Children with melanoma, to receive the best possible care and have a higher quality of life, require early identification. Nevertheless, there are issues with the deep learning technologies used today, which result in poor model robustness and less than ideal precision. This study presents YOLOv8-MNC, a unique algorithm created especially for melanoma diagnosis in pediatrics. This approach addresses issues by adding a dedicated stratum for small target identification, building upon the YOLOv8 network. Three main tactics are used by YOLOv8-MNC to improve its performance. First, it increases training accuracy by using NWD Loss. Second, to improve global feature learning, it incorporates the multi-head self-attention mechanism. Finally, to reduce information loss while upsampling, it uses the lightweight CARAFE upsampling operator. Based on a medical dataset, the experimental results demonstrate a notable improvement in detection accuracy. With 81.887% accuracy, YOLOv8-MNC significantly outperforms the prior method in terms of mean Average Precision (mAP@0.51), increasing by 5.65%. With potential implications in related sectors, this achievement represents a major advancement in addressing the challenges related to melanoma identification. In pediatric health. The efficacy of YOLOv8-MNC in improving early identification and intervention for cases of pediatric melanoma appears promising. Subsequent endeavors will center on enhancing this methodology and investigating its wider implications, underscoring its possible influence on augmenting results for juveniles afflicted by melanoma.