基于空谱融合特征主动学习的高光谱图像分类

针对高光谱图像分类过程中存在的样本量少和分类精度低的问题,提出一种基于空谱融合特征主动学习的高光谱图像分类方法。主要包括构造三通道图像,全卷积网络提取空间特征,空谱特征结合,主动学习方法选择训练样本几个部分。通过结合像素的光谱特性和相邻像素间的空间关联,提取出可以反映像素空谱联合特性的综合特征,提高了像素特征的表达能力。为克服高光谱图像标注数据少、缺乏训练样本的问题,应用主动学习算法,充分选择更具有代表性的样本进行训练,达到少样本情况下较高的分类正确率。通过在标准数据集上进行实验,结果表明:该方法可以达到在总样本数1%作训练样本的情况下,分类正确率达到99.79%,优于传统的高光谱分类算法。

Active Learning Based on Spatial-Spectral Feature for Hyperspectral Image Classification

In order to solve the problem that the number of samples is limited and the accuracy of the classification is low, the authors propose a hyperspectral classification method which extracts the spatial feature with fully conventional networks in three-channels image built by themselves. The following parts are included in this method:three channel image building, spatial feature extracting, spatial and spectral feature fusion and training examples selecting. Active learning method is used to select training sample after the fusion of spatial and spectral feature. By combining the spectral characteristic of the pixel and the spatial correlation of adjacent pixels, the integrated feature which can reflect associated spatial-spectral characteristic of the pixel is obtained. The combination can also enhance the expression ability of the pixel. In order to overcome the limits of the labelled data of hyperspectral image and lack of training samples, the active learning algorithm is put into use. The algorithm selects the most representative sample for training, and occurs ideal accuracy using limited samples. Experiment on the standard dataset shows that the accuracy of the method is as high as 99.79% when one percent samples are selected to train.