Method and system for assessing biological impacts of marine micro-ecosystems
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 崂山国家实验室
- Filing Date
- 2026-04-15
- Publication Date
- 2026-06-30
Smart Images
Figure CN122020136B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of marine ecological observation and data analysis technology, specifically, it relates to a method and system for assessing the biological impact of marine micro-ecosystems. Background Technology
[0002] In marine ecological research (especially in the island sector), in-situ, real-time, and long-term ecological observation is a core requirement for analyzing biogeochemical processes (such as nutrient cycling and plankton community succession). Traditional research relies on field sample collection, requiring significant investment of manpower and resources (such as research vessel rental, sample transportation, and pretreatment), resulting in high costs, long processing times, and short observation cycles (unable to cover the complete life cycle of organisms). Furthermore, the ecosystems of islands and surrounding waters are complex, with significant environmental fluctuations, making it difficult to precisely control single variables to explore the causal relationship between environmental parameters and biological responses.
[0003] Existing studies mostly rely on qualitative descriptions (such as "decreased algal density") and have not established a quantitative model of "changes in environmental parameters - biological response - degree of impact", making it impossible to accurately determine the threshold of ecological stress.
[0004] The information disclosed in this background section is only intended to enhance the understanding of the background technology of this application, and therefore may include prior art that is not known to those skilled in the art. Summary of the Invention
[0005] This invention proposes a method and system for assessing the biological impact of marine micro-ecosystems, in order to solve the technical problem that existing technologies cannot accurately determine the ecological stress threshold based on environmental parameters.
[0006] To achieve the above-mentioned invention / design objectives, the present invention adopts the following technical solution:
[0007] A method for assessing the biological impact of marine micro-ecosystems, obtaining environmental parameters x of marine micro-ecosystems. i Calculate environmental parameter x i The difference Δx between the baseline environmental parameters and the baseline i ;
[0008] Calculate the function of environmental parameter variation , where ω i As the weights of environmental parameters, , where n is the number of environmental parameters. This refers to the range of values for environmental parameters;
[0009] Images of marine micro-ecosystems are captured and input into a CNN model to obtain morphological feature vectors V. I[Community density ρ (number of organisms / mm²), plankton feeding rate F (times / h), morphological distortion rate D (%)];
[0010] The spectra of marine micro-ecosystems are acquired using a multispectral camera, and then input into a spectral CNN model to obtain the spectral feature vector V. λ [Chlorophyll (Chl, μg / L), Carotenoids (Car, μg / L)];
[0011] Calculate biological response function Among them, V I (t) represents the image feature vector extracted by the CNN model at time t, V I (0) is the image feature vector extracted by the CNN model of the baseline state image, V λ (t) represents the spectral feature vector extracted by the spectral CNN model at time t, V λ (0) is the spectral feature vector extracted by the baseline state spectral CNN model; a and b are the weights, a+b=1;
[0012] Calculate the BI index, a quantitative index of the impact of environmental parameters on organisms. ; where ω E For the environmental parameter weights, ω B For biological response weights, ω E +ω B =1, ε is the error term.
[0013] The marine micro-ecosystem bioimpact assessment method described above uses an image CNN model and a spectral CNN model trained with a loss function LOSS1, where LOSS1 = α × MSE(ρ pred ,ρ true )+β×MAE(V λpred V λrue )+γ×(1–cos(V I V Iaug Where MSE is the mean squared error, MAE is the mean absolute error, and ρ is the mean squared error. pred ρ is the predicted value of community density. true V represents the measured community density. λpred V is the predicted value of biochemical characteristics. λrue V represents the measured value of biochemical characteristics. I V is the feature vector of the original image. Iaug To enhance the feature vector of the image, α, β, and γ are task weights, which are dynamically adjusted based on the validation set error during training.
[0014] The above-mentioned method for assessing the biological impact of marine micro-ecosystems was used to verify the accuracy of CNN feature extraction. The verification indicators must meet the following requirements: ① coefficient of determination R² ≥ 0.96 (goodness of fit between predicted and actual values); ② chlorophyll concentration MAE ≤ 0.5 μg / L; ③ carotenoid content MAE ≤ 0.2 μg / L; ④ cosine similarity of feature vectors ≥ 0.95.
