基于声呐图像的水下目标检测算法

Underwater Object Detection Algorithm Based on Sonar Images

摘要:

基于声呐图像的水下目标检测是水下智能感知系统的核心任务之一。然而,声呐图像低分辨率造成的目标细节缺失、低对比度以及强噪声对目标轮廓的掩盖,使得模型难以学习到稳定且具有判别力的目标特征。为了缓解这些问题,提出了一种高精度的声呐图像特征提取网络。该方法基于单阶段目标检测框架,通过引入空间映射到空间的轻量化卷积模块,在实现无损下采样的同时显著减少了网络参数冗余；采用底层特征增强模块,提升特征提取的有效性；最后引入空间与通道协同注意力模块,提升特征表示的空间敏感性与通道选择性。在声呐通用目标检测数据集的实验结果表明：所提出的方法对比现有模型不仅减少了参数量,还获得了更高的检测精度。

Abstract:

Underwater object detection based on sonar images is one of the core tasks of underwater intelligent sensing systems. However, the low resolution of sonar images leads to a lack of object details, low contrast, and strong noise masking the target contours, making it difficult for models to learn stable and discriminative object features. To alleviate these problems, a high-precision sonar image feature extraction network is proposed. This method is based on a single-stage target detection framework. By introducing the lightweight space-to-depth convolution module, lossless downsampling is achieved while significantly reducing network parameter redundancy. The lightweight feature enhancement module is used to enhance low-level features, improving the effectiveness of feature extraction. Finally, the spatial and channel synergistic Attention module is introduced to improve the spatial sensitivity and channel selectivity of feature representation. Experimental results on the sonar common target detection dataset show that the proposed method not only reduces the number of parameters but also achieves higher detection accuracy compared to the counterparts.