Dual alignment interactive network for infrared small target detection

To integrate the relationship between small targets and high-intensity structured backgrounds in infrared images, we propose a dual alignment interactive network (DAINet) for infrared small target detection. DAINet treats the low-rank background features of infrared images as complementary data, guiding the network to consistently focus and localize infrared small targets. First, the dual interactive block is built by receiving an infrared image and a background image, which interactively combines the target and background attention for better detection performance. Then, a multi-level feature fusion strategy is designed to get the final robust result. Experimental results on two public datasets reveal that DAINet can process images with high detection accuracy compared to various state-of-the-art (SOTA) methods.