利用多尺度特征与深度网络对遥感影像进行场景分类

针对因样本量小而导致的遥感图像场景分类精度不高的问题,结合非下采样Contourlet变换(NSCT)、深度卷积神经网络(DCNN)和多核支持向量机(MKSVM),提出了一种基于多尺度深度卷积神经网络(MS-DCNN)的遥感图像场景分类方法。首先利用非下采样Contourlet变换方法对遥感图像多尺度分解,然后对分解后的高频子带和低频子带分别用DCNN训练得到了不同尺度的图像特征,最后采用MKSVM综合多尺度特征并实现遥感图像场景分类。对标准遥感图像分类数据集的试验结果表明,本算法能够结合低频和高频子带对不同类别场景的识别优势,对遥感图像场景取得较好的分类结果。

Scene Classification of Remote Sensing Image Based on Multi-scale Feature and Deep Neural Network

Aiming at low precision of remote sensing image scene classification owing to small sample sizes, a new classification approach is proposed based on multi-scale deep convolutional neural network (MS-DCNN), which is composed of nonsubsampled Contourlet transform (NSCT), deep convolutional neural network (DCNN), and multiple-kernel support vector machine (MKSVM). Firstly, remote sensing image multi-scale decomposition is conducted via NSCT. Secondly, the decomposing high frequency and low frequency subbands are trained by DCNN to obtain image features in different scales. Finally, MKSVM is adopted to integrate multi-scale image features and implement remote sensing image scene classification. The experiment results in the standard image classification data sets indicate that the proposed approach obtains great classification effect due to combining the recognition superiority to different scenes of low frequency and high frequency subbands.