Method and system for predicting concentration of acesulfame potassium serving as artificial sweetener in offshore culture seawater based on behavior characteristics of juvenile black sea bream

Abstract

The invention discloses a method and a system for predicting the concentration of an artificial sweetener acesulfame potassium in offshore culture seawater based on behavior characteristics of juvenile black sea bream. The method comprises the following steps: collecting motion video data of the juvenile black sea bream in different acesulfame potassium concentration environments and preprocessing the motion video data; training based on an improved YOLOv8 network model to obtain a juvenile fish behavior recognition model and a behavior track data set corresponding to the black sea bream juvenile fish movement video data set; and constructing an initial space-time association network model based on GNN and GRU networks, training the initial space-time association network model by using the behavior trajectory data set, and predicting the acesulfame concentration by analyzing the behavior trajectory of the black sea bream juvenile fish to obtain a completely trained space-time association network model. According to the method, the concentration of the acesulfame potassium serving as an artificial sweetener in the environment where the juvenile black sea bream is located can be predicted according to the behavior characteristics of the juvenile black sea bream, and powerful support can be provided for monitoring and early warning of the offshore culture seawater environment.

Description

Technical Field

[0001] The present invention relates to the technical field of offshore marine environment monitoring, and particularly to a method and system for predicting the concentration of acesulfame in seawater for offshore aquaculture based on the behavioral characteristics of juvenile black porgy. Background Art

[0002] Sweeteners are substances that impart sweetness to foods and belong to a category of food additives. Artificial sweeteners, due to their high sweetness, low dosage, low cost, and strong market competitiveness, occupy half of China's sweetener market. In recent years, artificial sweeteners present in the environment have been regarded as a new type of pollutant, and their potential biological toxicity, ecological risks, and health risks are a cause for concern. Among them, acesulfame is not only favored in the field of food additives due to its simple production process, low price, and excellent performance, but is also widely used in pharmaceuticals, personal care products, and feeds. However, acesulfame cannot be metabolized by the human body, and it is difficult to completely remove it during wastewater treatment. Moreover, it has relatively high stability and a long half-life, and is prone to accumulate in the environment. Currently, acesulfame has been commonly detected in the marine environment, with a concentration range of 1.9×10 -4 ~9.9 μg / L, posing non-negligible potential hazards to the marine ecosystem environment. Therefore, accurately predicting the concentration of acesulfame, an artificial sweetener, in offshore seawater is of great significance for seawater environmental monitoring and early warning.

[0003] As a widely used artificial synthetic sweetener, current detection techniques for acesulfame include high performance liquid chromatography, ion chromatography, liquid chromatography-mass spectrometry, etc. For acesulfame with extremely low concentrations, the detection limit of high performance liquid chromatography may still be relatively high, resulting in the inability to accurately detect low-concentration acesulfame in seawater. The results of high performance liquid chromatography are affected by the selection and stability of the chromatographic column. Different brands and models of chromatographic columns may have different separation effects on acesulfame, and the separation effect may decline due to a decrease in column efficiency or contamination of the chromatographic column. Ion chromatography is mainly applicable to ionizable compounds. Although acesulfame exists in the form of anions under acidic conditions, it may not be completely ionized under certain specific conditions, thus affecting the accuracy of the detection results. Liquid chromatography-mass spectrometry combines two techniques, high performance liquid chromatography and mass spectrometry. Its operation and maintenance are more difficult, and the equipment cost is higher.

[0004] In order to accurately establish an early warning monitoring system for acesulfame in the offshore marine ecosystem, the present invention proposes a method and system for predicting the concentration of acesulfame, an artificial sweetener, in seawater for offshore aquaculture based on the behavioral characteristics of juvenile black porgy. By means of the behavioral responses of aquatic organisms, the concentration of acesulfame in seawater is predicted, and the ecological risks of acesulfame, a new pollutant, are evaluated from the perspective of ecological behavior. Summary of the Invention

[0005] In view of this, the present invention provides a method for predicting the concentration of acesulfame potassium, an artificial sweetener, in seawater for offshore aquaculture based on the behavioral characteristics of juvenile black porgy, so as to solve the technical problems of insufficient detection sensitivity, complex operation, and high equipment cost for low-concentration acesulfame potassium in seawater for offshore aquaculture.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0007] On the one hand, the present invention provides a method for predicting the concentration of acesulfame potassium, an artificial sweetener, in seawater for offshore aquaculture based on the behavioral characteristics of juvenile black porgy, including:

[0008] Collect the motion video data of juvenile black porgy in environments with different acesulfame potassium concentrations and perform preprocessing to obtain a motion video dataset of juvenile black porgy;

[0009] Construct an initial juvenile fish behavior recognition model based on an improved YOLOv8 network model, and use the motion video dataset of juvenile black porgy to train the initial juvenile fish behavior recognition model to obtain a well-trained juvenile fish behavior recognition model, as well as a behavior trajectory dataset corresponding to the motion video dataset of juvenile black porgy;

[0010] Construct an initial spatio-temporal correlation network model based on the GNN and GRU networks, use the behavior trajectory dataset to train the initial spatio-temporal correlation network model, and predict the acesulfame potassium concentration by analyzing the behavior trajectories of juvenile black porgy to obtain a well-trained spatio-temporal correlation network model;

[0011] Obtain the actual motion video of juvenile black porgy in seawater for offshore aquaculture, and use the juvenile fish behavior recognition model and spatio-temporal correlation model to predict the acesulfame potassium concentration of the seawater for offshore aquaculture according to the actual motion video.

[0012] Furthermore, collecting the motion video of juvenile black porgy in environments with different acesulfame potassium concentrations and performing preprocessing includes:

[0013] Collect the motion video of juvenile black porgy in environments with different acesulfame potassium concentrations based on a preset shooting perspective, resolution, and frame rate to obtain original video data;

[0014] Divide the original video data into segments with equal durations and label each video with the acesulfame potassium concentration to obtain a motion video dataset of juvenile black porgy;

[0015] Adopt data augmentation technology to expand the video dataset to obtain an expanded dataset;

[0016] Mark the juvenile black porgy in each video in the expanded dataset and perform format conversion on the marked videos to obtain a preprocessed motion video dataset;

[0017] Divide the preprocessed sports video dataset into a training set, a validation set, and a test set.