[0015] The image enhancement method for assessing the biological impact of marine micro-ecosystems as described above is as follows: image rotation of 0-180°, scaling of 0.8-1.2 times, and brightness adjustment of ±10%.
[0016] The above-described method for assessing the biological impact of marine micro-ecosystems, in ω E ≥0, ω B ≥0 and ω E +ω B =1, and ω i ≥0 and Σω i Under the constraint that = 1, the loss function LOSS2 = MSE(BI) is used. pred BI true )+λ×(1-r(BI pred BI true )) for ω i ω B and ω E Optimize, where r is the BI pred with BI true Pearson correlation coefficient, λ=0.5, BI pred The predicted value of the biological impact index (BI) calculated by the model. true To obtain the true values of the measured biological indicators, the weights are iteratively updated using the training set during the optimization process, and the convergence is evaluated using the validation set. The convergence criterion is: the Pearson correlation coefficient r of the validation set is ≥0.85 and the MSE is ≤0.02.
[0017] A marine micro-ecosystem bioimpact assessment system, the system comprising:
[0018] The enclosure includes a transparent partition, and light sources are arranged around the enclosure.
[0019] The cultivation tank is located inside the container;
[0020] Several cameras were used to acquire images and spectra of the culture tank;
[0021] Sensors are used to collect environmental parameters within the cultivation tank;
[0022] The data processing module is used to conduct marine biological impact assessments on the marine micro-ecosystems within the cultivation tanks based on the aforementioned assessment methods.
[0023] The marine micro-ecosystem biological impact assessment system described above includes cameras such as microscopic imaging cameras, high-speed dynamic cameras, and multispectral cameras.
[0024] The marine micro-ecosystem biological impact assessment system described above includes inlet and outlet pipes to achieve water circulation and nutrient replenishment.
[0025] The marine micro-ecosystem biological impact assessment system described above includes a water control valve for controlling water flow velocity and replacement frequency to simulate the marine water flow environment.
[0026] The marine micro-ecosystem biological impact assessment system described above includes an observation window that provides an observation channel for the camera, and the observation window includes high-transmittance glass.
[0027] Compared with existing technologies, the advantages and positive effects of this invention are as follows: This invention's method for assessing the biological impact of marine micro-ecosystems involves acquiring environmental parameters, capturing images, obtaining morphological feature vectors through an image CNN model, calculating environmental parameter change functions, acquiring spectral data, obtaining spectral feature vectors through a spectral CNN model, calculating biological response functions, and calculating the Biometric Index (BI) of the environmental parameters' impact on organisms based on the environmental parameter change function and the biological response function. This allows for the assessment of the degree of impact on organisms based on the BI index. This invention can dynamically capture instantaneous changes in the environment and organisms, simultaneously extracting biological morphology, behavior, and biochemical parameters, and quantifying the degree of impact of environmental factors on organisms through an established mathematical model.
[0028] This invention relates to a marine micro-ecosystem biodiversity impact assessment system, comprising a housing, a cultivation tank, cameras, a light source, sensors, and a data processing module. The housing includes a transparent partition, and the light source is arranged around the housing. The cultivation tank is located inside the housing. Several cameras are used to acquire images and spectra within the cultivation tank. Sensors are used to collect environmental parameters within the cultivation tank. The data processing module is used to assess the marine biodiversity impact of the marine micro-ecosystem within the cultivation tank according to the aforementioned assessment methods. This invention can dynamically capture instantaneous changes in the environment and organisms, simultaneously extract morphological, behavioral, and biochemical parameters of organisms, and quantify the degree of impact of environmental factors on organisms through an established mathematical model. It can achieve high-precision coupled control of environmental parameters in the target sea area, while also considering equipment miniaturization and modularity.
[0029] Other features and advantages of the present invention will become clearer after reading the detailed embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description
[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0031] Figure 1 This is a schematic diagram of the detection system according to a specific embodiment of the present invention.
