate of the distribution aggregation level parameter β is obtained by solving the numerical optimization algorithm, and the spatial distribution pattern of sea cucumber population is determined based on the β value.

[0052] The second objective of this invention is to provide a sea cucumber detection and population ecological dynamic monitoring and assessment system, using the aforementioned sea cucumber detection and population ecological dynamic monitoring and assessment method, comprising:

[0053] The image acquisition module acquires underwater calibration images of sea cucumbers in the target sea area based on a preset survey trajectory, as well as the vertical shooting height of the images from the seabed surface.

[0054] The sea cucumber instance segmentation module uses a pre-trained SCYOLO-seg model to extract the pixel boundaries of sea cucumbers in the underwater calibration image;

[0055] The area calculation module calculates the actual physical area of ​​the sea cucumber based on the pixel boundaries and the pre-calibrated actual physical area per unit pixel.

[0056] The weight prediction module estimates the true weight of sea cucumbers based on a sea cucumber weight trait prediction model that maps the area of ​​sea cucumbers to their weight.

[0057] The output module provides visualizations of sea cucumber area, weight, population density, biomass per unit area, and spatial aggregation distribution patterns.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0059] 1. This invention addresses the biological morphology and underwater imaging characteristics of sea cucumbers by constructing a dedicated SCYOLO-seg instance segmentation model based on the YOLO architecture. It replaces the conventional downsampling layer with a Haar wavelet downsampling (HDown) module, preserving the high-frequency edge information of the sea cucumber's spines while reducing dimensionality. This solves the problems of blurred edges and lost boundaries caused by the fusion of sea cucumbers with sediment backgrounds. Furthermore, it replaces the conventional feature extraction module with a C2f_DCNv4 module, enabling the convolutional kernel to adaptively fit the irregular shapes of sea cucumbers as they contract, stretch, and twist, significantly improving the model's ability to extract features from the varied postures of sea cucumbers. Improved generalization performance; through the heterogeneous multi-scale attention network HSAN module, accurate capture of multi-scale sea cucumber targets was achieved, while the dual attention mechanism of space and channel effectively suppressed background noise interference such as underwater suspended objects and water ripples; through the dynamic snake convolution segmentation head DSCsegment, high-precision continuous fitting of slender and curved sea cucumber boundaries was achieved, solving the problems of mask breakage and low edge fitting in traditional segmentation heads. SCYOLO-seg achieved an mAP50 index of 0.96448 for instance segmentation, which is 2.45% higher than YOLO11-seg.

[0060] 2. This invention obtains sea cucumber pixel boundaries based on instance segmentation, calculates the sea cucumber pixel area using the polygon area formula, calculates the true physical area of ​​the sea cucumber using the real physical area of ​​the unit pixel obtained through calibration, and calculates the predicted weight value of the sea cucumber based on the constructed sea cucumber weight trait prediction model. The entire process does not require harvesting or removing the sea cucumber from the water, realizing a completely non-contact individual size measurement. This avoids the stress damage caused to live sea cucumbers by traditional contact measurement, ensures the normal growth of sea cucumbers, and greatly improves the efficiency of sea cucumber measurement.

[0061] 3. This invention uses a symmetric Dirichlet-polynomial model to construct a log-likelihood function to calculate the maximum likelihood estimate of the distribution aggregation level parameter, thereby realizing a quantitative assessment of the spatiotemporal evolution of the spatial pattern of sea cucumber populations. Attached Figure Description

[0062] Figure 1 This is an overall flowchart of the method of the present invention;

[0063] Figure 2 The overall network structure diagram of the SCYOLO-seg instance segmentation model;

[0064] Figure 3 This is a structural diagram of the Haar wavelet downsampling (HDown) module;

[0065] Figure 4 This is a structural diagram of the deformable convolution (C2f_DCNv4) module;

[0066] Figure 5 This is a diagram of the Heterogeneous Multiscale Attention Network (HSAN) module structure.

[0067] Figure 6 This is a diagram showing the overall structure of the Dynamic Snake Convolution Segment (DSCsegment).

[0068] Figure 7 This is a diagram of the Dynamic Snake Convolution (DSCConv) module structure.

[0069] Figure 8 This is a comparison chart showing the segmentation results of sea cucumber instances using different segmentation models.

