Single-Beam Sonar Small Target Recognition Algorithm Based on Underwater Unmanned Platform
-
摘要: 针对水下无人平台载荷能力有限、声呐数据样本稀缺导致小目标识别困难的问题, 提出一种适配水下无人平台小样本条件下的单波束声呐信号目标识别算法。该算法基于主动声呐目标单波束回波信号, 提取波形的时域和频域多维特征, 通过相关性分析与主成分分析降维完成有效特征选择, 并结合随机森林分类器, 实现了小样本训练集下的目标高精度识别。水池实验数据测试结果表明, 相较多种基于多波束声呐图像联合深度学习的方法, 文中算法在更小的训练集下, 精确率达99.42%、召回率 99.39%、F1 值 99.39%、准确率 99.39%。文中方法具有计算量小、运行速度快、可解释性强等优势, 更适合在算力与存储资源受限的水下无人平台上部署, 为水下无人平台在资源受限条件下小目标识别提供了一种高效可行的方案。Abstract: Aiming at the difficulty of small target recognition caused by the limited payload capacity of underwater unmanned platforms and the scarcity of sonar data samples, this paper proposes a single-beam sonar signal target recognition algorithm adapted to the few-shot condition of underwater unmanned platforms. Based on the single-beam echo signal of active sonar targets, the algorithm realizes high-accuracy target recognition under the few-shot condition by extracting multi-dimensional time-frequency features of the signal, optimizing feature selection via correlation analysis and PCA dimensionality reduction, and integrating the random forest classifier. Test results on water tank experimental data show that, compared with various methods combining multi-beam sonar images with deep learning, the proposed algorithm achieves performance indicators of 99.42% precision, 99.39% recall, 99.39% F1-score, and 99.39% accuracy with a smaller training set. The proposed method has the advantages of low computational cost, fast running speed and strong interpretability, making it more suitable for deployment on underwater unmanned platforms with limited computing and storage resources. It provides an efficient and feasible scheme for small target recognition by underwater unmanned platforms under resource-constrained conditions.
-
Key words:
- sonar signal processing /
- small target recognition /
- feature selection /
- machine learning /
- few-shot
-
表 1 特征平均相关系数
Table 1. Average correlation coefficient of features
特征名称 平均
相关系数特征名称 平均
相关系数频谱峭度 0.24 裕度因子 0.62 带宽 0.27 波形峰度 0.63 频谱平坦度 0.33 峰值因子 0.63 上升时间 0.34 脉冲因子 0.66 双谱熵 0.49 香农熵 0.66 波形因子 0.62 波形熵 0.66 
下载: 导出CSV
表 2 模型性能评估
Table 2. Model performance evaluation
数据类型 模型 训练数
据(组)测试数
据(组)P(%) R(%) Acc(%) F1(%) 声呐图像 矩+SVM 385 110 97.95 97.54 97.59 97.60 声呐图像 AlexNet 385 110 99.35 99.28 99.29 99.28 声呐图像 VGG16 385 110 96.09 92.79 92.80 90.50 声呐图像 ResNet50 385 110 99.36 99.39 99.36 99.36 GAF编码 GAF-D3Net 385 110 99.64 99.63 99.65 99.63 单波束数据 本方法 10 110 99.42 99.39 99.39 99.39
下载: 导出CSV
表 3 对比实验结果
Table 3. The results of the comparative experiment
数据类型 模型 训练数
据(组)测试数
据(组)P(%) R(%) Acc(%) F1(%) 声呐图像 矩+SVM 10 110 56.77 57.58 57.58 57.71 声呐图像 AlexNet 10 110 2.78 16.67 16.67 1.77 声呐图像 VGG16 10 110 2.78 16.67 16.67 1.77 声呐图像 ResNet50 10 110 4.66 9.09 9.09 6.16 GAF编码 GAF-D3Net 10 110 15.40 18.03 18.03 16.61 单波束数据 本方法 10 110 99.42 99.39 99.39 99.39 单波束数据 SVM 10 110 97.82 97.58 97.58 97.57 单波束数据 KNN 10 110 98.61 98.48 98.48 98.48 单波束数据 LDA 10 110 98.12 97.88 97.88 97.57
下载: 导出CSV
表 4 特征贡献度
Table 4. Feature contribution degree
特征名称 PC1贡献度 PC2贡献度 PC3贡献度 波形因子 0.55 0.27 −0.12 上升时间 −0.06 0.76 0.63 波形峰度 0.48 0.39 −0.55 频谱峭度 0.49 −0.25 0.69 频谱平坦度 −0.46 0.37 −0.37