[0032] Figure 2 This is a cross-sectional view of the detection system according to a specific embodiment of the present invention.
[0033] In the diagram, 1. Box; 2. Light source; 3. Transparent partition; 4. Cultivation tank; 5. Water control valve; 6. Inlet and outlet water pipes; 7. Observation window; 8. Camera; 9. Sensor; 10. External box. Detailed Implementation
[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0035] In the description of this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0036] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. In the description of embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.
[0037] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.
[0038] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.
[0039] A method for assessing the biological impact of marine micro-ecosystems, with the "Bioimpact Index (BI)" as its core, automates the entire process from "environmental data → biological feature extraction → impact quantification," and specifically includes the following three parts:
[0040] (1) Data preprocessing:
[0041] Environmental data preprocessing: Obtaining environmental parameters x of marine micro-ecosystems i Calculate environmental parameter x i The difference Δx between the baseline environmental parameters and the baseline i .
[0042] Among them, environmental parameter x i This includes data collected by sensors such as temperature (T), salinity (S), light intensity (L(λ)), pH, dissolved oxygen (DO), and nutrients (N).
[0043] Preprocessing includes filtering to remove outliers.
[0044] Calculate environmental parameter x i The difference Δx between the baseline environmental parameters and the baseline i =x i (t)-x i (0); x i (t) represents the environmental parameter value at time t.
[0045] Image / spectral preprocessing:
[0046] Data preprocessing is performed before feature extraction to ensure input quality.
[0047] To avoid interference from non-biological factors (such as water scattering and image noise) in feature extraction, the input data to the CNN needs to be preprocessed first:
[0048] 1) Microscopic image preprocessing (2D-CNN input preparation):
[0049] For images captured by a 20x objective lens of a microscope imaging camera (224×224 pixels) and images acquired by a high-speed dynamic camera, the following three operations are performed: ① Geometric enhancement: The images are randomly rotated from 0 to 180° and scaled from 0.8 to 1.2 times to avoid single field-of-view bias; ② Brightness adjustment: The brightness fluctuation of ±10% simulates changes in laboratory lighting to improve the robustness of the model; ③ Pixel normalization: The pixel values are mapped to the [0,1] range to eliminate numerical differences caused by different imaging brightness.
[0050] Microscopic imaging camera (20x objective lens, 224×224 pixels): captures biological morphology; high-speed dynamic camera: records biological behavior.
[0051] 2) Multispectral data preprocessing (1D-CNN input preparation):
[0052] For reflectance / absorbance data acquired by a multispectral camera (400-750nm band, 1nm resolution), two key corrections were performed: ① Water scattering correction: Mie scattering model was used to eliminate the interference of suspended particles on the spectral signal (to avoid mistaking particle scattering as biological features); ② Fluorescence removal: a Gaussian difference filter (focusing 685nm±10nm bandpass) was used to separate the chlorophyll fluorescence signal from the spectral features of the organism itself, and then the spectral values were compressed to [0,1] by min-max normalization and reshaped into a 1D vector of (61,1) (61 is the number of wavelength points in the 400-750nm range at 1nm intervals).
[0053] Multispectral camera (400-750nm, 1nm resolution): Acquires biochemical parameters.
[0054] (2) CNN feature extraction:
[0055] Biological morphological and biochemical features were extracted using a dual-CNN architecture, as detailed below:
[0056] 1) 2D-CNN: Extracting image feature vector V I (256 dimensions):
[0057] Images of marine micro-ecosystems are captured and input into a CNN model to obtain morphological feature vectors V. I [Community density ρ (number of organisms / mm²), plankton feeding rate F (times / h), morphological distortion rate D (%)];
[0058] Model architecture: The ResNet50 architecture is adopted, combined with transfer learning optimization—loading ImageNet pre-trained weights, freezing the weights of the first four residual blocks (a total of 143 convolutional layers) (preserving general image feature extraction capabilities), and training only the last residual block and the fully connected layer; the fully connected layer is designed with a "256-dimensional output", corresponding to the image feature vector V.I .