[0070] Where: a is the test result graph of the SCYOLO-seg model, and b is the test result graph of the YOLO11-seg model;

[0071] Figure 9 A comparison of the fitting relationship between sea cucumber area and weight under different models;

[0072] Figure 10 A schematic diagram showing the segmentation and area measurement of a sea cucumber example;

[0073] Figure 11 A schematic diagram of the "S"-shaped full-scan survey route for a sea cucumber population ecological survey;

[0074] Figure 12 A schematic diagram of the quadrat setup and survey route in the transect survey method for sea cucumber population ecological investigation. Detailed Implementation

[0075] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0076] This invention provides a method for detecting and monitoring and evaluating the dynamic ecology of sea cucumber populations, achieving a fully integrated monitoring and evaluation process from non-contact, precise underwater individual measurements to macroscopic population density, biomass estimation, and environmentally driven spatiotemporal evolution analysis.

[0077] S1: Acquisition of sea cucumber image set and mapping of real physical features.

[0078] Collect at least 300 sea cucumbers from the same aquaculture area or marine ranch, acquire images of the sea cucumbers, and simultaneously weigh them to obtain their actual live weight. Use Zhang's calibration method to calibrate the internal parameters of the underwater camera, establishing a mapping relationship between pixel-level space and actual physical area.

[0079] S2: Extracting sea cucumber pixel boundaries based on a pre-trained accurate instance segmentation model.

[0080] This invention addresses the characteristics of sea cucumbers, such as their elongated shape, tendency to shrink and twist, and tendency to blend into the muddy background. It proposes a YOLO11-seg instance segmentation model, SCYOLO-seg, suitable for sea cucumbers, with the structure shown below. Figure 2 As shown.

[0081] S2.1: In this invention, the conventional downsampling layer Conv in the backbone and neck networks is replaced with a Haar wavelet downsampling (HDown) module to preserve the high-frequency edge information of the sea cucumber parapodia.

[0082] Because sea cucumbers are covered with irregular parapodia and their body color often blends seamlessly with complex backgrounds such as mud and sand, traditional stride convolution downsampling results in a significant loss of high-frequency edge details. This invention utilizes the HDown module, leveraging the characteristics of Haar wavelet transform, to preserve the parapodia outlines and high-frequency boundary components of sea cucumbers while reducing spatial resolution. This solves the segmentation problem of blurred edges and easy background confusion in complex underwater environments. Furthermore, Haar wavelets are computationally simple and provide accurate edge localization, making them suitable for real-time segmentation tasks.

[0083] To reduce the loss of sea cucumber edge texture information during downsampling, the conventional downsampling layers in the backbone and neck networks were replaced with Haar wavelet downsampling (HDown) modules, the structure of which is shown in the diagram below. Figure 3 .

[0084] Input feature map ,

[0085] Where C represents the number of channels; H and W represent the height and width of the feature map, respectively.

[0086] Two-dimensional Haar discrete wavelet transform (DWT) is used to decompose the feature into four sub-bands with half the spatial resolution. Low-frequency components are then extracted. And three high-frequency detail components, including the horizontal detail component. Vertical detail components Diagonal detail components For any spatial coordinates The formulas for calculating each component are as follows:

[0087] ,

[0088] ,

[0089] ,

[0090] ,

[0091] All component dimensions are .

[0092] Subsequently, the four components are concatenated along the channel dimension to generate an intermediate feature map. This helps preserve edge detail information:

[0093] ,

[0094] Finally, the downsampled features are obtained by mapping the target output channel number C' to a 1×1 pointwise convolution, followed by batch normalization and ReLU activation. :

[0095] ,

[0096] This downsampling module reduces spatial resolution while preserving high-frequency edge information, thus mitigating the loss of sea cucumber boundary details caused by downsampling.

[0097] S2.2: In this invention, the feature extraction module C3k2 in the backbone and neck networks is replaced with the deformable convolutional module C2f_DCNv4 to adapt to the irregular deformation of sea cucumbers.

[0098] In seawater, sea cucumbers exhibit irregular morphological changes such as contraction, stretching, and twisting. Fixed square convolutional kernels cannot effectively extract these dynamic features. The C2f_DCNv4 module introduced in this invention, through adaptive prediction of spatial offset, enables the sampling points of the convolutional kernel to dynamically fit the actual bending morphology of the sea cucumber, improving the model's generalization ability to the varied postures of sea cucumbers and the completeness of feature extraction.

[0099] To improve the network's ability to model sea cucumber posture changes, this invention replaces all feature extraction modules C3k2 in the backbone and neck networks with deformable convolutional modules C2f_DCNv4, the structure of which is shown in the diagram below. Figure 4 .

