摘要
针对高光谱图像数据分布不均匀、空谱特征提取不够充分以及随着网络层数增加而导致的网络退化等问题,提出一种基于多尺度混合卷积网络的高光谱图像分类方法。首先,使用主成分分析对高光谱数据进行降维处理;接着,利用邻域提取将邻域内的像素点作为一个样本,补充相应的空间信息;然后,使用多尺度混合卷积网络对预处理后的样本数据进行特征提取,并加入混合域注意力机制来加强空间和光谱维中有用的信息;最后,使用Softmax分类器对每个像素样本进行类别划分。实验结果表明:将所提出的模型在Indian Pines和Pavia University两个高光谱数据集中进行实验,其总体分类精度、平均分类精度、Kappa系数分别能达到0.9879、0.9833、0.9862和0.9990、0.9969、0.9986。该算法能够更加充分地提取高光谱图像的特征信息,与其他分类方法相比取得了更好的分类效果。
To solve the problems of uneven distribution of hyperspectral image data,insufficient spatialspectral feature extraction,and network degradation caused by the increase of network layers,a hyperspectral image classification algorithm based on multi-scale hybrid convolutional network is proposed.Firstly,principal component analysis is applied to reduce the dimension of hyperspectral data.Then,the neighborhood extraction is applied to take all pixels in the neighborhood as a sample to supplement the corresponding spatial information.Next,an improved multi-scale hybrid convolutional network is applied to extract features from the preprocessed sample data,and the mixed domain attention mechanism is added to enhance the useful information in the spatial and spectral dimensions.Finally,the Softmax classifier is used to classify each pixel sample.The proposed model is tested on hyperspectral datasets of Indian Pines and Pavia University.Experiments show that the overall classification accuracy,average classification accuracy and Kappa coefficient can reach 0.9879,0.9833,0.9862 and 0.9990,0.9969,0.9986,respectively.Compared with other classification methods,this algorithm can extract the feature information of hyperspectral images more fully,and achieves better classification results.
作者
杨云
周瑶
陈佳宁
YANG Yun;ZHOU Yao;CHEN Jia-ning(School of Electronic Information and Artificial Intelligence,Shaanxi University of Science&Technology,Xi'an 710021,China)
出处
《液晶与显示》
CAS
CSCD
北大核心
2023年第3期368-377,共10页
Chinese Journal of Liquid Crystals and Displays
基金
国家自然科学基金(No.61971272,No.61601271)
国家重点研发重点专项(No.2019YFC1520204)。
关键词
高光谱图像
混合卷积网络
多尺度特征
注意力机制
hyperspectral image
hybrid convolutional network
multi-scale features
attention mechanism