[0059] Extraction target: V I The vector encodes three types of key biological response parameters (which can be obtained through vector decoding):
[0060] ①Algal density ρ (cells / mm²): The outline of algal cells is captured by convolutional layers, and the number of cells per unit area is counted by global average pooling (GAP).
[0061] ②Plankton feeding rate F (times / h): The frequency of contact between plankton movement trajectories and food particles is analyzed by using high-speed dynamic camera image sequences and encoded as temporal features in a vector;
[0062] ③ Biomorphological distortion rate D (%): The degree of deviation between the cell outline and the baseline state (such as the circular outline of normal algae). The greater the deviation, the higher the distortion rate coding value.
[0063] Output format: 256-dimensional floating-point vector V I Each bit corresponds to an abstract image feature (such as "the 32nd bit corresponds to the smoothness of the cell outline" and "the 128th bit corresponds to the frequency of the motion trajectory").
[0064] 2) 1D-CNN: Extracting spectral feature vector V λ (128 dimensions):
[0065] The spectra of marine micro-ecosystems are acquired using a multispectral camera, and then input into a spectral CNN model to obtain the spectral feature vector V. λ [Chlorophyll (Chl, μg / L), Carotenoids (Car, μg / L)].
[0066] Model architecture: The ResNet1D-18 architecture is adopted, with 1 input channel, a kernel size of 3 (stride 1, padding=1), and 4 residual blocks; the output layer is designed as a 128-dimensional floating-point vector V. λ It uses a three-layer coding feature.
[0067] Extraction target: V λ Vectors are divided into three parts according to their function, directly related to biological and biochemical states:
[0068] ① The first 32 dimensions: key band reflectance intensity (such as the 685nm chlorophyll a absorption peak and the 490nm carotenoid characteristic peak), reflecting the basic pigment content;
[0069] ② 64-dimensional: High-dimensional abstract features extracted by CNN convolutional layers (such as pigment ratio, cell structure changes - such as cell wall thickening leading to increased reflectivity in specific bands);
[0070] ③ The last 32 dimensions: Quantitative encoding of biochemical parameters (mapped to specific values such as chlorophyll a concentration Chl (μg / L) and carotenoid content Car (μg / L) through fully connected layers).
[0071] Output format: 128-dimensional floating-point vector V λ The measured values of biochemical parameters can be obtained directly through "post-32-dimensional decoding", or the full vector can be used to participate in the calculation of biological response similarity.
[0072] Calculate biological response function Among them, V I (t) represents the image feature vector extracted by the CNN model at time t, V I (0) is the image feature vector extracted by the CNN model of the baseline state image, V λ (t) represents the spectral feature vector extracted by the spectral CNN model at time t, V λ (0) is the spectral feature vector extracted by the baseline state spectral CNN model; a and b are weights, a+b=1.
[0073] In some embodiments, a=b=0.5.
[0074] Cosine similarity (value range [-1, 1], the closer the cosine similarity is to 1, the more consistent the biological state is with the initial state, the closer the Bio is to 0, and the more significant the influence is.
[0075] If the organisms are unaffected (e.g., algal density and chlorophyll content remain consistent with the baseline), then cos(V I (t),V I (0))≈1、cos(V λ (t),V λ (0))≈1, at this time ≈0; if organisms are subjected to severe stress (such as mass algal mortality), the similarity decreases. Approaching 1. Output range: ∈[0,1], the closer the value is to 1, the more significant the influence of the environment on the organism.
[0076] The training process for the CNN model is as follows:
[0077] 1. Dataset construction: ≥500 samples (300 for training, 100 for validation, 100 for testing), covering different biological densities and morphologies;
[0078] 2. Loss function:
[0079] The image CNN model and the spectral CNN model are trained using the loss function LOSS1.
[0080] Multi-task weighted LOSS1 = α × MSE(ρ pred ,ρ true )+β×MAE(V λpred V λrue )+γ×(1–cos(V I V Iaug )).
