Research on Camouflaged Object Detection Based on Transformer Feature

A camouflaged object detection network based on transformer feature association (NTFA) is proposed to address the issues of insufficient utilization of semantic features and lack of prominent spatial features in camouflaged object detection. The NTFA extracts the correlation of multi-scale feature maps obtained by the network through the attention feature association (AFA) module, and strengthens the connection between pixels inside the camouflaged object to suppress background noise. By using the decoder of Transformer and combining the associated feature sequence obtained from AFA module for global feature reconstruction, object detection information is obtained through a fully connected network to solve the problems of camouflaged boundary disturbance and feature loss. Comparative experiments on the COD10K datasets show that the proposed method outperforms three state-of-the-art models: SINet, PFNet, and C2FNet.