Soft-Attention Mask Branch Network for Infrared Small Target Detection

Due to its insensitivity to environmental impact factors like heavy fog or cloud, the application of infrared small target detection technology is widespread in fields including remote sensing, drone vision systems, and UAV infrared guidance. The recognition of small objects in complex backgrounds with poor contrast and low Signal-to-Noise Ratio(SNR) in infrared images has remained a challenge despite recent advances in deep learning for machine vision. In this paper, a Soft-Attention Mask Branch Network (SAMBnet) is proposed, which consists of a Maxpooling-Based Soft-attention Mask module (MSAM) and a context fusion module. These modules can optimize the fusion of context information features and enhancing the spatial information and regional features of small targets. The proposed SAMBnet shows good performance in infrared small target identification in complicated settings, according to experimental findings on the NUAA-SIRST dataset.