[0100] The input feature map is After adjusting the number of channels using 1×1 convolution, the channels are evenly divided into reserved branches. and processing branches , The system sequentially enters n Bottleneck structures containing DCNv4. Each Bottleneck structure first extracts local features through a standard 3×3 convolution. Then, the offset is dynamically predicted using DCNv4. And modulation coefficient (modulation scalar) For the reference sampling point corresponding to the output position Adaptive sampling formula for:

[0101] ,

[0102] in, This represents the modulation coefficient of the k-th sampling point in group G. This indicates the reference sampling point at the current output position. This indicates a predefined sampling point offset. This represents the dynamic offset obtained through learning, and K represents the total number of sampling points.

[0103] offset and modulation scalar Predicted through additional convolutional branches:

[0104] .

[0105] Let the output of the nth Bottleneck be denoted as Finally, the branch will be preserved. Initial processing branch and the output of each Bottleneck By concatenating the data along the channel dimension and then performing 1×1 convolutional fusion for dimensionality reduction, a feature map is obtained that enhances the ability to represent target deformation. .

[0106] ,

[0107] The convolution kernel in the C2f_DCNv4 module can adaptively adjust the sampling position according to the target shape, thereby enhancing the ability to represent the irregular curved contours of sea cucumbers.

[0108] S2.3: This invention replaces the C2PSA module in the feature extraction and fusion stage with a heterogeneous multi-scale attention network (HSAN) to filter effective information and suppress underwater suspended object background interference in multi-scale feature fusion.

[0109] Due to differences in shooting distance and individual sea cucumber size, the sea cucumbers in the images vary in size, and background interference such as suspended objects may appear. This invention employs a heterogeneous multi-scale attention network (HSAN). On one hand, it accurately captures sea cucumbers of different sizes through multi-scale convolutional modules. On the other hand, it uses a dual attention mechanism to coordinate spatial and channel attention, focusing on the key directional distribution of the sea cucumber body and suppressing underwater environmental noise and false target interference. Its structural diagram is shown below. Figure 5 .

[0110] To accurately locate the spatial position of deformation features, the C2PSA module in the feature extraction and fusion stage is replaced with a heterogeneous multi-scale attention network (HSAN), thereby improving the model's ability to jointly model multi-scale local structures, directional spatial distributions, and channel importance.

[0111] Input feature map After adjusting the number of channels using 1×1 convolution, the features are divided into g subgroups along the channel dimension. Each subgroup passes through a multi-scale convolutional module (MSCM) in parallel. The MSCM contains at least three parallel branches, employing at least 1×1, 3×3, and 5×5 convolutional kernels respectively, to extract features from different receptive fields, compressing the output channel count to [number missing]. The features output from each convolutional branch are concatenated along the channel dimension and then fused using a 1×1 convolution to obtain the multi-scale features corresponding to that subgroup. The features of each subgroup are then input into the channel attention module and the spatial attention module respectively, and the attention weights are calculated in parallel.

[0112] Channel attention module for the J-th subgroup features Perform global average pooling and global max pooling respectively. Two were obtained. The two descriptors are fed into a shared multilayer perceptron (MLP), the outputs are summed, and then activated by a sigmoid function to obtain the channel weights. :

[0113] ;

[0114] The spatial attention module employs a direction decomposition mechanism, encoding features in both the height and width directions, and then processing the subgroup features. Perform max pooling and average pooling along the channel axis respectively to obtain two results. The feature maps are concatenated and then subjected to 7×7 convolution and sigmoid activation to obtain spatial weights. :

[0115] .

[0116] The attention-enhanced features are then summed to obtain the attention-enhanced features of the Jth subgroup. :

[0117] ,

[0118] In the formula, This indicates element-wise multiplication.

[0119] Finally, attention enhancement features for all subgroups The data is concatenated and connected to the input feature map via residual connections. Adding them together yields the global output feature map of the HSAN module. ,

[0120] .

[0121] This module jointly models multi-scale local structures and directional spatial distributions, effectively suppressing underwater noise and background interference.

[0122] S2.4: This invention replaces the original segmentation head with a dynamic snake convolution segmentation head (DSCsegment) to improve the ability to finely fit the slender and curved sea cucumber boundary.

[0123] Sea cucumbers exhibit a typical elongated, continuously tortuous structure morphologically. Conventional segmentation heads are prone to mask breakage or poor edge fit when processing such slender and curved targets. This invention employs a dynamic serpentine convolutional segmentation head (DSCsegment), which continuously iterates and accumulates the sampling point coordinates, ensuring that the receptive field strictly follows the tortuous path of the sea cucumber in space, thereby generating an instance segmentation mask with high edge fit and structural continuity.

[0124] To enhance the detection head's ability to fit the elongated, curved, and complex boundaries of sea cucumbers, a dynamic serpentine convolution (DSCConv) is introduced in the mask coefficient prediction branch to replace the standard convolution, constructing a dynamic serpentine convolution segmentation detection head (DSCsegment). This improves the modeling ability for curved, elongated, and complex boundary targets. Its structure diagram is shown below. Figure 6 and Figure 7 .

