AUTHOR=Zhu Jinjing , Li Ling 

TITLE=Advancements in image classification for environmental monitoring using AI

JOURNAL=Frontiers in Environmental Science

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/environmental-science/articles/10.3389/fenvs.2025.1562287

DOI=10.3389/fenvs.2025.1562287

ISSN=2296-665X

ABSTRACT=IntroductionAccurate environmental image classification is essential for ecological monitoring, climate analysis, disaster detection, and sustainable resource management. However, traditional classification models face significant challenges, including high intra-class variability, overlapping class boundaries, imbalanced datasets, and environmental fluctuations caused by seasonal and lighting changes.MethodsTo overcome these limitations, this study introduces the Multi-Scale Attention-Based Environmental Classification Network (MABEC-Net), a novel deep learning framework that enhances classification accuracy, robustness, and scalability. MABEC-Net integrates multi-scale feature extraction, which enables the model to analyze both fine-grained local textures and broader environmental patterns. Spatial and channel attention mechanisms are incorporated to dynamically adjust feature importance, allowing the model to focus on key visual information while minimizing noise.In addition to the network architecture, we propose the Adaptive Environmental Training Strategy (AETS), a robust training framework designed to improve model generalization across diverse environmental datasets. AETS employs dynamic data augmentation to simulate real-world variations, domain-specific regularization to enhance feature consistency, and feedback-driven optimization to iteratively refine the model‘s performance based on real-time evaluation metrics.ResultsExtensive experiments conducted on multiple benchmark datasets demonstrate that MABEC-Net, in conjunction with AETS, significantly outperforms state-of-the-art models in terms of classification accuracy, robustness to domain shifts, and computational efficiency. DiscussionBy integrating advanced attention-based feature extraction with adaptive training strategies, this study establishes a cutting-edge AI-driven solution for large-scale environmental monitoring, ecological assessment, and sustainable resource management. Future research directions include optimizing computational efficiency for deployment in edge computing and resource-constrained environments, as well as extending the framework to multimodal environmental data sources, such as hyperspectral imagery and sensor networks.