Position Search and Auxiliary Supervision for Infrared Small Target Detection

Infrared small target detection, as an important task, has a wide range of application areas, such as military reconnaissance, security monitoring, and fire warning. However, due to the special nature of small targets in infrared images, some of the existing methods for infrared small target detection have some challenges, which require specific methods and techniques to solve. Firstly, small IR targets usually have a small size in the image and may blend with the background, making them difficult to be easily detected by the human eye or traditional methods. Second, small targets in IR images usually have limited features available for detection. This may lead to poor results in conventional target detection methods and require finer feature representations. Aiming at the small size of infrared small targets, we propose a small target search module that achieves position search of small targets by the interaction of features at the high-level and low-level during iterative training, allowing the network to focus on the characteristic information of the area where the small target is located. For the weak feature expression ability of infrared small targets, we build a hard region supervision module, which aims to capture the features in the hard-to-detect regions of the target and enhance the feature learning ability of the network in the hard-to-detect regions. The validation results on the NUDT-SIRST dataset and the NUAA-SIRST dataset demonstrate that our infrared small target detection method achieves the best accuracy.