[0018] Furthermore, the improved YOLOv8 network model includes:

[0019] Replace the C2f module in the YOLOv8 backbone network with a C2f-MSADCN module formed by stacking the C2f module and the MSADCN module;

[0020] Integrate the MSSimAM attention module into the C2f module in the YOLOv8 neck network;

[0021] Replace the CIoU loss function in the YOLOv8 head network with the SEIoU loss function.

[0022] Furthermore, the MSADCN module introduces multi-scale and attention mechanisms, and the convolution operation of the module is expressed by the formula:

[0023]

[0024] where y(p) is the feature value at position p in the output feature map, S represents the number of multi-scales, α s is the learnable weight on scale s, used to adjust the feature contributions of different scales, K is the number of sampling points of the convolution kernel, is the convolution weight at the k-th sampling position under the s-th scale, is the offset reference at the k-th sampling position, is the learnable offset, is the modulation coefficient at the k-th sampling point, used to adjust the feature importance at this position.

[0025] Furthermore, the MSSimAM attention module includes a multi-scale convolutional layer, a similarity adjustment layer, an attention layer, and a feature fusion layer, which are used to capture local and global information of the image using the receptive fields of multiple scales;

[0026] The multi-scale convolutional layer is used to extract features of different scales using convolution kernels of different sizes;

[0027] The similarity measurement layer is used to measure the similarity between features of different scales;

[0028] The attention layer is used to generate an attention map according to the similarity measurement, weight the multi-scale features, and highlight the key features;

[0029] The feature fusion layer is used to fuse the weighted multi-scale features into the final output of the module.

[0030] Furthermore, the SEIoU loss function is expressed by the formula:

[0031]

[0032] Among them, 1 - IoU represents the intersection over union loss, which is used to measure the overlap difference between the predicted bounding box and the ground truth bounding box, and ρ 2 (b, b gt ) represents the square of the Euclidean distance between the center points of the predicted bounding box and the ground truth bounding box, w and w gt are the widths of the predicted bounding box and the ground truth bounding box respectively, and h and h gt represent the heights of the predicted bounding box and the ground truth bounding box respectively, w c and h c are the width and height of the smallest bounding rectangle that can enclose the predicted bounding box and the ground truth bounding box respectively.

[0033] Furthermore, the initial larval behavior recognition model is trained using the Sparus macrocephalus larval movement video dataset to obtain a fully trained larval behavior recognition model, including:

[0034] The initial larval behavior recognition model is trained using the training set and validation set of the Sparus macrocephalus larval movement video dataset to obtain a trained larval behavior recognition model;

[0035] The trained larval behavior recognition model is tested using the test set of the Sparus macrocephalus larval movement video dataset;

[0036] When the prediction result of the model reaches the training expectation, a fully trained larval behavior recognition model is obtained, and the coordinate time series of all Sparus macrocephalus larvae in the training set, validation set, and test set videos are extracted respectively.

[0037] Furthermore, an initial spatio - temporal correlation network model is constructed based on GNN - GRU, including:

[0038] The initial spatio - temporal correlation model includes a GNN module, a multi - layer GRU module, and a fully connected module;

[0039] The GNN module is used to obtain spatial features based on the input of the position features of Sparus macrocephalus larvae at different times, for capturing the spatial correlation between Sparus macrocephalus larvae;

[0040] The multi - layer GRU module contains multiple cascaded GRU network models, and calculates the hidden states of each GRU network model according to the spatial features at each time; it is used to capture the temporal correlation of Sparus macrocephalus larval behavior;

[0041] The fully connected network model predicts the acesulfame potassium concentration according to the hidden states output by the multi - layer GRU module.

[0042] Further, the initial spatio-temporal correlation network model is trained using the behavioral trajectory dataset, and the acesulfame concentration is predicted by analyzing the behavioral trajectories of juvenile black porgy, resulting in a fully trained spatio-temporal correlation network model, including:

[0043] The initial spatio-temporal correlation network model is supervised-trained using the coordinate time series of juvenile black porgy in the training set and the validation set and their corresponding acesulfame concentrations as labels;

[0044] The trained spatio-temporal correlation network model is tested using the coordinate time series of juvenile black porgy in the test set and their corresponding acesulfame concentrations as labels. When the accuracy of the prediction results reaches the expectation, a fully trained spatio-temporal correlation network model is obtained.

[0045] On the other hand, the present invention also provides a prediction system for the acesulfame concentration in seawater for offshore aquaculture based on the behavioral characteristics of juvenile black porgy, including:

[0046] A data acquisition module for collecting the motion video data of juvenile black porgy in environments with different acesulfame concentrations and preprocessing it to obtain a motion video dataset of juvenile black porgy;

[0047] A behavioral recognition model construction module for constructing an initial juvenile fish behavioral recognition model based on an improved YOLOv8 network model, training the initial juvenile fish behavioral recognition model using the motion video dataset of juvenile black porgy to obtain a fully trained juvenile fish behavioral recognition model, and the corresponding behavioral trajectory dataset of the motion video dataset of juvenile black porgy;

[0048] A spatio-temporal correlation network model construction module for constructing an initial spatio-temporal correlation network model based on GNN-GRU, training the initial spatio-temporal correlation network model using the behavioral trajectory dataset, and predicting the acesulfame concentration by analyzing the behavioral trajectories of juvenile black porgy to obtain a fully trained spatio-temporal correlation network model;

[0049] A concentration prediction module for obtaining the actual motion video of juvenile black porgy in seawater for offshore aquaculture and predicting the acesulfame concentration of the seawater for offshore aquaculture using the juvenile fish behavioral recognition model and the spatio-temporal correlation model according to the actual motion video.

