Infrared Airport Scene Segmentation Based on Aggregation Networks

Abstract

The idea of convergent networks is adopted to implement infrared airport scene segmentation. First, an infrared airport scene image dataset containing five types of foreground targets and one type of background for semantic segmentation research is built. Second, an end-to-end infrared semantic segmentation algorithm framework is built and trained on the basis of deep aggregation networks. After experiments, we use the mean value filtering operation to further improve the segmentation accuracy in order to solve the problem of complex image detail information affecting the segmentation. Finally, the output results of the algorithm framework are evaluated and analyzed on an infrared test set. The experimental results show that the semantic segmentation of infrared images using aggregated networks can achieve pixel-level classification of images and high prediction accuracy. Thus, we can obtain information on the shape, type, and location distribution of the scenes in the infrared image and realize the semantic understanding of the infrared scene.