双层视图筛选下多视图主动学习的高光谱图像分类  

Hyperspectral image classification based on multi-view active learning with double-layer view screening

作　　者：陈立伟[1] 崔玉婕 房赫 佟志勇 CHEN Liwei;CUI Yujie;FANG He;TONG Zhiyong(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;Heilongjiang Provincial Military Command,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [2]黑龙江省军区,黑龙江哈尔滨150001

出　　处：《应用科技》2022年第1期27-32,38,共7页Applied Science and Technology

基　　金：国家自然科学基金项目(61675051).

摘　　要：针对高光谱图像分类已标记样本稀缺的问题,研究如何高效利用多视图选取更高质量的未标记样本。首先采用不同方向、尺度的3D-Gabor滤波器产生兼具空间-光谱信息的数据立方体;然后进行双层视图筛选:第1层筛选最具充分性的多视图,接着针对传统视图多样性筛选计算量大、耗时长的问题,提出在低维空间计算视图的多样性强度值DI,利用DI值从最具充分性的多视图中再次筛选多样性视图;最后,利用自适应最大不一致采样策略从未标记样本集挑选样本。将提出的方法在Indian Pines数据集和Salinas数据集上进行仿真实验,实验结果表明与传统算法相比,基于多样性强度值视图筛选方法在保持相似的分类精度同时,有效减少耗时,节省了大量时间成本。Aiming at the scarcity of labeled samples in hyperspectral image classification,this paper studies how to efficiently use multiple views to select highly informative unlabeled samples.Firstly,3D-Gabor filters with different directions and scales are used to generate cubes with spatial and spectral information.Then there is a stage for double-layer view screening.The first layer selects the most sufficient views.The diversity intensity value DI is calculated in the low-dimensional space to solve the problem of time consuming in traditional algorithm.The DI value is an indicator to screen the views from the most sufficient multiple views again.Finally,the adaptive maximum diversity sampling strategy(AMD)is used to select samples from unlabeled sample sets.Experiments are carried out on Indian Pines dataset and Salinas dataset,showing that the proposed method can effectively reduce time consumption while maintaining similar classification accuracy.

关 键 词：高光谱图像分类 多视图学习 主动学习 3D-Gabor 视图筛选 充分性 多样性 采样策略 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]