[0081] The loss function LOSS1 measures the model's prediction error and guides parameter optimization, and is crucial for ensuring the accuracy of feature extraction. The smaller the LOSS1 value, the smaller the deviation between the model's predicted values and the true values, and the stronger the feature stability.
[0082] In some embodiments, CNN training is completed when the validation set Loss1≤0.08 and the corresponding R²≥0.96 and MAE≤0.5μg / L.
[0083] Wherein: MSE is the mean squared error, ensuring "high accuracy" in biological density prediction. It is the core term in the loss function for "community density prediction," and its function is to quantify the deviation between the CNN's predicted values of biological community density and the actual values, forcing the model to reduce serious misjudgments. MAE is the mean absolute error, ensuring "stability" in biochemical parameter prediction. It is the core term in the loss function for "spectral biochemical parameters" (chlorophyll a, carotenoids), and its function is to quantify the CNN's prediction bias of biochemical indicators, while avoiding interference from outliers (such as measurement errors) on model training.
[0084] ρ pred ρ is the predicted value of community density. true This represents the measured value of the community density.
[0085] The measured community density was obtained by manually counting three times under a microscope and averaging the results. The unit is individuals / mm².
[0086] V λpred V represents the predicted values for biochemical characteristics (chlorophyll a, b, Car). λtrue These are the measured values of biochemical characteristics.
[0087] The measured values of biochemical characteristics were determined by high performance liquid chromatography (HPLC).
[0088] V I V is the feature vector of the original image. Iaug This is the feature vector of the enhanced image.
[0089] Image enhancement methods: image rotation 0-180°, scaling 0.8-1.2 times, brightness adjustment ±10%, and addition of Gaussian noise with a standard deviation of 0.005 to the spectral data.
[0090] This is for calculating cosine similarity.
[0091] α, β, and γ are task weights, initially set to 1.0, 0.8, and 0.5, respectively. During training, they are dynamically adjusted based on the validation set error. For example, when the morphological feature similarity does not meet the standard, γ is increased to 0.8.
[0092] To verify the accuracy of CNN feature extraction, the verification metrics must meet the following: ① Coefficient of determination R² ≥ 0.96 (goodness of fit between predicted and true values); ② Chlorophyll concentration MAE (mean absolute error) ≤ 0.5 μg / L; ③ Carotenoid content MAE ≤ 0.2 μg / L; ④ Feature vector cosine similarity ≥ 0.95.
[0093] Calculate the function of environmental parameter variation .
[0094] Where, ω i As the weights of environmental parameters, n is the number of environmental parameters, such as temperature ω. T =0.4, salinity ω S =0.3, Nutrients ω N =0.2, pHω pH =0.1.
[0095] This refers to the range of values for environmental parameters; for example, the temperature range is 52℃.
[0096] (3) Mathematical model of the biological impact quantification index (BI):
[0097] Calculate the BI index, a quantitative index of the impact of environmental parameters on organisms. ; where ω E For the environmental parameter weights, ω B For biological response weights, ω E +ω B =1, ε is the error term.
[0098] Weight optimization: The process of "freezing CNN → training weights → end-to-end fine-tuning" is adopted.
[0099] ① Multi-factor orthogonal experiment (L 16 (4 4 (Orthogonal array + 24 extreme environments, ≥100 experiments, cycle ≥90 days)
[0100] ②In ω E ≥0, ω B ≥0 and ω E +ω B =1, and ω i ≥0 and Σω i Under the constraint that = 1, the loss function LOSS2 = MSE(BI) is used.pred BI true )+λ×(1-r(BI pred BI true )) for ω i ω B and ω E Optimize, where r is the BI pred with BI true Pearson correlation coefficient, λ=0.5, BI pred The predicted value of the biological impact index (BI) calculated by the model. true To obtain the true values of the measured biological indicators, the weights are iteratively updated using the training set during the optimization process, and the convergence is evaluated using the validation set. The convergence criterion is: the Pearson correlation coefficient r of the validation set is ≥0.85 and the MSE is ≤0.02.
