| 摘要: |
| 舰船目标检测是进行海上目标监管，保障海上权益的重要手段。本文在SSD（single shot multibox detector）算法的基础上，利用残差网络（ResNet，residual network）作为骨干网络构建SSD模型，将改进后的SSD算法应用于卫星视频舰船目标检测，该算法采用残差连接替换原本的级联方式，加强前后特征之间的联系，减少模型参数，在保证检测精度的同时提高检测速度。在本文构建的数据集上进行实验，结果表明，改进后的SSD算法在测试集上的均值平均精度（mAP，mean average precision）为93%，比原始SSD算法提高了5.31%，充分证明了该方法对于提升SSD模型性能的有效性。使用“吉林一号”视频03星图像进行验证，结果表明，该算法能够较准确地检测到舰船目标，可为海上复杂环境条件下的舰船实时检测提供参考。 |
| 关键词: 卫星视频 目标检测 舰船 深度学习 |
| DOI：10.11759/hykx20201108006 |
| 分类号:TP79 |
| 基金项目:国家重点研发计划项目（2017YFC1405600） |
|
| Research on ship target detection based on satellite video |
|
ZHANG Chi1, WANG Zhao2, SHENG Hui1
|
|
1.College of ocean and space information, China University of Petroleum (East China), Qingdao 266580, China;2.Qingdao Geotechnical Investigation and Surveying Research Institute, Qingdao 266000, China
|
| Abstract: |
| Ship target detection is an important means to supervise and control maritime targets and protect maritime rights and interests. In this paper, based on SSD (single shot multibox detector) algorithm, using ResNet as the backbone network to build SSD model, the improved SSD algorithm is applied to satellite video ship target detection. The algorithm uses residual connection to replace the original cascade mode, strengthens the connection between the front and rear features, reduces model parameters, and improves the detection speed while ensuring the detection accuracy. The experimental results show that the map of the improved SSD algorithm on the test set is 93%, which is 5.31% higher than the original SSD algorithm, which fully proves the effectiveness of this method for improving the performance of SSD model. The results show that the algorithm can detect the ship target more accurately, which has a certain theoretical significance for real-time ship detection in complex marine environment. |
| Key words: satellite video target detection ship deep learning |
