Skin Cancer Detection Using GAN and SqueezeNet

Abstract

Skin cancer positions among the most common and possibly dangerous illnesses universally, with melanoma being the most aggressive and unsafe form. Early and precise discovery is fundamental for improving survival rates. Routine computer-aided diagnosis (CAD) Systems regularly depend on convolutional neural networks (CNNs) and object detection models like YOLO (You Only Look Once) for skin injury classification. In spite of the fact that viable, these strategies ordinarily require considerable computational assets and handling control. In this ponder, we present an effective and lightweight skin cancer location System that combines Generative Adversarial Networks (GANs) for information enlargement with SqueezeNet for classification. GANs produce manufactured skin injury pictures to grow the dataset and address class imbalance, improving show performance. For classification, we utilize SqueezeNet a compact deep learning design that keeps up tall precision with a negligible parameter count, making it well-suited for deployment on versatile and edge gadgets. The proposed System forms input pictures, expands the dataset utilizing GAN-generated tests, and classifies injuries as generous or harmful utilizing SqueezeNet. Testscomes about appear that our modelachieves exactness comparable to larger CNN designs while essentially decreasing memory utilization and computational cost. This makes it particularly reasonable for utilize in resource-limited situations, such as country clinics and versatile wellbeing stages.