The determination of hair and scalp diseases is a challenging dermotological activity which requires accurate and automated image-based analysis to make early diagnosis and to treat the patient in the right way. Conventional manual diagnosis is lengthy, saubjective and limited by the fact that it is restricted by large amounts of scalp image. This study aims art resolving these problems by introducing a new scalp. Deep learning system ConViCap that is a blend of ConvNeXt and Vision Transformer (ConViNeXt) in order to collect local texture details as well as global contextual information. The framework uses BiGaBor pre-processing, graph-based feature refinement (GNN), and Capsule Network classification (CapsNet). For the experimental analysis, the proposed model attains an overall accuracy of 99.47% and an F1 score of 95.47% for efficient hair and scalp disease detection. The proposed Scalp ConViCap model enhances overall accuracy by 7.51%,21.28%’ 8.21%, and 9.62% compared to Xception with ReLU, ViT-B/16, EfficientNet-based CNN, and R-CNN with CLAHE, respectively.