高光谱图像分类已被公认为是高光谱数据处理的基础性和挑战性任务,丰富的光谱信息和空间信息为有效描述和识别地表物质提供了契机。卷积神经网络(CNN)中的参数较多,为了避免过拟合问题,需要大量的训练样本。Log-Gabor滤波器可以有效地提取包括边缘和纹理在内的空间信息,降低CNN特征提取的难度。为了充分利用CNN和Log-Gabor滤波器的优点,提出了一种将Log-Gabor滤波器和CNN相结合的高光谱图像分类方法,并利用两个真实的高光谱图像数据集进行了对比实验。实验结果表明,所提方法比传统的支持向量机和CNN方法具有更高的分类精度。

Hyperspectral image classification has been recognized as a basic and challenging task in hyperspectral data processing, wherein the rich spectral and spatial information provides an opportunity for the effective description and identification of the surface materials of the earth. There are many parameters in convolutional neural network (CNN). In order to avoid overfitting problem, a large number of training samples are needed in CNN. In addition, the Log-Gabor filtering can effectively extract spatial information, such as the edge and texture, which reduces the difficulty of CNN feature extraction. To leverage the advantages of CNN and Log-Gabor filtering, a hyperspectral image classification method that combines the Log-Gabor filtering and CNN is proposed herein, and two real hyperspectral image datasets are used for comparison experiments. Experimental results show that the proposed method has a higher classification accuracy than that of the traditional support vector machine and CNN.