摘要
对卷积神经网络架构进行改进,提高其对遥感图像处理的识别能力。以端到端的形式处理图像数据,为提取更加完善的图像特征数据,对图像数据进行打补丁操作并利用小训练集训练模型参数,达到优化网络的目的,降低过拟合的概率。为验证该方法的有效性,采用遥感数据集进行实验验证,实验结果表明,该卷积神经网络架构对遥感图像的处理有更好的效果和更高的精确度。
The convolutional neural network architecture was improved to improve the ability of recognizing remote sensing image processing. The way to process image data was processed in an end-to-end format. To improve the image feature data extraction, the image data were patched and the model parameters were trained using a small training set to achieve the purpose of optimizing the network and reducing the over-fitting the probability. To verify the effectiveness of the proposed method, experimental verification of the remote sensing dataset was carried out. The results show that the convolutional neural network architecture proposed has better effects on processing remote sensing images with higher accuracy.
作者
宋超峰
杨剑
宋文爱
张涛
SONG Chao-feng;YANG Jian;SONG Wen-ai;ZHANG Tao(College of Software,North University of China,Taiyuan 030051,China)
出处
《计算机工程与设计》
北大核心
2019年第3期819-824,共6页
Computer Engineering and Design
基金
山西省回国留学人员科研基金项目(2014-053)
山西省科技公关基金项目(20090322004)
关键词
卷积神经网络
遥感图像
图像特征
补丁
过拟合
convolutional neural network
remote sensing image
image feature
patch
overfitting