[0050] Compared with the prior art, the present invention proposes a method for predicting the concentration of acesulfame potassium, an artificial sweetener, based on the behavioral characteristics of juvenile black porgy. This method constructs a behavior recognition model and a spatio-temporal joint model based on the improved YOLOv8 and GNN-GRU respectively, improving the classification accuracy of different concentrations of acesulfame potassium according to the behavioral characteristics of juvenile black porgy, thereby providing a scientific basis for evaluating the environmental pollution of acesulfame potassium to the marine ecosystem. The present invention also has broad application potential, such as in the fields of fish ecological behavior research, environmental monitoring, underwater robot navigation, etc. The present invention has important application and theoretical value in promoting technological progress in the fields of marine environmental protection and ecological monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic flowchart of the method for predicting the concentration of acesulfame potassium, an artificial sweetener, in seawater for offshore aquaculture based on the behavioral characteristics of juvenile black porgy provided by the present invention;

[0052] Figure 2 It is a schematic structural diagram of constructing an initial juvenile fish behavior recognition model based on the improved YOLOv8 network model provided by the present invention;

[0053] Figure 3 It is a schematic structural diagram of the MSADCN module provided by the present invention;

[0054] Figure 4 It is a schematic structural diagram of the MSSimAM attention module provided by the present invention;

[0055] Figure 5 It is a schematic structural diagram of the initial spatio-temporal correlation network model provided by the present invention;

[0056] Figure 6 It is a schematic diagram of the mAP results of the combined training of the improved YOLOv8 model and different modules provided by the present invention;

[0057] Figure 7 It is a schematic diagram of the training time consumption results of the combined training of the improved YOLOv8 model and different modules provided by the present invention;

[0058] Figure 8 It is a schematic diagram for comparing the prediction accuracy between the spatio-temporal correlation model and the random forest algorithm provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The following specifically describes the preferred embodiments of the present invention in conjunction with the drawings, wherein the drawings form a part of this application and are used together with the embodiments of the present invention to explain the principle of the present invention, and are not used to limit the scope of the present invention.

[0060] Before introducing the embodiments of the present invention, the inventive concept of this application is first explained.

[0061] Behavioral changes are the most direct and sensitive responses of organisms to external environmental stimuli and internal environmental changes. The behavioral responses of aquatic organisms have been used as a method for environmental monitoring, but they still receive much less attention than other fields such as animal toxicology. Among the organisms used for behavioral assessment, fish are particularly noteworthy. Due to their wide distribution in aquatic ecosystems and their key ecological position in the food web, and the fact that fish have a longer iteration cycle compared to other aquatic organisms, fish play a crucial role in ecological toxicology monitoring.

[0062] Sparus macrocephalus is a typical offshore fish, widely distributed in the coastal waters of China and the offshore waters of other Asian countries. In these regions, the important position of Sparus macrocephalus in the food chain makes it a key species in the aquatic ecosystem. As a benthic fish, juvenile Sparus macrocephalus is highly sensitive to physicochemical changes in the seawater environment. Especially when factors such as dissolved oxygen, temperature, and pollutants in the seawater change, the behavior of Sparus macrocephalus will change significantly. By monitoring the behavioral changes of juvenile Sparus macrocephalus, not only can the impact of acesulfame concentration on this species be reflected, but also the impact of pollutants on the entire ecosystem can be indirectly reflected, thus providing data support for ecological risk assessment.

[0063] Artificial sweeteners can have toxic effects on the biological behavior of fish. Existing studies have shown that artificial sweeteners can cause abnormal fish behavior, such as social, anxiety, and memory and cognitive behaviors. For example, adult zebrafish exposed to aspartame and acesulfame show increased anxiety behavior and reduced activity; high-dose aspartame increases the phototaxis of zebrafish and their preference for blue, reducing their exploration ability and social behavior. Zebrafish fed with aspartame and a high-fat diet also show a decrease in acute swimming ability. Similarly, zebrafish exposed to saccharin also show anxiety-like behavior, reduced exploration ability and sociality, and their color-based associative learning and memory abilities are also disturbed. Generally speaking, fish behavior has advantages such as high sensitivity and easy observability, and it plays an indispensable role in the ecological risk assessment of pollutants and is an effective entry point for establishing an ecological early warning monitoring system.

[0064] Therefore, evaluating the ecological risk of the new pollutant acesulfame from the perspective of ecological behavior helps to establish an offshore marine ecological early warning monitoring system for acesulfame and provides scientific support for its ecological risk and pollution control.

[0065] Based on the potential connection between fish ethology and artificial sweeteners, the present invention identifies the trajectories of juvenile Sparus macrocephalus in the offshore aquaculture seawater environment polluted by the artificial sweetener acesulfame through deep learning technology, classifies the concentration of the artificial sweetener acesulfame in the environment where the juvenile Sparus macrocephalus is located according to the behavioral characteristics of the juvenile Sparus macrocephalus, and predicts the concentration of the artificial sweetener acesulfame in the seawater to achieve the monitoring of the offshore aquaculture seawater environment.

[0066] Please refer toFigure 1 , this embodiment provides a method for predicting the acesulfame potassium concentration in seawater for offshore aquaculture based on the behavioral characteristics of juvenile black porgy, including:

[0067] Step S101: Collect the motion video data of juvenile black porgy in environments with different acesulfame potassium concentrations and perform preprocessing to obtain a motion video dataset of juvenile black porgy;

[0068] Step S102: Construct an initial juvenile fish behavior recognition model based on the improved YOLOv8 network model, and use the motion video dataset of juvenile black porgy to train the initial juvenile fish behavior recognition model to obtain a well-trained juvenile fish behavior recognition model and the corresponding behavior trajectory dataset of the motion video dataset of juvenile black porgy;

[0069] Step S103: Construct an initial spatio-temporal association network model based on the GNN and GRU networks, use the behavior trajectory dataset to train the initial spatio-temporal association network model, and predict the acesulfame potassium concentration by analyzing the behavior trajectory of juvenile black porgy to obtain a well-trained spatio-temporal association network model;

[0070] Step S104: Obtain the actual motion video of juvenile black porgy in seawater for offshore aquaculture, and use the juvenile fish behavior recognition model and spatio-temporal association model to predict the acesulfame potassium concentration of the seawater for offshore aquaculture according to the actual motion video.

[0071] The method of this embodiment uses the improved YOLOv8 network model to identify the behavior of juvenile black porgy. Through the trained behavior recognition model, the motion behavior data of juvenile black porgy can be accurately extracted. Combining the GNN (Graph Neural Network) and GRU (Gated Recurrent Unit) networks for spatio-temporal association analysis of the behavior trajectory can more accurately predict the acesulfame potassium concentration in seawater. By continuously collecting the behavior data of juvenile fish and performing real-time analysis, the real-time monitoring and prediction of the acesulfame potassium concentration in the offshore aquaculture environment can be realized. This method avoids the cumbersome process of traditional manual sampling and laboratory analysis, improves the prediction efficiency and timeliness, and can predict the acesulfame potassium concentration in the offshore aquaculture environment in real time, efficiently and accurately. By predicting the acesulfame potassium concentration in the offshore aquaculture waters, water quality changes can be detected in advance and measures can be taken in time for adjustment. This helps to improve the efficiency of aquaculture water area management, ensure the health of the aquaculture environment, and thus improve the growth efficiency of black porgy and the aquaculture economic benefits. In addition, this method can reduce the need for manual operations, reduce labor costs, and reduce the impact of human factors on the prediction results through automated video monitoring and data analysis. For large-scale farms, it can effectively reduce the workload of management personnel and improve the aquaculture management efficiency.

