A Survey of Active Learning Algorithms for Supervised Remote Sensing Image Classification

By Devis Tuia, Michele Volpi, Loris Copa, Mikhail F. Kanevski, Jordi Muñoz-Marí. IEEE Journal on Selected Topics in Signal Processing, 2011

553 citations recorded by the source corpus.

Defining an efficient training set is one of the most delicate phases for the success of remote sensing image classification routines. The complexity of the problem, the limited temporal and financial resources, as well as the high intraclass variance can make an algorithm fail if it is trained with a suboptimal dataset. Active learning aims at building efficient training sets by iteratively improving the model performance through sampling. A user-defined heuristic ranks the unlabeled pixels according to a function of the uncertainty of their class membership and then the user is asked to provide labels for the most uncertain pixels. This paper reviews and tests the main families of active learning algorithms: committee, large margin, and posterior probability-based. For each of them, the most recent advances in the remote sensing community are discussed and some heuristics are detailed and tested. Several challenging remote sensing scenarios are considered, including very high spatial resolution and hyperspectral image classification. Finally, guidelines for choosing the good architecture are provided for new and/or unexperienced user.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/d10add2f135c6627f0aabf0685f5702ee02d7021

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Semisupervised Classification of Remote Sensing Images With Active Queries (2012, background): “Active learning is a recent field of research in remote sensing data analysis aiming at defining efficient training sets, usually by user–machine interaction (for a review of the literature in remote sensing, see [24]).”
Remote Sensing Image Scene Classification Meets Deep Learning: Challenges, Methods, Benchmarks, and Opportunities (2020, background): “sensing images at pixel level or subpixel level [23]–[25], through labeling each pixel in the remote sensing images with a semantic class, because the spatial resolution of remote sensing images is very low—the size of a pixel is similar to the sizes of the objects of interest [26].”
Transformers in Remote Sensing: A Survey (2022, background): “[13] compare and evaluate different active learning algorithms for supervised remote sensing image classification task.”
Automatic Feature Learning for Spatio-Spectral Image Classification With Sparse SVM (2014, background): “This type of validation is standard when considering active learning methods [49], [50], which sample the most informative samples (contrarily to features here) among a large amount of unlabeled pixels.”
Monitoring Land-Cover Changes: A Machine-Learning Perspective (2016, background): “In a different line of research, active learning approaches have been devised to operate in land-cover classification scenarios with a paucity of labeled data [25], [26], [33]–[35], where the emphasis is on parsimoniously requesting training labels from an expert annotator during the process of lear”
Domain Adaptation for the Classification of Remote Sensing Data: An Overview of Recent Advances (2016, methodology): “Surveys on AL methods applied to remote sensing can be found in [69]–[71].”
Advances in Hyperspectral Image Classification: Earth Monitoring with Statistical Learning Methods (2013, methodology): “ADVANCES IN HYPERSPECTRAL IMAGE CLASSIFICATION [EARTH MONITORING WITH STATISTICAL LEARNING METHODS] Gustavo Camps-Valls, Devis Tuia, Lorenzo Bruzzone and Jón Atli Benediktsson”
Towards better exploiting convolutional neural networks for remote sensing scene classification (2016, background): “This drawback makes almost impracticable to fully design and train a network from scratch for most remote sensing problems, since large datasets in this domain are unusual given that training data may require high costs with travel and other logistics [21, 22].”
Advances in Hyperspectral Image and Signal Processing: A Comprehensive Overview of the State of the Art (2017, background): “For a complete survey on the use of active learning for remote-sensing image analysis, see [126] and [127].”
AID: A Benchmark Data Set for Performance Evaluation of Aerial Scene Classification (2016, background): “In the literature, primary studies have devoted to classifying aerial images at pixel level, by assigning each pixel in an aerial image with a thematic class [43], [44].”
Accurate Annotation of Remote Sensing Images via Active Spectral Clustering with Little Expert Knowledge (2015, methodology): “To achieve this purpose, there are three common types of strategies for intelligent sampling to select new labeled samples from a candidate pool of unlabeled samples [24]: (1) large-margin-based methods [32], which select candidates lying within the margin of the current support vector machine (SVM)”
Comparison of Classification Algorithms and Training Sample Sizes in Urban Land Classification with Landsat Thematic Mapper Imagery (2014, background): “This kind of algorithms adds unlabeled samples as training samples from the sample pool through human-machine interaction [40].”