Infusion-Net: Inter- and Intra-Weighted Cross-Fusion Network for Multispectral Object Detection

Object recognition is conducted using red, green, and blue (RGB) images in object studies. However, RGB low-light environments or where other objects occlude the target cause poor performance. In contrast, infrared (IR) provide acceptable performance these because they detect IR waves rather than visible illumination. this paper, we propose an inter- intra-weighted cross-fusion network (Infusion-Net), which improves by combining strengths of RGB-IR image pairs. Infusion-Net connects dual detection models a high-frequency (HF) assistant (HFA) to combine advantages images. To extract HF components, HFA transforms input into discrete cosine transform domain. The extracted components are weighted via pretrained intra-weights for feature-domain cross-fusion. inter-weighted fused features transmitted each other’s networks complement limitations modality. also used enhance any insufficient objects. Thus, experimental results present superiority proposed improved multispectral task.