Increasing the Effectiveness of Active Learning: Introducing Artificial Data Generation in Active Learning for Land Use/Land Cover Classification

Abstract

In remote sensing, Active Learning (AL) has become an important technique to collect informative ground truth data “on-demand” for supervised classification tasks. Despite its effectiveness, it is still significantly reliant on user interaction, which makes it both expensive and time consuming to implement. Most of the current literature focuses on the optimization of AL by modifying the selection criteria and the classifiers used. Although improvements in these areas will result in more effective data collection, the use of artificial data sources to reduce human–computer interaction remains unexplored. In this paper, we introduce a new component to the typical AL framework, the data generator, a source of artificial data to reduce the amount of user-labeled data required in AL. The implementation of the proposed AL framework is done using Geometric SMOTE as the data generator. We compare the new AL framework to the original one using similar acquisition functions and classifiers over three AL-specific performance metrics in seven benchmark datasets. We show that this modification of the AL framework significantly reduces cost and time requirements for a successful AL implementation in all of the datasets used in the experiment.

Improving Active Learning Performance through the Use of Data Augmentation

“…In addition, the resulting classifcation performance of AL-trained classifers is frequently inconsistent and marginally improves the classifcation performance when compared to classifers trained over the entire training set. In addition, there is also signifcant variability in the data selection efciency during diferent runs of the AL iterative process [20].…”

“…(2) Generalization of the generator module proposed in [20] from oversampling techniques to any other data augmentation mechanism and/or policy.…”

“…Tis process is shown in Figure 2. Before the start of the iterative process, assuming D t�0 lab � ∅, the data used to populate D t�1 lab are typically collected randomly from D � D t�0 pool and are labeled by a supervisor [20][21][22].…”

“…Learning. Te only AL model found that uses data augmentation outside of the computer vision or NLP domains implements a pipelined approach, described in [20]. In this study, the AL model proposed was applied for tabular data using an oversampling data augmentation policy (i.e., the artifcial data was only generated to balance the target class frequencies).…”

Research Trends and Applications of Data Augmentation Algorithms

“…In addition, in contexts of limited data availability (i.e., small datasets), deep learning approaches are not appropriate since the amount of parameters to be tuned during the learning phase often exceeds the number of observations in the dataset, making it overparametrized. The augmentation of small datasets using heuristic approaches is explored in an Active Learning context in [30]. The authors found that significantly smaller amounts of curated data using Active Learning, along with heuristic data augmentation methods, achieved a classification performance comparable to classifiers trained over the full dataset.…”