 This paper proposes a new deep learning technique for detecting ships in synthetic aperture radar, SAR, images. It uses a combination of atrial spatial pyramid pooling and shuffle attention to improve the detection accuracy of small ships while also focusing on relevant features and ignoring irrelevant ones. Additionally, it utilizes the SIOU loss function to optimize the model's performance. Tests on two datasets show that the proposed method outperforms the baseline YOLOV7, achieving a 98.01% detection accuracy, a 96.18% recall rate, and a mean average precision, map, of 98.6%. Furthermore, compared to other deep learning-based methods, the proposed method still performs better in terms of overall performance. This article was authored by Zhu Chen, Chang Lu, V.F. Filler-Ritoff and others.