基于深度卷积迁移学习网络算法的遥感图像分类
Remote Sensing Image Classification Based on Deep Convolution Transfer Learning Network Algorithm
王晓卓 1王楷 1曹澍1
作者信息
- 1. 国网新疆电力有限公司信息通信公司,乌鲁木齐 830000
- 折叠
摘要
由于缺少对图形特征的重采样处理,导致最终分类精度较低,分类性能不理想.为此,提出基于深度卷积迁移学习网络算法的遥感图像分类.通过划分原始遥感图像的几何区域,提取畸变几何像素的描述符,结合尺度空间与像素配准,求取图像的旋转角度,完成遥感图像的几何校正.创新性地采用双流网络架构对像素频带进行通道拼接,结合图像一致性联合特征的交互结果,输出图像融合特征,并利用新像素值指标对特征进行重采样处理.采用深度卷积迁移学习网络算法计算输入图像在参考类别中的所属概率,以此实现遥感图像分类.实际应用结果显示,所设计的方法在遥感图像分类中具有较高的分类精度,分类性能更好.
Abstract
Due to the lack of resampling processing for graphic features,the final classification accuracy and the classification performance are not ideal.To this end,a remote sensing image classification based on deep convolution transfer learning network algorithm is proposed.By dividing the geometric region of the original remote sensing image,extracting the descriptors for the distorted geometric pixel,combining the scale space and pixel registration and the rotation angle of the image is calculated,and the geometric correction of the remote sensing image is completed.The dual stream network architecture is innovatively adopted to splice the channel of pixel frequency bands,combining the interaction results of image consistency joint features,outputting image fusion features,and resampling the features with new pixel value index.The deep convolutional transfer learning network algorithm is used to calculate the probability of the input image in the reference class,so as to realize the remote sensing image classification.The application results of the example show that the designed method has high classification accuracy and better classification performance in remote sensing image classification.
关键词
迁移学习/遥感图像/识别分类/特征融合Key words
Transfer learning/Remote sensing image/Identification and classification/Features fusion引用本文复制引用
出版年
2024
NETL
NSTL
万方数据