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

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. Howev...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Pei Zhang, Guoliang Fan, Chanyue Wu, Dong Wang, Ying Li
Formato: article
Lenguaje:EN
Publicado: MDPI AG 2021
Materias:
Q
Acceso en línea:https://doaj.org/article/4991804dda80475e995865ceb5225731
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
id oai:doaj.org-article:4991804dda80475e995865ceb5225731
record_format dspace
spelling oai:doaj.org-article:4991804dda80475e995865ceb52257312021-11-11T18:49:46ZTask-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification10.3390/rs132142002072-4292https://doaj.org/article/4991804dda80475e995865ceb52257312021-10-01T00:00:00Zhttps://www.mdpi.com/2072-4292/13/21/4200https://doaj.org/toc/2072-4292The 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 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 Dynamic Kernel Fusion Network (DKF-Net) 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-adaptive 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.Pei ZhangGuoliang FanChanyue WuDong WangYing LiMDPI AGarticleremote-sensing classificationscene classificationfew-shot learningmeta-learningvision transformersmultiscale feature fusionScienceQENRemote Sensing, Vol 13, Iss 4200, p 4200 (2021)
institution DOAJ
collection DOAJ
language EN
topic remote-sensing classification
scene classification
few-shot learning
meta-learning
vision transformers
multiscale feature fusion
Science
Q
spellingShingle remote-sensing classification
scene classification
few-shot learning
meta-learning
vision transformers
multiscale feature fusion
Science
Q
Pei Zhang
Guoliang Fan
Chanyue Wu
Dong Wang
Ying Li
Task-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification
description 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 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 Dynamic Kernel Fusion Network (DKF-Net) 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-adaptive 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.
format article
author Pei Zhang
Guoliang Fan
Chanyue Wu
Dong Wang
Ying Li
author_facet Pei Zhang
Guoliang Fan
Chanyue Wu
Dong Wang
Ying Li
author_sort Pei Zhang
title Task-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification
title_short Task-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification
title_full Task-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification
title_fullStr Task-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification
title_full_unstemmed Task-Adaptive Embedding Learning with Dynamic Kernel Fusion for Few-Shot Remote Sensing Scene Classification
title_sort task-adaptive embedding learning with dynamic kernel fusion for few-shot remote sensing scene classification
publisher MDPI AG
publishDate 2021
url https://doaj.org/article/4991804dda80475e995865ceb5225731
work_keys_str_mv AT peizhang taskadaptiveembeddinglearningwithdynamickernelfusionforfewshotremotesensingsceneclassification
AT guoliangfan taskadaptiveembeddinglearningwithdynamickernelfusionforfewshotremotesensingsceneclassification
AT chanyuewu taskadaptiveembeddinglearningwithdynamickernelfusionforfewshotremotesensingsceneclassification
AT dongwang taskadaptiveembeddinglearningwithdynamickernelfusionforfewshotremotesensingsceneclassification
AT yingli taskadaptiveembeddinglearningwithdynamickernelfusionforfewshotremotesensingsceneclassification
_version_ 1718431699259359232