[0072] As a preferred embodiment, in step S101, collecting the motion video of juvenile black porgy in environments with different acesulfame potassium concentrations and performing preprocessing includes:

[0073] Collect the movement videos of juvenile black porgy in environments with different acesulfame potassium concentrations based on preset shooting perspectives, resolutions, and frame rates to obtain the original video data;

[0074] Divide the original video data into segments of equal duration and label each video with the acesulfame potassium concentration to obtain the movement video dataset of juvenile black porgy;

[0075] Use data augmentation techniques to expand the video dataset to obtain an augmented dataset;

[0076] Label the juvenile black porgy in each video in the augmented dataset and convert the format of the labeled videos to obtain the preprocessed movement video dataset;

[0077] Divide the preprocessed movement video dataset into a training set, a validation set, and a test set.

[0078] As a specific embodiment, to ensure the quality and diversity of the video data for better analysis of the movement behavior of juvenile black porgy, first construct four environments with different acesulfame potassium concentrations (0 μg / L, 10 μg / L, 100 μg / L, and 1000 μg / L), and place 10 juvenile black porgy in each environment. Use a Hikvision camera to shoot from a top-down perspective, with a video resolution of 720p and a frame rate of 30 frames per second. Divide all videos into segments of 15 seconds in duration, label each video with the acesulfame potassium concentration, and form the video dataset of the movement of juvenile black porgy.

[0079] Use data augmentation techniques to expand the video dataset. Specifically: achieve data augmentation of the video dataset by rotating the video 90 degrees, 180 degrees, and 270 degrees, as well as performing horizontal and vertical mirror inversions, cropping, mirroring, brightness adjustment, etc. on the video. This can enable the training model to generalize better, avoid overfitting, and at the same time increase the number of training samples and improve the robustness of the model.

[0080] For the augmented video dataset, use LabelMe to label the juvenile black porgy in the video. The labeled data can help researchers analyze the movement patterns of fish and identify factors affecting movement behavior (such as acesulfame potassium concentration). Convert it to the YOLOv8 format and divide the dataset into a training set, a validation set, and a test set according to different acesulfame potassium concentrations.

[0081] YOLOv8 is a new version of the YOLO (You Only Look Once) series of target detection models launched after YOLOv5. It inherits the core idea of ​​the YOLO series, that is, based on a convolutional neural network (CNN), it directly regresses the image to predict the category and position of the target. YOLOv8 performs well in target detection, and is particularly good at processing the simultaneous detection of multiple targets in an image. The present invention uses the YOLOv8 model as the basic architecture of the model, and improves and optimizes the model based on the dynamic moving video characteristics of black sea bream fry in water, so as to meet the actual needs of identifying the behavior trajectory of black sea bream fry through motion video analysis of black sea bream fry.

[0082] When identifying black sea bream fry in offshore aquaculture seawater, YOLOv8 has the following defects:

[0083] 1. Since black sea bream usually appear in complex offshore waters, factors such as reflection of offshore water bodies, changes in illumination, and transparency of fish schools will result in low contrast between the school of fish and the background. Since the school of young fish is small, and the fluctuating seawater and refraction of underwater light cause a cluttered video background, YOLOv8, as a general target detection network, is not capable of extracting key features (such as fast swimming or dense aggregation), and cannot distinguish between the school of fish and the background well, and cannot complete the task of identifying the school of fish behavior.

[0084] 2. Black sea bream groups are often dynamic groups, with small distances between individuals and even overlapping. The size, speed and movement direction of black sea bream schools vary greatly, so the movement trajectories may appear at different spatial scales. YOLOv8 has difficulty identifying target features of different sizes, which affects the effect of fish school behavior recognition.

[0085] 3. The loss function of the YOLOv8 basic model is CIoU. The CIoU (Complete Intersection over Union) loss function is usually used to improve the accuracy of bounding box regression in target detection tasks. It combines the center distance, aspect ratio and overlapping area to optimize the regression of the target bounding box. In a dynamic environment, the behavior of the fish school will cause the bounding box to change continuously, and the movement trajectory of the black sea bream school may involve the interaction of multiple individuals, and the target shape and scale change greatly. At this time, the CIoU loss function cannot accurately reflect the real-time position and changes of the fish school.

[0086] In order to solve the above problems, we optimize and improve the YOLOv8 network model to better identify the behavior of the black sea bream fry motion video obtained in the offshore aquaculture seawater scene. As a preferred embodiment, in step S102, the specific improvement method of the YOLOv8 network model includes:

[0087] Replace the C2f module in the YOLOv8 backbone network with the C2f-MSADCN module formed by stacking the C2f module and the MSADCN module;

[0088] Integrate the MSSimAM attention module into the C2f module in the YOLOv8 neck network;

[0089] Replace the CIoU loss function in the YOLOv8 head network with the SEIoU loss function.

[0090] The following is a detailed introduction to the specific improvement methods:

[0091] (1) Replace the C2f module in the YOLOv8 backbone network with the C2f-MSADCN module formed by stacking the C2f module and the MSADCN module:

[0092] The strategies of MSADCN mainly include multi-scale feature extraction, attention mechanism, and dynamic context learning.

[0093] As a specific embodiment, as Figure 2 shown, Figure 2 is a schematic diagram of the structure of the initial fry behavior recognition model constructed based on the improved YOLOv8 network model. In this embodiment, the C2f module in the original YOLOv8 backbone network is replaced with the C2f-MSADCN module formed by stacking the C2f module and the MSADCN module.

