激光与光电子学进展, 2019, 56 (5): 052801, 网络出版: 2019-07-31
基于改进的全卷积神经网络的资源三号遥感影像云检测 下载: 1290次
Cloud Detection of ZY-3 Satellite Remote Sensing Images Based on Improved Fully Convolutional Neural Networks
遥感 资源三号影像 深度学习 全卷积网络 云检测 remote sensing ZY-3 images deep learning fully convolutional networks cloud detection
摘要
提出了基于改进的深度学习全卷积神经网络的资源三号遥感影像云检测方法。将预训练后的深层卷积神经网络全连接层改为全卷积层,采用反卷积方法对特征图进行上采样,优化改进网络结构,并采用Adam梯度下降法加速收敛。利用资源三号云区影像数据集对网络进行训练,将上采样后的影像特征输入sigmoid分类器进行分类。实验结果表明,该方法检测精度和速度均优于传统方法,准确率可达90.11%,单张影像检测耗时可缩短至0.46 s。
Abstract
A method for cloud detection of ZY-3 satellite remote sensing images is proposed based on improved deep learning fully convolutional neural network. In pre-trained deep convolutional neural network, full convolution layer is used instead of full connection layer, and deconvolution method is used to up-sample feature map to optimize and improve network structure, then the Adam gradient descent method is adopted to accelerate convergence. The network is trained by using the resource image database of ZY-3 satellite, and the up-sampled image features are input into the Sigmoid classifier . Experimental results show that the proposed method performs better than the traditional methods in terms of detection accuracy and speed. The accuracy achieves 90.11%, and detection time can be reduced to 0.46 s.
裴亮, 刘阳, 谭海, 高琳. 基于改进的全卷积神经网络的资源三号遥感影像云检测[J]. 激光与光电子学进展, 2019, 56(5): 052801. Liang Pei, Yang Liu, Hai Tan, Lin Gao. Cloud Detection of ZY-3 Satellite Remote Sensing Images Based on Improved Fully Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2019, 56(5): 052801.