A fishery intelligent monitoring analysis and dynamic optimization method based on federated learning

Abstract

The present application relates to the technical field of fishery artificial intelligence and federated learning, and more particularly to a fishery intelligent monitoring analysis and dynamic optimization method based on federated learning.S1. Each fishery participant initializes;S2. Local contrast learning;S3. Gradient compression;S4. Central server processing;S5. Personalized federated learning;S6. Iterative optimization.The present application adopts a federated contrast learning method to improve the representation learning ability of distributed training under the premise of protecting data privacy, solves the problem of poor accuracy of fishery artificial intelligence based on federated learning, adopts a federated gradient compression method to reduce communication load and improve robustness, solves the problem of poor real-time performance of fishery artificial intelligence based on federated learning, and adopts a personalized federated learning method to balance the consistency of the global model and the personalized adaptation of local data, solves the problem of model updating difficulty of fishery artificial intelligence based on federated learning.

Description

Technical Field

[0001] This invention relates to the technical fields of artificial intelligence and federated learning in fisheries, and more specifically, to a method for intelligent monitoring, analysis and dynamic optimization of fisheries based on federated learning. Background Technology

[0002] With the increasing scarcity of freshwater resources and intensifying ecological protection efforts globally, freshwater fisheries are accelerating their transformation towards intelligent manufacturing. Artificial intelligence (AI) technologies are being applied to scenarios such as fish school monitoring (e.g., sonar imaging, drone inspections), water quality early warning (e.g., ammonia nitrogen concentration prediction), intelligent feeding (e.g., feed optimization based on fish density), and illegal fishing detection (e.g., electronic fence monitoring). Although Federated Learning (FL), as a distributed learning framework, alleviates the data silo problem through a "data-stationary-model-dynamic" approach, it still faces the following technical bottlenecks in practical implementation:

[0003] First, the accuracy of fisheries AI based on federated learning is poor due to the complexity of freshwater fisheries data sources and the large differences in data quality among participating parties. Second, the real-time performance of fisheries AI based on federated learning is poor due to the large differences in the performance of edge devices and poor signal coverage. Third, the fisheries ecosystem is highly dynamic and subject to strong policy intervention, making it difficult to update fisheries AI models based on federated learning. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for intelligent monitoring, analysis, and dynamic optimization of fisheries based on federated learning. To solve the accuracy problem of fisheries AI based on federated learning, a federated contrastive learning method is adopted to enhance the representation learning capability of distributed training while protecting data privacy. To address the poor real-time performance of fisheries AI based on federated learning, a federated gradient compression method is employed to reduce communication load and improve robustness. To address the difficulty of model updating in fisheries AI based on federated learning, a personalized federated learning method is used to balance the consistency of the global model with the personalized adaptation of local data.

[0005] In this framework, contrastive learning is primarily used to enhance the representation learning capabilities of the local model, gradient compression is used to reduce communication load and improve robustness, and personalized federated learning is used to balance the consistency of the global model with the personalized adaptation of local data. By combining these three methods, the system can achieve efficient distributed training and optimization while protecting data privacy.

[0006] The specific technical solution is as follows:

[0007] A method for intelligent monitoring, analysis, and dynamic optimization of fisheries based on federated learning includes the following steps:

[0008] S1. Initialization of each fishery participant. Each participant (monitoring station, aquaculture farm, research institution) initializes its local model based on its own data characteristics.

[0009] Monitoring sites: Lightweight networks (such as CNNs) are used for feature extraction, which is suitable for real-time monitoring and data analysis.

[0010] Large-scale farms: Use pre-trained deep learning models (such as ResNet) and fine-tune them for large-scale data and high-precision prediction.

[0011] Small family farms: Use a simple multilayer perceptron (MLP) model, suitable for small-scale data and basic management needs.

[0012] Fisheries research institutions: Use complex model structures, including time series processing modules and feature interaction layers, suitable for in-depth research on fish growth mechanisms.

