The all-day and all-weather characteristics of the synthetic aperture radar (SAR) images make them be widely applied in the maritime monitoring field. Recently, convolution neural networks-based (CNNs) SAR ship detection algorithms are hot research topics. However, owing to the indistinctive ship features and complex backgrounds, outstanding feature extraction ability is required. Moreover, it is challenging to balance the detection effect and the inference speed. Therefore, a novel hybrid attention-synthetic aperture radar ships detector (HA-SARSD) based on the You Only Look Once version 5 (YOLOv5) is proposed in this paper. The local hybrid attention residual module (LHARM) is designed to optimize the feature extraction ability. Owing to the abundant channels in the deep-level feature, LHARM is developed in the fifth layer of HA-SARSD. Experimental results on Large-Scale SAR Ship Detection Dataset-v1.0 (LS-SSDD-v1.0) and SSDD datasets show that HA-SARSD optimizes the SAR ship feature extraction ability and obtains the balance of detection effect and speed.