Multispectral Object Detection via Edge-Enhanced and Frequency-Aware Fusion Network

Multispectral object detection is essential for applications including autonomous driving, surveillance, security systems, and Earth observation using airborne or satellite platforms, where illumination conditions often vary dramatically. While significant progress has been made, existing approaches primarily focus on fusion strategies while neglecting in-depth exploration of modality-specific characteristics. To address these limitations, we propose a novel Edge-Enhanced and Frequency-Aware Fusion network (EFAF), an end-to-end framework that integrates visible and infrared modalities for robust multispectral object detection. Our framework leverages the strengths of edge-enhanced features, frequency-domain analysis, and multi-modal fusion to achieve superior performance. The key innovations of EFAF lie in its Edge Enhancement Module (EEM) and Frequency-Aware Fusion Module (FAFM). The EEM integrates Sobel and Laplacian operators with convolutional layers to enhance target contours and provide precise spatial guidance for fusion. Meanwhile, the FAFM employs the Discrete Wavelet Transform (DWT) to decompose features into multi-band frequency components and applies self-attention to refine global representations. It further incorporates dual cross-attention and channel attention to adaptively fuse complementary modality information. Extensive experiments on public datasets demonstrate that EFAF exhibits superior performance in visible-infrared multispectral object detection, representing a significant advancement for practical applications requiring reliable detection under varying conditions.