Many supervised Machine learning (ML) and Deep learning (DL) algorithms succeeded in the Hyperspectral image classification for various applications. Scientific findings reveal that supervised learning methods' performance heavily depends on training set size, i.e., labelled by the ground truth ...
1.Variational Adversarial Active Learning 提出了一种 pool-based 半监督主动学习算法,主要是一种对抗学习的方式:通过一个 variational autoencoder (VAE) 提取图片特征,一个判别网络判断图片是标注过的还是未标注的,VAE 希望欺骗判别网络对所有样本都判断为 labeled 数据,然而判别网络希望要准确分辨出 data pool 中的...
文献“Multi-class active learning for image classification(2009)”提出了基于最优标号和次优标号的准则(BvSB),考虑样例所属概率最高的前2个类别,忽略剩余类别对样例选择标准产生的干扰。 文献“基于主动学习和半监督学习的多类图像分类(2011)”将BvSB和带约束的自学习(Constrained self-training,CST)引入到基于SV...
(1)根据 labeled sample 生成更多 new sample, 目标是用尽可能少的labeled sample来train 网络,并获得尽可能多的accuracy. 创新点: 可能存在的不足: (1)data augmentation的不足? (2)mask的生成过程?随机噪声生成的不足? (3)original labeled sample 对accuracy的影响比较大,(希望通过对网络部分的改进,解决此问...
It is noteworthy that DifABAL exhibits a certain degree of superiority over iterative active learning. Similar findings have been reported in the domains of medical segmentation [5,27], natural language processing [52], and image classification [35,36,53]. The literature [20] provides theoretical...
Adaptive Active Learning for Image ClassificationXin LiYuhong GuoDepartment of Computer and Information SciencesTemple UniversityPhiladelphia, PA 19122{xinli,yuhong}@temple.eduAbstractRecently active learning has attracted a lot of attentionin computer vision field, as it is time and cost consumingto ...
Bayesian Active Learning for Classification and Preference Learning(论文2011年)通过贪婪地找到一个能使当前模型熵最大程度减少的数据点x,但由于模型参数维度很高,直接求解困难,因此在给定数据D和新增数据点x条件下,模型预测和模型参数之间的互信息。 Deep Bayesian Active Learning with Image Data(论文,代码2017年)...
Bayesian Active Learning for Classification and Preference Learning(论文2011年)通过贪婪地找到一个能使当前模型熵最大程度减少的数据点x,但由于模型参数维度很高,直接求解困难,因此在给定数据D和新增数据点x条件下,模型预测和模型参数之间的互信息。 Deep Bayesian Active Learning with Image Data(论文,代码2017年)...
Recent successes in learning-based image classification, however, heavily rely on the large number of annotated training samples, which may require considerable human efforts. In this paper, we propose a novel active learning framework, which is capable of building a competitive classifier with optimal...
Active learning-based hyperspectral image classification: a reinforcement learning approach In the last few years, deep neural networks have been successful in classifying hyperspectral images (HSIs). However, training deep neural networks needs a... U Patel,V Patel - 《Journal of Supercomputing...