.png){kind=icon}
.png){kind=icon}
Return
Neutrosophic soft-computing ensemble for high-accuracy satellite image scene classification
.png){kind=icon}
OA
AI
E
M
S
V
B
A
J
DOI：10.3389/frsen.2026.1882437.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Satellite image scene classification is a fundamental task in remote sensing. It underpins land-use monitoring; urban planning; disaster response; and environmental management. Despite substantial progress through deep learning; complex aerial scenes remain challenging owing to high inter-class visual similarity; intra-class spatial variance; and inherent prediction uncertainty across heterogeneous model architectures. This paper proposes a novel triple-branch ensemble framework that combines three architecturally complementary deep learning backbones-ConvNeXt-Small; HRNet-W18; and Swin Transformer (Small) - to jointly exploit local texture hierarchies; high-resolution spatial representations; and global self-attention context. To advance beyond conventional probability averaging; three uncertainty-aware soft-computing fusion strategies are developed and compared: Normal Fuzzy logic; Intuitionistic Fuzzy logic; and Neutrosophic logic. The proposed Neutrosophic fusion decomposes each branch output into Truth; Indeterminacy; and Falsity components; explicitly down-weighting predictions characterised by high inter-branch disagreement. Experiments are conducted on the UC Merced Land Use Dataset (21 classes; 2; 100 images) using a stratified 80/20 split with a comprehensive eight-stage preprocessing pipeline and ImageNet transfer learning. The proposed Neutrosophic Ensemble achieves 99.29% accuracy and the highest precision of 99.31%; outperforming all individual backbones and simpler ensemble baselines. These results suggest that architectural complementarity combined with neutrosophic uncertainty modelling is a promising approach for satellite image scene classification on this benchmark; though validation on larger and more diverse datasets remains an important next step.
Keywords:
ensemble learning
transfer learning
swin transformer
neutrosophic logic
HRNet
convNeXt
satellite image classification
UC merced
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
F
IF:
3.7
Papers:
560
Citations:
993
Organization
Cited Papers