Skin Lesion Classification Using Convolution Neural Networks With an Enhanced Accuracy

Skin Lesion categorization is a fault-finding task in dermatology for early discovery of skin afflictions, containing melanoma and additional forms of skin tumor. In this study, it is established that the DenseNet architecture enhanced with data augmentation techniques. The approach begins with the collection of a diverse dataset, preprocess the data and apply augmentation improving techniques to a degree rotation, flipping, and scaling to improve the dataset's robustness and variety. A CNN based on the DenseNet architecture, known for its dense connectivity patterns, which facilitate feature reuse and alleviate the vanishing gradient problem is constructed. The proposed model is trained using the augmented dataset, leveraging transfer knowledge by initializing the combined pre-trained weights on a big concept dataset. After training, the accomplishment of proposed model on an additional test set, weighing metrics to a degree of veracity, precision, recall, and F1 score is judged. Additionally, accomplishment of model accompanying other brand-new approaches to evaluate its influence is compared.