Local Salient Feature-focused Learning for SAR Ship Detection

‍ ‍Ship target detection in Synthetic Aperture Radar （SAR） images has gained substantial attention due to its broad application prospects in the military and civilian domains. With the recent rapid advancements in deep learning technologies， SAR ship target detection based on deep learning methods has demonstrated promising performance across various scenarios. It is thus becoming a research hotspot nationally and internationally. However， due to the unique imaging characteristic of SAR， ship targets usually have similar scattering characteristics with the background environment， making it difficult to identify actual ship targets， and the ships’ scales are relatively small， making it challenging to detect ship targets accurately. In this study， we propose a novel local salient feature-focused learning approach for SAR ship detection， which can effectively improve the accuracy of ship target detection and address these issues. First， a dual-attention module （DAM） is designed to provide a dual-attention weighting mechanism for features on both channel and spatial levels. This facilitates thoroughly exploring the critical semantic information associated with ship targets， thereby increasing the model’s ability to understand the nuances between the ship’s target and the background environment. Subsequently， to further enhance the model’s representation capability of ship target features， a balanced feature pyramid network （BFPN） module is introduced. This module achieves a balanced distribution of semantic and spatial information among different feature scales， by scaling enhancing， and aggregating multi-scale features of ship targets. As a result， the model can capture the ship target features contained in different feature scales more comprehensively， and the semantic clues of ship targets are effectively integrated， thus improving ship detection accuracy. Extensive experimental analyses are conducted on the SAR Ship Detection Dataset （SSDD） to validate the method’s effectiveness. For the evaluation metrics， Precision， Recall， F1-score， and mean Average Precision （mAP） are selected as the evaluation metrics for algorithm performance. In addition， visualization of the detection results of different methods under different scenarios is provided. The quantitative and qualitative results consistently show the effectiveness of the model in improving the detection accuracy of ship targets. This study contributes to the development of SAR ship detection methods and improves their practicality and realistic usability. In conclusion， this study presents a SAR ship detection method that leverages local salient feature-focused learning. The method demonstrates enhanced performance in detecting ship targets in SAR imagery through dual-attention and balanced feature pyramid network modules. This holds significant implications for military and civilian applications and contributes to the broader field of deep learning and computer vision. It is worth noting that the proposed method only focuses on enhancing and refining the features of ship targets at each layer. While the topological features of ship targets are important physical structure information of SAR images， using them to enhance the performance of ship target detection will be the focus of future studies.