A customized MobileNetV2-based lightweight CNN for monkeypox detection and classification.

Monkeypox (Mpox) is an infectious disease that can spread through physical contact or from infected animals and has historically been endemic to Africa, with major outbreaks recorded in 2022. More recently, it has spread worldwide, becoming a significant public health concern. Early detection, robust surveillance, and effective contact tracing are essential for controlling the spread of Mpox. In response, various approaches have been proposed for early detection; however, most existing methods lack both efficiency and high accuracy, making them less suitable for deployment on edge devices. This study introduces a customized MobileNetV2 (CMBNV2) lightweight model for the early detection of Mpox, which delivers excellent detection performance while maintaining efficiency. A customization to the original MobileNetV2 was applied by replacing the original classifier with a lightweight pooling dropout head. To enable the proposed model to capture Mpox features while preserving its efficiency for edge deployment, the researcher fine-tuned only the last 20 layers. The researcher utilized the publicly available Monkeypox Skin Image Dataset (MSID) from the Kaggle repository, consisting of 770 images across four categories: Monkeypox, Measles, Chickenpox, and Normal. The dataset was split into 70% for training and 30% for validation/testing. To address the challenges of limited data, overfitting, and generalization, data augmentation, weighted cross-entropy, and fine-tuning techniques were applied during training. The CMBNV2 was evaluated using multiple performance metrics, including accuracy, F1-score, precision, recall, and a confusion matrix. CMBNV2, with a size of only 8.63 MB, achieved an outstanding detection accuracy of 99%. Due to its lightweight design, the model can be easily deployed on edge devices such as smartphones, playing a crucial role in the early detection and control of Mpox spread.