Striae gravidarum is a kind of striae atrophicae commonly observed in primigravid women. Striae gravidarum appear irregular purple or light red at the beginning of pregnancy, and gradually fade to silver bright stripes in the late pregnancy or postpartum. Currently, laser therapy can help with striae gravidarum, so it is crucial to know how to effectively assess striae gravidarum both during and after treatment. This research proposed a feature enhanced U-net architecture model for identifying severity of stretch marks. This approach initially extracts the single-channel grayscale from RGB images of striae gravidarum and enhances their contrast and saturation by using optimal adjustment parameters. Subsequently, a U-net model is employed to perform striae gravidarum extraction based on the enhanced images. The train dataset is 848 frames, and the test dataset is 95 frames. Compared to the traditional threshold segmentation methods, the proposed approach achieves higher accuracy in segmenting stretch marks across different pregnancies. Compared to the full-connection model method, the proposed approach not only demonstrates faster calculation speed and reduced data requirements but also achieves superior results in terms of segmentation and classification accuracy, which is of great significance for the construction of auxiliary diagnostic instrument for striae gravidarum treatment.
|