[0125] Input feature map First, the initial offset is predicted using a 3×3 convolution. , Where K is the number of sampling points, and is scaled using the tanh function to... The offset prediction formula is:

[0126] ,

[0127] in, This represents the initial coordinate offset matrix for prediction; α is the scale factor used to control the maximum magnitude of the offset, with a default value of 2; BN indicates batch normalization operation. This represents a standard convolution with a kernel size of 3×3.

[0128] Coordinates of the original sampling point Based on this, and according to the shape and orientation of the target to be segmented, the horizontal offset component of the initial prediction is adjusted along the X-axis or Y-axis. or vertical offset Iterative accumulation is performed to generate a continuously curved sampling path that fits the target contour.

[0129] Let the center index of the convolution kernel be... Taking the horizontal direction as an example, the cumulative offset of the center sampling point in the initial state , ,

[0130] Cumulative offset of the vth sampling point to the right of the center point for:

[0131] ,

[0132] Cumulative offset of the v-th sampling point to the left of the center point for:

[0133] ,

[0134] Cumulative offset of the vth sampling point above the center point for:

[0135] ,

[0136] Cumulative offset of the vth sampling point below the center point for:

[0137] ,

[0138] Coordinates of the original sampling point The horizontal offset component is accumulated in the previous iteration. and vertical offset Obtain the final sampling point coordinates .

[0139] Since the cumulative coordinates may be non-integer, bilinear interpolation is used to extract the final interpolation features corresponding to consecutive sampling points p. :

[0140] ,

[0141] In the formula, p represents continuous sampling points. Let p be the four integer pixels surrounding p, and q be the set. Integer pixels in the data. The original feature value corresponding to the integer pixel q. For distance-based interpolation weights, interpolation weights The calculation is as follows:

[0142] .

[0143] The sampled features are weighted and summed according to the convolution kernel weights to obtain the final output features. Finally, the extracted deformation features are processed through 1×1 convolution to generate bounding box parameters (Box), classification scores (Cls), and mask coefficients (MaskCoef), which are then combined with the prototype mask to obtain the instance segmentation result. This module ensures that the receptive field extends continuously along the curved path of the target, thereby avoiding mask breaks.

[0144] refer to Figure 8Using the same training and validation sets and under the same hardware configuration, both the SCYOLO-seg model of this invention and the basic YOLO11-seg model were trained using the minimum parameter version (n-series), with 100 training epochs for each model. Experimental results show that SCYOLO-seg significantly outperforms YOLO11-seg in all performance metrics for object detection and instance segmentation tasks. In terms of object detection performance, SCYOLO-seg achieves precision, recall, mAP50, and mAP50-95 of 0.96462, 0.93253, 0.95025, and 0.66519, respectively, representing improvements of 0.03948, 0.00371, 0.00916, and 0.03573 compared to YOLO11-seg. In terms of instance segmentation performance, SCYOLO-seg achieved precision, recall, mAP50, and mAP50-95 of 0.97305, 0.94340, 0.96448, and 0.74387, respectively, which are improvements of 0.04791, 0.01058, 0.02309, and 0.05210 compared to YOLO11-seg.

[0145] Among them, SCYOLO-seg achieved an mAP50 score of 0.96448 for instance segmentation, which is 2.45% higher than YOLO11-seg, indicating that it has better localization and mask fitting capabilities. This shows that SCYOLO-seg has stronger feature representation and boundary recovery capabilities for slender, curved, and complex targets such as sea cucumbers, making it more suitable for sea cucumber instance segmentation tasks.

[0146] S3: Construct a measurement model for sea cucumber growth traits.

[0147] The image to be tested is input into the SCYOLO-seg model to extract the pixel boundaries of the sea cucumber, and the true physical area of ​​the sea cucumber is calculated based on the Shoelace polygon theorem. Multidimensional outlier data is cleaned using the interquartile range (IQR) algorithm based on the allometric growth law, and then input into a network that integrates a multi-model regression algorithm to output the predicted live weight of the sea cucumber.

[0148] S3.1: Extraction and calculation of sea cucumber area based on SCYOLO-seg model.

[0149] The image to be tested is input into the pre-trained SCYOLO-seg instance segmentation model, which outputs the set of vertex coordinates of the boundary polygon corresponding to each sea cucumber.

