Camouflaged personnel segmentation based on feature fusion

Camouflage is widely used in military warfare due to its concealment, but this greatly increases the difficulty of segmenting camouflaged personnel. To address issues such as blurred edges in camouflaged object segmentation, this paper emphasizes strengthening the network's feature - extraction capacity and designs a texture perception module and a context feature fusion module. The texture perception module enhances the feature capture ability of each level of the network through a multi-scale attention mechanism. The context feature fusion module fuses adjacent features through a residual structure to complement each other, improving the network's feature extraction ability. Compared with other advanced methods, the network proposed in this paper has higher accuracy on the dataset.