RS-SSKD: Self-Supervision Equipped with Knowledge Distillation for Few-Shot Remote Sensing Scene Classification

While growing instruments generate more and more airborne or satellite images, the bottleneck in remote sensing (RS) scene classification has shifted from data limits toward a lack of ground truth samples. There are still many challenges when we are facing unknown environments, especially those with insufficient training data. Few-shot classification offers a different picture under the umbrella of meta-learning: digging rich knowledge from a few data are possible. In this work, we propose a method named RS-SSKD for few-shot RS scene classification from a perspective of generating powerful representation for the downstream meta-learner. Firstly, we propose a novel two-branch network that takes three pairs of original-transformed images as inputs and incorporates Class Activation Maps (CAMs) to drive the network mining, the most relevant category-specific region. This strategy ensures that the network generates discriminative embeddings. Secondly, we set a round of self-knowledge distillation to prevent overfitting and boost the performance. Our experiments show that the proposed method surpasses current state-of-the-art approaches on two challenging RS scene datasets: NWPU-RESISC45 and RSD46-WHU. Finally, we conduct various ablation experiments to investigate the effect of each component of the proposed method and analyze the training time of state-of-the-art methods and ours.

Download Full-text

Few-shot Classification of Aerial Scene Images via Meta-learning

10.20944/preprints202010.0033.v1 ◽

2020 ◽

Cited By ~ 1

Author(s):

Pei Zhang ◽

YIng Li ◽

Dong Wang ◽

Yunpeng Bai

Keyword(s):

Classification Problem ◽

Training Data ◽

Scene Classification ◽

Shot Classification ◽

Training Stage ◽

Testing Stage ◽

Meta Learning ◽

Cosine Distance ◽

Query Sample

CNN-based methods have dominated the field of aerial scene classification for the past few years. While achieving remarkable success, CNN-based methods suffer from excessive parameters and notoriously rely on large amounts of training data. In this work, we introduce few-shot learning to the aerial scene classification problem. Few-shot learning aims to learn a model on base-set that can quickly adapt to unseen categories in novel-set, using only a few labeled samples. To this end, we proposed a meta-learning method for few-shot classification of aerial scene images. First, we train a feature extractor on all base categories to learn a representation of inputs. Then in the meta-training stage, the classifier is optimized in the metric space by cosine distance with a learnable scale parameter. At last, in the meta-testing stage, the query sample in the unseen category is predicted by the adapted classifier given a few support samples. We conduct extensive experiments on two challenging datasets: NWPU-RESISC45 and RSD46-WHU. The experimental results show that our method outperforms three state-of-the-art few-shot algorithms and one typical CNN-based method, D-CNN. Furthermore, several ablation experiments are conducted to investigate the effects of dataset scale and support shots; the experiment results confirm that our model is specifically effective in few-shot settings.

Download Full-text

Meta-FSEO: A Meta-Learning Fast Adaptation with Self-Supervised Embedding Optimization for Few-Shot Remote Sensing Scene Classification

Remote Sensing ◽

10.3390/rs13142776 ◽

2021 ◽

Vol 13 (14) ◽

pp. 2776

Author(s):

Yong Li ◽

Zhenfeng Shao ◽

Xiao Huang ◽

Bowen Cai ◽

Song Peng

Keyword(s):

Remote Sensing ◽

Deep Learning ◽

Remote Sensing Data ◽

Scene Classification ◽

Generalization Capability ◽

Urban Scenes ◽

Shot Classification ◽

Meta Learning ◽

Land Surfaces ◽

Fast Adaptation

The performance of deep learning is heavily influenced by the size of the learning samples, whose labeling process is time consuming and laborious. Deep learning algorithms typically assume that the training and prediction data are independent and uniformly distributed, which is rarely the case given the attributes and properties of different data sources. In remote sensing images, representations of urban land surfaces can vary across regions and by season, demanding rapid generalization of these surfaces in remote sensing data. In this study, we propose Meta-FSEO, a novel model for improving the performance of few-shot remote sensing scene classification in varying urban scenes. The proposed Meta-FSEO model deploys self-supervised embedding optimization for adaptive generalization in new tasks such as classifying features in new urban regions that have never been encountered during the training phase, thus balancing the requirements for feature classification tasks between multiple images collected at different times and places. We also created a loss function by weighting the contrast losses and cross-entropy losses. The proposed Meta-FSEO demonstrates a great generalization capability in remote sensing scene classification among different cities. In a five-way one-shot classification experiment with the Sentinel-1/2 Multi-Spectral (SEN12MS) dataset, the accuracy reached 63.08%. In a five-way five-shot experiment on the same dataset, the accuracy reached 74.29%. These results indicated that the proposed Meta-FSEO model outperformed both the transfer learning-based algorithm and two popular meta-learning-based methods, i.e., MAML and Meta-SGD.

