A Convolutional Neural Network Based Infrared-Visible Image Fusion Method and its Application in Aerospace Measurement and Control

In aerospace measurement and control missions, optical equipment collects infrared images and visible images with different characteristics individually. To acquire better image quality, image fusion method is studied in this paper. Firstly, based on the idea of dense connection in image classification, a feature encoding module is designed and the resolution of the feature map is kept unchanged to preserve the position information. Secondly, a image reconstruction module based on the symmetrical reconstruction in semantic segmentation is designed to fully merge the high-level and low-level feature. Then, a lightweight encoding-feature fusion-decoding image fusion neural network with only 9 CNN layers is proposed. The experimental results demonstrated that our proposed method achieve great performance improvement whether in subjective or objective evaluation. Besides, a real-time video capture and fusion system is built based on the Black Magic video capture card Declink Duo 2 and the high performance inference toolkit tensorRT. And the built system is successfully deployed in many aerospace measurement and control missions.