一种基于双注意力机制的水下侧扫声纳图像识别方法及系统
By introducing a dual attention mechanism into the underwater side-scan sonar image recognition model, EDA-Net solves the problem of insufficient feature representation for multi-scale target recognition with small samples, achieving higher recognition accuracy and stability while reducing model complexity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV OF INFORMATION SCI & TECH
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing underwater side-scan sonar image recognition models suffer from scale variation and geometric distortion problems in small-sample, multi-scale target recognition tasks. They also lack feature representation and robust detection capabilities, and rely on large-scale datasets for training, making it difficult to learn the essential features and discriminative representations of targets in small-sample scenarios.
The EDA-Net model based on a dual attention mechanism is adopted, including an improved backbone network and a neck network. It utilizes a dual-path heterogeneous attention module and a bottleneck convolution-parameterless attention module. Through feature extraction by lightweight branches and enhanced branches, combined with channel and spatial attention mechanisms, multi-dimensional feature extraction and adaptive weighting are achieved, reducing model complexity and improving robustness.
It significantly improves the accuracy and stability of target recognition under small sample conditions, alleviates the feature loss problem caused by lack of information, enhances the generalization performance and cross-scale robustness of the model, and reduces computational complexity while maintaining high accuracy.
Smart Images
Figure CN122116109B_ABST