[0150] Let the sea cucumber dividing boundary be... The pixel area of ​​the sea cucumber is composed of vertices. The area is calculated using the polygon area formula:

[0151] ,

[0152] in, The total number of vertices. , Let a be the coordinates of the a-th vertex. , Let the coordinates be the (a+1)th vertex;

[0153] Based on the actual physical area per pixel obtained from previous camera calibration Further calculate the true physical area of ​​the sea cucumber. :

[0154] ,

[0155] in, The preferred unit is cm²;

[0156] The above method enables automatic conversion from sea cucumber boundary segmentation results to the actual area characteristics of sea cucumbers.

[0157] S3.2: Construct a prediction model for sea cucumber weight traits.

[0158] To establish the mapping relationship between sea cucumber area and weight, at least 300 sea cucumber samples were collected from the same sea area, covering different growth stages and individual sizes as much as possible. For each sea cucumber, the SCYOLO-seg model was used for instance segmentation, and its true physical area was calculated according to step 3.1. Meanwhile, the true weight of each sea cucumber was obtained by manual weighing. This forms a sample set with a one-to-one correspondence between area and weight.

[0159] Based on the growth pattern of sea cucumbers, a physical proportion characteristic variable P is constructed:

[0160] ,

[0161] In the formula, E is the allometric growth parameter for the transformation of the two-dimensional projected area of ​​the sea cucumber into the three-dimensional equivalent volume, which is preferably 1.5.

[0162] To improve the stability of the weight prediction model, the interquartile range (IQR) algorithm is used to remove extreme outliers from the physical proportion P.

[0163] Let the first quartile be Q1 and the third quartile be Q3, then we have:

[0164] ,

[0165] Samples that meet the following criteria are retained as high-quality training samples:

[0166] ,

[0167] Samples that do not meet the above conditions will be considered abnormal samples and removed to reduce interference caused by sea cucumber evisceration, extreme shrinkage, measurement errors, or abnormal individuals.

[0168] After outlier cleaning, the high-quality sample set is divided into training and testing sets, and a feature matrix (sea cucumber area) is constructed. ) and target vector (true weight of sea cucumber) ).

[0169] Based on this, various machine learning regression models were established for fitting, referencing... Figure 9 The regression model includes at least one or more of the following: power function regression model, quadratic polynomial regression model, Huber robust polynomial regression model, support vector regression model, random forest regression model, generalized additive model, etc.

[0170] To evaluate the performance of each regression model, the mean absolute percentage error (MAPE) and the coefficient of determination were used. As an evaluation indicator:

[0171] ,

[0172] ,

[0173] in, Indicates the actual weight. Indicates the predicted weight. This represents the average actual weight.

[0174] Preferably, the constructed model satisfies: MAPE < 10%. >0.95.

[0175] Two to three regression models with the best performance can be selected based on the evaluation results of the test set. Their prediction results can be averaged and fused to obtain the final predicted value of sea cucumber weight, thereby improving the generalization ability and robustness of the model.

[0176] In practical applications, the image of the sea cucumber to be measured is input into the SCYOLO-seg model to achieve accurate segmentation of each sea cucumber instance. Subsequently, the true area of ​​the sea cucumber is calculated based on the segmentation boundaries. Finally, the true area is input into the weight trait prediction model to obtain the predicted weight value of the sea cucumber, thereby achieving high-throughput, non-contact, and automated measurement of growth traits such as sea cucumber area and weight.

[0177] S4: Construct a method for monitoring and evaluating the dynamics of sea cucumber populations.

[0178] Sea cucumber distribution images were acquired through in-situ, non-destructive surveys using an underwater camera system. The density and biomass per unit area of ​​sea cucumbers within each quadrat were automatically calculated using the SCYOLO-seg model and a trait measurement model. The maximum likelihood estimate β was calculated using a symmetric Dirichlet-multinomial (SDM) model to quantify the spatial aggregation pattern of the sea cucumber population. Furthermore, combined with simultaneously acquired multi-parameter water quality data, principal component analysis (PCA) and analysis of variance were employed to comprehensively assess the impact of environmental factors on the spatiotemporal evolution and distribution of the sea cucumber population.

[0179] S4.1: Ecological survey and data acquisition of sea cucumber populations.

