摘要
本文提出卷积神经网络的船舶遥感图像目标检测方法。采用拉普拉斯算子增强处理船舶遥感图像,使得船舶目标特征信息更加清晰;基于Snake模型分割出遥感图像中的船舶目标,通过Gabor滤波器提取船舶遥感图像目标特征向量;基于卷积神经网络搭建船舶遥感图像目标检测架构,统一化处理分支网络置信度,对卷积神经网络进行训练,获取最优权重系数;将遥感图像输入至训练好的卷积神经网络中,即可实现船舶目标的检测。实验数据显示:应用本文方法获得的F1 Score参量与IoU参量数值全部大于给定标准数值,充分证实本文方法具有较好的船舶遥感图像目标检测效果。
A convolutional neural network based ship remote sensing image target detection method is proposed. The ship remote sensing image is enhanced and processed by the Laplace operator to make the ship target feature information clearer. The ship target in the remote sensing image is segmented based on the Snake model, and the target feature vector of the ship remote sensing image is extracted by the Gabor filter. Based on the convolutional neural network Build a ship remote sensing image target detection architecture, uniformly process the confidence of the branch network, train the convolutional neural network, obtain the optimal weight coefficient, and input the remote sensing image into the trained convolutional neural network to realize the ship target detection. detection. The experimental data show that the values of F1 Score parameters and IoU parameters obtained by the proposed method are all larger than the given standard values, which fully confirms that the proposed method has a good target detection effect in ship remote sensing images.
作者
李礁
钟乐海
邢伟寅
LI Jiao;ZHONG Le-Hai;XING Wei-yin(School of Graduates Studies of Management and Science University,Shah Alam 40170,Malaysia;School of Electronics and Information of Mianyang Polytechnic,Mianyang 621000,China)
出处
《舰船科学技术》
北大核心
2022年第7期146-149,共4页
Ship Science and Technology
基金
四川省科技计划重点研发项目(2019YFG0112)
四川省科技攻关项目(05GG009-18)。
关键词
卷积神经网络
船舶
图像目标
遥感图像
目标检测
检测精度
convolutional neural network
ship
image target
remote sensing image
target detection
detection accuracy