下载: 导出CSV
-
[1] 李斌, 郑思远, 曹秀岭, 等. 一种AUV移动OFDM水声通信系统设计[J]. 水下无人系统学报, 2018, 26(6): 612-617. doi: 10.11993/j.issn.2096-3920.2018.06.016Li B, Zheng S Y, Cao X L, et al. Design on mobile OFDM underwater acoustic communication system for an AUV[J]. Journal of Unmanned Undersea Systems, 2018, 26(6): 612-617. doi: 10.11993/j.issn.2096-3920.2018.06.016 [2] Yang H, Zhou T, Jiang H, et al. A lightweight underwater target detection network for forward-looking sonar images[J]. IEEE transactions on instrumentation and measurement, 2024, 73: 1-13. [3] Zhou T, Si J, Wang L, et al. Automatic detection of underwater small targets using forward-looking sonar images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-12. [4] XU H, YUAN H. An SVM-based AdaBoost cascade classifier for sonar image[J]. IEEE Access, 2020, 8: 115857-115864. doi: 10.1109/ACCESS.2020.3004473 [5] Cheng P, Yikun L, Xinpan Y, et al. Research of image recognition method based on enhanced inception-ResNet-V2[J]. Multimedia Tools and Applications, 2022, 81(24): 34345-34365. [6] 韩婧祺, 南明星, 张鹏, 等. 基于自训练YOLO 11模型的低虚警率声呐图像目标检测方法[J]. 水下无人系统学报, 2025, 33(2): 238-248. doi: 10.11993/j.issn.2096-3920.2024-0165Han J Q, Nan M X, Zhang P, et al. A sonar image target detection method with low false alarm aate based on self-trained YOLO11 model[J]. Journal of Unmanned Undersea Systems, 2025, 33(2): 238-248. doi: 10.11993/j.issn.2096-3920.2024-0165 [7] 廖宝浩. 基于注意力机制的UUV声呐目标识别方法研究[D]. 哈尔滨工程大学, 2024, 47-50. [8] 李耀波, 孙常存, 任志良, 等. 一种基于快速傅里叶变换反鱼雷鱼雷主动声引信回波检测方法[J]. 鱼雷技术, 2011, 19(3): 176-180.Li Y B, Sun C C, Ren Z L, et al. An FFT-based detection method of active acoustic fuse echo for Anti-torpedo torpedo[J]. Torpedo Technology, 2011, 19(3): 176-180. [9] 田璐. 基于时频特征的主动声呐回波信号分类研究[D]. 东南大学, 2023, 25-27. [10] 舒象兰, 孙荣光, 马鑫. 强混响背景下LFM信号回波检测[J]. 电讯技术, 2016, 56(1): 82-87. doi: 10.16798/j.issn.1003-0530.2019.09.008Shu X L, Sun R G, Ma X. Echo detection of LFM signal under strong reverberation background[J]. Telecommunication Technology, 2016, 56(1): 82-87. doi: 10.16798/j.issn.1003-0530.2019.09.008 [11] 江铖, 祁丹. 高速光电探测器上升时间校准方法研究[J]. 工业计量, 2025, 35(5): 31-34.Jiang C, Qi D. Calibration techniques for rise time measurement in high-speed photodetectors[J]. Industrial Measurement, 2025, 35(5): 31-34. [12] Feng S, Zhu X, Ma S, et al. Adversarial Attacks in Underwater Acoustic Target Recognition with Deep Learning Models[J]. Remote Sensing, 2023, 15(22): 5386. doi: 10.3390/rs15225386 [13] 赵玉新, 赵廷. 海底声呐图像智能底质分类技术研究综述[J]. 智能系统学报, 2020, 15(3): 14. doi: 10.11992/tis.202004026Zhao Y X, Zhao T. Survey of the intelligent seabed sediment classification technology based on sonar images[J]. CAAI Transactions on Intelligent Systems, 2020, 15(3): 14. doi: 10.11992/tis.202004026 [14] 马晓东. 基于加权决策树的随机森林模型优化[D]. 华中师范大学, 2017, 29-30. [15] 周天, 司吉坤, 杜伟东, 等. 采用GAF-D3Net深度学习网络的水下目标有源识别方法[J]. 声学学报, 2023, 48(5): 950-958. doi: 10.12395/0371-0025.2022016Zhou T, Si J K, Du W D, et al. Active recognition of underwater targets using GAF-D3Net deep learning network[J]. Acta Acustica, 2023, 48(5): 950-958. doi: 10.12395/0371-0025.2022016 -
下载:
点击查看大图
计量
- 文章访问数: 12
- HTML全文浏览量: 7
- PDF下载量: 4
- 被引次数: 0
图(7) / 表(4)