[0013] S2. Local Comparative Learning. Monitoring stations collect local data, including water flow velocity, water temperature changes, and fish activity frequency. Positive and negative sample pairs are generated through data augmentation (e.g., rotation, cropping) or sampling strategies. For example, positive sample pairs (similar water flow velocities) and negative sample pairs (different water flow velocities) are generated from water flow velocity data. Lightweight neural networks (such as Convolutional Neural Networks (CNNs) or Multilayer Perceptrons (MLPs) are used to encode features of the samples. These networks can map input data to a high-dimensional feature space to better capture the intrinsic features of the data. Monitoring stations can use CNNs to extract spatial and temporal features from water flow velocity data.

[0014] InfoNCE Loss (Information-Noise Contrast Estimation Loss) is used to maximize the similarity of positive sample pairs and minimize the similarity of negative sample pairs. The formula is as follows:

[0015]

[0016] This represents the contrast loss for the i-th sample. Adding a negative sign before the logarithmic function is a common mathematical transformation. The logarithmic function has properties such as converting multiplicative relationships into additive relationships; using -log here is to transform the similarity ratio into an optimizable loss value. exp is the exponential function, i.e., e^(-1 / 2). x (e≈2.71828). d(·,·) is the similarity function (cosine similarity). h i (x q ) and h i (x p () are the model's response to the query sample x. q and positive sample x pExtracted feature representation. k is the number of negative samples. This indicates that the model is working on the j-th negative sample. Extracted feature representation, The calculation is based on the query sample feature h. i (x q ) and the features of the j-th negative sample The similarity between them.

[0017] The local model parameters are updated using the gradient descent algorithm. Monitoring sites can use the Adam optimizer to update the CNN weights, improving the accuracy of water flow velocity predictions. Through these steps, local contrastive learning effectively enhances the model's representation learning ability, thus better utilizing local data for training within a federated learning framework.

[0018] S3. Gradient Compression. Each participant quantizes its local gradient with 8-bit precision, mapping floating-point gradient values ​​to a finite set of discrete values. For example, if the gradient range is [-50, 50], it can be compressed using the formula... The quantized gradient value q is calculated. Differential encoding is performed on the quantized gradient, calculating the difference between adjacent gradient values ​​and encoding them for transmission to reduce data transmission volume. The compressed and encoded gradient is then uploaded to the central server via an encrypted channel.

[0019] S4. Central Server Processing. Upon receiving the compressed gradient, the central server decodes it to reconstruct the quantized gradient values, then dequantizes them to convert the discrete values ​​back to floating-point numbers. The central server globally aggregates the decoded and dequantized gradients using a weighted average method, determining the weights based on the data volume and requirements of each participant. The central server updates the global model parameters based on the aggregated gradients and then encrypts and distributes them to all participants.

[0020] S5. Personalized Federated Learning. Model Update: After receiving the global model parameters, each participant updates the model according to its individual needs.

[0021] Large-scale fish farms: Based on their large data volume and focus on fish growth prediction, an adaptive learning rate algorithm is used for model updates.

[0022] Small-scale family farms: Given their small data volume and focus on cost control, the Adagrad optimization algorithm is used for model updates.

[0023] Fisheries research institutions: Based on their need for in-depth research into fish growth mechanisms, they use Bayesian optimization algorithms to update models.

[0024] S6. Iterative Optimization. After each round of training, the system evaluates the model to determine if it has met the predetermined performance metrics. Evaluation is typically performed on a validation set, using various metrics (such as accuracy, mean squared error (MSE), etc.) to measure model performance. For fish growth prediction models, MSE might be used to assess the difference between predicted and actual values; for fish activity frequency prediction models, accuracy might be used to assess classification accuracy. If the model's error on the validation set does not reach a predetermined threshold (error rate below 5%, accuracy above 95%), the system continues with the next round of training. In personalized federated learning, the system dynamically adjusts the weights of model parameters based on the data quality and contribution of each participant. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the steps of a fisheries intelligent monitoring system and optimization method based on federated learning, as described in this invention.

[0026] Figure 2 This is a schematic diagram of the system data flow of the present invention. Detailed Implementation

[0027] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0028] To better illustrate this embodiment, some parts of the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions;

[0029] It is understandable to those skilled in the art that some well-known details may be omitted from the accompanying drawings.

[0030] The positional relationships depicted in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0031] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] like Figure 1 As shown:

[0033] S1. Initialization of each fishery participant. Each participant (monitoring station, aquaculture farm, research institution) initializes its local model based on its own data characteristics.

