FSTNet: frequency–spatial transformer network for camouflaged object detection

Camouflaged object detection (COD) is a challenging task in the field of computer vision, aiming to accurately segment and localize objects that are highly similar to their surroundings. Existing methods are often difficult to effectively capture the subtle differences between camouflaged targets and the background and especially perform poorly in complex scenes. To address the above problems, a novel frequency–spatial transformer network (FSTNet) is proposed in this paper. The network first uses the pre-trained PVTv2 as the backbone network to extract multi-level features and then captures rich information in both spatial and frequency domains simultaneously by the innovatively designed multi-scale feature aggregation (MSFA) module. In addition, a multi-scale transform fusion (MSTF) module is designed in this paper to efficiently fuse the feature representations at different scales by utilizing wavelet transform and Transformer decoder. Finally, the features are refined by Wavelet-Enhanced Feature Integration Module (WFIM) to generate the final camouflaged target prediction results. Extensive experiments on four publicly available benchmark datasets show that the proposed FSTNet outperforms 10 existing state-of-the-art methods in several evaluation metrics, such as average absolute error, enhanced alignment metric, structure metric, and average F metric. In addition, detailed ablation experiments further validate the effectiveness of each key component of the network. The visualization results show that the proposed method has significant advantages in dealing with challenging scenarios such as small objects, multiple objects, and objects highly similar to the background.