全文下载: 202204016.pdf
文章编号: 1672-6987(2022)04-0120-07; DOI: 10.16351/j.1672-6987.2022.04.016
刘启明1, 杨树国1, 赵莉2*(1.青岛科技大学 数理学院,山东 青岛 266061; 2.青岛市中医医院(市海慈医院),山东 青岛 266033)
摘要: 提出了一种基于深度卷积神经网络的海洋多目标涡旋检测方法。首先利用改进的密集卷积精确提取海洋涡旋特征,并使用跨层融合技术提高特征的利用率,来充分捕捉边缘信息;然后结合转置卷积和跳过连接构建上采样路径得到检测结果,以获得更高的检测准确率;最后在CMEMS(哥白尼海洋环境监测服务中心)发布的公开数据集上对本工作提出的方法进行了对比实验,实验结果表明:与EddyNet、PET等方法相比,本工作的方法能够有效分离、检测出距离较近的涡旋,具有更好的检测结果。
关键词: 海洋涡旋; 深度学习; 特征融合; 目标检测
中图分类号: S 513文献标志码: A
引用格式: 刘启明, 杨树国, 赵莉. 基于深度卷积神经网络的海洋多目标涡旋检测方法[J]. 青岛科技大学学报(自然科学版), 2022, 43(4): 120-126.
LIU Qiming, YANG Shuguo, ZHAO Li. Ocean multi-eddy detection method based on deep convolution neural network[J]. Journal of Qingdao University of Science and Technology(Natural Science Edition), 2022, 43(4): 120-126.
.
Ocean Multi-Eddy Detection Method Based on Deep
Convolution Neural Network
LIU Qiming1, YANG Shuguo1, ZHAO Li2
(1.College of Mathematics and Physics, Qingdao University of Science and Technology, Qingdao 266061, China;
2.Qingdao Hospital of Traditional Chinese Medicine (Qingdao Hiser Hospital), Qingdao 266033, China)
Abstract: This paper proposes a deep convolutional neural network-based ocean multi-eddy detection method. Firstly, we use an improved dense convolution to accurately extract ocean eddy features, and use a cross-layer fusion technique to improve feature utilization, aiming to capture edge information fully. Secondly, we construct an upsampling path combining transposed convolutions and skip connections for higher detection accuracy. Finally, we compare the proposed method with other methods on a public dataset published by Copernicus Marine Environmental Monitoring Services (CMEMS). The experimental results show that compared with EddyNet and PET, this method can more effectively separate and detect eddy currents in close range, so this method has better detection effect.
Key words: ocean eddy; deep learning; feature fusion; object detection
收稿日期: 2021-08-26
基金项目: 国家自然科学基金山东省联合基金项目(U1906215).
作者简介: 刘启明(1997—),男,硕士研究生.*通信联系人