A Transfer Learning Based Approach for Human Scar Detection

In dermatology, human scar detection systems play a vital role in identification of different types of scars and timely treatment for them. The early and accurate scar detection can help in preventing further complications in treatment and promoting faster healing. Therefore, this paper aims to propose a human scar detection approach using transfer learning. In this paper, we have used three different transfer learning models: EfficientNetB1, MobileNetV3Small and Xception, and each model is trained on the dataset having two categories of scars: keloid scar and hypertrophic scar. First, in the paper there is a discussion about the related work, followed by the discussion about the proposed methodology and at last there is a discussion about the obtained results. The performance of all the models have been analyzed on important performance parameters: accuracy, precision, recall and f1-score. From the results, it has been observed that EfficientNetB1 has shown remarkable results with ${9 0 \%}$ accuracy and 0.89 f1-score.