CDFGAN: A SELF-SUPERVISED INFRARED SMALL TARGET DETECTION NETWORK

Infrared small target detection is widely applied across various fields, but the development of deep learning models is often hindered by the limited availability of infrared data and the high cost of labeling. To address the issue of data scarcity, this paper proposes a self-supervised infrared small target detection network: Cycle Diffusion GAN (CDFGAN). To enable self-supervised training, the proposed method employs two branches to cyclically generate pseudo-images and labels. Additionally, Spatially-Adaptive Normalization (SPADE) is used to guide the generation of more accurate target edges. To aid the model in more precisely segmenting the target, high-frequency components in the image are separated through wavelet transform. We evaluated the algorithm on the public infrared dataset IRSTD-1k and compared it with other methods. The results demonstrate that our algorithm is a highly accurate self-supervised model for infrared small target detection.