[0034] Monitoring sites: Lightweight networks (such as CNNs) are used for feature extraction, which is suitable for real-time monitoring and data analysis.

[0035] Large-scale farms: Use pre-trained deep learning models (such as ResNet) and fine-tune them for large-scale data and high-precision prediction.

[0036] Small family farms: Use a simple multilayer perceptron (MLP) model, suitable for small-scale data and basic management needs.

[0037] Fisheries research institutions: Use complex model structures, including time series processing modules and feature interaction layers, suitable for in-depth research on fish growth mechanisms.

[0038] S2. Local Comparative Learning. Monitoring stations collect local data, including water flow velocity, water temperature changes, and fish activity frequency. Positive and negative sample pairs are generated through data augmentation (e.g., rotation, cropping) or sampling strategies. For example, positive sample pairs (similar water flow velocities) and negative sample pairs (different water flow velocities) are generated from water flow velocity data. Lightweight neural networks (such as Convolutional Neural Networks (CNNs) or Multilayer Perceptrons (MLPs) are used to encode features of the samples. These networks can map input data to a high-dimensional feature space to better capture the intrinsic features of the data. Monitoring stations can use CNNs to extract spatial and temporal features from water flow velocity data.

[0039] InfoNCE Loss (Information-Noise Contrast Estimation Loss) is used to maximize the similarity of positive sample pairs and minimize the similarity of negative sample pairs. The formula is as follows:

[0040]

[0041] This represents the contrast loss for the i-th sample. Adding a negative sign before the logarithmic function is a common mathematical transformation. The logarithmic function has properties such as converting multiplicative relationships into additive relationships; using -log here is to transform the similarity ratio into an optimizable loss value. exp is the exponential function, i.e., e^(-1 / 2). x (e≈2.71828). d(·,·) is the similarity function (cosine similarity). h i (x q ) and h i (x p () are the model's response to the query sample x. q and positive sample x p Extracted feature representation. k is the number of negative samples. This indicates that the model is effective for the j-th negative sample. Extracted feature representation, The calculation is based on the query sample feature h. i (x q ) and the features of the j-th negative sample The similarity between them.

[0042] The local model parameters are updated using the gradient descent algorithm. Monitoring sites can use the Adam optimizer to update the CNN weights, improving the accuracy of water flow velocity predictions. Through these steps, local contrastive learning effectively enhances the model's representation learning ability, thus better utilizing local data for training within a federated learning framework.

[0043] S3. Gradient Compression. Each participant quantizes its local gradient with 8-bit precision, mapping floating-point gradient values ​​to a finite set of discrete values. For example, if the gradient range is [-50, 50], it can be compressed using the formula... The quantized gradient value q is calculated. Differential encoding is performed on the quantized gradient, calculating the difference between adjacent gradient values ​​and encoding them for transmission to reduce data transmission volume. The compressed and encoded gradient is then uploaded to the central server via an encrypted channel.

[0044] S4. Central Server Processing. Upon receiving the compressed gradient, the central server decodes it to reconstruct the quantized gradient values, then dequantizes them to convert the discrete values ​​back to floating-point numbers. The central server globally aggregates the decoded and dequantized gradients using a weighted average method, determining the weights based on the data volume and requirements of each participant. The central server updates the global model parameters based on the aggregated gradients and then encrypts and distributes them to all participants.

[0045] S5. Personalized Federated Learning. Model Update: After receiving the global model parameters, each participant updates the model according to its individual needs.

[0046] Large-scale fish farms: Based on their large data volume and focus on fish growth prediction, an adaptive learning rate algorithm is used for model updates.

[0047] Small-scale family farms: Given their small data volume and focus on cost control, the Adagrad optimization algorithm is used for model updates.

[0048] Fisheries research institutions: Based on their need for in-depth research into fish growth mechanisms, they use Bayesian optimization algorithms to update models.

[0049] S6. Iterative Optimization. After each round of training, the system evaluates the model to determine if it has met the predetermined performance metrics. Evaluation is typically performed on a validation set, using various metrics (such as accuracy, mean squared error (MSE), etc.) to measure model performance. For fish growth prediction models, MSE might be used to assess the difference between predicted and actual values; for fish activity frequency prediction models, accuracy might be used to assess classification accuracy. If the model's error on the validation set does not reach a predetermined threshold (error rate below 5%, accuracy above 95%), the system continues with the next round of training. In personalized federated learning, the system dynamically adjusts the weights of model parameters based on the data quality and contribution of each participant.