Download Full-text

Task-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification

10.20944/preprints202108.0389.v1 ◽

2021 ◽

Author(s):

Pei Zhang ◽

Guoliang Fan ◽

Chanyue Wu ◽

Dong Wang ◽

Ying Li

Keyword(s):

Remote Sensing ◽

State Of The Art ◽

Adaptive Strategy ◽

Experimental Results ◽

Similarity Metrics ◽

Scene Classification ◽

Dual Roles ◽

Kernel Fusion ◽

Adaptive Embedding ◽

Meta Learning

The central goal of few-shot scene classification is to learn a model that can generalize well to a novel scene category (UNSEEN) from only one or a few labeled examples. Recent works in the remote sensing (RS) community tackle this challenge by developing algorithms in a meta-learning manner. However, most prior approaches have either focused on rapidly optimizing a meta-learner or aimed at finding good similarity metrics while overlooking the embedding power. Here we propose a novel Task-Adaptive Embedding Learning (TAEL) framework that complements the existing methods by giving full play to feature embedding’s dual roles in few-shot scene classification - representing images and constructing classifiers in the embedding space. First, we design a lightweight network that enriches the diversity and expressive capacity of embeddings by dynamically fusing information from multiple kernels. Second, we present a task-adaptive strategy that helps to generate more discriminative representations by transforming the universal embeddings into task-specific embeddings via a self-attention mechanism. We evaluate our model in the standard few-shot learning setting on two challenging datasets: NWPU-RESISC4 and RSD46-WHU. Experimental results demonstrate that, on all tasks, our method achieves state-of-the-art performance by a significant margin.

Download Full-text

Few-Shot Classification of Aerial Scene Images via Meta-Learning

Remote Sensing ◽

10.3390/rs13010108 ◽

2020 ◽

Vol 13 (1) ◽

pp. 108

Author(s):

Pei Zhang ◽

Yunpeng Bai ◽

Dong Wang ◽

Bendu Bai ◽

Ying Li

Keyword(s):

Classification Problem ◽

Training Data ◽

Scene Classification ◽

Shot Classification ◽

Training Stage ◽

Testing Stage ◽

Meta Learning ◽

Cosine Distance ◽

The Impact

Convolutional neural network (CNN) based methods have dominated the field of aerial scene classification for the past few years. While achieving remarkable success, CNN-based methods suffer from excessive parameters and notoriously rely on large amounts of training data. In this work, we introduce few-shot learning to the aerial scene classification problem. Few-shot learning aims to learn a model on base-set that can quickly adapt to unseen categories in novel-set, using only a few labeled samples. To this end, we proposed a meta-learning method for few-shot classification of aerial scene images. First, we train a feature extractor on all base categories to learn a representation of inputs. Then in the meta-training stage, the classifier is optimized in the metric space by cosine distance with a learnable scale parameter. At last, in the meta-testing stage, the query sample in the unseen category is predicted by the adapted classifier given a few support samples. We conduct extensive experiments on two challenging datasets: NWPU-RESISC45 and RSD46-WHU. The experimental results show that our method yields state-of-the-art performance. Furthermore, several ablation experiments are conducted to investigate the effects of dataset scale, the impact of different metrics and the number of support shots; the experiment results confirm that our model is specifically effective in few-shot settings.

Download Full-text

Improved SinGAN Integrated with an Attentional Mechanism for Remote Sensing Image Classification

Remote Sensing ◽

10.3390/rs13091713 ◽

2021 ◽

Vol 13 (9) ◽

pp. 1713

Author(s):

Songwei Gu ◽

Rui Zhang ◽

Hongxia Luo ◽

Mengyao Li ◽

Huamei Feng ◽

...

