EfficientShip: A Hybrid Deep Learning Framework for Ship Detection in the River

Optical image-based ship detection can ensure the safety of ships and promote orderly management in offshore waters. Current deep learning researches on optical mainly focus improving one-stage detectors for real-time but sacrifices accuracy detection. To solve this problem, we present a hybrid framework which is named EfficientShip paper. The core parts are DLA-backboned object location (DBOL) CascadeRCNN-guided classification (CROC). DBOL responsible finding potential objects, CROC used to categorize objects. We also design pixel-spatial-level data augmentation (PSDA) reduce risk model overfitting. compare proposed with state-of-the-art (SOTA) literature dataset called Seaships. Experiments show our achieves result 99.63% (mAP) at 45 fps, much better than 8 SOTA approaches meet requirements application scenarios.