摘要
X频段航海雷达图像广泛应用于溢油探测中,但是目前仅用于判断是否有溢油,对油污区域的精确计算较少。针对上述问题,引入了基于柔性注意力机制的图像分割的模型。首先建立以基于特征金字塔的目标检测模型为主干模型的语义分割模型,将多尺度特征进行充分融合,包括高层次的语义信息和低层次的位置信息;然后针对每一个特征层分别引入通道注意力机制,计算特征层中通道之间的权重关系,提高模型对溢油特征提取的表达能力,同时采用像素级别的语义分割打分函数作为评价分割区域质量的指标。对大连区域海面溢油数据集进行检测的实验结果表明,该方法能够提高溢油区域的分割精度,同时与PSPNet、DeepLab V3+和Attention U-net等分割模型相比,基于像素级别的分割准确率提高至85.77%。
X-band navigational radar images are widely employed in oil spill detection.However,they are currently only used to assess whether there are oil spills or not,and the greasy area is calculated with less accuracy.For above problems,a model of image segmentation based on soft attention mechanism is introduced.First,the object detection model based on the feature pyramid structure is established to fully integrate multi-scale features,such as high-level semantic information and low-level location information.Then the channel attention method is included for each of the feature layers to calculate the weight relationship between channels in the feature layer to boost the model s expressive ability for extracting oil spill features.Simultaneously,the pixel-level semantic segmentation score function is used as an indicator to assess the segmentation area s quality.The data set of oil spills on the sea surface in the Dalian area is used for detection and the experimental results show that the proposed method improves the segmentation accuracy of the oil spill area.At the same time,compared with segmentation models such as PSPNet,DeepLab V3+,and Attention U-net,the segmentation accuracy based on the pixel level is improved to 85.77%.
作者
周慧
严凤龙
褚娜
陈明华
刘振宇
ZHOU Hui;YANG Fenglong;CHU Na;CHEN Minghua;LIU Zhenyu(School of Software,Dalian Neusoft Information University,Dalian 116023,China;School of Computer,Dalian Neusoft Information University,Dalian 116023,China)
出处
《电讯技术》
北大核心
2022年第12期1774-1779,共6页
Telecommunication Engineering
基金
辽宁省自然科学基金项目(JH1102)。
关键词
X频段航海雷达
海面溢油检测
图像分割
柔性注意力机制
X-band navigational radar
oil spill identification
image segmentation
soft attention mechanism
