基于深度语义分割的多源遥感图像海面溢油监测

针对遥感图像海面溢油区域通常受到斑噪声以及强度不均等因素的影响,从而导致溢油区域监测效果较差的问题,本文引入了深度语义分割的方法,将深度卷积神经网络与全连接条件随机场相结合,形成端对端连接。以Resnet结构为基础,首先通过深度卷积神经网络对多源遥感图像粗分割并作为输入,然后经过改进的全连接条件随机场,利用高斯成对势和平均场近似定理,建立条件随机场形成递归神经网络作为输出。通过多源遥感图像对海面溢油区域进行监测,并利用可见光图像估计溢油区域面积。实验在所建立的多源遥感图像数据集上与其它先进模型进行对比,结果表明本文方法提高了溢油区域的分割精度以及精细细节程度,平均交并比为82.1%,监测效果具有明显地改善。

Research on oil spill monitoring of multi-source remote sensing image based on deep semantic segmentation

In remote sensing images,oil spill areasareusually affected by spot noise and uneven intensity,which leads to poor segmentation.A deep semantic segmentation method was introduced to combine a deep convolution neural network with a full connection conditional random field to form an end-to-end connection.Based on Resnet,first,the multi-source remote sensing image was roughly segmented as input by the deep convolutional neural network.Then,using Gaussian pairwise and mean field approximation,the conditional random field was established as the output of the recurrent neural network.The oil spill area on the sea surface was monitored by amulti-source remote sensing image and estimated by optical images.Experimental results show that the proposed method improves class ification accuracy and captures finer details of oil spill are ascompared with other models using the dataset established by the multi-source remote sensing image.The mean intersection over the union is 82.1%,and the monitoring effect is significantly improved.