Marine Ship Detection Based on Twin Feature Pyramid Network and Spatial Attention

Recently, ship detection technology has been applied extensively in the marine security monitoring field. However, achieving accurate marine ship detection still poses significant challenges due to factors such as varying scales, slightly occluded objects, uneven illumination, and sea clutter. To address these issues, we propose a novel ship detection approach, i.e., the Twin Feature Pyramid Network and Data Augmentation (TFPN-DA), which mainly consists of three modules. First, to eliminate the negative effects of slightly occluded objects and uneven illumination, we propose the Spatial Attention within the Twin Feature Pyramid Network (SA-TFPN) method, which is based on spatial attention to reconstruct the feature pyramid. Second, the ROI Feature Module (ROIFM) is introduced into the SA-TFPN, which is used to enhance specific crucial details from multi-scale features for object regression and classification. Additionally, data augmentation strategies such as spatial affine transformation and noise processing, are developed to optimize the data sample distribution. A self-construct dataset is used to train the detection model, and the experiments conducted on the dataset demonstrate the effectiveness of our model.