[0101] The experimental procedure is as follows:
[0102] Test organisms: Common planktonic organisms (such as copepods) and algae (such as Chlorella) from the surrounding waters of the islands were selected and domesticated to a stable state.
[0103] Baseline parameter settings: temperature 25±0.5℃, salinity 35±0.2PSU, DO 8±0.3mg / L, nutrient N 0.5±0.05mg / L, illumination L(λ) simulates the noon spectrum of the target sea area (PAR 1000μmol / m² / s).
[0104] Equipment calibration: CCD camera focal length calibration (to ensure clear imaging with 20x objective lens), sensor accuracy calibration (temperature error ≤0.1℃), CNN model pre-training (trained with 500 baseline samples to R²≥0.96).
[0105] 4.2 Environmental Factor Disturbance Experiment
[0106] Using L 16 (4 4 Sixteen basic experiments were designed using an orthogonal array (temperature 20 / 25 / 30 / 35℃, salinity 30 / 33 / 35 / 38 PSU, nutrient N 0.1 / 0.5 / 1.0 / 1.5 mg / L, pH 7.2 / 7.5 / 8.0 / 8.5), with six parallel samples in each group. Additionally, 24 extreme environments were added (e.g., 35℃ + 38 PSU, 0.1 mg / L N + pH 7.2), for a total of 100 experimental groups over a period of 90 days.
[0107] 4.3 Data Acquisition and Analysis
[0108] Data collection: ① Environmental factors: Daily average values were collected every 24 hours; ② Images / spectroscopy: Data were collected every 7 days (3 fields of view / sample for microscopic images, 3 replicates / sample for multispectral data); ③ Measured biological indicators: Mortality was counted daily, and reproductive capacity was calculated every 14 days as the basis for BItrue calibration.
[0109] Model training: ① Freeze the CNN and train ω E ω B ω i (Learning rate 1e-4, batch size=32, 80 rounds); ② Unfreeze the last 2 residual blocks of ResNet and fine-tune end-to-end (convolutional layer 1e-5, fully connected layer 5e-5, 100 rounds).
[0110] Verification results:
[0111] The weight optimization phase is performed according to the following convergence criteria: if the Pearson correlation coefficient r on the validation set is ≥0.85 and the MSE is ≤0.02, the initial weight optimization is completed; after subsequent end-to-end fine-tuning, the overall model must meet the final validation criteria: the MAE of the test sets BIpred and BItrue is ≤0.015, the Pearson correlation coefficient is ≥0.9, and the MAE change rate under noise interference is ≤10%, which can be considered as the model training being completed.
[0112] like Figure 1-2 As shown in the figure, this embodiment also proposes a marine micro-ecosystem biological impact assessment system, which includes: a container 1, a cultivation tank 4, several cameras 8, sensors 9 and a data processing module.
[0113] The enclosure 1 is the main frame of the equipment, providing a sealed experimental environment, isolating external interference, and supporting all internal components.
[0114] The cultivation tank 4 is located inside the enclosure 1. The cultivation tank 4 is used to contain the experimental water and test organisms (such as plankton and algae), serving as the core area for ecological simulation. Sensor placement points and inlet / outlet pipe interfaces are reserved on the bottom or side of the cultivation tank 4.
[0115] The enclosure 1 includes a transparent partition 3, and a light source 2 is arranged around the enclosure 1. The transparent partition 3 is fixed horizontally or vertically inside the enclosure 1, located between the light source 2 and the cultivation tank 4. The transparent partition 3 separates the illumination area from the observation area, avoiding direct light interference with imaging, while ensuring uniform light transmission.
[0116] Light source 2 is installed on the top or side of enclosure 1 and connected to a computer terminal via a control line to achieve remote adjustment of spectral parameters. Light source 2 is used to provide tunable spectrum (PAR / UVR) to simulate the illumination conditions of the target sea area, and the spectral intensity distribution function is L(λ).
