Text this: Semantic segmentation of camouflage objects via fusing reconstructed multispectral and RGB images