光学学报, 2019, 39 (4): 0428004, 网络出版: 2019-05-10
类别非均衡遥感图像语义分割的全卷积网络方法 下载: 1444次
Fully Convolutional Network Method of Semantic Segmentation of Class Imbalance Remote Sensing Images
图像处理 遥感图像 语义分割 类别非均衡 全卷积网络 深度学习 image processing remote sensing images semantic segmentation class imbalance fully convolutional network deep learning
摘要
基于U-Net模型,提出了一个全卷积网络(FCN)模型,用于高分辨率遥感图像语义分割,其中数据预处理采用了数据标准化和数据增强,模型训练过程采用Adam优化器,模型性能评估采用平均Jaccard指数。为提高小类预测的准确率,模型中采用了加权交叉熵损失函数和自适应阈值方法。在DSTL数据集上进行了实验,结果表明所提方法将预测结果的平均Jaccard指数从0.611提升到0.636,可实现对高分辨率遥感图像端到端的精确分类。
Abstract
A fully convolutional network (FCN) model based on U-Net is proposed to implement the semantic segmentation of remote sensing images with high resolution, in which the data standardization and data augmentation are adopted for data preprocessing. In addition, the Adam optimizer is used for the model training and the average Jaccard index is used as the evaluation metric. A weighted cross entropy loss function and an adaptive threshold algorithm are employed to improve the classification accuracy of small classes. The experimental results on the DSTL dataset show that the proposed method can increase the average Jaccard index of prediction results from 0.611 to 0.636, and produces an accurate end-to-end classification for high-resolution remote sensing images.
吴止锾, 高永明, 李磊, 薛俊诗. 类别非均衡遥感图像语义分割的全卷积网络方法[J]. 光学学报, 2019, 39(4): 0428004. Zhihuan Wu, Yongming Gao, Lei Li, Junshi Xue. Fully Convolutional Network Method of Semantic Segmentation of Class Imbalance Remote Sensing Images[J]. Acta Optica Sinica, 2019, 39(4): 0428004.