信号处理2025，Vol.41Issue(8)：1303-1322,20.DOI:10.12466/xhcl.2025.08.001
基于深度学习的无人机高光谱图像智能解译方法综述
Review of Deep Learning Methods for Interpreting UAV-Borne Hyperspectral Imagery
摘要
Abstract
With the rapid development of unmanned aerial vehicle(UAV)platforms and hyperspectral imaging technolo-gies,UAV-based hyperspectral remote sensing has shown great promise in precision agriculture,ecological monitoring,and natural resource management.However,its practical application still faces several technical challenges,including unstable data acquisition,severe image distortion,and poor spectral consistency,necessitating systematic research and methodological optimization.In recent years,deep learning has gradually emerged as the core technological approach for intelligent interpretation in this field,owing to its advantages in automatic feature extraction and complex scenario modeling.To comprehensively outline recent research progress and technical trends,this paper presents a systematic re-view of UAV hyperspectral imaging systems,deep learning interpretation methods,and typical application areas.Spe-cial emphasis is placed on the evolution and implementation of deep learning in core tasks such as classification,seg-mentation,and target detection.Representative models-including convolutional neural networks,graph neural net-works,and transformers-are thoroughly analyzed.The study highlights the high processing complexity and significant uncertainty of UAV hyperspectral data,primarily caused by illumination variations,attitude instability,and background interference.Current approaches remain limited by poor adaptability to varied scenarios,high demands for real-time per-formance,and the cost of extensive data preprocessing in real-world applications.Future research should focus on effi-cient feature representation,lightweight model architectures,and improving robustness across diverse environments.By systematically organizing mainstream deep learning approaches,this review summarizes representative techniques and interpretation strategies for key tasks,including target detection,classification,and segmentation,offering a theoretical foundation for continued optimization and practical implementation of UAV hyperspectral image analysis.In addition,the discussion of public datasets provides practical references to support experimental validation and facilitate further re-search in the field.关键词
无人机平台/高光谱图像/分类/分割/目标检测Key words
UAV platform/hyperspectral image/classification/segmentation/object detection分类
信息技术与安全科学引用本文复制引用
李伟,宋璐杰,马晓虎..基于深度学习的无人机高光谱图像智能解译方法综述[J].信号处理,2025,41(8):1303-1322,20.基金项目
天基智能信息处理全国重点实验室基金(TJ-01-22-03) （TJ-01-22-03）
河南省通用航空技术重点实验室开放基金(ZHKF-230211)National Key Laboratory of Science and Technology on Space-Born Intelligent Information Processing(TJ-01-22-03) （ZHKF-230211）
Open Fund of Henan Key Laboratory of Gen eral Aviation Technology(ZHKF-230211) （ZHKF-230211）