Keyword(s):

Remote Sensing ◽

Real Life ◽

Attention Mechanism ◽

Training Data ◽

Generative Adversarial Networks ◽

Natural Image ◽

Remote Sensing Images ◽

Training Time ◽

Adversarial Networks ◽

Remote Sensing Image Classification

Deep learning is an important research method in the remote sensing field. However, samples of remote sensing images are relatively few in real life, and those with markers are scarce. Many neural networks represented by Generative Adversarial Networks (GANs) can learn from real samples to generate pseudosamples, rather than traditional methods that often require more time and man-power to obtain samples. However, the generated pseudosamples often have poor realism and cannot be reliably used as the basis for various analyses and applications in the field of remote sensing. To address the abovementioned problems, a pseudolabeled sample generation method is proposed in this work and applied to scene classification of remote sensing images. The improved unconditional generative model that can be learned from a single natural image (Improved SinGAN) with an attention mechanism can effectively generate enough pseudolabeled samples from a single remote sensing scene image sample. Pseudosamples generated by the improved SinGAN model have stronger realism and relatively less training time, and the extracted features are easily recognized in the classification network. The improved SinGAN can better identify sub-jects from images with complex ground scenes compared with the original network. This mechanism solves the problem of geographic errors of generated pseudosamples. This study incorporated the generated pseudosamples into training data for the classification experiment. The result showed that the SinGAN model with the integration of the attention mechanism can better guarantee feature extraction of the training data. Thus, the quality of the generated samples is improved and the classification accuracy and stability of the classification network are also enhanced.

Download Full-text

A Multi-Branch Feature Fusion Strategy Based on an Attention Mechanism for Remote Sensing Image Scene Classification

Remote Sensing ◽

10.3390/rs13101950 ◽

2021 ◽

Vol 13 (10) ◽

pp. 1950

Author(s):

Cuiping Shi ◽

Xin Zhao ◽

Liguo Wang

Keyword(s):

Remote Sensing ◽

Feature Extraction ◽

Classification Accuracy ◽

Feature Fusion ◽

State Of The Art ◽

Rapid Development ◽

Remote Sensing Image ◽

Classification Performance ◽

Attention Mechanism ◽

Scene Classification

In recent years, with the rapid development of computer vision, increasing attention has been paid to remote sensing image scene classification. To improve the classification performance, many studies have increased the depth of convolutional neural networks (CNNs) and expanded the width of the network to extract more deep features, thereby increasing the complexity of the model. To solve this problem, in this paper, we propose a lightweight convolutional neural network based on attention-oriented multi-branch feature fusion (AMB-CNN) for remote sensing image scene classification. Firstly, we propose two convolution combination modules for feature extraction, through which the deep features of images can be fully extracted with multi convolution cooperation. Then, the weights of the feature are calculated, and the extracted deep features are sent to the attention mechanism for further feature extraction. Next, all of the extracted features are fused by multiple branches. Finally, depth separable convolution and asymmetric convolution are implemented to greatly reduce the number of parameters. The experimental results show that, compared with some state-of-the-art methods, the proposed method still has a great advantage in classification accuracy with very few parameters.

Download Full-text

Multi-Scale Meta-Learning-Based Networks for High-Resolution Remote Sensing Scene Classification

10.1109/igarss47720.2021.9555134 ◽

2021 ◽

Author(s):

Xu Tang ◽

Weiquan Lin ◽

Chao Liu ◽

Xiao Han ◽

Wenjing Wang ◽

...

Keyword(s):

Remote Sensing ◽

High Resolution ◽

Scene Classification ◽

Multi Scale ◽

Meta Learning

Download Full-text

PulseNetOne: Fast Unsupervised Pruning of Convolutional Neural Networks for Remote Sensing

Remote Sensing ◽

10.3390/rs12071092 ◽

2020 ◽

Vol 12 (7) ◽

pp. 1092

Author(s):

David Browne ◽

Michael Giering ◽

Steven Prestwich

Keyword(s):

