Nearshore vessel detection based on Scene-mask R-CNN in remote sensing image

Object detection method of vessels based on deep learning technology can extract the vessel position and category information in remote sensing image. However, the buildings and artificial facilities in the port area will cause false alarms in the vessel recognition results, which decrease the accuracy. Therefore, containing the object detection and semantic segmentation tasks of deep convolutional neural networks, this paper proposes a ship identification method based on scene-mask R-CNN. Based on the deep convolutional neural network, location information extraction, target classification and target scene discrimination are organically combined to a unified object detection framework. It works in the scene that contains the object, and the context information is used to suppress false alarms that appear in the non-target scene area. Finally, the effectiveness of the proposed method is verified on vessel datasets.