基于3D-CNN的高光谱遥感图像分类算法

Hyperspectral remote sensing image classification algorithm based on 3D-CNN

现有的三维卷积神经网络(Three-dimensional convolutional neural network,3D-CNN)模型常有参数过多和特征提取不全面的情况,并且对样本标签有限问题的处理存在不足。针对样本标签有限的问题,采用生成式对抗网络模型对原始数据进行数据增强,解决了个别样本标签少导致分类模型出现过拟合的现象;针对3D-CNN网络提取特征不全面的问题,所设计高效的3D-CNN网络模型,在网络中加入纹理信息增强模型,使网络能更好地提取图像的空谱特征。实验表明,算法在小样本数据情况下比原始网络分类精度更高,能自适应提取高光谱图像的空谱联合特征。

Existing three-dimensional convolutional neural network(3 D-CNN)models have too many parameters and incomplete feature extraction,and they have insufficient handling of the problem of limited sample labels.Aiming at the problem of limited sample labels,the generative confrontation network model is used to enhance the original data and solve the problem of overfitting of the classification model due to the few labels of individual samples.In view of the problem of incomplete extraction of features from the 3 D-CNN network,an efficient 3 D-CNN network model is designed.The texture information enhancement model is added to the network,so that the network can better extract the spatial spectrum characteristics of the image.Experiments show that the algorithm can classify more accurately than the original network in the case of small sample data,and can adaptively extract the space-spectrum joint features of hyperspectral images.