Remote Sensing ◽

Neural Networks ◽

Deep Learning ◽

Convolutional Neural Networks ◽

Data Augmentation ◽

Recognition Task ◽

Scene Recognition ◽

Training Data ◽

Learning Approach ◽

Scene Classification

Scene classification is an important aspect of image/video understanding and segmentation. However, remote-sensing scene classification is a challenging image recognition task, partly due to the limited training data, which causes deep-learning Convolutional Neural Networks (CNNs) to overfit. Another difficulty is that images often have very different scales and orientation (viewing angle). Yet another is that the resulting networks may be very large, again making them prone to overfitting and unsuitable for deployment on memory- and energy-limited devices. We propose an efficient deep-learning approach to tackle these problems. We use transfer learning to compensate for the lack of data, and data augmentation to tackle varying scale and orientation. To reduce network size, we use a novel unsupervised learning approach based on k-means clustering, applied to all parts of the network: most network reduction methods use computationally expensive supervised learning methods, and apply only to the convolutional or fully connected layers, but not both. In experiments, we set new standards in classification accuracy on four remote-sensing and two scene-recognition image datasets.

Download Full-text

Effectiveness, Explainability and Reliability of Machine Meta-Learning Methods for Predicting Mortality in Patients with COVID-19: Results of the Brazilian COVID-19 Registry

10.1101/2021.11.01.21265527 ◽

2021 ◽

Author(s):

Bruno Barbosa Miranda de Paiva ◽

Polianna Delfino Pereira ◽

Claudio Moises Valiense de Andrade ◽

Virginia Mara Reis Gomes ◽

Maria Clara Pontello Barbosa Lima ◽

...

Keyword(s):

Machine Learning ◽

Prediction Models ◽

State Of The Art ◽

Laboratory Data ◽

Machine Learning Algorithms ◽

Training Data ◽

Learning Models ◽

Learning Methods ◽

Meta Learning ◽

Machine Learning Models

Objective: To provide a thorough comparative study among state ofthe art machine learning methods and statistical methods for determining in-hospital mortality in COVID 19 patients using data upon hospital admission; to study the reliability of the predictions of the most effective methods by correlating the probability of the outcome and the accuracy of the methods; to investigate how explainable are the predictions produced by the most effective methods. Materials and Methods: De-identified data were obtained from COVID 19 positive patients in 36 participating hospitals, from March 1 to September 30, 2020. Demographic, comorbidity, clinical presentation and laboratory data were used as training data to develop COVID 19 mortality prediction models. Multiple machine learning and traditional statistics models were trained on this prediction task using a folded cross validation procedure, from which we assessed performance and interpretability metrics. Results: The Stacking of machine learning models improved over the previous state of the art results by more than 26% in predicting the class of interest (death), achieving 87.1% of AUROC and macroF1 of 73.9%. We also show that some machine learning models can be very interpretable and reliable, yielding more accurate predictions while providing a good explanation for the why. Conclusion: The best results were obtained using the meta learning ensemble model Stacking. State of the art explainability techniques such as SHAP values can be used to draw useful insights into the patterns learned by machine-learning algorithms. Machine learning models can be more explainable than traditional statistics models while also yielding highly reliable predictions. Key words: COVID-19; prognosis; prediction model; machine learning

Download Full-text

Convolutional Neural Networks with Deep Supervised Feature Learning for Remote Sensing Scene Classification

10.20944/preprints202008.0113.v1 ◽

2020 ◽

Author(s):

Grigorios Tsagkatakis ◽

Panagiotis Tsakalides

Keyword(s):

Remote Sensing ◽

Feature Learning ◽

Ground Truth ◽

Classification Performance ◽

Cross Entropy ◽

Scene Classification ◽

Feature Representations ◽

Benchmark Datasets ◽

Low Dimensional ◽

Fully Connected

State-of-the-art remote sensing scene classification methods employ different Convolutional Neural Network architectures for achieving very high classification performance. A trait shared by the majority of these methods is that the class associated with each example is ascertained by examining the activations of the last fully connected layer, and the networks are trained to minimize the cross-entropy between predictions extracted from this layer and ground-truth annotations. In this work, we extend this paradigm by introducing an additional output branch which maps the inputs to low dimensional representations, effectively extracting additional feature representations of the inputs. The proposed model imposes additional distance constrains on these representations with respect to identified class representatives, in addition to the traditional categorical cross-entropy between predictions and ground-truth. By extending the typical cross-entropy loss function with a distance learning function, our proposed approach achieves significant gains across a wide set of benchmark datasets in terms of classification, while providing additional evidence related to class membership and classification confidence.

Download Full-text