基于级联卷积神经网络的港口多方向舰船检测与分类

港口舰船目标自动检测的定位和类型分类是一个重要而具有挑战性的问题。针对高分辨率光学遥感影像中多方向性排列密集的近岸舰船目标定位和识别困难的问题,提出基于级联区域卷积神经网络和手工提取特征相结合的近岸舰船检测识别框架。首先,使用级联的区域卷积神经网络对舰船位置进行粗定位并对类别进行估计，得到一系列粗定位的垂直预测框。然后,设计一个可以准确定位舰船的斜框旋转回归器,其将第一阶段所得粗定位垂直矩形框转变为带方向的斜矩形框。最后,使用非极大值抑制的方法去除冗余的预测框。实验采用谷歌地球上采集的数据集进行训练和预测,实验结果表明所提算法在精准率和召回率上均具有较大优势。

Oriented inshore ship detection and classification based on cascade RCNN

Automatic inshore ship recognition, including target localization and type classification, is an important and challenging problem. However, arbitrarily rotated ships are always moored inshore densely. This makes it very difficult to recognize and locate ship targets. To resolve this problem, a multiclass oriented ship localization and recognition framework is proposed based on a cascade region convolutional neural network (RCNN) and feature designed manually. Firstly, cascade RCNN is adopted to localize and classify the positive regions of ships-a set of bounding boxes (BBox). Secondly, a novel procedure which transforms a bounding box to a rotated bounding box is designed and applied to each BBox. Finally, non-maximum suppression (NMS) is adopted to remove the redundant rotated BBoxes (RBoxes). Extensive experimental results conducted on the dataset collected from Google Earth demonstrate the effectiveness of the proposed approach, compared to the other approaches.