[0094] The MSADCN module (Multiscale Attention Deformable Convolutional Network) is an improvement of the deformable convolutional network DCNv2. This network introduces multi-scale and attention mechanisms on the basis of DCNv2, enabling it to more effectively process irregular deformations and multi-scale features. The structure of the MSADCN module is as Figure 3 shown. The core convolution operation of MSADCN can be expressed by the following formula:

[0095]

[0096] where y(p) is the feature value at position p in the output feature map, S represents the number of multi-scales, α s is the learnable weight at scale s, used to adjust the feature contributions of different scales, K is the number of sampling points of the convolution kernel, is the convolution weight at the kth sampling position in the sth scale, is the offset reference at the kth sampling position, is the learnable offset, is the modulation coefficient of the k-th sampling point, which is used to adjust the feature importance at this position.

[0097] By replacing the C2f module in the YOLOv8 neck network with the C2f-MSADCN module formed by stacking the C2f module and the MSADCN module, it is possible to more accurately identify the irregular deformations presented by the black porgy fish school during swimming, as well as the fast swimming of the fish school, the blurring caused by water flow, and the changing background in underwater video surveillance, and improve the overall recognition effect of the behavior of juvenile black porgy. Pay attention to the behavior of the fish school covered by the interference factors of the surrounding environment (such as water flow fluctuations, plants, sand and stones, etc.) in the complex background of the video, and automatically adjust the position and weight of the convolutional sampling points through dynamic context learning, so as to better improve the performance of YOLOv8 in identifying the behavior trajectory of the black porgy fish school.

[0098] (2) Integrate the MSSimAM attention module into the C2f module in the YOLOv8 neck network:

[0099] MSSimAM (Multi-Scale Simulated Attention Module) is an improvement of the SimAM attention module. This module can use multi-scale receptive fields to better capture the local and global information of images. The structure of the MSSimAM attention module is as Figure 4 shown.

[0100] For the input feature map where C represents the number of channels, H represents the height of the input feature map, and W represents the width of the input feature map. In some embodiments, C = 256, H = 1280, and W = 720. The energy function of MSSimAM at position (h, w) at scale s can be expressed as:

[0101]

[0102] where, y j represents the output of the j-th neuron, is the mean of the outputs of all neurons at scale s, is the variance of the neuron outputs at scale s, λ is a parameter that controls the balance between the mean term and the variance term, and M is the number of neurons in the neighborhood.

[0103] The attention weight of MSSimAM at position (h, w) at different scales s can be expressed as:

[0104]

[0105] where, ∈ is a constant to prevent division by zero.

[0106] The attention weight at the position (h, w) after multi-scale fusion of MSSimAM can be expressed as:

[0107]

[0108] where S represents the number of different scales, and w s represents the fusion weight of scale s, which is used to control the contribution of each scale to the final feature.

[0109] The eigenvalue at the position (h, w) after enhancing MSSimAM can be expressed as:

[0110] y h,w = A(h, w)x h,w

[0111] where x h,w represents the eigenvalue at the position (h, w) in the input feature map, and y h,w represents the eigenvalue at the enhanced position (h, w).

[0112] Through the above improvements, MSSimAM can identify the movements of juvenile black porgy during fast swimming, turning, or moving between different water layers, thereby capturing the dynamic behavior information at these different scales. For example, when the black porgy is moving at a relatively long distance or in a small background, MSSimAM can identify the overall movement pattern of the fish through a larger receptive field; when the black porgy is swimming fast at a relatively close distance or in a complex background, the smaller-scale receptive field helps the model focus on local details and rapidly changing features. At the same time, MSSimAM can calculate the energy function at different scales, balance the influence of local and global information, and avoid missing important movement information due to the bias towards a certain scale, so as to help the model identify the subtle changes during group interaction. By combining local and global information, the ability to capture complex behaviors is enhanced. For example, when the fish school is moving collectively, it is necessary to capture their global movement trend; when an individual black porgy is performing detailed actions (such as jumping, pausing, fast swimming, etc.), it is necessary to pay attention to local changes. By calculating the attention weight at each scale, local and global information is organically fused, thereby enhancing the ability to capture these complex behaviors.

[0113] Therefore, integrating the MSSimAM attention module into the C2f module in the YOLOv8 neck network can significantly improve the behavior recognition accuracy in black porgy movement videos, especially when facing dynamic scenes, complex backgrounds, and small object detection. Through multi-scale feature fusion and optimization of the attention mechanism, MSSimAM can not only capture richer feature information but also improve the robustness and generalization ability of the model, thereby enhancing the behavior classification, localization, and tracking ability of black porgy.

[0114] (3) Replace the CIoU loss function in the YOLOv8 head network with the SEIoU loss function:

[0115] The SEIoU loss function, known as the Simplified Efficient Intersection over Union, is an improvement over the EIoU loss function. The SEIoU loss function can be expressed by the following formula:

[0116]

[0117] Among them, 1 - IoU represents the intersection over union loss, which is used to measure the overlap difference between the predicted bounding box and the ground truth bounding box. ρ 2 (b, b gt ) represents the square of the Euclidean distance between the center points of the predicted bounding box and the ground truth bounding box. w and w gt are the widths of the predicted bounding box and the ground truth bounding box respectively, and h and h gt represent the heights of the predicted bounding box and the ground truth bounding box respectively. w c and h c are the width and height of the smallest bounding rectangle that can enclose the predicted bounding box and the ground truth bounding box respectively.

[0118] Using SEIoU as the loss function of the model not only enables the model to focus on the overlap degree (IoU) between the bounding boxes, but also introduces the Euclidean distance of the center points and the scale adjustment term, which can more accurately locate the behavior patterns of black porgy fish with fast movement or small - scale changes. By increasing the constraint on the position of the center point of the bounding box, SEIoU makes the predicted bounding box closer to the ground truth bounding box, reducing the error caused by the position deviation of the predicted bounding box. Especially in the scenes of fast - moving or overlapping porgy fish movements, it effectively improves the localization accuracy of the model. In addition, the scale adjustment term in SEIoU, especially by calculating the ratio of the width and height differences to the smallest bounding rectangle, can improve the accuracy of behavior recognition affected by the fish body size differences. For example, small black porgy fish at a long distance or black porgy fish blocked by other fish groups often produce bounding boxes of different sizes. Juvenile black porgy fish that appear relatively small under long - distance or light - condition influence. The scale adjustment part in the SEIoU loss function can more effectively handle small object detection, especially in the case of being far from the camera or blocked by other objects, maintaining high - precision detection, and finally enabling the black porgy fish behavior recognition model based on YOLOv8 to more accurately and reliably identify and track various behavior patterns of black porgy fish.