[0180] To regularly monitor and assess sea cucumbers in marine ranches or aquaculture areas, in-situ non-destructive survey methods are used to obtain seabed images and environmental data. The specific steps are as follows:

[0181] Divers carrying underwater camera equipment, or waterproof camera systems mounted on unmanned vessels or underwater robots, conduct vertically oriented filming of the surveyed sea area. The waterproof cameras are pre-calibrated. The survey route is selected based on the size of the area to be surveyed, see [link to relevant documentation]. Figure 11 and Figure 12 If the survey area is small, an "S"-shaped full-scan survey method is used. The camera lens is controlled to be vertically downward, aimed at the seabed, and the entire survey area is filmed and recorded along a continuous "S"-shaped serpentine trajectory. If the survey area is large, a transect survey method is used. The survey area is divided into multiple parallel survey transects, and multiple square survey plots are set at equal intervals or randomly within each transect. Within each plot, complete image sampling is performed using the aforementioned "S"-shaped full-scan survey trajectory. During the above filming process, the vertical distance from the camera to the seabed (i.e., the shooting height) corresponding to each frame of image is recorded and stored in real time.

[0182] While taking images, the onboard water quality multi-parameter measuring instrument can also be used to simultaneously measure the environmental physicochemical factors of the sea cucumber distribution area at the bottom. The environmental factors include at least: pH value, dissolved oxygen, water temperature and salinity.

[0183] S4.2: Ecological monitoring and comprehensive assessment of sea cucumber populations.

[0184] For raw images or video frames acquired using an "S"-shaped full-scan survey, the actual physical area represented by a single pixel in the image is calculated based on the real-time shooting height *h* of each image and the camera's intrinsic calibration information. Subsequently, the images are input into a pre-trained SCYOLO-seg sea cucumber instance segmentation model to segment all sea cucumber targets in the image and extract the mask for each sea cucumber. By counting the number of pixels occupied by the mask for each sea cucumber and combining this with the physical area of ​​a single pixel, the projected area of ​​each sea cucumber, i.e., its size, is calculated. The projected area of ​​each sea cucumber is then input into a constructed sea cucumber weight prediction model to estimate the live weight of each sea cucumber. Finally, all monitored sea cucumbers within a single sample plot are summed to obtain the total number, individual size distribution, and weight information of sea cucumbers within that sample plot.

[0185] Based on the data collected from the quadrats, the basic population indices of sea cucumbers in this surveyed sea area were calculated, including population density D and biomass per unit area B:

[0186] ,

[0187] in, This represents the total number of sea cucumbers detected within the quadrat. This represents the actual physical area of ​​the quadrat, in units of: ,

[0188] ,

[0189] in, This represents the sum of estimated wet weights of all sea cucumbers within the sample plot, in units of... .

[0190] The symmetric Dirichlet-polynomial (SDM) model was used to assess the distribution and aggregation level β of sea cucumbers within the sea area.

[0191] Suppose the survey contains t quadrats, and the total number of sea cucumbers in the quadrat set is T, where... Let represent the number of sea cucumber individuals observed in the u-th quadrat (u = 1, 2, ..., t). Then, given the distribution clustering level β and the total number of individuals T, the observed count vector... The probability mass function is:

[0192] ,

[0193] in, For gamma function, This is the probability mass function.

[0194] For the observed dataset, construct its log-likelihood function. :

[0195] ,

[0196] Numerical optimization algorithms (such as the Newton-Raphson method or the Brent method) are used to optimize the above log-likelihood function within the constraint β>0. The maximum likelihood estimate β is obtained by maximizing the probability. The distribution pattern is determined as follows: if β ≫ 1 (much greater than 1), the spatial distribution of the sea cucumber population tends to be uniform; if β ≈ 1 (around 1), the spatial distribution of the sea cucumber population is close to random; if β → 0 (close to 0), the spatial distribution of the sea cucumber population exhibits a highly clustered distribution. Furthermore, the obtained β value is compared with β values ​​from different periods or other sea areas to assess the spatiotemporal evolution dynamics of the sea cucumber population spatial pattern.

[0197] A multi-dimensional ecological assessment was conducted by combining the acquired environmental factor data:

[0198] Principal component analysis (PCA) was used to reduce the dimensionality and extract features of environmental factors such as pH, dissolved oxygen, temperature, and salinity. By calculating the correlation between sea cucumber distribution quadrats and principal component scores, the driving force and influence weight of each environmental factor on sea cucumber habitat selection were comprehensively evaluated.

[0199] If it is necessary to compare and evaluate sea cucumbers from different periods or different sea areas, when the sea cucumber density or environmental factor data conform to a normal distribution and have homogeneity of variance, one-way ANOVA should be used to test the significance of differences between different sea areas or different habitats; when the data do not conform to a normal distribution, the Kruskal-Wallis H nonparametric test should be used to evaluate the differences.

[0200] For the sea cucumber population weight structure data obtained from different survey periods, the normality of its weight distribution was first tested using the Kolmogorov-Smirnov single-sample test. Then, the least significant difference test (LSD test) was used for multiple comparisons to quantitatively assess the succession and growth status of the sea cucumber population weight structure at different time points.

