Underwater Acoustic Target Recognition With RNN Differential Learning Rates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current UATR methods face challenges such as low signal-to-noise ratio, limited datasets, and complexity due to underwater acoustic data characteristics, which affect accuracy and reliability, especially in military applications where data confidentiality limits training data availability.
Innovation Solution
A UATR method using a recurrent neural network (RNN) structure and differential learning rate (LR) retraining, involving preprocessing, constructing a UATR depth model with a pre-trained model, and setting different LRs for the pre-training part and RNN structure to enhance feature extraction and classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a pre-trained model is used for UATR, then training data requirement is reduced, but model adaptability to underwater acoustic signals deteriorates
Solution Approach 1:
The patent applies parameter changes by differentiating learning rates for different model components. The pre-trained model parameters use a first learning rate while the RNN structure parameters use a second learning rate, allowing each component to be optimized independently for its specific function in the underwater acoustic signal processing task.
Solution Approach 2:
The patent segments the model into two distinct parts: a pre-trained model component for feature extraction and an RNN structure component for temporal modeling. This segmentation allows each part to be trained and optimized separately, with the pre-trained model providing general acoustic features and the RNN providing task-specific temporal processing for underwater acoustic signals.
2Device complexity
If a single learning rate is used for the entire model, then training simplicity is improved, but training efficiency and convergence deteriorate
Solution Approach 1:
The patent applies local quality by assigning different learning rates to different parts of the model based on their specific requirements. The pre-trained model uses a first learning rate optimized for its feature extraction capabilities, while the RNN structure uses a second learning rate optimized for temporal pattern learning, allowing each component to converge at the appropriate speed.
3Speed
If the pre-trained model is fine-tuned with high learning rate, then convergence speed is improved, but model stability and feature preservation deteriorate
Solution Approach 1:
The patent uses parameter changes by setting a first learning rate for the pre-trained model that is lower than the second learning rate for the RNN structure. This differentiated approach allows the pre-trained model to preserve its stable features while still adapting to the new task, preventing excessive updates that would destabilize the model.
Data Source
AI summary
Provided is an underwater acoustic target recognition (UATR) method based on a recurrent neural network (RNN) structure and differential learning rate (LR) retraining, which is specifically aimed at identification and classification issues of ship underwater acoustic targets. Specific implementation steps include: 1. A ship underwater acoustic signal is preprocessed. 2. A UATR depth model is constructed based on a pre-trained model. 3. Retraining configuration. 4. The model is retrained to implement migration to a target domain. 5. A high-performance classification model for UATR is trained. In a model structure of this application, the pre-trained model is combined with the newly added RNN structure and a classification layer, so that a model obtained after retraining can more accurately identify an underwater acoustic target.