[0050] The system data flow diagram of the entire federated learning-based intelligent fisheries monitoring system and optimization method embodiment is shown below. Figure 2 As shown.

[0051] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent monitoring, analysis, and dynamic optimization of fisheries based on federated learning, characterized in that, Includes the following steps: S1. Initialization of each fishery participant; each participant initializes its local model based on its own data characteristics; participants include monitoring stations, aquaculture farms, and research institutions; S2. Local contrastive learning; monitoring stations collect local data and generate positive and negative sample pairs through data augmentation or sampling strategies; lightweight neural networks are used to encode the features of the samples; These networks are able to map input data to a high-dimensional feature space, capturing the intrinsic features of the data; The monitoring site uses CNN to extract spatial and temporal features of water flow velocity data; The InfoNCE Loss method is used to maximize the similarity of positive sample pairs and minimize the similarity of negative sample pairs; the formula is as follows: ， Let represent the contrast loss of the i-th sample; This is to convert the similarity ratio into an optimizable loss value form; d(·,·) is the similarity function; and These are the model's response to the query sample. and positive samples Extracted feature representation; k is the number of negative samples. This indicates that the model is working on the j-th negative sample. Extracted feature representation, The calculation is based on the features of the query sample. Features of the j-th negative sample The similarity between them; Update the local model parameters using the gradient descent algorithm; The monitoring site uses the Adam optimizer to update the weights of the CNN; S3. Gradient compression: Each participant quantizes its local gradient, sets 8-bit precision, and maps floating-point gradient values ​​to a finite set of discrete values; differentially encodes the quantized gradient, calculates the difference between adjacent gradient values ​​and encodes them for transmission to reduce data transmission volume; the compressed and encoded gradient is then uploaded to the central server through an encrypted channel. S4. Central server processing: After receiving the compressed gradient, the central server performs a decoding operation to restore the quantized gradient value, and then performs dequantization to convert the discrete value back to a floating-point number. The central server performs global aggregation of the decoded and dequantized gradients using a weighted average method, determining the weights based on the data volume and requirements of each participant; it then updates the global model parameters based on the aggregated gradients and encrypts and distributes them to each participant. S5. Personalized Federated Learning; Model updates: After receiving the global model parameters, each participant updates the model according to its individual needs. Large-scale fish farms: Based on their large data volume and focus on fish growth prediction, an adaptive learning rate algorithm is used for model updates. Small-scale family farms: Given their small data volume and focus on cost control, the Adagrad optimization algorithm is used for model updates. Fisheries research institutions: Based on their need for in-depth research into fish growth mechanisms, they use Bayesian optimization algorithms to update models; S6. Iterative optimization; After each round of training, the model is evaluated to determine whether it has reached the predetermined performance metrics; The evaluation is performed on the validation set, using multiple metrics to measure the model's performance.

2. The method for intelligent monitoring, analysis, and dynamic optimization of fisheries based on federated learning according to claim 1, characterized in that, In S1; Monitoring sites: Lightweight networks are used for feature extraction, making them suitable for real-time monitoring and data analysis; Large-scale farms: Use pre-trained deep learning models and fine-tune them for large-scale data and high-precision prediction; Small family farms: A simple multilayer perceptron model is used, suitable for small-scale data and basic management needs; Fisheries research institutions: Use complex model structures, including time series processing modules and feature interaction layers, suitable for in-depth research on fish growth mechanisms.

3. The method for intelligent monitoring, analysis, and dynamic optimization of fisheries based on federated learning according to claim 1, characterized in that, In S6: For fish growth prediction models, mean squared error is used to evaluate the difference between predicted and actual values; for fish activity frequency prediction models, accuracy is used to evaluate the accuracy of classification. If the model's error on the validation set does not reach the predetermined threshold, the next round of training continues; in personalized federated learning, the weights of the model parameters are dynamically adjusted based on the data quality and contribution of each participant.

Images