Enhancing Monkeypox Detection with Efficientnet-B5 And Image Augmentation Fusion Technique

Abstract

The recent surge of monkeypox infections worldwide has underscored the need for rapid, accurate diagnostic tools, particularly in regions with limited access to laboratory-based tests. This study employs deep learning, utilizing a pre-trained efficientNet-B5 model through transfer learning, to classify monkeypox from digital skin lesion images. Data was compiled from Kaggle, web scraping, and hospital records, covering both monkeypox and similar skin conditions such as chickenpox, measles and smallpox. The dataset was preprocessed using advanced augmentation fusion techniques, enhancing image diversity and maintaining diagnostic features critical for the model's efficacy. The efficientNet-B5 model achieved impressive results, demonstrating 99.47% accuracy, 99.19% precision and a recall of 99.72 for monkeypox. These findings suggest that the efficientNet-B5 model, supported by augmentation fusion, can serve as a reliable tool for detecting monkeypox, providing a scalable solution for early identification and public health intervention in resource-constrained settings.