For　medical　images,　domain　shift　is　a　very　common　phenomenon.　To　address　this　issue,　researchers　have　proposed　unsupervised　domain　adaptation　and　multi-source　domain　generalization.　However,　these　methods　are　sometimes　impractical for　clinical　applications　since　they　need　multi-domain　data.　To this　end,　single-source　domain　generalization　has　been　further　proposed.　However,　most　single-source　domain　generalization　methods are　designed　for　grayscale　medical　images,　making　them　unsuitable for　color　images　such　as　fundus　images.　In　this　paper,　we　first　propose　a　novel　and　effective　Fourier　transform-based　data　augmentation　method　for　single-source　domain　color　medical　images,　named　spatial　amplitude　perturbation　module　(SAPM).　The　SAPM　uses　different　Gaussian　distributions　to　perturb　different　regions　of　the　amplitude　map　obtained　by　FFT　decomposition,　thereby　avoiding　the　need for　information　from　other　domains　and　ensuring　the　diversity　of the　augmented　images.　Then,　we　use　feature　sensitivity　to　guide the　network　to　learn　domain-invariant　features,　which　can　suppress　feature　channels　sensitive　to　domain　shift　and　emphasize　feature　channels　insensitive　to　domain　shift.　We　evaluate　our　method　on a　multi-domain　fundus　segmentation　benchmark,　and　the　results　demonstrate　the　effectiveness　of　our　proposed　method.　©　The　Author(s),　under　exclusive　license　to　Springer　Nature　Switzerland　AG　2025.