[0119] As a preferred embodiment, use the black porgy juvenile fish movement video dataset to train the initial juvenile fish behavior recognition model to obtain a well - trained juvenile fish behavior recognition model, including:

[0120] The initial juvenile fish behavior recognition model is trained using the training set and validation set of the Sparus macrocephalus juvenile fish movement video dataset to obtain the trained juvenile fish behavior recognition model;

[0121] The trained juvenile fish behavior recognition model is tested using the test set of the Sparus macrocephalus juvenile fish movement video dataset;

[0122] When the prediction result of the model reaches the training expectation, a trained complete juvenile fish behavior recognition model is obtained, and the coordinate time series of all Sparus macrocephalus juvenile fish in the training set, validation set, and test set videos are extracted respectively.

[0123] Specifically, the training expectation is measured by mAP (mean Average Precision). If the mAP does not meet the expectation (usually set to 0.95), it may be because there are some problems in model training or improper hyperparameter adjustment. The model needs to be continuously trained until the mAP meets the preset standard threshold. Then, a trained complete juvenile fish behavior recognition model can be obtained, and the coordinate time series of all Sparus macrocephalus juvenile fish in the training set, validation set, and test set videos are extracted respectively. These data are position time series, that is: the position information of Sparus macrocephalus juvenile fish at each moment, and their positions (coordinates) and possible labels (such as behavior categories) are output.

[0124] As a specific embodiment, the coordinate time series of all Sparus macrocephalus juvenile fish in the training set, validation set, and test set videos are extracted respectively. The 1st and 16th frame images per second are taken to extract the position coordinates, and the coordinate time series of each video is represented as a three-dimensional matrix

[0125] Among them, T represents the time length, N represents the number of Sparus macrocephalus juvenile fish in the video, and 2 represents the x and y coordinates of each Sparus macrocephalus juvenile fish. In this embodiment, T and N take values of 30 and 10 respectively.

[0126] As a preferred embodiment, in step S103, an initial spatio-temporal correlation network model based on GNN-GRU is constructed, including:

[0127] The initial spatio-temporal correlation model includes a GNN module, a multi-layer GRU module, and a fully connected module;

[0128] The GNN module is used to obtain spatial features based on the input of the position features of Sparus macrocephalus juvenile fish at different moments, and is used to capture the spatial correlation between Sparus macrocephalus juvenile fish;

[0129] The multi-layer GRU module contains multiple cascaded GRU network models, and calculates the hidden states of each GRU network model according to the spatial features at each moment; it is used to capture the temporal correlation of Sparus macrocephalus juvenile fish behavior;

[0130] The fully connected network model predicts the acesulfame concentration based on the hidden states output by the multi-layer GRU module.

[0131] As a specific embodiment, as Figure 5 shown, Figure 5 The initial spatio-temporal correlation network model structure is shown. The spatio-temporal correlation network model of GNN-GRU consists of a GNN network model, a multi-layer GRU network model, and a fully connected network model. Among them, the GNN network model is used to process the position feature data X t at each moment t and outputs the spatial feature H with a dimension of N×d. t The multi-layer GRU network model is composed of T GRU network models connected in series, and sequentially receives the spatial feature H t at each moment t, and gradually calculates the hidden states of each GRU network model. Finally, the hidden state h T output by the T-th GRU network model is used as the output result of the entire multi-layer GRU network model. The fully connected network model uses the hidden state h T to predict the acesulfame concentration.

[0132] In the spatio-temporal correlation model, the GNN network model is used to capture the spatial correlation between juvenile black porgy. For the input data at moment t (1≤t≤T), the N×2 position feature represents the node position matrix at this moment, and the GNN is used to capture the spatial relationship between nodes at moment t. The GNN network structure includes an input node structure, an adjacency matrix calculation, and a graph convolution module. Among them, the GNN input node structure is the N×2 position feature at moment t, and the position feature can be represented as a matrix where each row represents the node coordinates x, y; the adjacency matrix is represented as a matrix It reflects the spatial relationship between nodes and is generated according to the Euclidean distance between nodes, that is The graph convolution module convolves the features of each node to capture the spatial correlation between targets, and its update rule is:

[0133]

[0134] Among them, is the node feature of the l-th layer, and the initial value is X t , A t is the normalized adjacency matrix, D is the degree matrix of the adjacency matrix, W (l) is the learned weight matrix, and σ is the non-linear activation function ReLU. The output result of the GNN is the N×d-dimensional feature matrix H t , where d represents the node feature embedding dimension.

[0135] The GRU network model is used to capture the temporal correlation of the behavior of juvenile black porgy. For the input feature matrix H t , where t = 1, 2, 3, … T, the GRU network model is used to capture the temporal dependence of nodes within T time. The GRU network module structure includes an input node structure and an update rule. Among them, the input node structure is an N×d-dimensional feature matrix H t , and the GRU update rule is expressed as the following formula:

[0136]

[0137] where z t represents the update gate, h t―1 represents the hidden state at the previous moment, represents the candidate hidden state.

[0138] As a preferred embodiment, the initial spatio-temporal correlation network model is trained using the behavior trajectory dataset, and the acesulfame concentration is predicted by analyzing the behavior trajectory of juvenile black porgy, obtaining a trained complete spatio-temporal correlation network model, including:

[0139] Using the coordinate time series of juvenile black porgy in the training set and the validation set and the corresponding acesulfame concentration as labels, the initial spatio-temporal correlation network model is trained in a supervised manner;

[0140] Using the coordinate time series of juvenile black porgy in the test set and the corresponding acesulfame concentration as labels, the trained spatio-temporal correlation network model is tested. When the accuracy of the prediction result reaches the expectation, a trained complete spatio-temporal correlation network model is obtained.

[0141] To test the actual effect of the present invention, we conducted the following experiments: The YOLOv8 base model and various combined models of the YOLOv8 base model with the MSADCN module, the MSSimAM module, and the SEIoU loss function were used to train the juvenile black porgy dataset respectively. The mAP and training time consumption of each model are as shown in Figure 6 and Figure 7 respectively.

[0142] Model A is the base YOLOv8 model, with an mAP of 87.6% and a training time of 11.04 hours.

