Research And Application Of Ship Detection In High-resolution Remote Sensing Images Based On Deep Learning

Along with the rapid development of remote sensing technology,optical remote sensing images are now capable of accurately revealing and locating surface entities.Therefore,high-resolution optical remote sensing images are widely used in the field of ship detection.In ship detection,target heading plays is key information in applications such as track prediction and port monitoring.In recent years,many advanced methods based on deep convolution neural networks(DCNNs)have acquired the target orientation by regressing the angle of rotated bounding boxes.However,due to the limitation of boundary discontinuity problem,these methods using angle regression model are not effective at determining the prow orientation.To address this problem,we propose a complex coordinates regression model(CPCRM).Experiments shows that this model effectively improves the detection accuracy of the existing DCNN-based ship detection methods.Furthermore,to improve the accuracy of prow detection,we propose a feature-enhancement-based prow detection network called OPD-Net(Omnidirectional Prow Detection Network).OPD-Net mainly consists of a feature refinement network(FRN),a prow attention network(PAN),and a complex plane coordinates regression model.First,the FRN balances the low-level location information and high-level semantic information from multiscale feature maps and then fits various geometric transformations regarding ship targets through deformable blocks.Next,the PAN is used to enhance the ship prow feature as well as suppress background noise,which improves the accuracy of ship heading predictions.Then,the CPCRM is able to uniquely and continuously represent ship heading in arbitrary orientation.In comparison with other DCNN-based prow detection methods,experiments on google earth dataset and HRSC2016 demonstrate the robustness and superiority of our method for ship heading prediction.Finally,to lower the threshold of algorithm application for practitioners in ship-detection-related industries,a ship detection software embedded in the proposed network is developed.The software covers the complete process of remote sensing image interpretation,with functions such as image browsing,object detection,data management and model training.Meanwhile,we design and develop a WebGIS-based intelligent ship detection platform to meet the needs of multi-user and cross-platform.It provides more convenient and intelligent user experience.