[0201] This invention also provides a sea cucumber detection and population ecological dynamic monitoring and assessment system, comprising:

[0202] The image acquisition module acquires underwater calibration images of sea cucumbers in the target sea area based on a preset survey trajectory, as well as the vertical shooting height of the images from the seabed surface.

[0203] The sea cucumber instance segmentation module uses a pre-trained SCYOLO-seg model to extract the pixel boundaries of sea cucumbers in the underwater calibration image;

[0204] The area calculation module calculates the actual physical area of ​​the sea cucumber based on the pixel boundaries and the pre-calibrated actual physical area per unit pixel.

[0205] The weight prediction module estimates the true weight of sea cucumbers based on a sea cucumber weight trait prediction model that maps the area of ​​sea cucumbers to their weight.

[0206] The population ecological assessment module calculates the population density and biomass per unit area of ​​sea cucumbers in the target sea area based on the sea parameter quantity and weight data in the survey plots, and uses a symmetric Dirichlet-polynomial model to quantitatively assess the spatial aggregation and distribution pattern of sea cucumber populations.

[0207] The output module provides visualizations of sea cucumber area, weight, population density, biomass per unit area, and spatial aggregation distribution patterns.

[0208] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention. The above are merely preferred embodiments 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 technical principles of the present invention, and these improvements and modifications should also be considered within the protection scope of the present invention.

Claims

1. A method for detecting and monitoring and evaluating the ecological dynamics of a sea cucumber population, characterized in that, include: Underwater calibration images of sea cucumbers in the target sea area were vertically acquired based on a preset survey trajectory. Use a pre-trained SCYOLO-seg model to extract the pixel boundaries of sea cucumbers; The pixel area of ​​the sea cucumber is calculated using the polygon area formula, and the real physical area of ​​the sea cucumber is calculated based on the pre-calibrated real physical area of ​​the unit pixel. The true weight of sea cucumbers was estimated using a sea cucumber weight trait prediction model based on the mapping relationship between sea cucumber area and weight.

2. The method according to claim 1, wherein, The SCYOLO-seg model, based on the original YOLO11-seg visual recognition model, replaces the conventional downsampling layer Conv in the backbone and neck networks with Haar wavelet downsampling modules, replaces the feature extraction module C3k2 in the backbone and neck networks with deformable convolutional C2f_DCNv4 modules, replaces the C2PSA module in the feature extraction and fusion stage with a heterogeneous multi-scale attention network, and replaces the original segmentation head with a dynamic snake-shaped convolutional segmentation head.

3. The sea cucumber detection and population ecological dynamic monitoring and assessment method according to claim 2, characterized in that, The Haar wavelet down-sampling module adopts two-dimensional Haar discrete wavelet transform to decompose the input feature map into low-frequency components , horizontal detail components , vertical detail components , and diagonal detail components ;​ The four components are concatenated along the channel dimension to generate an intermediate feature map. ; The output features are downsampled by mapping to the target number of output channels C' through 1×1 pointwise convolution, followed by batch normalization and ReLU activation. .

4. The sea cucumber detection and population ecological dynamic monitoring and assessment method according to claim 2, characterized in that, The C2f_DCNv4 module will input the feature map After adjusting the number of channels using 1×1 convolution, the channels are evenly divided into reserved branches. and processing branches The processing branch There are n Bottleneck structures containing DCNv4; The Bottleneck structure first extracts local features through a 3×3 standard convolution. Then, the offset is dynamically predicted using DCNv4. and modulation coefficient For the reference sampling point corresponding to the output position Adaptive sampling formula for: , in, This represents the modulation coefficient of the k-th sampling point in group G. This indicates the reference sampling point at the current output position. This indicates a predefined sampling point offset. This represents the learned dynamic offset, and K represents the total number of sampling points; Predicting offsets using additional convolutional branches and modulation scalar : ； Keep the branch Initial processing branch and the output of each Bottleneck The concatenation is performed along the channel dimension, followed by 1×1 convolutional fusion and dimensionality reduction to output the feature map. .

5. The sea cucumber detection and population ecological dynamic monitoring and assessment method according to claim 2, characterized in that, The heterogeneous multi-scale attention network module will input the feature map First, the number of channels is adjusted using 1×1 convolution, and the features are then divided into g subgroups along the channel dimension. Each subgroup passes through a multi-scale convolutional module in parallel. This module contains at least three parallel branches, each extracting features from a different receptive field, and compressing the output channel count to [number missing]. The features output from each convolutional branch are concatenated along the channel dimension and then fused through a 1×1 convolution to obtain the multi-scale features corresponding to that subgroup. The features of each subgroup are input into the channel attention module and the spatial attention module respectively, and all subgroups are attention-enhanced features. The data is concatenated and connected to the input feature map via residual connections. Add them together to output the feature map. .