[0143] After integrating the MSADCN into the base model, Model B has an mAP increase of 1.6% compared to Model A, and the training time increases by 1.42 hours. Although the training time increases slightly, the MSADCN enhances the feature extraction and representation ability of the model in low-light water environments and improves the detection accuracy of the model.

[0144] Model C integrates MSSimAM into the base model. Similar to Model B, MSSimAM further improves the mAP while the increase in training time is very small. This is mainly because MSSimAM enhances the local detailed features, thus improving the accuracy of the juvenile black porgy behavior recognition task.

[0145] Model D improves the loss function of the base model, resulting in a slight increase in mAP and a reduction in training time compared to Model A. This is mainly because SEIoU accelerates the model's convergence and reduces the training time.

[0146] Model E integrates MSADCN and MSSimAM into the base model, leading to a significant improvement in accuracy. The fusion of these three significantly enhances the model's ability to describe features.

[0147] Model F integrates MSADCN into the base model and uses SEIoU, with an mAP of 89.9% and a training time of 12.39 hours. The fusion of these technologies enables the model to identify targets more accurately and focus on high-quality priors.

[0148] Model G utilizes MSSimAM and SEIoU technologies, improving the accuracy of identifying juvenile black porgy behavior while reducing the training time. Model H integrates MSADCN, MSSimAM, and SEIoU, with an mAP of 95.7% and the training time slightly increasing to 13.04 hours. By supplementing different modules, the newly constructed improved YOLOv8 model achieves better performance improvement and ensures a good trade-off between the recognition accuracy and computational speed of juvenile black porgy behavior detection.

[0149] For the time series of the positions of juvenile black porgies extracted by the improved YOLOv8 model, the GNN-GRU spatio-temporal correlation model and traditional classification methods are respectively used to predict the acesulfame concentration. The traditional classification method calculates indicators such as the group movement speed, nearest neighbor distance, individual swimming speed, individual swimming speed synchrony, and polarity of juvenile black porgies, and realizes the prediction of the acesulfame concentration through the random forest algorithm. The prediction results of the two methods are as Figure 8 shown. Among them, the prediction accuracy of the random forest is 89.97%, while the prediction accuracy of GNN-GRU is 95.46%, which is much higher than the traditional prediction method.

[0150] From the above experimental results, it can be seen the excellent performance of the method of the present invention in the tasks of juvenile black porgy behavior recognition and acesulfame concentration prediction. The comprehensive effect of its target recognition and classification prediction is better than the existing methods, and its adaptability and detection accuracy in complex environments have also been further improved. This method not only has significant theoretical value but also provides strong support for the practical applications of the aquatic ecological environment.

[0151] This embodiment also provides a prediction system for the concentration of acesulfame in offshore aquaculture seawater based on the behavioral characteristics of juvenile black porgy, including:

[0152] A data acquisition module, which is used to collect the motion video data of juvenile black porgy in environments with different acesulfame concentrations and perform preprocessing to obtain a motion video dataset of juvenile black porgy;

[0153] A behavior recognition model construction module, which is used to construct an initial juvenile fish behavior recognition model based on an improved YOLOv8 network model, and use the motion video dataset of juvenile black porgy to train the initial juvenile fish behavior recognition model to obtain a well-trained juvenile fish behavior recognition model, as well as a behavior trajectory dataset corresponding to the motion video dataset of juvenile black porgy;

[0154] A spatio-temporal correlation network model construction module, which is used to construct an initial spatio-temporal correlation network model based on GNN-GRU, use the behavior trajectory dataset to train the initial spatio-temporal correlation network model, and predict the acesulfame concentration by analyzing the behavior trajectory of juvenile black porgy to obtain a well-trained spatio-temporal correlation network model;

[0155] A concentration prediction module, which is used to obtain the actual motion video of juvenile black porgy in offshore aquaculture seawater, and use the juvenile fish behavior recognition model and spatio-temporal correlation model to predict the acesulfame concentration of the offshore aquaculture seawater according to the actual motion video.

[0156] The prediction method for the concentration of acesulfame, an artificial sweetener, in offshore aquaculture seawater based on the behavioral characteristics of juvenile black porgy proposed by the present invention has its core advantages reflected in that an improved YOLOv8 and GNN-GRU models are respectively used to construct a juvenile fish behavior recognition model and a spatio-temporal correlation network model. The two models are optimized for the recognition task of the behavioral characteristics of juvenile black porgy, improving the model performance. By introducing a multi-scale attention deformable convolutional network (MSADCN) into YOLOv8, it is possible to more accurately identify the irregular deformations presented by the black porgy fish school during swimming in underwater video monitoring, as well as the fast swimming of the fish school, the blurring caused by water flow, the changing background, and multi-scale features. In addition, a multi-scale simulated attention module (MSSimAM) is added to enhance the feature fusion ability, and at the same time, through a simple and efficient intersection over union (SEIoU) loss function, the accuracy of bounding box regression is improved, further optimizing the detection effect and calculation speed.

[0157] On the other hand, the GNN-GRU spatiotemporal joint model captures the complex relationship between targets by using GNN to model the spatial features in each time step, and gradually updates the hidden state in the time series through GRU to capture long-term and short-term dependencies, forming a joint modeling of spatiotemporal dependencies. Combining the spatial feature extraction of GNN and the temporal dynamic modeling of GRU, the classification accuracy of different concentrations of acesulfame potassium based on the behavioral characteristics of black sea bream fry has been improved, thus providing a scientific basis for evaluating the pollution of acesulfame potassium to the marine ecological environment.

[0158] In summary, the present invention has broad application potential, such as fish ecological behavior research, environmental monitoring, underwater robot navigation and other fields, and has important application and theoretical value in promoting technological progress in the fields of marine environmental protection and ecological monitoring.

[0159] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles, characterized in that: include: The motion video data of black porgy juveniles under different acesulfame potassium concentrations were collected and preprocessed to obtain the black porgy juvenile motion video data set. An initial juvenile behavior recognition model is constructed based on the improved YOLOv8 network model, and the initial juvenile behavior recognition model is trained using a black sea bream juvenile motion video dataset to obtain a fully trained juvenile behavior recognition model and a behavior trajectory dataset corresponding to the black sea bream juvenile motion video dataset; An initial spatiotemporal association network model is constructed based on the GNN and GRU networks, the initial spatiotemporal association network model is trained using the behavioral trajectory data set, and the concentration of acesulfame potassium is predicted by analyzing the behavioral trajectory of black sea bream fry to obtain a fully trained spatiotemporal association network model; The actual movement video of black sea bream fry in offshore aquaculture seawater is obtained, and the fry behavior recognition model and the spatiotemporal association model are used to predict the acesulfame potassium concentration in the offshore aquaculture seawater according to the actual movement video.

2. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles according to claim 1, characterized in that: The motion videos of black porgy fry under different acesulfame potassium concentrations were collected and preprocessed, including: Based on the preset shooting angle, resolution, and frame rate, the motion videos of black sea bream juveniles in different acesulfame potassium concentration environments were collected to obtain the original video data; The original video data is divided into segments of equal length, and the concentration of acesulfame potassium is marked for each video to obtain a motion video dataset of black sea bream fry; The video data set is expanded by using data enhancement technology to obtain an expanded data set; The black sea bream fry in each video in the expanded data set are marked, and the format of the marked videos is converted to obtain a preprocessed motion video data set; The preprocessed motion video dataset is divided into training set, validation set and test set.

3. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles according to claim 1, characterized in that: The improved YOLOv8 network model includes: Replace the C2f module in the YOLOv8 backbone network with a C2f-MSADCN module formed by stacking the C2f module and the MSADCN module; Integrate the C2f module in the YOLOv8 neck network into the MSSimAM attention module; Replace the CIoU loss function in the YOLOv8 head network with the SEIoU loss function.

4. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream fry according to claim 3, characterized in that: The MSADCN module introduces multi-scale and attention mechanisms, and the convolution operation of the module is expressed as follows: Among them, y(p) is the eigenvalue of position p in the output feature map, S represents the number of multi-scales, and α s is the learnable weight at scale s, which is used to adjust the feature contribution of different scales. K is the number of sampling points of the convolution kernel. is the convolution weight of the kth sampling position at the sth scale, is the offset reference of the kth sampling position, is the learnable offset, is the modulation coefficient of the kth sampling point, which is used to adjust the feature importance of this position.

5. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles according to claim 3, characterized in that: The MSSimAM attention module includes a multi-scale convolution layer, a similarity adjustment layer, an attention layer and a feature fusion layer, which is used to capture the local and global information of the image using a multi-scale receptive field; The multi-scale convolution layer is used to extract features of different scales using convolution kernels of different sizes; The similarity measurement layer is used to measure the similarity between features of different scales; The attention layer is used to generate an attention map based on the similarity measurement and weight the multi-scale features to highlight the key features; The feature fusion layer is used to fuse the weighted multi-scale features into a module for final output.

6. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles according to claim 3, characterized in that: The SEIoU loss function is expressed as follows: Among them, 1-IoU represents the intersection-over-union loss, which is used to measure the overlap difference between the predicted box and the real box, ρ 2 (b,b gt ) represents the square of the Euclidean distance between the center point of the predicted box and the center point of the real box, w and w gt are the widths of the predicted box and the true box, h and h respectively. gt Represents the height of the predicted box and the real box, w c and h c They are the width and height of the minimum bounding rectangle that can wrap the predicted box and the true box, respectively.

7. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles according to claim 2, characterized in that: The initial juvenile behavior recognition model was trained using the black sea bream juvenile motion video dataset to obtain a fully trained juvenile behavior recognition model, including: The initial juvenile behavior recognition model is trained using the training set and validation set of the black porgy juvenile motion video dataset to obtain a trained juvenile behavior recognition model; The trained juvenile behavior recognition model was tested using the test set of the black porgy juvenile motion video dataset. When the prediction results of the model meet the training expectations, a fully trained juvenile behavior recognition model is obtained, and the coordinate time series of all black sea bream juveniles in the training set, validation set and test set videos are extracted respectively.

8. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles according to claim 1, characterized in that: The initial spatiotemporal correlation network model based on GNN-GRU is constructed, including: The initial spatiotemporal correlation model includes a GNN module, a multi-layer GRU module and a fully connected module; The GNN module is used to obtain spatial features based on the position feature input of the black porgy fry at different times, so as to capture the spatial correlation between the black porgy fry; The multi-layer GRU module includes multiple GRU network models connected in series, and the hidden state of each GRU network model is calculated according to the spatial characteristics at each moment; it is used to capture the temporal correlation of the behavior of black sea bream fry; The fully connected network model predicts the concentration of acesulfame potassium according to the hidden state output by the multi-layer GRU module.

9. The method for predicting the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles according to claim 2, characterized in that: The initial spatiotemporal association network model is trained using the behavioral trajectory data set, and the concentration of acesulfame potassium is predicted by analyzing the behavioral trajectory of black porgy fry to obtain a fully trained spatiotemporal association network model, including: The coordinate time series of black sea bream fry in the training set and validation set and their corresponding acesulfame potassium concentrations were used as labels to conduct supervised training on the initial spatiotemporal association network model. The trained spatiotemporal association network model was tested using the coordinate time series of black sea bream fry in the test set and their corresponding acesulfame potassium concentrations as labels. When the accuracy of the prediction results reached the expected level, a fully trained spatiotemporal association network model was obtained.

10. A prediction system for the concentration of artificial sweetener acesulfame potassium in offshore aquaculture seawater based on the behavioral characteristics of black sea bream juveniles, characterized in that: include: A data acquisition module is used to collect and pre-process the motion video data of black porgy fry in different acesulfame potassium concentration environments to obtain a black porgy fry motion video data set; A behavior recognition model construction module is used to construct an initial juvenile behavior recognition model based on an improved YOLOv8 network model, and train the initial juvenile behavior recognition model using a black sea bream juvenile motion video dataset to obtain a fully trained juvenile behavior recognition model and a behavior trajectory dataset corresponding to the black sea bream juvenile motion video dataset; A spatiotemporal association network model construction module is used to construct an initial spatiotemporal association network model based on GNN-GRU, train the initial spatiotemporal association network model using the behavioral trajectory data set, predict the concentration of acesulfame potassium by analyzing the behavioral trajectory of black sea bream fry, and obtain a fully trained spatiotemporal association network model; The concentration prediction module is used to obtain the actual movement video of black sea bream fry in offshore aquaculture seawater, and use the fry behavior recognition model and the spatiotemporal association model to predict the acesulfame potassium concentration of the offshore aquaculture seawater according to the actual movement video.

Images