[0117] A number of cameras 8 are used to obtain images and spectra inside the cultivation tank 4. Among them, the cameras 8 include a microscopic imaging camera, a high-speed dynamic camera, and a multispectral camera. The cameras 8 can be placed on the side and / or top surface and / or bottom surface of the box body 1. When the cameras 8 are located on the top surface and / or bottom surface of the box body 1, they are opposite to the cultivation tank 4.
[0118] The sensor 9 is used to collect the environmental parameters inside the cultivation tank 4; the sensor 9 is inserted into the cultivation tank 4, and the data is transmitted to the computer terminal through the sensor network.
[0119] The data processing module, that is, the computer terminal, is used to conduct an assessment of the impact on marine organisms of the marine microecosystem inside the cultivation tank according to the above assessment method.
[0120] The system includes an inlet and outlet water pipe 6 to achieve water body circulation and nutrient salt supplementation. One end of the inlet and outlet water pipe 6 is connected to an external water body supply system, and the other end is connected to the observation cultivation tank 4 and is linked with the water control valve 5.
[0121] The system includes a water control valve 5, which is used to control the water flow rate and replacement frequency, simulate the sea area water flow environment, accurately adjust the water volume inside the cultivation tank 4, and control the opening and closing and opening degree of the water control valve 5 through the computer terminal.
[0122] The system includes an external box body 10, and the external box body 10 is wrapped for heat preservation and light shielding. The external box body 10 is provided with an observation window 7 to provide an observation channel for the cameras 8. The observation window 7 includes high-transmission glass to reduce the interference of light reflection.
[0123] The camera 8 is fixed outside the observation window 7 and is connected to the computer terminal through a data cable to achieve second-level sampling and data transmission.
[0124] In this embodiment, the bay-level ecosystem is compressed into an observation cultivation tank, and at the same time, the fidelity of the biogeochemical process is maintained (the environmental parameter accuracy is ±0.1°C / PSU), realizing controllable simulation in the laboratory;
[0125] Multi-camera - dual CNN collaborative technology: For the first time, through the coupling of microscopic, high-speed, and multispectral cameras, combined with 2D-CNN (morphology) and 1D-CNN (biochemistry), the contradiction of "seeing clearly / measuring accurately" is broken through, and the extraction accuracy of biological parameters R²≥0.96;
[0126] BI index quantification model: Construct a mathematical closed-loop of "environment - organism - impact" to realize the quantification of the ecological stress threshold (BI≤0.2 has no impact, 0.2<BI≤0.6 observable change, BI>0.6 significant stress);
[0127] Internet of Things data interconnection: Directly connect to the marine Internet of Things through the OPC-UA protocol to realize the linkage of laboratory data and field data, and support the iterative optimization of ecological models.
[0128] This embodiment achieves real-time performance, high efficiency, and sensitivity.
[0129] Real-time performance: Second-level sampling frequency + real-time sensor monitoring, dynamically capturing instantaneous changes in the environment and organisms;
[0130] High efficiency: No sample pretreatment required, modular operation (simple operation and maintenance, low loss rate), ≥90 days of continuous experimentation without manual intervention;
[0131] Sensitivity: Centimeter-level observation accuracy + high resolution (0-1) BI index, capable of identifying the impact of minute environmental changes (such as 0.1℃ temperature fluctuations) on organisms;
[0132] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions claimed by the present invention.
Claims
1. A method for assessing the biological impact of marine micro-ecosystems, characterized in that, Obtain environmental parameters x of marine micro-ecosystems i Calculate environmental parameter x i The difference Δx between the baseline environmental parameters and the baseline i ; Calculate the function of environmental parameter variation , where ω i As the weights of environmental parameters, , where n is the number of environmental parameters. This refers to the range of values for environmental parameters; Images of marine micro-ecosystems are captured and input into a CNN model to obtain morphological feature vectors V. I V I The vector encodes three key biological response parameters: community density ρ (organisms / mm²), plankton feeding rate F (times / h), and morphological distortion rate D (%). The spectra of marine micro-ecosystems are acquired using a multispectral camera, and then input into a spectral CNN model to obtain the spectral feature vector V. λ V λ The 32-dimensional quantitative biochemical parameter encodings are mapped to chlorophyll (Chl, μg / L) and carotenoids (Car, μg / L) through a fully connected layer. Calculate biological response function Among them, V I (t) represents the image feature vector extracted by the CNN model at time t, V I (0) is the image feature vector extracted by the CNN model of the baseline state image, V λ (t) represents the spectral feature vector extracted by the spectral CNN model at time t, V λ (0) is the spectral feature vector extracted by the baseline state spectral CNN model; a and b are the weights, a+b=1; Calculate the BI index, a quantitative index of the impact of environmental parameters on organisms. ; where ω E For the environmental parameter weights, ω B For biological response weights, ω E +ω B =1, ε is the error term.
