To improve the accuracy of multi-spectral semantic segmentation, an attention fusion network (AFNet) based on deep learning is proposed. Different from current methods, the AFNet uses a co-attention mechanism by designing an attention fusion module to calculate the spatial correlation between the red-green-blue (RGB) image and infrared (IR) image feature maps to guide the fusion of features from different spectra. This approach enhances the feature presentation and makes full use of the complementary characteristics of multi-spectral sources. The proposed network is tested on RGB-IR datasets and compared with relevant state-of-the-art networks. The experimental analyses prove that the proposed AFNet can improve multi-spectral semantic segmentation results with good visual definition and high accuracy in classification and localization.

为了提高多谱语义分割的准确性，提出了一种基于深度学习的注意力融合网络（AFNet）。与目前的方法不同，AFNet通过设计注意力融合模块来使用共同注意机制，以计算红绿蓝（RGB）图像和红外（IR）图像特征图之间的空间相关性，以指导融合来自不同的光谱。这种方法增强了特征表示，并充分利用了多光谱源的互补特性。拟议的网络在RGB-IR数据集上进行了测试，并与相关的最新网络进行了比较。实验分析表明，所提出的AFNet可以提高多光谱语义分割的效果，视觉效果好，分类和定位精度高。