Unsupervised Representation Learning with Deep Convolutional Neural Network for Remote Sensing Images

Yu, Yang; Gong, Zhiqiang; Zhong, Ping; Shan, Jiaxin

doi:10.1007/978-3-319-71589-6_9

Cited by 131 publications

(80 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A GAN-based method for estimating arbitrary shaped occluded regions of sea surface temperature images by exploiting historical observations is proposed in [134]. The authors consider the deep convolutional generative adversarial network [135] in order to address the unavailability of observations due to cloud occlusion, where the generator network is tasked with producing realistic estimation while the discriminator network must classify the inputs as real or synthesized. Estimating of missing regions is subsequently carried out by trying to estimating the closest vector representation of the uncorrupted image through the minimization of a loss function encapsulating the ℓ2 norm on the available observation, the adversarial error and the deviation of an average values.…”

Section: Restorationmentioning

confidence: 99%

Survey of Deep-Learning Approaches for Remote Sensing Observation Enhancement

Tsagkatakis

Aidini

Fotiadou

et al. 2019

Sensors

111

View full text Add to dashboard Cite

Deep Learning, and Deep Neural Networks in particular, have established themselves as the new norm in signal and data processing, achieving state-of-the-art performance in image, audio, and natural language understanding. In remote sensing, a large body of research has been devoted to the application of deep learning for typical supervised learning tasks such as classification. Less yet equally important effort has also been allocated to addressing the challenges associated with the enhancement of low-quality observations from remote sensing platforms. Addressing such channels is of paramount importance, both in itself, since high-altitude imaging, environmental conditions, and imaging systems trade-offs lead to low-quality observation, as well as to facilitate subsequent analysis, such as classification and detection. In this paper, we provide a comprehensive review of deep-learning methods for the enhancement of remote sensing observations, focusing on critical tasks including single and multi-band super-resolution, denoising, restoration, pan-sharpening, and fusion, among others. In addition to the detailed analysis and comparison of recently presented approaches, different research avenues which could be explored in the future are also discussed.

show abstract

Section: Restorationmentioning

confidence: 99%

Survey of Deep-Learning Approaches for Remote Sensing Observation Enhancement

Tsagkatakis

Aidini

Fotiadou

et al. 2019

Sensors

111

View full text Add to dashboard Cite

show abstract

“…Recently, generative adversarial networks (GAN) [5] have gained a lot of attention due to their capability to generate complex data without explicitly modelling the probability density function. GAN models proved their power and flexibility by achieving state-of-the-art performance in multiple hard generation tasks, like plausible sample generation for datasets [15], realistic photograph generation [2], text-to-image synthesis [14], image-to-image translation [16], super-resolution [10] and many more.…”

Section: Generative Adversarial Networkmentioning

confidence: 99%

Learning to Generate Ambiguous Sequences

Iclănzan

Szilágyi

2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

In this paper, we experiment with methods for obtaining binary sequences with a random probability mass function and with low autocorrelation and use it to generate ambiguous outcomes. Outputs from a neural network are mixed and shuffled, resulting in binary sequences whose probability mass function is non-convergent, constantly moving and changing. Empirical comparison with algorithms that generate ambiguity shows that the sequences generated by the proposed method have a significantly lower serial dependence. Therefore, the method is useful in scenarios where observes can see and record the outcome of each draw sequentially, by hindering the ability to make useful statistical inferences.

show abstract

“…We propose to use a denoising autoencoder architecture as a discriminator in which the energy is a reconstruction error. Our experimental design selection is based on our attempt to get the pair of models to converge [14] and to exhibit more stable behavior than regular GANs during training [15].…”

Section: Generative Adversarial Network For Nlp Tasksmentioning

confidence: 99%

“…Even though there are a large variety of approaches to representation learning in general, the underlying concept is to learn some set of features from data, and then use these features to solve, for example, a separate (possibly unrelated) task for which we have a large number of labeled examples. As a result, the emergence of large-scale datasets, such as ImageNet [1], which contains 14,197,122 manually labeled images, has allowed the wider-spread use and popularity of convolutional neural networks (CNNs) even in the unrelated task of medical imaging. Currently, the majority of existing classifiers cannot perform as expected when the size of the training dataset is small.…”

Section: Introductionmentioning

confidence: 99%

“…Additionally, we decided to replace the standard probabilistic GAN that we cast into the energy model (using Gibbs distributions [12]), as has been proposed by the work of Kim and Bengio [13], and we decided to present the same Nash equilibrium as a standard GAN, but through a different and more generalized class of loss functionals, such as hinge loss. This experimental design selection is based on our attempt in get the pair of models to converge [14] and to exhibit more stable behavior than regular GANs during training [15].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Representations of Natural Language Texts with Generative Adversarial Networks at Document, Sentence, and Aspect Level

2018

View full text Add to dashboard Cite

The ability to learn robust, resizable feature representations from unlabeled data has potential applications in a wide variety of machine learning tasks. One way to create such representations is to train deep generative models that can learn to capture the complex distribution of real-world data. Generative adversarial network (GAN) approaches have shown impressive results in producing generative models of images, but relatively little work has been done on evaluating the performance of these methods for the learning representation of natural language, both in supervised and unsupervised settings at the document, sentence, and aspect level. Extensive research validation experiments were performed by leveraging the 20 Newsgroups corpus, the Movie Review (MR) Dataset, and the Finegrained Sentiment Dataset (FSD). Our experimental analysis suggests that GANs can successfully learn representations of natural language texts at all three aforementioned levels.

show abstract

Unsupervised Representation Learning with Deep Convolutional Neural Network for Remote Sensing Images

Cited by 131 publications

References 11 publications

Survey of Deep-Learning Approaches for Remote Sensing Observation Enhancement

Survey of Deep-Learning Approaches for Remote Sensing Observation Enhancement

Learning to Generate Ambiguous Sequences

Learning Representations of Natural Language Texts with Generative Adversarial Networks at Document, Sentence, and Aspect Level

Contact Info

Product

Resources

About