2. The method for assessing the biological impact of marine micro-ecosystems according to claim 1, characterized in that, The image CNN model and the spectral CNN model are trained using the loss function LOSS1, where Loss1 = α × MSE(ρ pred ,ρ true )+β×MAE(V λpred V λrue )+γ×(1–cos(V I V Iaug Where MSE is the mean squared error, MAE is the mean absolute error, and ρ is the mean squared error. pred ρ is the predicted value of community density. true V represents the measured community density. λpred V is the predicted value of biochemical characteristics. λrue V represents the measured value of biochemical characteristics. I V is the feature vector of the original image. Iaug To enhance the feature vector of the image, α, β, and γ are task weights, which are dynamically adjusted based on the validation set error during training.
3. The method for assessing the biological impact of marine micro-ecosystems according to claim 2, characterized in that, To verify the accuracy of CNN feature extraction, the verification metrics must meet the following: ① Coefficient of determination R² ≥ 0.96 (goodness of fit between predicted and true values); ② Chlorophyll concentration MAE ≤ 0.5 μg / L; ③ Carotenoid content MAE ≤ 0.2 μg / L; ④ Cosine similarity of feature vectors ≥ 0.
95.
4. The method for assessing the biological impact of marine micro-ecosystems according to claim 2, characterized in that, The image enhancement methods are: image rotation 0-180°, scaling 0.8-1.2 times, and brightness adjustment ±10%.
5. The method for assessing the biological impact of marine micro-ecosystems according to any one of claims 1-4, characterized in that, In ω E ≥0, ω B ≥0 and ω E +ω B =1, and ω i ≥0 and Σω i Under the constraint that = 1, the loss function Loss2 = MSE(BI) is used. pred BI true )+λ×(1-r(BI pred BI true )) for ω i ω B and ω E Optimize, where r is the BI pred with BI true Pearson correlation coefficient, λ=0.5, BI pred The predicted value of the biological impact index (BI) calculated by the model. true To obtain the true values of the measured biological indicators, the weights are iteratively updated using the training set during the optimization process, and the convergence is evaluated using the validation set. The convergence criterion is: the Pearson correlation coefficient r of the validation set is ≥0.85 and the MSE is ≤0.
02.
6. A marine micro-ecosystem bioimpact assessment system, characterized in that, The system includes: The enclosure includes a transparent partition, and light sources are arranged around the enclosure. The cultivation tank is located inside the container; Several cameras were used to acquire images and spectra of the culture tank; Sensors are used to collect environmental parameters within the cultivation tank; A data processing module is used to conduct a marine biological impact assessment of a marine micro-ecosystem in a cultivation tank using the assessment method according to any one of claims 1-5.
7. The marine micro-ecosystem bioimpact assessment system according to claim 6, characterized in that, The cameras include microscopic imaging cameras, high-speed dynamic cameras, and multispectral cameras.
8. The marine micro-ecosystem bioimpact assessment system according to claim 6, characterized in that, The system includes inlet and outlet pipes to realize water circulation and nutrient replenishment.
9. The marine micro-ecosystem bioimpact assessment system according to claim 8, characterized in that, The system includes a water control valve for controlling the water flow rate and replacement frequency to simulate the marine water flow environment.
10. The marine micro-ecosystem bioimpact assessment system according to claim 6, characterized in that, The system includes an observation window that provides an observation channel for the camera, and the observation window includes high-transmittance glass.