Efficient remote sensing image classification using the novel STConvNeXt convolutional network

Abstract Remote sensing images present formidable classification challenges due to their complex spatial organization, high inter-class similarity, and significant intra-class variability. To address the balance between computational efficiency and feature extraction capability in existing methods,...

Full description

Saved in:
Bibliographic Details
Main Authors: Bo Liu, Chenmei Zhan, Cheng Guo, Xiaobo Liu, Shufen Ruan
Format: Article
Language:English
Published: Nature Portfolio 2025-03-01
Series:Scientific Reports
Subjects:
Online Access:https://doi.org/10.1038/s41598-025-92629-x
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850023317612789760
author Bo Liu
Chenmei Zhan
Cheng Guo
Xiaobo Liu
Shufen Ruan
author_facet Bo Liu
Chenmei Zhan
Cheng Guo
Xiaobo Liu
Shufen Ruan
author_sort Bo Liu
collection DOAJ
description Abstract Remote sensing images present formidable classification challenges due to their complex spatial organization, high inter-class similarity, and significant intra-class variability. To address the balance between computational efficiency and feature extraction capability in existing methods, this paper innovatively proposes a lightweight convolutional network, STConvNeXt. In its architectural design, the model incorporates a split-based mobile convolution module with a hierarchical tree structure. It employs parameterized depthwise separable convolutions to reduce computational complexity and constructs a multi-level feature tree to facilitate cross-scale feature fusion. For feature enhancement, a fast pyramid pooling module replaces the traditional spatial pyramid structure, effectively reducing the number of parameters while preserving large-scale contextual awareness. In terms of training strategy, a dynamic threshold loss function is introduced, utilizing a learnable inter-class margin to improve the model’s ability to distinguish difficult-to-classify samples. Systematic experiments on the UCMerced, AID, and NWPU-RESISC45 benchmark datasets validate the effectiveness of the proposed approach: compared with the ConvNeXt baseline, STConvNeXt reduces both parameter count (by 56.49%) and FLOPs (by 49.89%), while improving classification accuracy by 1.2–2.7%. Furthermore, compared with the current state-of-the-art remote sensing scene classification models, our method still exhibits significant advantages. Ablation studies further confirm the effectiveness of each module design, particularly demonstrating that the model maintains excellent classification accuracy despite a substantial reduction in parameters.
format Article
id doaj-art-bfa4baa67c4f4101bb1e5b1fd6521f8f
institution DOAJ
issn 2045-2322
language English
publishDate 2025-03-01
publisher Nature Portfolio
record_format Article
series Scientific Reports
spelling doaj-art-bfa4baa67c4f4101bb1e5b1fd6521f8f2025-08-20T03:01:23ZengNature PortfolioScientific Reports2045-23222025-03-0115111810.1038/s41598-025-92629-xEfficient remote sensing image classification using the novel STConvNeXt convolutional networkBo Liu0Chenmei Zhan1Cheng Guo2Xiaobo Liu3Shufen Ruan4Mathematical and Physical Sciences, Wuhan Textile UniversityMathematical and Physical Sciences, Wuhan Textile UniversityMathematical and Physical Sciences, Wuhan Textile UniversityAutomation, China University of GeoscienceMathematical and Physical Sciences, Wuhan Textile UniversityAbstract Remote sensing images present formidable classification challenges due to their complex spatial organization, high inter-class similarity, and significant intra-class variability. To address the balance between computational efficiency and feature extraction capability in existing methods, this paper innovatively proposes a lightweight convolutional network, STConvNeXt. In its architectural design, the model incorporates a split-based mobile convolution module with a hierarchical tree structure. It employs parameterized depthwise separable convolutions to reduce computational complexity and constructs a multi-level feature tree to facilitate cross-scale feature fusion. For feature enhancement, a fast pyramid pooling module replaces the traditional spatial pyramid structure, effectively reducing the number of parameters while preserving large-scale contextual awareness. In terms of training strategy, a dynamic threshold loss function is introduced, utilizing a learnable inter-class margin to improve the model’s ability to distinguish difficult-to-classify samples. Systematic experiments on the UCMerced, AID, and NWPU-RESISC45 benchmark datasets validate the effectiveness of the proposed approach: compared with the ConvNeXt baseline, STConvNeXt reduces both parameter count (by 56.49%) and FLOPs (by 49.89%), while improving classification accuracy by 1.2–2.7%. Furthermore, compared with the current state-of-the-art remote sensing scene classification models, our method still exhibits significant advantages. Ablation studies further confirm the effectiveness of each module design, particularly demonstrating that the model maintains excellent classification accuracy despite a substantial reduction in parameters.https://doi.org/10.1038/s41598-025-92629-xConvolutional neural networksDeep learningRemote sensingSMConvTree structures
spellingShingle Bo Liu
Chenmei Zhan
Cheng Guo
Xiaobo Liu
Shufen Ruan
Efficient remote sensing image classification using the novel STConvNeXt convolutional network
Scientific Reports
Convolutional neural networks
Deep learning
Remote sensing
SMConv
Tree structures
title Efficient remote sensing image classification using the novel STConvNeXt convolutional network
title_full Efficient remote sensing image classification using the novel STConvNeXt convolutional network
title_fullStr Efficient remote sensing image classification using the novel STConvNeXt convolutional network
title_full_unstemmed Efficient remote sensing image classification using the novel STConvNeXt convolutional network
title_short Efficient remote sensing image classification using the novel STConvNeXt convolutional network
title_sort efficient remote sensing image classification using the novel stconvnext convolutional network
topic Convolutional neural networks
Deep learning
Remote sensing
SMConv
Tree structures
url https://doi.org/10.1038/s41598-025-92629-x
work_keys_str_mv AT boliu efficientremotesensingimageclassificationusingthenovelstconvnextconvolutionalnetwork
AT chenmeizhan efficientremotesensingimageclassificationusingthenovelstconvnextconvolutionalnetwork
AT chengguo efficientremotesensingimageclassificationusingthenovelstconvnextconvolutionalnetwork
AT xiaoboliu efficientremotesensingimageclassificationusingthenovelstconvnextconvolutionalnetwork
AT shufenruan efficientremotesensingimageclassificationusingthenovelstconvnextconvolutionalnetwork