Human cancer is one of the worst illnesses and is mainly caused by inherited instability and a variety of molecular disruptions. Worldwide, skin cancer is one of the most common types of cancer, and patient survival depends on an early and precise diagnosis. Although it is crucial, clinical evaluation of skin lesions is fraught with difficulties, including significant wait times and individual interpretations. Dermatologists have difficulties in early skin cancer detection, and deep learning has recently been widely used in supervised and unsupervised learning tasks. Defining the lesion region is crucial to the effectiveness of these techniques because it allows for accurate segmentation of skin cancer lesions at various stages, which helps in early diagnosis and treatment. In this study, we have proposed an Xception model-based U_Net (XcUNet) model for skin lesion image segmentation. The proposed model’s lower computational load from separable convolutions and residuals usually results in better detection times and more sophisticated comprehension of spatial hierarchies and patterns. This attribute is extremely important in clinical fields where quick outcomes are required, such as skin lesion segmentation.