Abstract:In remote sensing images, ship objects have the characteristics of small size, slender shape, close arrangement of multiple objects and high similarity between classes. The existing deep learning object detection algorithms have low detection accuracy for small ship objects, and are prone to error detections and missed detections. In order to effectively utilize the remote sensing image information and improve the accuracy of small object detection, the SDNGV ship data set is constructed, and an improved single short multiBOX detector (SSD) ship object detection and recognition method based on concatenated rectified linear unit (CReLU) and feature pyramid networks (FPN) is proposed. Firstly, CReLU is added to the shallow layer of the SSD network to improve the transmission efficiency of its shallow layer features. Secondly, FPN is used to fuse the multiscale feature map used for detection in SSD step by step from the deep layer to the shallow layer of the network to improve the positioning accuracy and classification accuracy of the network. Experiments demonstrate that the proposed detection algorithm has good detection accuracy, the improved method has obvious effect, and the detection accuracy of small ship objects has 10 percent improvement.