多光谱图像
计算机视觉
目标检测
人工智能
计算机科学
遥感
融合
传感器融合
图像融合
对象(语法)
多光谱模式识别
合成孔径雷达
特征提取
变更检测
模式识别(心理学)
图像分割
航空影像
雷达跟踪器
高光谱成像
图像处理
雷达成像
可视化
作者
Qi Cao,Anan Du,Jue Zhang,Zhigang Lian,Yanqin Ma,Shuchao Pang,Fengchao Xiong
标识
DOI:10.1109/tgrs.2026.3751730
摘要
The integration of multispectral imagery, particularly visible (RGB) and infrared (IR) modalities, has attracted increasing attention in aerial object detection. However, complementary target cues in RGB and IR images are often entangled with redundant responses, while semantic inconsistency across feature channels hinders effective cross-modal fusion, ultimately limiting detection performance. To address these challenges, we propose SCPFusion, a Synergistic Complementary Progressive Fusion framework for multispectral aerial object detection. SCPFusion first models modality-complementary information and subsequently establishes semantic correspondence across heterogeneous feature representations before fusion. Specifically, a Cross-Modal Spatial Decoupling Module (CMSDM) is developed to jointly model global modality reliability and local complementary compensation. By adaptively adjusting modality contributions according to scene-level illumination conditions and preserving high-confidence modality-exclusive target cues, CMSDM constructs reliable and complementary representations for subsequent cross-modal interaction. Building upon these representations, a Channel-aligned Feature Aggregation Module (CFAM) is introduced to establish semantic correspondence across heterogeneous RGB-IR feature channels and progressively integrate aligned primary representations with complementary information. Through the synergy between complementary-information modeling and semantic channel alignment, SCP-Fusion effectively bridges the cross-modal representation gap arising from the heterogeneous sensing mechanisms of RGB and IR modalities, thereby substantially improving detection performance in complex aerial environments. Extensive experiments on benchmark datasets, including DroneVehicle, VEDAI, and ES-CVehicle, demonstrate that SCPFusion consistently outperforms existing state-of-the-art methods.
科研通智能强力驱动
Strongly Powered by AbleSci AI