EAAU-Net: Enhanced Asymmetric Attention U-Net for Infrared Small Target Detection

Detecting infrared small targets lacking texture and shape information in cluttered environments is extremely challenging. With the development of deep learning, convolutional neural network (CNN)-based methods have achieved promising results generic object detection. However, existing CNN-based with pooling layers may lose and, thus, cannot be directly applied for target To overcome this problem, we propose an enhanced asymmetric attention (EAA) U-Net. Specifically, present efficient powerful EAA module that uses both same-layer feature exchange cross-layer fusion to improve representation. In proposed approach, spatial channel exchanges occur between same reinforce primitive features targets, a bottom-up global focuses on enable dynamic weighted modulation high-level under guidance low-level features. The detailed ablation studies empirically validate effectiveness each component architecture. Compared state-of-the-art methods, method superior performance, intersection-over-union (IoU) 0.771, normalised IoU (nIoU) 0.746, F-area 0.681 publicly available SIRST dataset.