Camouflage target segmentation based on reverse attention and self-interaction fusion

Camouflage Target Segmentation (COS) aims to segment targets hidden in complex environment. When the existing COS algorithm fuses multi-level features, it ignores the expression and positioning of the edge features of the camouflage target, and pays more attention to the influence of the fusion of features on the segmentation performance. Therefore, a COS algorithm based on disguised target segmentation based on reverse attention and self-interaction fusion is proposed. First, multi-scale features are extracted through the backbone network; Then, in order to improve the expression ability of edge features, the features extracted by the backbone network are enhanced using a network composed of a reverse attention module (RAM); Finally, the self-interaction fusion module (SIM) drives the features of different scales to achieve layer-by-layer fusion, while suppressing noise interference and obtaining more accurate target information. Experimental results show that on the three commonly used natural camouflage datasets of CHAMELEON, CAMO and CODIOK, the model shows better segmentation effect than other typical models.