Dual Spatial Attention Network for Underwater Object Detection With Sonar Imagery

Underwater object detection with sonar imagery has gained significant attention in the object detection community. However, sonar images have less visual information than optical images, so the detailed features between the object and the background are difficult to distinguish in sonar images. In addition, the object features are sparsely caused by low-resolution characteristics, which seriously affects the detection performance. It is crucial to effectively utilize global and local image information to address this issue for extracting object features in sonar images. In this work, we propose a novel dual spatial attention network (DSA-Net) for underwater object detection with sonar imagery. DSA-Net consists of three key components: Firstly, a convolutional neural network (CNN) with a feature pyramid network (FPN) are used to extract discriminative features in sonar images. Second, an efficient dual spatial attention module (DSAM) comprises a global and local spatial attention (LSA) branches. DSAM aims to enhance detection accuracy by extracting and fusing local detailed information and global semantic features. Finally, the Generalized Focal Loss (GFL) is introduced to optimize the proposed network, enabling rapid regression of the location network by learning the probabilities of bounding box values. Experimental evaluations are conducted on two public underwater detection datasets, demonstrating that the proposed method outperforms other state-of-the-art (SOTA) methods.