6. The sea cucumber detection and population ecological dynamic monitoring and assessment method according to claim 5, characterized in that, The channel attention module focuses on the features of the Jth subgroup. Perform global average pooling and global max pooling respectively. Two were obtained. The two descriptors are fed into a shared multilayer perceptron (MLP), the outputs are summed, and then activated by a sigmoid function to obtain the channel weights. ; The spatial attention module employs a directional decomposition mechanism, encoding features in both the height and width directions, and then processing the subgroup features. Perform max pooling and average pooling along the channel axis respectively to obtain two results. The feature maps are concatenated and then subjected to 7×7 convolution and sigmoid activation to obtain spatial weights. .

7. The sea cucumber detection and population ecological dynamic monitoring and assessment method according to claim 2, characterized in that, The dynamic snake-shaped convolution segmentation head will input the feature map Predict the initial offset using a 3×3 convolution. Where K is the number of sampling points, and is scaled using the tanh function to... The offset prediction formula is: ， in, This represents the initial coordinate offset matrix for prediction, α is the scaling factor, and BN represents the batch normalization operation. This represents a standard convolution with a kernel size of 3×3; Coordinates of the original sampling point Based on this, and according to the shape and orientation of the target to be segmented, the horizontal offset component of the initial prediction is adjusted along the X-axis or Y-axis. or vertical offset The process involves iterative accumulation, followed by bilinear interpolation to extract the final interpolated features corresponding to consecutive sampling points p. : ， In the formula, p represents continuous sampling points. Let p be the four integer pixels surrounding p, and q be the set. Integer pixels in the data. The original feature value corresponding to the integer pixel q. For distance-based interpolation weights, interpolation weights The calculation is as follows: ； The sampled features are weighted and summed according to the convolution kernel weights to obtain the final output features.

8. The method for sea cucumber detection and population ecological dynamic monitoring and assessment according to claim 1, characterized in that, The sea cucumber pixel area The calculation formula is: ， in, The total number of vertices. , Let a be the coordinates of the a-th vertex. , Let the coordinates be those of the (a+1)th vertex. The sea cucumber weight trait prediction model includes: Based on the growth patterns of sea cucumbers, a physical proportion characteristic variable P is constructed. ， In the formula, E is the allometric growth parameter for converting the two-dimensional projected area of ​​the sea cucumber into its three-dimensional equivalent volume; The interquartile range algorithm is used to remove extreme outliers from the physical ratio P. Let the first quartile be Q1 and the third quartile be Q3, then: , Samples that meet the following criteria are retained as high-quality training samples: , Samples that do not meet the above conditions will be considered abnormal samples and removed. Multiple machine learning regression models were built for fitting, and 2-3 mean absolute percentage error (MAPE) models were selected from them. And the coefficient of determination The regression model with a value greater than 0.95 was used to average and fuse the prediction results to obtain the final predicted value of sea cucumber weight.

9. The method for detecting and monitoring and evaluating the dynamic ecology of sea cucumber populations according to claim 1, characterized in that, Also includes: Obtain the number of quadrats t in the underwater calibration image, the total number of sea cucumber individuals T in the quadrat set, and the number of sea cucumber individuals observed in the u-th quadrat. A log-likelihood function is constructed based on the symmetric Dirichlet-polynomial model. ， in, For gamma function, Let be the probability mass function. For the observed dataset, construct its log-likelihood function. : , The maximum likelihood estimate of the distribution aggregation level parameter β is obtained by solving the numerical optimization algorithm, and the spatial distribution pattern of sea cucumber population is determined based on the β value.

10. A sea cucumber detection and population ecological dynamic monitoring and assessment system, using the sea cucumber detection and population ecological dynamic monitoring and assessment method as described in any one of claims 1-9, characterized in that, include: The image acquisition module acquires underwater calibration images of sea cucumbers in the target sea area based on a preset survey trajectory, as well as the vertical shooting height of the images from the seabed surface. The sea cucumber instance segmentation module uses a pre-trained SCYOLO-seg model to extract the pixel boundaries of sea cucumbers in the underwater calibration image; The area calculation module calculates the actual physical area of ​​the sea cucumber based on the pixel boundaries and the pre-calibrated actual physical area per unit pixel. The weight prediction module estimates the true weight of sea cucumbers based on a sea cucumber weight trait prediction model that maps the area of ​​sea cucumbers to their weight. The output module provides a visualization of the sea cucumber's area and weight.

Images