Federated Ship Detection Training With Adaptive Weighted Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing federated learning methods struggle to reasonably utilize imbalanced and uneven-quality training data, leading to reduced model accuracy in ship detection networks, which is crucial for maritime surveillance.
Innovation Solution
A training method for ship detection networks using federated learning that involves constructing local detection networks, performing dual-branch attention enhancement and feature fusion detection, determining a prediction total loss with a dynamic non-monotonic focusing method, and adaptively weighted aggregation of local parameters to obtain global parameters, iteratively updating until performance improves.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If federated learning is used to train ship detection networks while ensuring data privacy, then data privacy and security are improved, but model accuracy deteriorates due to uneven data distribution and varying data quality
Solution Approach 1:
The patent applies local quality by enabling each client to train their local detection network using their own local ship image data, allowing each participant to leverage their specific data characteristics. The local detection networks process data independently according to local conditions, and only share updated parameters, not the actual data, thus maintaining privacy while adapting to local data quality variations.
Solution Approach 2:
The patent employs parameter changes through adaptive weighted aggregation, where the server dynamically adjusts aggregation weights based on data quality metrics. Clients with higher data quality receive higher weights in the global parameter update, allowing the system to optimize model accuracy across heterogeneous data distributions without compromising privacy.
2Measurement precision
If centralized machine learning training is used, then model accuracy can be improved through comprehensive data access, but data privacy and security requirements of maritime regulatory departments are violated
Solution Approach 1:
The patent segments the centralized training process into distributed local training processes. Each maritime regulatory department operates as an independent client with its own detection network, training locally on its data. This segmentation eliminates the need for centralized data access while maintaining collaborative model improvement through parameter sharing.
Solution Approach 2:
The patent introduces a server as an intermediary that facilitates collaborative training without accessing actual data. The server receives model parameters from clients, performs adaptive weighted aggregation based on data quality, and returns updated global parameters. This intermediary enables coordination and accuracy improvement while keeping data private at each client.
3Reliability
If existing federated learning methods are used with imbalanced data, then data privacy is maintained, but reasonable utilization of training data is reduced leading to lower model accuracy
Solution Approach 1:
The patent applies dynamics through adaptive weighted aggregation, where aggregation weights are not fixed but dynamically adjusted based on real-time data quality assessments. The server evaluates data quality metrics from each client and adjusts weights accordingly, allowing the system to adapt to changing data conditions and optimize training data utilization efficiency while maintaining privacy.
Solution Approach 2:
The patent changes the parameter of aggregation weights dynamically based on data quality. Clients with higher data quality receive higher weights, ensuring their contributions are adequately reflected in the global model. This parameter adjustment optimizes the utilization of imbalanced data across clients while preserving privacy through distributed training.
Data Source
AI summary
The present invention provides a training method and application method for ship detection network based on federated learning. The training method comprises: constructing a ship detection network and a plurality of local detection networks for a plurality of clients; inputting ship image data into each local detection network, performing dual-branch attention enhancement and feature fusion detection on the ship image data to obtain a ship detection output, determining a prediction total loss based on a dynamic non-monotonic focusing method, and updating local parameters of the local detection network; performing adaptively weighted aggregation on the local parameters to obtain global parameters of the ship detection network, updating the local parameters to obtain a new round of local detection networks, and iteratively updating the local parameters and the global parameters until network performance no longer improves. The present invention enhancing the accuracy of the ship detection network.