Infrared Small Object Detection Using Deep Interactive U-Net

Xin Wu, Danfeng Hong, Zhanchao Huang, Jocelyn Chanussot

Abstract: Infrared objects acquired from a long distance have small sizes and are easily submerged by a complex and variable background. The existing deep network detection framework suffers greatly from the feature spatial resolution loss caused by the networks’ depth and multiple downsampling operations, which is extremely detrimental to small object detection. So, a crucial and urgent goal is how to trade-off network depth and feature spatial resolution while learning feature context representation and interaction to distinguish from the background. To this end, we propose a deep interactive U-Net (DI-U-Net) architecture with high feature learning and feature interaction ability. First, feature learning is first achieved through a multilevel and high-resolution (ML-HR) network structure. This structure ensures feature resolution as the network depth increases, and also focuses on the object’s global context information. Then, the dense feature interactive (DFI) is further achieved by the dense feature encoder module to learn object local context information. The proposed method yields strong object context representation and well discriminability, as well as a good fit for infrared small object detection. Extensive experiments are conducted on the SISRT dataset and the synthetic infrared small target detection data (Synthetic dataset), demonstrating the superiority and effectiveness of the proposed deeper U-Net compared with the previous state-of-the-art detection methods.