Unsupervised Deep Feature Extraction for Remote Sensing Image Classification

By Adriana Romero, Carlo Gatta, Gustau Camps-Valls. 2015

717 citations recorded by the source corpus.

This paper introduces the use of single-layer and deep convolutional networks for remote sensing data analysis. Direct application to multi- and hyperspectral imagery of supervised (shallow or deep) convolutional networks is very challenging given the high input data dimensionality and the relatively small amount of available labeled data. Therefore, we propose the use of greedy layerwise unsupervised pretraining coupled with a highly efficient algorithm for unsupervised learning of sparse features. The algorithm is rooted on sparse representations and enforces both population and lifetime sparsity of the extracted features, simultaneously. We successfully illustrate the expressive power of the extracted representations in several scenarios: classification of aerial scenes, as well as land-use classification in very high resolution or land-cover classification from multi- and hyperspectral images. The proposed algorithm clearly outperforms standard principal component analysis (PCA) and its kernel counterpart (kPCA), as well as current state-of-the-art algorithms of aerial classification, while being extremely computationally efficient at learning representations of data. Results show that single-layer convolutional networks can extract powerful discriminative features only when the receptive field accounts for neighboring pixels and are preferred when the classification requires high resolution and detailed results. However, deep architectures significantly outperform single-layer variants, capturing increasing levels of abstraction and complexity throughout the feature hierarchy.

Subfields: Atmospheric Science, Computer Vision and Pattern Recognition, Ecology, Media Technology

How researchers see this paper

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

In Atmospheric Science

In Computer Vision and Pattern Recognition

In Ecology

In Media Technology

Citations in the corpus

Remote Sensing Image Scene Classification: Benchmark and State of the Art (2017, methodology): “Then, we searched the keywords of "OBIA", "GEOBIA", "land cover classification", "land use classification", "geographic image retrieval", and "geospatial object detection" on Web of Science and Google Scholar to carefully select 15 meaningful scene classes.”
A hybrid MLP-CNN classifier for very fine resolution remotely sensed image classification (2017, methodology): “Similar to the MLP, the CNN is a network stacked into a number 170 of layers, where the output of the previous layer is connected sequentially to the input 171 of the next one by a set of learnable weights and biases (Romero et al., 2016).”
3-D Deep Learning Approach for Remote Sensing Image Classification (2018, background): “Autoencoders are also the basic concept behind the model introduced in [38], presenting a spatiospectral framework that”
Deep Learning for Remote Sensing Data: A Technical Tutorial on the State of the Art (2016, background): “These shallow models could be considered to stack into deep versions in a hierarchical manner [31], [106].”
Self-Supervised Learning in Remote Sensing: A review (2022, background): “, in scene classification [38], [40], [41]; semantic segmentation [42]–[44]; change detection [45], [46]; and multisensor data analysis [47], [48].”
Deep learning in remote sensing: a review (2017, background): “The authors of [35] propose an unsupervised convolutional network for learning spectral-spatial features using sparse learning to estimate the network weights in a greedy layerwise fashion instead of end-to-end learning.”
Deep Learning Meets Hyperspectral Image Analysis: A Multidisciplinary Review (2019, background): “In [194], Romero et al. presented a layer-wise unsupervised pre-training for CNN, which leads to both performance gains and improved computational efﬁciency.”
Hyperspectral Image Classification Using Deep Pixel-Pair Features (2017, methodology): “Furthermore, convolutional neural network (CNN) [23] was employed to exploit deep representation based on spectral signatures and the performance proved to be superior to that of SVM; unsupervised sparse features were learned via deep CNN in a greedy layerwise fashion for pixel classification in [24”
Learning to Diversify Deep Belief Networks for Hyperspectral Image Classification (2017, methodology): “The typical deep learning methods are based on the convolutional neural networks (CNNs) [20]–[24].”
Deep Learning for Hyperspectral Image Classification: An Overview (2019, background): “Recently, more and more research works [47], [97], [104], [105], [111], [116], [117] focus on designing a robust and effective unsupervised/semisupervised feature learning framework based on deep learning to classify HSIs.”
Deep Learning for Classification of Hyperspectral Data: A Comparative Review (2019, positive): “[53] introduced 2D multiscale convolutional autoencoders for simultaneous embedding learning of hyperspectral data, including both spatial and spectral information and classification, while [54] proposed the use of CNNs for unsupervised feature extraction: it performed dimension reduction of the spe”
Remote Sensing Image Scene Classification Meets Deep Learning: Challenges, Methods, Benchmarks, and Opportunities (2020, neutral): “scene classification methods have emerged [58]–[66], and made substantial progress for scene classification.”
Geography-Aware Self-Supervised Learning (2020, background): “Most of the studies utilize small-scale datasets specific to a small geographical region [3, 16, 29, 14, 18], a few classes [24] or a highly-specific modality, i.”
Multi-scale 3D deep convolutional neural network for hyperspectral image classification (2017, methodology): “In [5], an unsupervised CNN was proposed for remote sensing image processing, in which they used greedy layerwise strategy to train their model.”
Deep learning for remote sensing image classification: A survey (2018, methodology): “For example, the use of deep CNNs for RS image classification was introduced in Romero, Gatta, and Camps-Valls (2016) with the network being trained by an unsupervised method that seeks sparse feature representation.”