Infrared Multi-Scale Small-Target Detection Algorithm Based on Feature Pyramid Network

Abstract

Technologies for the detection of dim and small targets in infrared images play an increasingly important role in various applications, including military early warning, precise guidance, military reconnaissance, environmental monitoring, and aerospace applications. This paper proposes a new approach for the detection of infrared multi-scale small targets based on a feature pyramid network. Three pyramid segmentation–connection modules are incorporated into the proposed pyramid network to capture both local and global context information across various layers. Furthermore, a dual attention fusion module is proposed to fuse the feature maps containing context information and the deep features that have been upsampled twice through the attention mechanism of the dual attention fusion module to highlight important semantic information. Experimental results on two benchmark datasets show that the proposed method can generate results with good accuracy on both datasets and outperforms several other state-of-the-art methods for small-target detection in terms of accuracy and robustness.

Keywords

• object detection
• deep learning
• feature pyramid network
• attention mechanism fusion